SVNY339-Tu April 6, 2007 10:15

Springer Series in
M AT E R I A L S S C I E N C E 92
SVNY339-Tu April 6, 2007 10:15
Springer Series in
M AT E R I A L S S C I E N C E
Editors: R. Hull R.M. Osgood, Jr. J. Parisi H. Warlimont
The Springer Series in Materials Science covers the complete spectrum of materials physics, including funda-
mental principles, physical properties, materials theory and design. Recognizing the increasing importance
of materials science in future device technologies, the book titles in this series ref lect the state-of-the-art
in understanding and controlling the structure and properties of all important classes of materials.
78 Macromolecular Nanostructured Materials
Editors: N. Ueyama and A. Harada
79 Magnetism and Structure
in Functional Materials
Editors: A. Planes, L. Mañosa,
and A. Saxena
80 Micro- and Macro-Properties of Solids
Thermal, Mechanical and Dielectric Properties
By D.B. Sirdeshmukh, L. Sirdeshmukh,
and K.G. Subhadra
81 Metallopolymer Nanocomposites
By A.D. Pomogailo and V.N. Kestelman
82 Plastics for Corrosion Inhibition
By V.A. Goldade, L.S. Pinchuk,
A.V. Makarevich and V.N. Kestelman
83 Spectroscopic Properties of Rare Earths
in Optical Materials
Editors: G. Liu and B. Jacquier
84 Hartree–Fock–Slater Method
for Materials Science
The DV–X Alpha Method for Design
and Characterization of Materials
Editors: H. Adachi, T. Mukoyama,
and J. Kawai
85 Lifetime Spectroscopy
A Method of Defect Characterization
in Silicon for Photovoltaic Applications
By S. Rein
86 Wide-Gap Chalcopyrites
Editors: S. Siebentritt and U. Rau
Volumes 30–77 are listed at the end of the book.
87 Micro- and Nanostructured Glasses
By D. Hülsenberg and A. Harnisch
88 Introduction to Wave Scattering,
Localization and Mesoscopic Phenomena
By P. Sheng
89 Magneto-Science
Magnetic Field Effects on Materials:
Fundamentals and Applications
Editors: M. Yamaguchi and Y. Tanimoto
90 Internal Friction in Metallic Materials
A Reference Book
By M.S. Blanter, I.S. Golovin,
H. Neuhäuser, and H.-R. Sinning
91 Time-dependent Mechanical Properties
of Solid Bodies
By W. Gräfe
92 Solder Joint Technology
Materials, Properties, and Reliability
By K.-N. Tu
93 Materials for Tomorrow
Theory, Experiments and Modelling
Editors: S. Gemming, M. Schreiber
and J.-B. Suck
94 Magnetic Nanostructures
Editors: B. Aktas, L. Tagirov
and F. Mikailov
95 Nanocrystals and Their Mesoscopic
Organization
By C.N.R. Rao, P.J. Thomas
and G.U. Kulkarni
SVNY339-Tu April 6, 2007 10:15

King-Ning Tu

Solder Joint Technology

Materials, Properties, and Reliability
SVNY339-Tu April 6, 2007 10:15

King-Ning Tu
Department of Materials Science and Engineering
University of California at Los Angeles, CA,
USA
6532 Boelter Hall
Los Angeles 90095–6595
Email, personal: kntu@ucla.edu

Library of Congress Control Number: 2007921097

ISBN-10: 0-387-38890-7 e-ISBN-10: 0-387-38892-3

ISBN-13: 978-0-387-38890-8 e-ISBN-13: 978-0-387-38892-2

Printed on acid-free paper.


C 2007 Springer Science+Business Media, LLC
All rights reserved. This work may not be translated or copied in whole or in part without the written
permission of the publisher (Springer Science+Business Media, LLC, 233 Spring Street, New York, NY 10013,
USA), except for brief excerpts in connection with reviews or scholarly analysis. Use in connection with
any form of information storage and retrieval, electronic adaptation, computer software, or by similar or
dissimilar methodology now known or hereafter developed is forbidden.
The use in this publication of trade names, trademarks, service marks, and similar terms, even if they are
not identified as such, is not to be taken as an expression of opinion as to whether or not they are subject
to proprietary rights.

9 8 7 6 5 4 3 2 1

springer.com
SVNY339-Tu April 6, 2007 10:15

Contents

1 Introduction 1
1.1 Introduction of Solder Joint................................................. 1
1.2 Lead-Free Solders............................................................... 4
1.2.1 Eutectic Pb-Free Solders.......................................... 4
1.2.2 High-Temperature Pb-Free Solders ............................ 8
1.3 Solder Joint Technology ...................................................... 9
1.3.1 Surface Mount Technology ....................................... 9
1.3.2 Pin-through-Hole Technology.................................... 10
1.3.3 C-4 Flip Chip Technology ....................................... 12
1.4 Reliability Problems in Solder Joint Technology...................... 16
1.4.1 Sn Whiskers........................................................... 16
1.4.2 Spalling of Interfacial Intermetallic Compounds in
Direct Chip Attachment ......................................... 17
1.4.3 Thermal-Mechanical Stresses .................................... 22
1.4.4 Impact Fracture...................................................... 25
1.4.5 Electromigration and Thermomigration...................... 25
1.4.6 Reliability Science on the Basis of Nonequilibrium
Thermodynamics .................................................... 26
1.5 Future Trends in Electronic Packaging .................................. 28
1.5.1 The Trend of Miniaturization ................................... 28
1.5.2 The Trend of Packaging Integration Evolution—SIP,
SOP, and SOC ....................................................... 29
1.5.3 Chip–Packaging Interaction...................................... 30
1.5.4 Solderless Joints ..................................................... 31
References................................................................................. 31

Part I Copper–Tin Reactions

2 Copper–Tin Reactions in Bulk Samples 37

2.1 Introduction...................................................................... 37
2.2 Wetting Reaction of Eutectic SnPb on Cu Foils...................... 38
SVNY339-Tu April 6, 2007 10:15

vi Contents

2.2.1 Crystallographic Relationship between Cu6 Sn5

Scallop and Cu....................................................... 43
2.2.2 Rate of Consumption of Cu in Soldering Reaction
with Eutectic SnPb................................................. 48
2.3 Wetting Reaction of SnPb on Cu Foil as a Function of
Solder Composition ............................................................ 48
2.4 Wetting Reaction of Pure Sn on Cu Foils............................... 52
2.5 Ternary Phase Diagram of Sn-Pb-Cu .................................... 53
2.5.1 Ternary SnPbCu Phase Diagrams at 200 and 170◦ C .... 56
2.5.2 5Sn95Pb/Cu Reaction and Ternary SnPbCu
Phase Diagrams at 350◦ C ........................................ 57
2.6 Solid-State Reaction of Eutectic SnPb on Cu Foils.................. 58
2.6.1 Formation of Cu3 Sn and Kirkendall Voids .................. 58
2.7 Comparison between Wetting and Solid-State Reactions .......... 60
2.7.1 Morphology of Wetting Reaction and
Solid-State Aging.................................................... 60
2.7.2 Kinetics of Wetting Reaction and
Solid-State Aging.................................................... 64
2.7.3 Reactions Controlled by Rate of Gibbs Free
Energy Change....................................................... 67
2.8 Wetting Reaction of Pb-Free Eutectic Solders on Thick
Cu UBM........................................................................... 68
References................................................................................. 70

3 Copper–Tin Reactions in Thin-Film Samples 73

3.1 Introduction...................................................................... 73
3.2 Room-Temperature Reaction in a Bilayer Thin Film
of Sn/Cu ......................................................................... 74
3.2.1 Phase Identification by Glancing Incidence
X-ray Diffraction .................................................... 74
3.2.2 Growth Kinetics of Cu6 Sn5 and Cu3 Sn....................... 75
3.2.3 Copper Is the Dominant Diffusing Species .................. 80
3.2.4 Kinetic Analysis of Sequential Formation of Cu6 Sn5
and Cu3 Sn............................................................. 81
3.2.5 Atomistic Model of Interfacial-Reaction Coefficient ...... 89
3.2.6 Measurement of Strain in Cu and Sn Thin Films......... 92
3.3 Spalling in Wetting Reaction of Eutectic SnPb on Cu
Thin Films........................................................................ 93
3.4 No Spalling in High-Pb Solder on Au/Cu/Cu-Cr
Thin Films........................................................................ 97
3.5 Spalling in Eutectic SnPb Solder on Au/Cu/Cu-Cr
Thin Films........................................................................ 100
3.6 No Spalling in Eutectic SnPb on Cu/Ni(V)/Al
Thin Films........................................................................ 100
SVNY339-Tu April 6, 2007 10:15

Contents vii

3.7 Spalling in Eutectic SnAgCu Solder on Cu/Ni(V)/Al

Thin Films........................................................................ 101
3.8 Enhanced Spalling Due to Interaction across a
Solder Joint....................................................................... 104
3.9 Wetting Tip Reaction on Thin-Film-Coated V-Grooves............ 105
References................................................................................. 108

4 Copper–Tin Reactions in Flip Chip Solder Joints 111

4.1 Introduction...................................................................... 111
4.2 Processing a Flip Chip Solder Joint and a Composite
Solder Joint....................................................................... 113
4.3 Chemical Interaction across a Flip Chip Solder Joint............... 116
4.4 Enhanced Dissolution of Cu-Sn IMC by Electromigration......... 117
4.5 Enhanced Phase Separation in Solder Alloys by Electromigration
and Thermomigration......................................................... 119
4.6 Thermal Stability of Bulk Diffusion Couples of
SnPb Alloys ...................................................................... 123
4.7 Thermal Stress Due to Chip–Packaging Interaction ................. 124
4.8 Design and Materials Selection of a Flip Chip
Solder Joint....................................................................... 124
References................................................................................. 125

5 Kinetic Analysis of Flux-Driven Ripening of

Copper–Tin Scallops 127
5.1 Introduction...................................................................... 127
5.2 Morphological Stability of Scallop-Type IMC Growth in
Wetting Reactions.............................................................. 128
5.2.1 Analysis of Morphological Stability of Scallops in
Wetting Reactions................................................... 131
5.3 A Simple Model for the Growth of Mono-Size
Hemispheres...................................................................... 135
5.4 Theory of Nonconservative Ripening with a Constant
Surface Area ..................................................................... 139
5.5 Size Distribution of Scallops ................................................ 142
5.5.1 Dependence of Cu6 Sn5 Morphology on
Solder Composition................................................. 143
5.5.2 Size Distribution and Average Radius of Scallops ........ 147
5.6 Nano Channels between Scallops .......................................... 150
References................................................................................. 150

6 Spontaneous Tin Whisker Growth: Mechanism

and Prevention 153
6.1 Introduction...................................................................... 153
6.2 Morphology of Spontaneous Sn Whisker Growth..................... 154
SVNY339-Tu April 6, 2007 10:15

viii Contents

6.3 Stress Generation (Driving Force) in Sn Whisker Growth by

Cu-Sn Reaction ................................................................ 157
6.4 Effect of Surface Sn Oxide on Stress Gradient Generation and
Whisker Growth ................................................................ 160
6.5 Measurement of Stress Distribution by Synchrotron Radiation
Micro-diffraction ................................................................ 163
6.6 Stress Relaxation (Kinetic Process) in Sn Whisker Growth
by Creep: Broken Oxide Model ............................................ 168
6.7 Irreversible Processes.......................................................... 169
6.8 Kinetics of Grain Boundary Diffusion-Controlled
Whisker Growth ................................................................ 170
6.9 Accelerated Test of Sn Whisker Growth ................................ 175
6.10 Prevention of Spontaneous Sn Whisker Growth ...................... 178
References................................................................................. 180

7 Solder Reactions on Nickel, Palladium, and Gold 183

7.1 Introduction...................................................................... 183
7.2 Solder Reactions on Bulk and Thin-Film Ni........................... 184
7.2.1 Reaction between Eutectic SnPb and
Electroless Ni(P) .................................................... 188
7.2.2 Reaction between Eutectic Pb-Free Solders and
Electroless Ni(P) .................................................... 191
7.2.3 Formation of (Cu, Ni)6 Sn5 versus (Ni, Cu)3 Sn4 ............ 193
7.2.4 Formation of Kirkendall Voids .................................. 194
7.3 Solder Reactions on Bulk and Thin-Film Pd .......................... 194
7.3.1 Reaction between Eutectic SnPb and Pd Foil.............. 194
7.3.2 Reaction between Eutectic SnPb and Pd
Thin Film.............................................................. 197
7.4 Solder Reactions on Bulk and Thin-Film Au .......................... 198
7.4.1 Reaction between Eutectic SnPb and Au Foil.............. 198
7.4.2 Reaction between Eutectic SnPb and Au
Thin Film.............................................................. 204
References................................................................................. 204

Part II Electromigration and Thermomigration

8 Fundamentals of Electromigration 211

8.1 Introduction...................................................................... 211
8.2 Electromigration in Metallic Interconnects ............................. 214
8.3 Electron Wind Force of Electromigration............................... 217
8.4 Calculation of the Effective Charge Number........................... 221
8.5 Effect of Back Stress on Electromigration and
Vice Versa ........................................................................ 222
SVNY339-Tu April 6, 2007 10:15

Contents ix

8.6 Measurement of Critical Length, Critical Product,

Effective Charge Number .................................................... 225
8.7 Why Is There Back Stress in Electromigration? ...................... 226
8.8 Measurement of the Back Stress Induced
by Electromigration............................................................ 229
8.9 Current Crowding and Current Density Gradient Force ........... 230
8.10 Electromigration in Anisotropic Conductor of Beta-Sn............. 235
8.11 Electromigration of a Grain Boundary in
Anisotropic Conductor........................................................ 238
8.12 AC Electromigration .......................................................... 240
References................................................................................. 241

9 Electromigration in Flip Chip Solder Joints 245

9.1 Introduction...................................................................... 245
9.2 Unique Behaviors of Electromigration in Flip Chip
Solder Joints ..................................................................... 247
9.2.1 Low Critical Product of Solder Alloys ........................ 247
9.2.2 Current Crowding in Flip Chip Solder Joints .............. 247
9.2.3 Phase Separation in Eutectic Solder Joints ................. 248
9.2.4 Narrow Range of Current Density ............................. 248
9.2.5 Effect of Under-Bump Metallization
on Electromigration ................................................ 249
9.3 Failure Mode of Electromigration in Flip Chip
Solder Joints ..................................................................... 249
9.4 Electromigration in Flip Chip Eutectic Solder Joints ............... 255
9.4.1 Electromigration in Eutectic SnPb Flip Chip
Solder Joints .......................................................... 256
9.4.2 Electromigration in Eutectic SnAgCu Flip Chip
Solder Joints .......................................................... 259
9.4.3 Marker Motion Analysis Using Area Array of
Nano-indentations................................................... 261
9.4.4 Mean-Time-to-Failure of Flip Chip Solder Joints ......... 264
9.4.5 Comparison between Eutectic SnPb and SnAgCu
Flip Chip Solder Joints............................................ 266
9.4.6 Kinetic Analysis of Pancake-Type Void Growth
along the Contact Interface ...................................... 267
9.4.7 Time-Dependent Melting of Flip Chip Solder Joints..... 270
9.5 Electromigration in Flip Chip Composite Solder Joints............ 272
9.5.1 Thin-Film Cu UBM in Composite Solder Joints .......... 272
9.5.2 Thick Cu UBM in Composite Solder Joints................. 274
9.6 Effect of Thickness of Cu UBM on Current Crowding and
Failure Mode..................................................................... 275
9.6.1 Cu Column Bumps ................................................. 276
9.6.2 The Design of a Near-Ideal Flip Chip Solder Joint ....... 280
SVNY339-Tu April 6, 2007 10:15

x Contents

9.7 Electromigration-Induced Phase Separation in Eutectic

Two-Phase Solder Alloy ...................................................... 281
9.7.1 Electromigration-Induced Back Stress in
Two-Phase Structure............................................... 283
9.7.2 Electromigration-Induced Kirkendall Shift in
Two-Phase Structure............................................... 285
9.7.3 Stochastic Tendency in Electromigration in
Two-Phase Structure............................................... 286
References................................................................................. 287

10 Polarity Effect of Electromigration on Solder Reactions 289

10.1 Introduction...................................................................... 289
10.2 Preparation of V-Groove Samples......................................... 289
10.2.1 Electromigration of Eutectic SnPb as a Function
of Temperature....................................................... 291
10.3 Polarity Effect on IMC Growth at the Anode......................... 293
10.3.1 IMC Growth without Electric Current ....................... 294
10.3.2 Growth of IMC at Anode and Cathode with
Electric Current...................................................... 294
10.3.3 IMC Thickness Change with Current Density
and Temperature .................................................... 295
10.3.4 Comparison among Electrodes of Cu, Ni, and Pd
in V-Groove Samples............................................... 297
10.4 Polarity Effect on IMC Growth at the Cathode ...................... 297
10.4.1 Dynamic Equilibrium .............................................. 299
10.5 Effect of Electromigration on the Competing
Growth of IMC.................................................................. 301
References................................................................................. 302

11 Ductile–to-Brittle Transition of Solder Joints Affected

by Copper–Tin Reaction and Electromigration 305
11.1 Introduction...................................................................... 305
11.2 Tensile Test Affected by Electromigration.............................. 306
11.3 Shear Test Affected by Electromigration................................ 309
11.4 Impact Test ...................................................................... 311
11.4.1 Charpy Test........................................................... 311
11.4.2 Mini Charpy Machine to Test Solder Joints ................ 314
11.5 Drop Test ......................................................................... 316
11.5.1 JEDEC-JESD22-B111 Standard of Drop Test ............. 316
11.5.2 Dropping of a Packaging Board Vertically and
the Torque on Solder Balls ....................................... 320
11.6 Converting a Mini Charpy Impact Machine to Perform
Drop Test ......................................................................... 322
11.6.1 Dropping of a Chip Size Package Horizontally in
Mini Charpy Machine.............................................. 323
SVNY339-Tu April 6, 2007 10:15

Contents xi

11.6.2 Dropping of a Chip Size Package Vertically in

Mini Charpy Machine.............................................. 325
11.7 Creep and Electromigration................................................. 325
References................................................................................. 326

12 Thermomigration 327
12.1 Introduction...................................................................... 327
12.2 Thermomigration in Flip Chip Solder Joints of SnPb............... 329
12.2.1 Thermomigration in Unpowered Composite
Solder Joints .......................................................... 329
12.2.2 In Situ Observation of Thermomigration .................... 332
12.2.3 Random States of Phase Separation in Eutectic
Two-Phase Structures.............................................. 333
12.2.4 Thermomigration in Unpowered Eutectic SnPb
Solder Joints .......................................................... 335
12.3 Fundamentals of Thermomigration ....................................... 338
12.3.1 Driving Force of Thermomigration ............................ 338
12.3.2 Entropy Production ................................................ 340
12.3.3 Effect of Concentration Gradient
on Thermomigration ............................................... 341
12.3.4 The Critical Length below Which No Thermomigration
Occurs in a Pure Metal............................................ 342
12.3.5 Thermomigration in a Eutectic Two-Phase Alloy......... 343
12.4 Thermomigration and DC Electromigration in Flip Chip
Solder Joints ..................................................................... 344
12.5 Thermomigration and AC Electromigration in Flip Chip
Solder Joints ..................................................................... 344
12.6 Thermomigration and Chemical Reaction in Solder Joints........ 345
12.7 Thermomigration and Creep in Solder Joints.......................... 345
References................................................................................. 346

Appendix A: Diffusivity of Vacancy Mechanism of

Diffusion in Solids 347

Appendix B: Growth and Ripening Equations

of Precipitates 351

Appendix C: Derivation of Huntington’s Electron

Wind Force 359

Subject Index ......................................................................... 365

SVNY339-Tu April 6, 2007 10:15

Preface

The trend in consumer electronic products will be more and more wireless,
portable, and handheld. To manufacture these multifunctional products, high-
density circuit interconnections between a Si chip and its substrate are needed.
Flip chip solder joint technology, by which an area array of solder bumps
is used to join a chip to its substrate, is growing rapidly in demand. Flip
chip technology is the only technology that can provide a large number of
such interconnections with reliability. Solder joints are ubiquitous in electronic
products.
Due to environmental concerns regarding the toxicity of Pb-based solders,
the European Union Parliament issued a directive to ban the use of Pb-based
solders in consumer products on July 1, 2006. The application of Pb-free sol-
der joints to a wide range of devices is urgent, and R&D of Pb-free solders
for electronic manufacturing is thus very active at the moment. While solder
joint technology is mature, Pb-free solder technology is not, hence its relia-
bility must be proven. For example, electrical shorting due to Sn whiskers,
electrical opening due to electromigration, and joint fracture due to drop-
ping of handheld devices to the ground are challenging reliability problems
in the application of Pb-free solders. To solve these problems in a largely
technology-based manufacturing industry, scientific understanding and solu-
tions are required. The copper–tin reaction is essential in the formation of a
solder joint and the failure of such joints is due to externally applied forces as
in electromigration. A fundamental understanding of the copper–tin reaction
and the effect of external forces on solder joint reliability is critical and is
emphasized in this book.
There are two themes in this book. The first is the copper–tin reaction as a
function of time and temperature, and the second is the effect of external forces
on the reaction. Actually the second theme also emphasizes phase transforma-
tions under an inhomogeneous boundary condition. Typically, metallurgical
phase transformations occur under constant temperature and constant pres-
sure so that Gibbs free energy is minimized. However, in thermomigration or
stress migration (creep) of a solder joint, the temperature or the pressure is
not constant because there exists a temperature gradient or a stress gradient
SVNY339-Tu April 6, 2007 10:15

xiv Preface

to drive the phase change, so an equilibrium state with a minimum free energy
will not be reached. In electromigration, a potential gradient exists across the
sample too. These are irreversible processes.
The contents of the book are divided into two parts. Part I, from Chapters
2 to 7, covers copper–tin reactions, and Part II, from Chapters 8 to 12, covers
electromigration and thermomigration of solder joints.
Chapter 1 is an overview of flip chip technology. Why it is important and
the known reliability problems are explained. The future trend in electronic
packaging technology and its effect on solder joint technology are covered.
Chapter 2 concerns wetting reactions between molten eutectic solder and bulk
Cu foils. The unique morphology of scallop-type Cu-Sn intermetallic com-
pound formation is emphasized and analyzed. Chapter 3 considers about Cu-
Sn reactions in thin films. Thin-film reactions are important since most met-
allization on Si devices to be joined by solder is in thin film form. Spalling of
thin-film intermetallic compounds is a unique reliability phenomenon. Chap-
ter 4 covers solder reaction in a flip chip configuration in which the reac-
tion occurs on two interfaces. The two interfacial reactions interact with each
other and the interaction is a reliability issue. Chapter 5 presents a theoretical
analysis of flux-driven ripening of scallop-type growth of Cu-Sn intermetal-
lic compounds under the constraint of a constant surface area. Theoretically
derived and experimentally measured distribution functions of scallops are
compared. Chapter 6 examines spontaneous Sn whisker growth which is a
creep phenomenon. The necessary and sufficient conditions of whisker growth
are discussed, and how to conduct an accelerated test of Sn whisker growth
and how to prevent its growth are presented. Chapter 7 discusses briefly solder
reactions on nickel, palladium, and gold surfaces. In addition to copper, these
metals are used as under-bump metallization in devices. Chapter 8 covers
the fundamentals of electromigration and the differences between electromi-
gration in solder alloys and in Al or Cu interconnects. Why electromigration
in solder joints has only recently become a reliability problem is explained.
Chapter 9 concerns the unique behavior of electromigration in flip chip solder
joints, especially the effect of current crowding. It is a key chapter of the book.
The unique failure model due to pancake-type void formation at the cathode
contact interface is presented. Chapter 10 examines the interaction between
electrical and chemical forces in solder joints. The polarity effect of electromi-
gration on intermetallic compound formation at the cathode and the anode
interfaces of a solder joint is presented. Chapter 11 describes the interaction
between electrical and mechanical forces. An accidental drop to the ground is
the most frequent cause of failure of portable devices. Impact test and drop
test of solder joints are analyzed, and the effect of electromigration on these
tests is discussed. Chapter 12 considers thermomigration in solder joints, and
the interaction between electrical and thermal forces is analyzed. Microstruc-
ture instability in a eutectic two-phase structure driven by a temperature
gradient is adressed.
SVNY339-Tu April 6, 2007 10:15

Preface xv

I started solder research in 1965, when I began my Ph.D. dissertation on

cellular precipitation of Sn lamellae in SnPb alloys. However, the contents of
this book are based on thirteen Ph.D. dissertations finished in the University
of California at Los Angeles since 1996, supported by the National Science
Foundation (Dr. Bruce MacDonald), Semiconductor Research Corporation
(Dr. Harold Hosack), and several microelectronic companies (especially Dr.
Paul A. Totta of IBM East Fishkill, NY, Dr. Fay Hua of Intel, Santa Clara,
CA, Dr. Luu Nguyen at NSC, Santa Clara, CA, Dr. Darrel Frear at Freescale,
Phoenix, AZ, and Dr. Yi-Shao Lai at ASE, Taiwan). The dissertations of H. K.
Kim, Patrick Kim, Cheng-Yi Liu, Taek Yeong Lee, Woo-Jin Choi, Hua Gan,
Albert T. Wu, Emily Shengquan Ou, Minyu Yan, Fei Ren, Jong-ook Suh,
Annie Huang, and Tiffany Fan-Yi Ouyang are acknowledged. Also included
are the work of several postdocs (Grant Pan, J. W. Jang, Everett C. C. Yeh,
Kejun Zeng, J. W. Nah, and L. Y. Zhang) and M.Sc. students (Wang Yang,
Ann A. Liu, Jessica P. Almaraz, Quyen Tang Huynh, Xu Gu, Rajat Agarwal,
Joanne Huang, and Jackie Preciado). I acknowledge the generous support of
these students and postdocs and thank them. It is their dedication and hard
work that made this book possible.
Many outstanding contributions to solder research have been made by
other researchers. Since this book is an introduction to solder joint technology
and covers only a small part of the literature on the subject, I was unable to
include much of the published work on solder joints in the literature.
I apologize to those colleagues whose work is not included here. Still I hope
the book may serve as a steppingstone to reach out to a much broader field
of solder and as a reference to future R&D in the field. Indeed, much work on
the reliability of Pb-free solder joints remains undone.
While this book is intended for engineers and scientists working on solder
joint technology in the electronic manufacturing industry, it might be used
as a reference book in a course on reliability science of electronic packaging
technology for seniors and graduate students. Because very few textbooks on
the subject of electronic packaging technology and reliability science are avail-
able, I include in the appendixes the derivations of the diffusion coefficient in
vacancy mechanism of diffusion in a face-centered-cubic lattice, the growth
and dissolution equation of a spherical particle in the ripening process, and
Huntington’s electron wind force in electromigration. They are convenient ref-
erences for analyzing the basic kinetic behaviors of solder joints discussed in
this book. This derivation on electron wind force has been taken from the lec-
ture notes of Professor A. M. Gusak at Cherkasy National University, Ukraine.
He has been very helpful regarding the kinetic analyses presented in this book,
for example, irreversible processes. I am also grateful to Dr. Yuhuan Xu at
UCLA and Professor Yiping Wu at Huazhong University of Science and Tech-
nology, China, for helpful comments on the drop test in Chapter 11. I thank
Professor Chih Chen, National Chiao Tung University, Hsinchu, Taiwan, ROC,
and Professor Cheng-Yi Liu, National Central University, Chungli, Taiwan,
ROC, for a critical review of the book, and Mr. Jong-ook Suh at UCLA for
SVNY339-Tu April 6, 2007 10:15

xvi Preface

preparation of all the figures and Miss Fan-Yi Ouyang for revision correc-
tions. Finally, I thank Professor Y. C. Chan of the Department of Electrical
Engineering, City University of Hong Kong, and Professor Weijia Wen of the
Department of Physics, Hong Kong University of Science and Technology, for
hosting my two-month visit to Hong Kong so that I could concentrate on
finishing the final version of the book.

June 2006
SVNY339-Tu April 5, 2007 17:44

Copper–Tin Reactions
SVNY339-Tu March 2, 2007 22:0

1
Introduction

1.1 Introduction of Solder Joint

Solder has been used to join copper pipes for plumbing in every modern house
and to join copper wires for circuitry connection in every electrical product.
Solder joints are ubiquitous. The essential process in solder joining is the
chemical reaction between copper and tin to form intermetallic compounds
having a strong metallic bonding. After the iron–carbon (Fe-C) binary system,
copper–tin (Cu-Sn) may be the second most important metallurgical binary
system that has impacted human civilization, as suggested by the bronze
(Cu-Sn alloy) age.
Typically a solder alloy of lead (Pb) and tin is used to join copper parts;
however, due to environmental concern of lead toxicity, the solder in plumbing
is already Pb-free, and Pb-free solders are being introduced into electronic and
electrical products. For example, there is a directive to ban Pb-based solders
in consumer products by the European Congress on July 1, 2006. For a large-
scale application of Pb-free solders to electronic products, the reliability is of
serious concern.
Reliability of solder joint technology in the microelectronic packaging in-
dustry has been a concern for a long time, for example, the low cycle fatigue
of tin–lead (SnPb) solder joints in flip chip technology due to the cyclic ther-
mal stress between a Si chip and its substrate. At present, the risk of fatigue
has been much reduced by the innovative application of underfill of epoxy
between the chip and its substrate. On the other hand, to replace SnPb sol-
ders by Pb-free solders, new reliability issues have appeared, mostly because
the Pb-free solders have a very high concentration of Sn. Furthermore, due
to the demand of greater functionality in portable consumer electronic prod-
ucts, electromigration is now a serious reliability issue. This is because of the
increase of current density to be carried by the power solder joints.
This book is dedicated to the understanding of the fundamentals of solder
joint reliability problems. The science of solder reaction and electromigration
are emphasized, in particular. In this chapter, solder joint technology and
related reliability problems will be briefly introduced. The rest of the chapters
SVNY339-Tu March 2, 2007 22:0

2 Chapter 1

will be divided into two parts. Part I will cover copper–tin reactions, while
Part II will cover electromigration and thermomigration and related reliability
issues in flip chip solder joints.
The structure of a solder joint is unique in that it has two joint inter-
faces. Reliability failures tend to occur at these interfaces. While intermetallic
compound formation at the interfaces is required for metallic bonding, the
interfacial intermetallic has affected greatly the properties and reliability of
the joint. Also, electromigration failure of a solder joint occurs typically at
the cathode interfaces.
Solder is preferred to have a eutectic composition. This is not only because
the melting point of a eutectic alloy is lower than its two components, but more
importantly because it has a single melting point. Therefore, the entire joint
will melt when the temperature has reached its eutectic point so that both of
its interfaces are joined at the same time. In the application of thousands of
solder balls as input/output interconnections on a silicon chip, all of them must
melt and join at the same time so that the very large number of solder joints
between the chip and its substrate can be formed simultaneously by a very
simple process of heating or “reflow.” To connect the large-scale integration
of circuits on a Si chip to the circuits on its packaging substrate, flip chip
technology (to be discussed later in this chapter and Chapter 4) provides
the rare advantage that thousands of solder joints or electrical leads can be
formed simultaneously by a low temperature heating process in forming gas.
Since many of the solder joints can be placed near the center of a chip in
order to avoid the voltage drop from an electrical lead which is located at the
edge of the chip, it saves power too. The technology does require that all the
joints melt or solidify at the same temperature, so a eutectic alloy is favored
as solder.
Solder is a low melting eutectic alloy. The eutectic tin–lead (SnPb) has
a melting point of 183◦ C. To be able to form a metallic bond with Cu at
such a low temperature is the key reason why SnPb solder joints have been
used worldwide for so long. On the other hand, the typical temperature of
solder joint application is either near room temperature or the device working
temperature around 100◦ C. They are high-temperature applications of the
solder alloy. Take the melting point of eutectic SnPb at 456 K as an example,
room temperature and 100◦ C are respectively 0.66 and 0.82 of it. At such
high homologous temperatures, thermally activated processes such as atomic
diffusion cannot be ignored. The mechanical properties of solder joints or
the measurement of the mechanical properties of solder joints at slow strain
rates or the concern of reliability due to low cycle fatigue must take into
account the contribution of thermally activated processes such as creep and
recovery.
To achieve a good solder joint, the wetting reaction between molten solder
and solid Cu depends on chemical flux. A proper and fresh chemical flux is
essential, so that flux is the most important factor in solder joint manufac-
turing. The flux is required in order to remove oxide on both the Cu surfaces
SVNY339-Tu March 2, 2007 22:0

Introduction 3

and the solder surface so that the molten solder can wet the clean Cu surfaces
to be joined.
Solder reaction typically involves three chemical elements, e.g., a binary
solder alloy of SnPb and a conductor of Cu. There are two kinds of solder
reactions: the solder alloy can be in the liquid state or solid state. In a wetting
reaction, the solder is in the liquid state and the conductor is in the solid state.
Wetting reaction is often called “reflow” in electronic package manufacturing.
In a “reflow,” the solder joint experiences a temperature cycle, in which part
of the cycle must be above the melting point of the solder for half a minute or
so. Since the manufacturing of a device needs to go through several reflows,
the total period of wetting reaction for a solder joint in devices may add up to
a few minutes. On the other hand, to meet the device reliability specification,
the solder joint must be aged at 150◦ C for 1000 hr. In the aging test, while
both the solder and conductor are in the solid state, chemical reaction does
occur between them. The solid-state aging is of special interest in under-
the-hood applications of automobiles where the temperature is quite high. In
Chapters 3 and 6, it will be shown that solid-state reaction between solder
and Cu occurs even at room temperature.
In the solid state, a eutectic alloy is below the eutectic temperature. It has
the unique thermodynamic property of a constant chemical potential indepen-
dent of composition when the temperature is constant. The microstructure of
the solid eutectic alloy is a two-phase mixture of two primary phases and they
are at equilibrium with each other, independent of the amount of each phase.
Therefore, in the two-phase microstructure, phase separation can occur with-
out chemical potential change. Consequently, driven by an external force as
in electromigration or in thermomigration, phase separation occurs readily in
a eutectic solder, and the resulting microstructure change is quite unique, to
be discussed in Chapters 9 and 12.
In the eutectic SnPb alloy, Pb does not react with Cu to form any in-
termetallic compound. The binary Pb-Cu system is immiscible, so the solder
reaction is between Sn and Cu. The purpose of alloying Pb with Sn is to
lower the melting point, to soften the alloy or to enhance the ductility, and
to provide a shiny appearance. Besides, eutectic SnPb is known to have no
Sn whisker growth. Due to the environmental concerns of Pb, there are four
anti-Pb bills pending in the U.S. Congress, including one from the Environ-
mental Protection Agency. In the European Union, the WEEE (Waste from
Electrical and Electronic Equipment) has issued a Directive which calls for a
ban of Pb-containing solders in all electronic consumer products on July 1,
2006. At the moment, no chemical element has been found to replace Pb and
to function as well as Pb. The most promising Pb-free solders which can re-
place the eutectic SnPb are the eutectic SnAgCu, eutectic SnAg, and eutectic
SnCu. These Sn-based solders have a very high concentration of Sn, e.g., the
eutectic SnCu has 99.3 wt% Sn and the eutectic SnAgCu has about 95 to 96
wt% of Sn. Hence, in essence the solder reaction between the Pb-free and Cu
is the Cu-Sn reaction.
SVNY339-Tu March 2, 2007 22:0

4 Chapter 1

In the following, the major solder joint technologies used in manufacturing

of electronic devices will be described briefly. They are the surface mount
technology, pin-through-hole technology, and flip chip technology [1, 2]. A
clear picture of where and why solder is used in microelectronic devices is
presented below in order to reveal the link to issues of reliability. Then the
cause of reliability in these technologies will be addressed briefly, including
Sn whisker growth, spalling of interfacial intermetallic compounds, low cycle
fatigue, and electromigration and thermomigration in flip chip solder joints.
Finally, the future trend in microelectronic packaging and the role of solder
joint in the trend will be discussed. Because of the trend of miniaturization
of devices and packaging technology, a continuing study of these reliability
issues is needed in the near future. Since the topic of thermal stress-induced
fatigue and creep has been well covered in many books, it will be briefly
mentioned here. The two important contributors to solder joint reliability to
be emphasized in this book are copper–tin reaction, and electromigration and
thermomigration.

1.2 Lead-Free Solders

1.2.1 Eutectic Pb-Free Solders

At present, nearly all the eutectic Pb-free solders are Sn-based. A special class
of them are the eutectic alloys consisting of Sn and noble metals: Au, Ag, and
Cu. Other elements to alloy with Sn such as Bi, In, Zn, Sb, and Ge have
been considered. The eutectic points of the binary Pb-free solder systems are
compared with that of eutectic SnPb in Table 1.1. It can be seen that there is
a large temperature gap between the eutectic temperatures of SnZn (198.5◦ C)
and SnBi (139◦ C), for which no known Pb-free solder system exists.
Zinc (Zn) is cheap and readily available, but it quickly forms a stable oxide,
resulting in excessive drossing during wave soldering, and more problemati-
cally, it shows very poor wetting behavior due to the stable oxide formation.
Hence, a forming gas ambient is required. The eutectic SnZn has a melting
point which is closest to that of eutectic SnPb among all the eutectic Pb-free
solders and it has received much attention in Japan, especially. Bismuth (Bi)
has very good wetting properties. The eutectic Sn-Bi solder has been used in
pin-through-hole technology which will be discussed in the next section. How-
ever, the availability of Bi could be limited by the restrictions on Pb, because
the primary source of Bi is a by-product in Pb refining. By restricting the use
of Pb, much less Bi will be available. Antimony (Sb) has been identified as a
harmful element by the United Nations Environment Program. Germanium
(Ge) is used only as a minor alloying element of multicomponent solders due
to its reactivity. Indium (In) is too scarce and too expensive to be considered
for broad applications, besides it forms oxides very easily.
A common characteristic of eutectic Sn-noble metal alloys is the high melt-
ing point and high concentration of Sn compared to that of eutectic SnPb.
SVNY339-Tu March 2, 2007 22:0

Introduction 5

Ag3Sn

Cu6Sn5

100μm

Fig. 1.1. SEM image of Ag3 Sn platelet in SnAgCu solder joint.

Hence the corresponding reflow temperature will be higher, by about 40◦ C.

It may increase the dissolution rate and solubility of Cu and Ni in the molten
solder as well as the rate of intermetallic compound (IMC) formation with Cu
and Ni under-bump metallization. If the surface and interfacial energies are
considered, the surface energies of these Pb-free solders are higher than that
of SnPb, so they form a larger wetting angle on Cu, about 35 to 40◦ .
Concerning the microstructure of these eutectic solders, they are a mix-
ture of Sn and IMC because of the high concentration of Sn, unlike that of
eutectic SnPb which has no IMC. Since metallic Sn has the body-centered
tetragonal lattice structure and tends to deform by twinning, its mechanical
properties are anisotropic. The electrical conductivity of metallic Sn is also
anisotropic. The mechanical and electrical properties of these eutectic solders
will be anisotropic, thus the dispersion of the IMC may lead to the formation
of inhomogeneous microstructures, especially in the case of Ag3 Sn. The image
of Ag3 Sn appears to be long needlelike crystals in the eutectic SnAg on the
cross-sectional image of a solder joint. But after the matrix of the solder is
removed by deep etching, they turn out to be platelike, as shown in Fig. 1.1. If
such Ag3 Sn crystals formed in a high-stress area, such as the corner of a solder
bump, cracks can be initiated and can propagate along the interface between
the Ag3 Sn and solder, leading to fracture failure as observed in Fig. 1.2. To
avoid the formation of such large plate-type IMC, the Ag concentration in
the solder should be less than 3%, below the eutectic composition of 3.5%
as shown in Table 1.1. In the case of eutectic SnCu, it has only 0.7 wt%
Cu, so the bulk of the solder is almost pure Sn. Then, the phenomenan of
Sn whisker [3–5], Sn pest [6], and Sn cry [7] are of concern. Furthermore, if
electroplating is used to prepare the solder, it is very difficult to have a con-
trolled composition within 1%. In the case of Au-Sn solder, the formation of
SVNY339-Tu March 2, 2007 22:0

6 Chapter 1

Fig. 1.2. SEM image of crack at the corner of SnAgCu solder joint.

Table 1.1. Binary Pb-free eutectic solders

Systems Eutectic temp. (◦ C) Eutectic composition (wt%)

Sn-Cu 227 0.7

Sn-Ag 221 3.5
Sn-Au 217 10
Sn-Zn 198.5 9
Sn-Pb 183 38.1
Sn-Bi 139 57
Sn-In 120 51
SVNY339-Tu March 2, 2007 22:0

Introduction 7

Fig. 1.3. Eutectic ternary phase diagram showing the eutectic point of SnAgCu.

AuSn4 means that one Au atom will attract four Sn atoms. Hence, a small
amount of Au can lead to the formation of a large amount of IMC. This is
the origin of brittle “cold” joint when the content of Au is more than 5 wt%
in a solder.
Ternary and higher order solders are most likely based on the binary eutec-
tic Sn-Ag, Sn-Cu, Sn-Zn, or Sn-Bi alloys. The most promising one is eutectic
Sn-Ag-Cu. The eutectic Sn-Ag-Cu alloy forms good quality joints with copper.
Its thermomechanical property is better than those of the conventional Sn-Pb
solder. Its eutectic temperature has been determined to be about 217◦ C, but
its eutectic composition has been a subject of controversy. Based on metal-
lographic examination, differential scanning calorimetry measurements, and
differential thermal analysis results, the eutectic composition was estimated
at 3.5±0.3Ag, 0.9±0.2Cu (wt%) [8–10]. These data are plotted together with
the thermodynamically calculated phase diagram in Fig. 1.3, where the equi-
librium eutectic point is calculated to be Sn-3.38Ag-0.84Cu (wt%). While the
accuracy of the eutectic point is of academic interest, it is an important issue
in application. Due to the higher melting point, the reflow temperature will
be around 240◦ C, which is higher than that of eutectic SnPb. It is a manu-
facturing issue, for the applications on those polymer substrates that have a
low glass transition temperature as well as for the flux used in solder paste
that has a high evaporation rate. Because the chemical composition of the
flux for Pb-free solders is not as yet optimized, the reflow of Pb-free solder
SVNY339-Tu March 2, 2007 22:0

8 Chapter 1

paste has produced many more residue voids in solder joints than the eutectic
SnPb paste. The migration of these voids to contact interfaces can become a
reliability issue.
Both Cu and Ni thin films are used widely as on-chip under-bump metal-
lization (UBM), but they can be reacted and dissolved away by molten solder
during reflow, resulting in spalling of IMC, which is a subject of serious re-
liability concern to be discussed in Section 1.4.2. Hence, a solder of eutectic
composition is undesirable when it is used with thin-film UBM. It must be
supersaturated with excess Cu and/or Ni, about 1%, in order to reduce the
dissolution. Therefore, the recommended composition for SnAgCu solder is
about Sn-3Ag-3Cu. While it is off the eutectic composition and will not have
a single melting point, the effect of the composition on melting temperature
is very small and will not be an issue in manufacturing.

1.2.2 High-Temperature Pb-Free Solders

At present, there is no high-temperature Pb-free solder to replace the 95Pb5Sn
high-Pb solder, except the 70Au30Sn. A Sn-based solder of Sn-Ag or Sn-Cu
having a higher Ag or Cu, respectively, than the eutectic composition has a
large temperature gap between the liquidus and solidus points, hence it is
undesirable as a solder due to partial or nonuniform melting when it goes
through the large temperature gap. The 70Au30Sn alloy (at.%) has a eutec-
tic point of 280◦ C, so it can be considered as a high temperature Pb-free
solder, yet it is known to have a poor reflow behavior besides high cost.
The binary system of Au-AuSb2 has a eutectic temperature at 360◦ C, yet
its wetting property on Cu is unknown. Alloys of Sb-Sn have been consid-
ered as a high-temperature Pb-free solder, yet they have a large gap between
the liquidus and solidus temperatures too. Sn-Zn alloys have the same prob-
lem.
In the periodic table of chemical elements, Ge and Si belong to the same
column as Sn and Pb, therefore, it is of interest to consider eutectic alloys of
them with noble metals, such as Au-Ge which has a eutectic point at 356◦ C.
Concerning their wetting behavior on Cu, an active flux must be available
to prevent the oxidation of Ge or Si. However, these eutectic alloys undergo
volume expansion upon solidification. In the molten state when Ge and Si are
alloying with noble metals, their liquid structure is rather closely packed. But
in solidification to form a two-phase eutectic structure, the Ge and Si have
the diamond structure, which is quite open. Mechanical properties of these
eutectic joints may be poor because Si and Ge are not metals. To overcome
this problem, we may consider replacing Si or Ge by silicide or germanide
phases. In other words, we consider the eutectic structures of silicide and
noble or near-noble metals.
The ternary system of Au-Ge-Co might be of interest since Au-Ge has a
eutectic temperature of 356◦ C with a eutectic composition of Ge at 27 at.%.
Also, Au-Co forms a eutectic system with a eutectic temperature at 996◦ C.
SVNY339-Tu March 2, 2007 22:0

Introduction 9

Co and Ge form several germanides such as Co2 Ge, CoGe, and CoGe2 . It
is unknown if a eutectic structure can be formed between Au and one of
the germanides with a low eutectic temperature. However, the ternary phase
diagram must be investigated first, the eutectic point of this ternary system
must be calculated, and an active flux must be available.
Owing to the trend of miniaturization, the size of flip chip solder joint will
be below 50 μm in diameter, say 25 μm. For such a small solder joint, it can be
transformed completely into intermetallic compound by solder reaction at the
reflow temperature in 1 min or so. In Section 7.3, the rapid reaction between
Pd and Sn to form Pd-Sn compounds will be discussed. Since the intermetallic
compound has a high melting point, it will not melt at the reflow temperature
after it is formed. Since it is Pb-free, it might serve as a high-temperature Pb-
free solder joint.

1.3 Solder Joint Technology

1.3.1 Surface Mount Technology
In most of the portable electronic consumer products, such as mobile hand-
held telephones, wire bonding technology is the most widely used method
to connect the circuits on a Si chip to a leadframe substrate, which is then
connected electrically to the outside via an antenna. Figure 1.4 shows wire
bonding between a Si chip and a Cu leadframe. The legs of the leadframe
are joined to bond pads on a packaging circuit board by using solder joints.

Si

Sealing glass

Leads

Fig. 1.4. Schematic diagram of wire bonding between a Si chip and a leadframe.
SVNY339-Tu March 2, 2007 22:0

10 Chapter 1

Sn Whisker

Cu Pb-free finish
Leadframe Sn, SnAg, SnCu,
and SnPb
e-SnPb
e-nAgCu
Surface Mount Solder
Board Pad
FR4 board

Fig. 1.5. A schematic diagram of the solder joint between a leg and a substrate board.

These solder joints were fabricated by printing a pattern of mounts of solder

paste on the bond pads and every leg is placed on a printed paste mount,
then the assembly will be placed on a belt moving through a tube furnace
in a forming gas ambient and heating to a temperature above the melting
point of the solder paste to achieve joining. Figure 1.5 is a schematic dia-
gram of the solder joint between a leg and a substrate board. The heating
process is called “reflow,” and typically the temperature of reflow goes above
the melting point of the solder by 30 to 40◦ C for half a minute or so. During
this period the solder paste melts and reacts with the leadframe as well as
the bond pad to form a metallic joint. To facilitate the joining of all the legs
within the half-minute period, the leadframe is coated with a layer of eutectic
SnPb. Owing to environmental concerns, the coating has been replaced with
eutectic SnCu or pure Sn. However, spontaneous Sn whiskers are found to
grow on these Pb-free coatings. The whiskers may become electrical shorts
between the legs, and represent a reliability issue at the present time. Be-
sides Sn whiskers, cracking may occur along the interface between the leg and
the solder, mostly due to solder reaction induced void formation or impurity
segregation.

1.3.2 Pin-through-Hole Technology

The legs of the leadframe are typically bent into the shape of a pair of gale-
wings or J-wings so that they can be placed on the solder paste mounts on
the surface of a board for joining in surface mount technology. However, for
devices in military applications, in order to enhance the mechanical reliability
of the joints, the legs are straight so that they are like pins and they can be
inserted into holes drilled in the board. The holes have been plated with Cu
and immersion Sn, so that when the board with the inserted pins is guided
over a fountain of molten solder, the molten solder touches the bottom surface
of the board, wets the immersion Sn, and rises along the plated holes by the
capillary force. Thus, the pins are soldered to the board through the holes. For
SVNY339-Tu March 2, 2007 22:0

Introduction 11

Fig. 1.6. Cross-sectional image of through holes plated with Cu and immersion Sn.

this reason, it is also called pin-through-hole technology. Compared to surface

mount technology, this technology has better mechanical reliability but also
higher cost.
The surface of the immersion Sn affects the capillary force and in turn
controls the wetting action. If the plating bath contains impurity of S, the
immersion Sn surface will appear gray and dull and cannot be wetted by the
molten solder. The immersion Sn from a good plating bath should appear
white and bright. Figure 1.6 is a cross-sectional image of the through holes
plated with Cu and immersion Sn and part of the holes was not filled with
solder. The incomplete filling was because of gray Sn [11].
The potential of the pin-through-hole technology is that it can be applied
to the plating of macro-through-holes in Si. This will be clear after we have
discussed the thermal stress issue in flip chip technology in the next section.
The diameter of the through hole will be 10–100 μm, i.e., nearly the same di-
ameter as flip chip solder balls. We can fill the macro-through-holes with solder
or Cu and use the Si with plated-through-holes as packaging substrate for flip
SVNY339-Tu March 2, 2007 22:0

12 Chapter 1

chip technology. When a Si chip is joined to a Si substrate, there will be no

or very little thermal stress.

1.3.3 C-4 Flip Chip Technology

In mainframe computers such as a server, the solder joints connecting Si chips
to their substrate are more complicated. To see the complication, we begin
with two simple facts about interconnection on a chip. Today’s aluminum or
copper wiring on a very-large-scale-integration (VLSI) Si chip is 0.5 μm (or
less) in width. Assuming the spacing between two parallel wires is also 0.5 μm,
the pitch is 1 μm. Therefore, on a 1 cm2 chip area, we can have 104 wires, each
being 1 cm in length. This means that in a single layer of wiring, the total
length of the wire is 100 meters. Since six to seven layers of wiring are made
on a chip and if we add the length of vias between the layers, we find that
the interconnect wiring on a 1 cm2 chip is about 1 kilometer! We need this
much wiring to connect millions of transistors on the chip in order for them
to function together. This is the first fact. To provide external electrical leads
to all these wires on the chip, we may need several thousands of input/output
(I/O) pads on the chip surface of a logic chip. At present, the only practical
way to provide such high density of I/O pads is to use area array of tiny
solder balls. We can have 50-μm-diameter solder balls with a spacing of 50 μm
between them, so the pitch is 100 μm. We place 100 of them along a length
of 1 cm or 10,000 of them on an area of 1 cm2 . Thus, the second fact of
interconnection between a chip and its substrate is that we can have about
10,000 I/Os or solder balls on a chip surface. Because of the expected use of
solder balls of such small size and large number, the International Technology
Roadmap for Semiconductors (ITRS) since 1999 has identified “solder joint
in flip chip technology” to be an important subject of study concerning its
yield in manufacturing and its reliability in use [12]. Actually, 25-μm-diameter
solder balls are being developed.
What is a flip chip? It refers to a method of achieving electrical connections
between a Si chip and a ceramic module or a printed circuit board. The Si
chip is flipped facing down so the circuit of very-large-scale integration faces
the substrate. The electrical connection is achieved through an area array
of solder bumps between the chip and its substrate. Unlike wire bonding, in
which the wire connections are made at the periphery of the chip, flip chip
connections are an area array of solder bumps covering all or a large part of
the surface of the chip.
Figure 1.4 is a schematic diagram of wire bonding of a Si chip to a lead-
frame. The leads (legs) of the leadframe are soldered to a circuit board by
surface mount technology or by pin-through-hole technique. The VLSI side of
the chip is upside-up in wire bonding. Because wire bonding requires ultra-
sonic vibration, the stress applied during the bonding process may damage
the structure around and underneath the bonded area. Therefore, it must be
done on the periphery of the chip, away from the active VLSI region in the
SVNY339-Tu March 2, 2007 22:0

Introduction 13

Fig. 1.7. An area array of solder balls on a chip surface.

central area of the chip. Even if we can use 20-μm wire with 20-μm spacing,
we can only have about 1000 I/Os on the periphery of a 1 cm2 chip, which is
much less than the 10,000 solder balls that can be deposited or electroplated
on the same chip area. Since the wire bondpads occupy the periphery of the
chip, it wastes a large fraction of the chip surface too.
In placing solder balls over the entire surface of a Si chip, it has been found
that there is no stress problem when solder balls are placed directly over an
active VLSI area. Figure 1.7 shows an area array of solder balls on a chip
surface. To join the chip to a substrate, the chip will be flipped over, so it is
a flip chip and the VLSI side of the chip is upside-down, facing the substrate.
Generally speaking, the advantages of flip chip technology are smaller
packaging size, large I/O lead count, and higher performance. In chip-size
packaging, where the packaging is of nearly the same size as the chip, the
small packaging size is achieved without the periphery bonding area. The
larger I/O lead count is due to area array. The higher performance occurs be-
cause the bumps in the central part of the chip will allow the device to operate
at lower voltage and higher speed. For devices that require these advantages
such as handheld devices, flip chip is the only existing technology that can
provide the reliability needed.
The International Technology Roadmap for Semiconductors [12] has pro-
jected that Si technology will still be able to advance a new generation every
two to three years in the foreseeable future, probably until 2015. To stay with
the Si chip technology, the packaging technology must advance too. Thus,
SVNY339-Tu March 2, 2007 22:0

14 Chapter 1

C4
Chip solder
Joints

**out Ceramic
module
Eutectic
solder BGA

FR4 90%Pb/10% Sn
card ball

Fig. 1.8. Schematic diagram of cross section of a ceramic module joined to a large
printed circuit board and resulting in the two-level packaging scheme for mainframe
computers.

the circuit density and the number of I/Os on the packaging substrate must
increase. For the projected technology, reliability issues need to be discussed.
Flip chip technology has been used for over 30 years in mainframe
computers. It originated from the “controlled collapse chip connection” or
“C-4” technology in packaging chips on ceramic modules beginning in the
1960s [13, 14]. For a detailed discussion of C-4 technology and the history
of its evolution, readers are referred to a recent review by Puttlitz and
Totta [1].
When the VLSI chip technology was developed, a high density of wiring
and interconnection was required for the packaging. This has led to the de-
velopment of multilevel metal-ceramic modules and multichip modules for
mainframe computers. In a multilevel metal-ceramic module, many levels of
Mo wires were buried in the ceramic substrate. Each of these modules could
carry up to a hundred chips. Several of these ceramic modules were joined to
a large printed circuit board and resulted in the two-level packaging scheme
for mainframe computers, shown in Fig. 1.8. It consisted of a first-level chip-
to-ceramic module packaging and a second-level ceramic module-to-polymer
board packaging. In the first-level packaging, the on-chip under-bump metal-
lization (UBM) is a trilayer thin film of Cr/Cu/Au. Actually, in the trilayer
the Cr/Cu has a phased-in microstructure for the purpose of improving the
adhesion between Cr and Cu and strengthening its resistance against solder
reaction so that it can last several reflows. The bond pad on the ceramic sur-
face is typically Ni/Au. The solder which joins the UBM and the bond pad
is a high-Pb alloy such as 95Pb5Sn or 97Pb3Sn. A schematic diagram of the
cross section of the joint is shown in Figure 1.9. The on-chip solder bumps
SVNY339-Tu March 2, 2007 22:0

Introduction 15

Via
Si
Chip side
SiO2 Al
Cr
Cu Phased-in
Pb-Sn SOLDER Au
95 5
Au
Ni
Mo Module
side
MULTI-LAYERED CERAMICS

Fig. 1.9. Schematic diagram of the cross-section of the flip chip solder joint is shown.

were deposited by evaporation and patterned by lift-off, but now they are
deposited by selective electroplating. The high-Pb bump has a melting point
over 300◦ C. During the first reflow (around 350◦ C), a ball shape of the bump
is obtained on the UBM. Since the surface of SiO2 cannot be wetted by molten
solder, the base of the molten solder bump is defined by the contact opening of
the UBM, thus the molten solder bump balls up on the UBM contact. There-
fore, the UBM controls the dimensions (height and diameter) of the solder
ball when its volume is given. Often, the UBM is called “ball-limiting metal-
lization” (BLM). The BLM controls the height of the fixed volume of a solder
ball when it melts; this is the control in “controlled collapse chip connection.”
Without the control, the solder ball will spread and then the gap between the
chip and the module is too small.
To join the chips to a ceramic module, a second reflow is used. During
the second reflow, the surface energy of the molten solder bumps provides a
self-aligning force to position the chip on the module automatically. When
the solder melts to join the chip to the module, the chip will drop slightly
and rotate slightly. The drop and rotation are due to the reduction of surface
tension of the molten solder ball, which achieve the alignment between the
chip and its module, so it is a controlled collapse process.
We note that the high-Pb solder is a high melting-point solder, yet both
the chip and the ceramic module can withstand the high temperature of reflow
without a problem. Additionally, the high-Pb solder reacts with Cu to form
a layer-type Cu3 Sn, which can last several reflows without failure. We note
that each of the metals in the trilayer of Cr/Cu/Au has been chosen for a
particular reason. First, solder does not wet the Al wire, so Cu is selected for
its reaction with Sn to form intermetallic compounds (IMC). Second, Cu does
not adhere well to the dielectric surface of SiO2 , so Cr is selected as a glue layer
for the adhesion of Cu to SiO2 . The phased-in Cu-Cr UBM was developed to
improve the adhesion between Cu and Cr. Since Cr and Cu are immiscible,
their grains form an interlocking microstructure when they are co-deposited.
SVNY339-Tu March 2, 2007 22:0

16 Chapter 1

A discussion of the lattice image of the phased-in Cu-Cr microstructure will be

given in Chapter 2. In such a phased-in microstructure, the Cu adheres better
to the Cr and also it will be harder for Cu to be leached out to form IMC
with Sn during reflow. Furthermore, the phased-in microstructure provides a
mechanical locking of the IMC. Finally, Au is used as a surface passivation
coating to prevent the oxidation or corrosion of Cu. It also serves as a surface
finish to enhance solder wetting.
In the second-level packaging of ceramic module to polymer board, i.e.,
to join the ceramic substrate to a printed polymer circuit board, another
area array of solder joints is placed on the backside of the ceramic substrate.
They are called ball-grid-array (BGA) solder balls which have a much large
diameter than the C-4 solder balls. Typically the diameter is about 760 μm.
They are eutectic SnPb solder with a lower melting point (183◦ C), which
is reflowed around 220◦ C. Sometimes, composite solder balls of high-Pb and
eutectic SnPb are used, with the high-Pb as the core of the ball. It is obvious
that during this reflow (the third), the high-Pb solder joints in the first-level
packaging will not melt.
In summary, there are three steps in C-4 flip chip joints. The first is solder
bumping on a chip surface by electroplating or by stencil printing of an area
array of solder bumps, followed by a reflow to transform the bumps into balls.
The second step is to bond the chip to its substrate in a flip chip configuration
by a second reflow process. The substrate has an area array of bond pads to
receive the balls. The third step is to underfill the gap between the chip and
the substrate with epoxy. However, the BGA in the second level of packaging
does not use underfill.

1.4 Reliability Problems in Solder Joint Technology

1.4.1 Sn Whiskers
In May 1998, the Galaxy 4 satellite was killed in space by the growth of
Sn whiskers; one of them bridged or shorted a pair of metal contacts in the
satellite’s control processor [3–5]. The loss of the satellite was just one of the
more visible examples of reliability problems caused by Sn whiskers in those
electronic devices that require a long term and high degree of reliability. As
eutectic SnPb solder is replaced by the Sn-based Pb-free solders, the whisker
problem is of concern. Figure 1.10 shows an SEM image of whiskers on the
legs of a leadframe, in which a very long whisker can be seen to have bridged
a pair of the legs.
The growth of whiskers on beta-tin (β-Sn) is a surface relief phenomenon
of creep. It is driven by a compressive stress gradient and occurs at ambient.
The whisker growth is spontaneous, indicating that the compressive stress is
self-generated; no external applied stress is necessary. Otherwise, we expect
a whisker to slow down and stop when the applied stress is exhausted, if
SVNY339-Tu March 2, 2007 22:0

Introduction 17

Whisker

Fig. 1.10. SEM image of Sn whisker shorting two legs.

not applied continuously. Therefore, it is of interest to ask where is the self-

generated driving force coming from? How can the driving force maintain
itself to achieve the spontaneous and continuous whisker growth? Also, how
large must the compressive stress be to grow a whisker?
Spontaneous Sn whisker growth occurs readily on matte Sn finishes on Cu.
Today, because of the wide application of Pb-free solders on Cu conductors
used in the electronic packaging in consumer electronic products, Sn whisker
growth has become a widely recognized reliability issue. When the solder finish
on Cu leadframe is eutectic SnCu or matte Sn, whiskers are observed. We will
show in Chapter 6 that the self-generated driving force is due to the room
temperature reaction between Cu and Sn to form Cu6 Sn5 in the Pb-free finish
on the Cu leadframe. Some whiskers can grow to several hundred micrometers,
which are long enough to become electrical shorts between neighboring legs of
a leadframe. As the dimension of packaging shrinks, shorter whiskers will cause
problems. Hence, how to perform systematic studies of Sn whisker growth in
order to understand the driving force, the kinetics, and the mechanism of
growth, and how to suppress Sn whisker growth are challenging tasks in the
electronic packaging industry today.

1.4.2 Spalling of Interfacial Intermetallic Compounds in Direct

Chip Attachment
In Section 1.3.3 on flip chip technology, we have shown a two-level packag-
ing scheme which has worked well for mainframe computers, yet the ceramic
module is too expensive for low-cost and large-volume production of consumer
SVNY339-Tu March 2, 2007 22:0

18 Chapter 1

goods. To save cost, the electronic industry removes the ceramic module (or
the first level packaging) so that chips can be attached directly to printed
polymer circuit boards. This is so-called “direct chip attachment” or “flip
chip on organic technology.” Since polymer substrates have a low glass tran-
sition temperature, the reflow temperature should be lower for flip chip on
organic substrates, therefore the high-Pb solder cannot be used.
Since the Cr/Cu/Au thin films have been successfully used as on-chip
metallization in C-4 technology, one naturally attempts to transfer it to direct
chip attachment, i.e., to join chips having the Cr/Cu/Au UBM to an organic
substrate with eutectic SnPb solder. However, when eutectic SnPb is used
to wet the Cr/Cu/Au or even the phased-in Cu-Cr, the joint fails in multiple
reflows. The low-melting-point eutectic solder contains a high concentration of
Sn (74 at.% Sn). It can consume all the Cu film very quickly (at a rate of about
1 μm/min at 200◦ C), resulting in spalling of the Cu-Sn compound [15–17].
Thus, the solder joint becomes mechanically weak because there is no adhesion
between the solder and the remaining Cr. Figure 1.11 shows cross-sectional
SEM images of the interface between eutectic SnPb and Au/Cu/Cu-Cr thin-
film UBM after a heat treatment at 200◦ C for (a) 1 min, (b) 1.5 min, and (c) 10
min. Many of the spherical grains in Fig. 1.11(c) have spalled (detached from
the Cr/SiO2 surface) into the molten solder [16, 17]. The spalling phenomenon
is extremely undesirable, because it leads to chemically and mechanically weak
joints since the solder is now in contact with the unwettable Cr. Indeed, when
we performed mechanical test of samples of solder balls sandwiched between
two Si chips with Cu/Cr UBM, the load needed to fracture the joints decreased
dramatically with increasing heat treatment time [18, 19].
In essence, if we replace the high-Pb solder joint as shown in Fig. 1.9
by the eutectic SnPb, the wetting reaction between molten eutectic SnPb
and Au/Cu/Cr UBM becomes a problem. To solve this problem which is
due to the reaction between a molten solder and a thin-film UBM, there
are two general approaches: we can either improve the solder or improve the
UBM. They are illustrated in Fig. 1.12(a) and (b). About the first approach,
since we know that the phased-in Au/Cu/Cr works well with the high-Pb
solder, we keep them as they are. But we will use a low-melting-point eutectic
solder to join the high-Pb solder, this is the “composite solder” approach,
as shown in Fig. 1.12(a). Detailed discussion about composite solder joints
will be given in Section 4.2. The eutectic solder bump can be deposited on
the organic substrate before joining. The key advantage is that the reflow
temperature is low since it only needs to melt the low-melting-point solder
which will wet the high-Pb solder. Nevertheless, we note that this approach
has a potential problem. During reflow, the molten eutectic solder can migrate
along the outer surface of the high-Pb solder bump to reach the circumference
of the Au/Cu/Cr. Again, a certain amount of spalling of IMC occurs in the
circumference of the UBM. More seriously, electromigration will drive the Sn
from the substrate side to the chip side to replace the high-Pb when electrons
flow from the chip to the substrate, to be discussed in Chapter 9.
SVNY339-Tu March 2, 2007 22:0

Introduction 19

1 min

1.5 min

10 min

Fig. 1.11. (a) to (c) show cross-sectional SEM images of the interface between eutectic
SnPb and Au/Cu/Cu-Cr thin film UBM after a heat treatment at 200◦ C for 1 min,
1.5 min, and 10 min, respectively.

In the second approach, a Ni-based UBM is used typically to replace the

Cu-based UBM in order to slow down the solder reaction, as shown in Figure
1.12(b). We recall that Ni has been used in the bond pad on the ceramic
module side, and indeed Ni has been found to have a much slower solder
reaction rate with the eutectic SnPb (to be discussed in Chapter 7). However,
evaporated or sputtered Ni films tend to have a high residue stress. The stress
may crack the dielectric layer of SiO2 on the chip surface. That is why Ni has
been used only on the module side where there is no active device. Two kinds
of low-stress Ni-based UBMs are currently being used in wafer bumping. One
is electroless Ni(P) UBM, which is quite thick, over 10 μm, and the other is a
sputtered thin film Cu/Ni(V)/Al, in which the Ni(V) thin film is about 0.3 μm
in thickness. We shall discuss their wetting reactions in Sections 3.6 and 3.7.
SVNY339-Tu March 2, 2007 22:0

20 Chapter 1

(a) (b)

Fig. 1.12. (a) Direct chip attachment by using a composite solder joint. (b) Direct
chip attachment by using Ni-based UBM.

Besides Ni-based UBM, we can also use a very thick Cu UBM provided that
stress is not an issue. A thick Cu UBM or a Cu column bump will survive
several reflows without IMC spalling. As long as there remains free Cu in the
UBM, the Cu6 Sn5 compound will stick to the free Cu and will not spall.
All Pb-free solders, as far as we know today, are high-Sn solders. For
example, the eutectic SnAg solder has about 96 at.% Sn. The problem of
IMC spalling as we have discussed above is worse with these Pb-free solders
due to the very high concentration of Sn.
Solder reaction in flip chip technology has a dilemma. On the one hand, we
require a very rapid solder reaction in order to achieve the joining of thousands
of them simultaneously on a chip. On the other hand, we also want the solder
reaction to stop right after the joining since the on-chip thin-film UBM is
too thin to allow a prolonged reaction. Nevertheless, the manufacturing of
a device requires these solder joints to survival several reflows, in which the
total period of a solder bump in the molten state is several minutes. When
the solder is Pb-free and has a high concentration of Sn, the Cu-Sn reaction
rate is enhanced. Furthermore, the diffusion of Cu and Ni across a solder joint
of diameter 100 μm is so fast during reflow or during solid-state aging that
chip-to-packaging interaction takes place and affects solder joint reliability.
This is illustrated below.
Figure 1.13(a) is a schematic diagram depicting the metallurgical struc-
tures across a flip chip solder joint. The thin-film under-bump metallization
(UBM) on the chip side consists of 300 nm of Cu/400 nm of Ni(V)/400 nm of
Al. The thick metallic bond pad on the board side, across the solder bump,
consists of 125 nm of Au/10 μm of Ni(P) on a very thick Cu trace. The Cu
film and the Au film are respectively the surface metallization on the chip
side and on the board side, and between them is the eutectic SnAgCu solder
bump. Figure 1.13(b) shows the cross-sectional scanning electron microscopy
image of such a solder joint which bonded a chip (the bottom) to a board
SVNY339-Tu March 2, 2007 22:0

Introduction 21

Fig. 1.13. (a) A schematic di-

agram depicting the metallur-
gical structures across a real
flip chip solder joint. The thin
film under-bump metallization
(UBM) on the chip-side con-
sists of 300 nm of Cu/400 nm
of Ni(V)/400 nm of Al. The
thick metallic bond-pad on the
board-side, across the solder
bump, consists of 125 nm of
Au/ 10 μm of Ni(P) on a very
thick Cu pad. (b) The cross-
sectional scanning electron mi-
croscopy image of such a solder
joint which bonded a chip (the
bottom) to a board (the top).
The formation of scallop-type
IMC at the two solder inter-
faces can be seen. (c) The same
image of the joint after 10 re-
flows. The IMC on the chip-
side has spalled into the solder.

(the top). The formation of scallop-type intermetallic compound (IMC) at

the two solder interfaces can be seen. Figure 1.13(c) shows the same image
of the joint after 10 reflows. The IMC on the chip side has spalled into the
solder. In other words, the IMC has detached from the chip and moved into
SVNY339-Tu March 2, 2007 22:0

22 Chapter 1

the solder. Consequently, there is very little adhesion at the interface between
the solder and the chip. In other words, it is a weak interface chemically and
mechanically.
Therefore, the interest in solder reactions, especially the solder reaction
on thin films, is to understand the problem of spalling of IMC and to prevent
the latter from happening so that the solder joint is strong and can last a
long time. The scallop-type morphology is itself of interest. Why IMC forms
with such a morphology and why the morphology is stable (except ripening)
under isothermal annealing at 200◦ C up to 40 min are intriguing. The elec-
tronic packaging industry would like to have flip chip solder joints that can
last several reflows during manufacturing, and then in field use, they should
possess the reliability to survive solid-state aging, thermal stress cycling, and
electromigration.

1.4.3 Thermal-Mechanical Stresses

The difference in coefficients of thermal expansion between a Si chip and its
substrate is the cause of thermal-mechanical stress. Low cycle fatigue due
to the thermal stress, or the Coffin–Manson mode of fatigue, has long been a
reliability problem in the C-4 technology. To overcome the problem, a ceramic
module which has nearly the same thermal expansion coefficient as Si was
synthesized. Also, the technology of using a Si wafer as substrate for a Si chip
was developed. Yet, for low-cost consumer products, the thermal stress in flip
chip on organic substrates is very large, as shown in Figure 1.14(a). Due to
the very large difference in coefficients of thermal expansion between Si (α =
2.6 ppm/◦ C) and the organic FR4 board (α = 18 ppm/◦ C), a very large shear
strain exists in those bumps at the corners of a chip in direct chip attachment.
When the solder is in the molten state [see Figure 1.14(b)], there is no thermal
stress although the board has expanded much more than the chip. But upon
cooling, when the solder solidifies, the thermal mismatch begins to interfere.
We shall take the temperature difference to be that between room temperature
and 183◦ C, which is the solidification temperature of eutectic SnPb solder.
And we consider a bump at the corner of a chip of size 1 cm × 1 cm. The
shear is equal
√ to Δl/l = ΔαΔT. We obtain a value of Δl = 18 μm if we
take l = ( 2)/2 cm, which is the distance of half diagonal of the chip. If
we assume the chip to be rigid, the board will bend and the curvature is
concave downward. This is because the solid bumps will prevent the upper
surface of the board from shrinking, so the board bends when its lower part
shrinks [see Fig. 1.14(c)].
Due to the bending and the fact that the solder joints and the chip are not
rigid, the actual Δl will be less than the calculated value of 18 μm given in
the last paragraph. Figure 1.15(a) shows the schematic diagram and diagonal
cross-sectional SEM images of eutectic SnPb solder bumps between a flip chip
and an FR4 board. Figure 1.15(b) shows the joints in the center part of the
chip, where the alignment of the bump between the top UBM on the chip
SVNY339-Tu March 2, 2007 22:0

Introduction 23

Solder ball
Si chip

Organic module (FR4) UBM (a)

Δl Δl

(b)

Δ l′ Δ l′

q (c)

Fig. 1.14. (a) No thermal stress in flip chip on organic substrates before solder joining.
(b) When the solder is in the molten state, there is no thermal stress although the
board has expanded much more than the chip, so there is misalignment between the
UBM and bond pad, (c) Upon cooling to room temperature, the board bent concave
downward. This is because the solid bumps will prevent the upper surface of the board
from shrinking, so the board bends when its lower part shrinks.

and the bottom bond pad on the board is good. Figure 1.15(c) shows the
right-hand side corner of the chip, where we see that the bottom bond pad
on the board has been displaced to the right by about 10 μm. Figure 1.15(d)
shows the left-hand side corner of the chip, and the same displacement of the
bond pad is to the left. The nominal shear strain is Δl/h = 10/60, where
h = 60 μm is the gap between the chip and the board. In addition, the
chip is bent and has a concave downward curvature of 57 cm. Clearly, the
chip, the board, and the bumps are stressed. Besides the shear strain, there
may be normal stresses in the solder joints, especially across those bumps in
the center of the chip. During reflow, such thermal cycle is repeated. During
normal device operation, the chip will experience a working temperature near
100◦ C due to joule heating, and produces a low cycle thermal stress between
room temperature and 100◦ C to fatigue the solder joints. While the electronic
industry has introduced epoxy underfill to redistribute the stresses, it remains
a reliability issue.
SVNY339-Tu March 2, 2007 22:0

24 Chapter 1

Cutting line

Si chip Si chip
1 cm
1 cm
UBM

Solder ball and UBM

Organic substrate
Center Organic substrate
(a) (b)

Δx Δx 100 μm
Right
(c) (d) Left

Fig. 1.15. (a) A schematic diagram and diagonal cross-sectional SEM images of eu-
tectic SnPb solder bumps between a flip chip and a FR4 board, (b) SEM image of the
joints in the center part of the chip, where the alignment of the bump between the top
UBM on the chip and the bottom UBM on the board is good, (c) SEM image of the
right-hand side corner of the chip, where we see that the bottom UBM on the board
has been displaced to the right by about 10 μm, and (d) SEM image of the left-hand
side corner of the chip, and the same displacement of the UBM is to the left.

If we introduce underfill in the configuration shown in Fig. 1.15(c), the

solder joints have already been strained. Hence, it is better to apply underfill
to unstrained solder joints. Assuming it can be done, we still cannot avoid
the problem of thermal stress. This is because in subsequent solid-state aging,
in thermal cycling, and in device operation, the stress returns. If the solder
joint itself or its interface is weak, the stress can break it. It is worth noting
that it is this large shear strain that limits the size of a Si chip in a flip chip
manufacturing! Until we can solve the thermal stress issue, the chip size is
limited to about 1 × 1 cm2 . Nevertheless, chips having a size of 2 × 2 cm2 are
being made.
In Fig. 1.15, it is obvious that if we keep the chip and the board to be
the same and if we reduce the gap “h” between the chip and the board (or
the diameter of the bump), the shear strain increases. What is not obvious is
that if we keep the gap unchanged but increase the thickness of the UBM and
bond pad, it reduces the actual thickness of solder in between them, and it
will greatly increase the shear strain of the solder joint. This is clearly shown
in Fig. 1.15(b) to (d), where the UBM and bond pad are quite thick, so the
solder layer in between them is about 23 μm. Assuming that the UBM and
bond pad are rigid and the solder takes all the shear, its shear strain will be
Δl/h = 10/23, rather than Δl/h = 10/60 as given before. It is known that a
tall solder joint suffers less thermal cycle fatigue, but the current trend is to
use a smaller size solder bump and thicker UBM.
In addition, a thick IMC formed between the UBM and solder will also
further reduce the thickness of the unreacted solder and add to the shear
SVNY339-Tu March 2, 2007 22:0

Introduction 25

strain. While the total thickness of IMC may be only a few micrometers, its
thickness effect cannot be ignored if we use a small and thin solder joint. While
a thick UBM such as a column of Cu can be used in order to overcome the
problem of spalling of IMC and to increase the total height of a joint, it reduces
the thickness of the solder and creates a new problem of large shear strain of
the joint. This is especially serious when we deal with solder bumps that are
less than 50 μm in diameter. We envisage the diameter of the solder bump in
Figure 1.15(d) to be 50 μm and the thickness of the UBM and bond pad to
remain the same, the solder layer in between them will be much thinner, and
the shear strain will be much larger. The thickness of the intermetallic formed
between the UBM and the solder cannot be reduced because the reaction
temperature and time cannot be reduced.

1.4.4 Impact Fracture

While epoxy underfill has been introduced to strengthen the bonding of C-
4 solder joints, no underfill is applied to assist the ball-grid array (BGA)
of solder joints. Typically, the diameter of the solder ball in BGA is about
760 μm, so its volume or weight is three orders of magnitude larger than that
of a C-4 solder bump 76 μm in diameter. Without underfill, the gravity of
the very heavy solder ball itself can exert a very large force to break from
its substrates during an impact or a shock. Consider that wireless, portable,
and hand-held electronic consumer products are dropped accidentally to the
ground very often. The impact of dropping may induce fracture of BGA solder
joint interfaces, which is now one of the most serious failure modes of these
devices. The high-speed shear stress in impact is as important as, if not more
important than, the low cycle thermal stress discussed above from the point
of view of reliability of consumer products. To characterize the toughness of a
solder joint against impact, a mini Charpy impact test machine has been used
to measure the impact toughness of solder joints and to study the ductile-
to-brittle transition in solder joints [20]. The transition occurs due to aging
when either a large number of Kirkendall voids are formed at a solder joint
interface or the segregation of a certain impurity occurs in the interface to
cause brittleness. More discussion of the impact and drop tests will be given
in Chapter 11.

1.4.5 Electromigration and Thermomigration

The present packaging design rule is to distribute 1 ampere over five power
solder bumps, or 0.2 ampere per bump. For a solder bump 100 μm in diameter,
the current density will be about 2 × 103 A/cm2 . Since the contact area of the
bump is much smaller than the cross section of the bump, the actual current
density may be a factor of 2 higher at the solder bump contact where the
current enters the bump. While this current density is about two orders of
magnitude less than that in Al or Cu interconnect lines, electromigration
SVNY339-Tu March 2, 2007 22:0

26 Chapter 1

in the solder bumps cannot be ignored [21–24]. This is because of the low
melting point and high atomic diffusivity of solder alloys. For the eutectic
SnPb solder with a melting point of 183◦ C, room temperature is about two-
thirds of its melting point on the absolute temperature scale. This is the same
for Pb-free solders. The second reason is the low “critical product” of solder
alloys, so that electromigration can occur in solder alloy under a very low
current density, even as low as 5 × 103 A/cm2 . This point will be discussed
in detail in Chapter 8. The third reason is the line-to-bump geometry, in
which a large change of current density exists between the interconnect line
and the solder bump. It leads to a unique current crowding at the line-to-
bump contact interface, where the current density is a factor of 10 to 20
higher than the average current density in the bump. Effect of the current
crowding is the most serious reliability issue in flip chip solder joints from
the point of view of electromigration failure [25]. The fourth reason is joule
heating from the Al or Cu interconnect line contacting the bump. The joule
heating not only will increase the temperature of the solder bump, which
in turn will increase the rate of electromigration, but also may produce a small
temperature difference across the solder bump to cause thermomigration. A
temperature difference of 10◦ C across a solder bump 100 μm in diameter
will give a temperature gradient of 1000◦ C/cm, which cannot be ignored [26].
Thermomigration will be covered in Chapter 12.
Another very unique and important aspect electromigration behavior in
solder joints is that it has two reactive interfaces. The polarity effect on
IMC growth exists at the cathode and the anode. Electromigration drives
atoms from the cathode to the anode. Therefore, it tends to dissolve or retard
the growth of IMC at the cathode but build up or enhance the growth of
IMC at the anode [27–29]. Figure 1.16 shows a SEM cross-sectional images of
electromigration-induced failure at the cathode contact interface. The effect
of current crowding can be recognized very clearly because the failure was
initiated at the place where the current enters the solder bump. A detailed
discussion will be given in Chapter 9.

1.4.6 Reliability Science on the Basis of Nonequilibrium

Thermodynamics
The examples of reliability problems given above indicate that the time-
dependent microstructure change or instability in solder joints is different from
the phase transformations in conventional metallurgical systems. The latter
occurs typically between two equilibrium end states, for example, in precipita-
tion of GP zones or in martensitic transformation of a memory alloy [30–32].
The Gibbs free energy of the end states are defined when enthalpy and entropy
at constant temperature and constant pressure are given. The time-dependent
kinetic behavior can be represented by the temperature–time–transformation
fraction (TTT) curve. However, in electronic reliability problems, it is the
SVNY339-Tu March 2, 2007 22:0

Introduction 27

Fig. 1.16. A set of SEM images of electromigration failure caused by the dissolution
of a thick Cu UBM of 14 μm due to current crowding at the cathode of a flip chip
solder joint. The nominal current density was about 2 × 104 A/cm2 and the testing
temperature was 100◦ C. The dissolution of Cu UBM and Cu conducting line at the
upper left corner of the contact increases with time as shown from panel (a) to panel
(c). Panel (d) shows a void formation in the Cu line. (Courtesy of Prof. C. R. Kao,
National Taiwan University)

effect of external force in electromigration, for example, where the electrical

potential is not constant, that leads to phase change, in turn an open in a
circuit due to void formation at the cathode end or a short by extrusion at
the anode end. Furthermore, in thermomigration, the phase change is due
to a temperature gradient, so the temperature is not constant and no equi-
librium state can be defined except a steady state. Stress-migration-induced
void formation is a creep phenomenon under a stress or pressure gradient, so
the pressure is not constant. Tin whisker growth at room temperature is a
creep phenomenon too. Therefore, these microstructure failures or instability
problems are driven by an external force; they do not have a uniform bound-
ary condition such as constant temperature and constant pressure. They are
irreversible processes in nonequilibrium thermodynamics. However, we should
ask what is new in these irreversible processes? We mention three interesting
features below.
SVNY339-Tu March 2, 2007 22:0

28 Chapter 1

The first is the effect of interfaces where flux divergence occurs and the
supersaturation of vacancies will lead to void nucleation and growth. The
second is current crowding in interconnects and in flip chip solder joints (to be
discussed in Chapters 8 and 9) so that even the driving force is not constant.
The third is the eutectic system which has a two-phase microstructure so
there are two fluxes interacting with each other (to be discussed in Chapters
9 and 12).
However, there are phase changes in solder joint technology without an ex-
ternal force except temperature change as in wetting reaction and solid-state
aging of solder joints. Therefore, the following chapters have been divided into
two parts; Part I discusses Cu-Sn reactions and Part II discusses electromi-
gration and thermomigration.

1.5 Future Trends in Electronic Packaging

There are three trends in microelectronic technology that deserve our con-
sideration; the trend of miniaturization, the trend of packaging integration
evolution, and the trend of more serious chip–packaging interaction from the
integration of Cu/ultralow k in interconnect technology. Another possible fu-
ture trend is the use of fluxless and solderless joints.

1.5.1 The Trend of Miniaturization

The trend of miniaturization of electronic devices may reach nanoscale dimen-
sion in the future. Today, the feature size of a field-effect transistor (FET) is
already in the nanometer region, for example, the gate width is below 100
nm. Since ultrahigh density of Si memory technology is practical, the hard
disk memory used in portable devices based on laser-drilling of holes on thin
film disks will be replaced by flash memory sticks based on FET Si technology
because of small size, no moving parts, and impact resistance. Therefore, it is
possible to produce hand-held-size computers. To package nanoscale devices,
it is likely that the dimensions of packaging structures will be scaled down by
about two orders of magnitude, e.g., solder bumps 1 μm in diameter. While
there is no challenge in producing such small solder bumps by lithographic
technology and deposition, the standard solder reaction in reflow and solid-
state aging, for example, at 150◦ C for 1000 hrs, will convert the entire bump
into IMC. In other words, the entire joint becomes IMC, hence the physical
properties of Cu-Sn IMC will be even more important in the future packaging
technology.
Even if we consider the shrinking of the diameter of a solder bump by a
factor of 4, from 100 μm to 25 μm, the volume of the bump will be reduced by
a factor of 64. Relatively speaking, the volume fraction of IMC in the smaller
solder bump will increase 64 times. It will change the physical properties of
the bump such as its mechanical strength drastically, and it will make the
SVNY339-Tu March 2, 2007 22:0

Introduction 29

Cu-Sn reaction a much more serious reliability problem. This is because we

are as yet unable to reduce the reaction temperature and reaction time in
processing smaller solder joints.
We have emphasized the reliability issues of flip chip solder joints when
they are acted upon by chemical, mechanical, or electrical force individually.
Due to the trend of miniaturization, these forces will be expected to combine
together to make the reliability problems even more problematic.

1.5.2 The Trend of Packaging Integration Evolution—SIP, SOP,

and SOC
In packaging integration evolution, dramatic change is expected in portable
consumer electronic products. The design of packaging and the effective use
of limited space in hand-held devices will be the most challenging problems
in packaging technology in the near future. There are system-in-packaging
(SIP), system-on-packaging (SOP), system-on-chip (SOC), and other varia-
tions. The packaging technology is no longer about connecting a single chip
to a single substrate or joining several of the same kind of chips and other
discrete components to a module.
SIP is the advanced multifunctional packaging. It is to integrate all kinds
of components in and on a single package, mostly a laminated substrate. In
other words, it is to take several differently packaged devices and components
and assemble them on a substrate board, and the passive components are
embedded in the substrate or surface mounted. It is a heterogeneous integra-
tion. Different components (logic IC, memory IC, passive components, etc.),
different semiconductor chips (Si, SiGe, GaAs, GaN, etc.), and different tech-
nologies (electronics, optoelectronics, MEMS, biosensor, etc.) will be merged
together on a high-density interconnected substrate. To do so, chip-size pack-
aging becomes essential, in which the size of the packaging substrate is close
to the size of the chip, so that each chip or device does not occupy extra space
for cost-efficient assembly on the substrate. How to use the substrate space
effectively in a handheld cellular phone, for example, is critical as the function
of the device increases. The trend in chip-size packaging will lead to finer and
finer pitch of solder bumps in flip chip technology. Besides chip-size packaging,
chip on flexible substrate and stacking of various chips in three dimensions
will be emphasized. In three-dimensional stacking of chips, a combination of
wire bonding and solder bumping, or a multilevel soldering process or a high–
low combination of two kinds of Pb-free solders will be needed. Eventually, a
stacking of Si chips interconnected by via holes in the chips will be the most
effective way of packaging. SIP will be the dominant technology for electronic
consumer products in the near future.
SOP is to integrate a system of various devices in the design stage on a
module or board, so the integration will be more efficient than SIP which will
just take individual components and interconnect them together on a sub-
strate. The design in SOP is driven by system needs for specific applications,
SVNY339-Tu March 2, 2007 22:0

30 Chapter 1

for example, microelectronic and optoelectronic devices can be integrated for

mixed signal applications. The difference between SOP and SIP is small and
will diminish in the future.
SOC is to design and integrate a system on a single chip since chip manu-
facturing is becoming cheaper and cheaper. In other words, it is to integrate
active devices such as memory and logic devices and some packaging com-
ponents on a Si chip, so that the chip technology and chip-size-packaging
technology may get closer and closer and merge together. It is more of a
homogeneous integration. SIP starts from a concept based on packaging tech-
nology using a substrate to join multiple chips and components on it, and
SOC starts from Si chip technology and uses a Si chip to build multiple func-
tions and packaging components on the chip. The trend of miniaturization
will drive SIP toward SOC eventually, as in the example of replacing hard
disk memory by flash memory.

1.5.3 Chip–Packaging Interaction

To reduce RC delay in a multilayered interconnect structure on a chip,
ultralow-k materials with a value of dielectric constant approaching 2 are be-
ing developed and integrated with Cu conductors. The mechanical properties
of ultralow-k materials are of concern owing to the thermal stress between Cu
and ultralow-k due to their difference in thermal expansion coefficient, but a
more serious thermal stress comes from chip–packaging interaction. The latter
will be a major reliability issue in the future. In flip chip technology used in
high-end devices such as a server, the Cu/ultralow-k multilayered structure
will be joined to a packaging substrate by an area array of solder joints. In
joining a chip to its packaging, there are under-bump metallization and the
multilayered structure of Cu/ultralow-k on the chip side and there are bond
pads on the substrate side, then in between the chip and the substrate there
are solder bumps. In chip-join and in operation, thermal stress develops be-
tween the chip and the substrate, as discussed in Section 1.4.3. The thermal
stress will affect not only the mechanical integrity of the solder bumps but
also the Cu/ultralow-k multilayered structure. It has led to the well-known
low cycle fatigue failure of solder joints, yet the effect of thermal stress on the
Cu/ultralow-k structure due to chip–packaging interaction is unclear.
The microelectronics industry has been able to live with the fatigue prob-
lem, especially with the help of underfill of epoxy between the chip and the
substrate, otherwise our computers would not work today. This is because in
the Al/SiO2 or Cu/low-k technology, the mechanical strength of the SiO2 and
low k materials (which are carbon-doped SiO2 with a value of k slightly below
3) is quite strong, so the thermal stress affects mostly the solder bumps and
its interfaces, but not the interlayer dielectrics. Yet this may not be true any
more with the ultralow-k materials; at the least its weaker mechanical proper-
ties are of concern. Ultralow-k materials tend to be porous or a mixture with a
certain amount of polymer, so they crack easily under stress. In the extreme,
SVNY339-Tu March 2, 2007 22:0

Introduction 31

if we consider an array of flip chip solder bumps placed on a soft ultralow-k

material, any displacement of the solder bumps will cause a strain in the soft
ultralow-k material and also the Cu lines embedded in it. This will be a very
serious reliability problem. To avoid the thermal stress from chip–packaging
interaction, it seems that a Si substrate is better for Si chips since there will
be no thermal stress between a Si chip and a Si substrate.

1.5.4 Solderless Joints

In principle, we can have solderless joints. An example is the anisotropic con-
ducting polymer, except that both its conduction and adhesion are not as yet
good enough for high-end packaging technology. There are three basic criteria
in solder joining reaction: it has to be low temperature, its electrical con-
duction should be metallic, and its interfacial adhesion or bonding strength
should be as strong as atomic bonds in metals. As long as we can find an
interfacial process that meets these criteria, we could have a replacement for
solder joints, [33] provided that it can be applied over the active device area
as in area array solder joints. Otherwise, we can use wire-bonding.
We may consider the principle of wafer-bonding technique to develop sol-
derless joints. The bonding of two Si wafers can occur at room temperature,
without applying flux, and without applying heat and pressure. Flux has
served as the poor-man’s vacuum to overcome the problem of surface oxide in
interfacial bonding. Interfacial bonding without flux in ultrahigh vacuum and
without interdiffusion and compound formation at room temperature can be
achieved, provided that the two surfaces to be joined are atomically flat and
clean as in wafer bonding. Without ultrahigh vacuum and without flux, we
may consider using pure Au bumps or Cu bumps with a very thick Au coat-
ing. To achieve a direct Au-to-Au bonding, we should polish two Au bumps
as flat as possible and bond them directly without chemical reaction at room
temperature. To improve the hardness of Au for polishing, we could use very
Au-rich alloys such as 18K gold or even Pt. However, the challenge is not to
bond just one pair of Au bumps, but an area array of them between a chip
and a substrate.

References
1. K. Puttlitz and P. Totta, “Area Array Technology Handbook for Micro-
electronic Packaging,” Kluwer Academic, Norwell, MA (2001).
2. J. H. Lau, “Flip Chip Technologies,” McGraw–Hill, New York (1996).
3. I. Amato, “Tin whiskers: The next Y2K problem?” Fortune Magazine,
Vol. 151, Issue 1, p. 27 (2005).
4. B. Spiegel, “Threat of tin whiskers haunts rush to lead-free,” Electronic
News, 03/17/2005.
5. http://www.nemi.org/projects/ese/tin whisker.html
SVNY339-Tu March 2, 2007 22:0

32 Chapter 1

6. Y. Kariya, C. Gagg, and W. J. Plumbridge, “Tin pest in lead-free solders,”

Sold. Surf. Mount Technol., 13, 39–40 (2001).
7. K. N. Tu and D. Turnbull, “Direct observation of twinning in tin lamellae,”
Acta Metall., 18, 915 (1970).
8. C. M. Miller, I. E. Anderson, and J. F. Smith, “A viable Sn-Pb solder
substitute: Sn-Ag-Cu,” J. Electron. Mater. 23, 595–601 (1994).
9. M. E. Loomans and M. E. Fine, “Tin-silver-copper eutectic temperature
and composition,” Metall. Mater. Trans., 31A, 1155–1162 (2000).
10. K.-W. Moon, W. J. Boettinger, U. R. Kattner, F. S. Biancaniello, and
C. A. Handwerker, “Experimental and thermodynamic assessment of Sn-
Ag-Cu solder alloys,” J. Electron. Mater., 29, 1122–1136 (2000).
11. Z. Kovac and K.N. Tu, “Immersion tin: its chemistry, metallurgy and
application in electronic packaging technology,” IBM J. Res. Dev. 28,
726–734 (1984).
12. “1999 International Roadmap for Semiconductor Technology,” Semi-
conductor Industry Association, San Jose, CA (1999). See Website
http://public.itrs.net/
13. L. F. Miller, “Controlled collapse reflow chip joining,” IBM J. Res. Dev.,
13, 239–250 (1969).
14. P. A. Totta and R. P. Sopher, “SLT device metallurgy and its monolithic
extensions,” IBM J. Res. Dev., 13, 226–238 (1969).
15. B. S. Berry and I. Ames, “Studies of SLT chip terminal metallurgy,” IBM
J. Res. Dev., 13, 286–296 (1969).
16. A. A. Liu, H. K. Kim, K. N. Tu, and P. A. Totta, “Spalling of Cu6 Sn5
spheroids in the soldering reaction of eutectic SnPb on Cr/Cu/Au thin
films,” J. Appl. Phys., 80, 2774–2780 (1996).
17. H. K. Kim, K. N. Tu, and P. A. Totta, “Ripening-assisted asymmetric
spalling of Cu-Sn compound spheroids in solder joints on Si wafers,” Appl.
Phys. Lett., 68, 2204–2206 (1996).
18. C. Y. Liu, C. Chih, A. K. Mal, and K. N. Tu, “Direct correlation between
mechanical failure and metallurgical reaction in flip chip solder joints,” J.
Appl. Phys., 85, 3882–3886 (1999).
19. J. W. Jang, C. Y. Liu, P. G. Kim, K. N. Tu, A. K. Mal, and D. R. Frear,
“Interfacial morphology and shear deformation of flip chip solder joints,”
J. Mater. Res., 15, 1679–1687 (2000).
20. M. Date, T. Shoji, M. Fujiyoshi, K. Sato, and K. N. Tu, “Ductile-to-brittle
transition in Sn-Zn solder joints measured by impact test,” Scr. Mater.
51, 641–645 (2004).
21. S. Brandenburg and S. Yeh, “Electromigration studies of flip chip bump
solder joints,” in Proc. Surface Mount International Conference and Ex-
position, SMTA, Edina, MN, 1998, p. 337–344.
22. S.-W. Chen, C.-M. Chen, and W.-C. Liu, “Electric current effects upon
the Sn/Cu and Sn/Ni interfacial reactions,” J. Electron. Mater., 27, 1193–
1197 (1998).
SVNY339-Tu March 2, 2007 22:0

Introduction 33

23. C. Y. Liu, C. Chih, C. N. Liao, and K. N. Tu, “Microstructure–

electromigration correlation in a thin stripe of eutectic SnPb solder
stressed between Cu electrodes,” Appl. Phys. Lett., 75, 58–60 (1999).
24. T. Y. Lee, K. N. Tu, S. M. Kuo, and D. R. Frear, “Electromigration
of eutectic SnPb solder interconnects for flip chip technology,” J. Appl.
Phys., 89, 3189–3194 (2001).
25. E. C. C. Yeh, W. J. Choi, K. N. Tu, P. Elenius, and H. Balkan, “Current
crowding induced electromigration failure in flip chip technology,” Appl.
Phys. Lett., 80, 580–582 (2002).
26. A. T. Huang, A. M. Gusak, K. N. Tu, and Y.-S. Lai, “Thermomigration
in SnPb composite flip chip solder joints,” Appl. Phys. Lett., 88, 141911
(2006).
27. Y. C. Hu, Y. L. Lin, C. R. Kao, and K. N. Tu, “Electromigration failure
in flip chip solder joints due to rapid dissolution of Cu,” J. Mater. Res.,
18, 2544–2548 (2003).
28. Y. H. Lin, C. M. Tsai, Y. C. Hu, Y. L. Lin, and C. R. Kao, “Electromi-
gration induced failure in flip chip solder joints,” J. Electron. Mater., 34,
27–33 (2005).
29. H. Gan and K. N. Tu, “Polarity effect of electromigration on kinetics of
intermetallic compound formation in Pb-free solder v-groove samples,”
J. Appl. Phys., 97, 063514-1 to -10 (2005).
30. P. G. Shewmon, “Transformations in Metals,” Indo American Books,
Delhi (2006).
31. D. A. Porter and K. E. Easterling, “Phase Transformation in Metals and
Alloys,” Chapman & Hall, London (1992).
32. J. W. Christian, “The Theory of Transformations in Metals and Alloys;
Part 1 Equilibrium and General Kinetic Theory,” 2nd ed., Pergamon
Press, Oxford (1975).
33. Chin C. Lee and Ricky Chuang, “Fluxless non-eutectic joints Fabricated
using Au-In multilayer composites,” IEEE Trans. Components and Pack-
aging Technology, 26, 416–422 (2003).
SVNY339-Tu April 5, 2007 17:44

2
Copper–Tin Reactions in Bulk Samples

2.1 Introduction

Solder reaction is the wetting of a molten solder on a solid Cu surface. Typi-

cally, when a small drop of molten solder touches a large Cu surface, it spreads
and forms a cap on the Cu surface. The cap has a stable wetting angle, which
is defined usually by Young’s equation for a triple point. The wetting reaction
forms Cu-Sn intermetallic compounds at the interface between the molten sol-
der and Cu, and the interface achieves metallic bonding after cooling, hence
two pieces of Cu can be joined by the solder in a solder joint. While the wet-
ting reaction is important from the point of view of yield in manufacturing,
we must also consider solid-state reaction between solid solder and Cu from
100◦ C to 150◦ C from the point of view of reliability. This is because the work-
ing temperature of Si devices is around 100◦ C and an aging at 150◦ C for 1000
hr is a required test in reliability specification.
There are two intermetallic compounds of Cu and Sn; they are Cu6 Sn5
and Cu3 Sn, and they form during the wetting reaction as well as the solid-
state reactions between Sn and Cu according to the binary phase diagram of
Cu-Sn. A very interesting finding is that rates of the intermetallic compound
formation are very different between wetting reaction and solid-state reaction;
they differ by four orders of magnitude although the temperature of reaction
differs by only 10◦ C or so between the two reactions.
Since Pb does not react with Cu to form compounds, the intermetallic
compound formation in the reaction between SnPb and Cu is similar to that
between Sn and Cu. The effect of Pb on the Cu-Sn reaction can be exam-
ined by using the ternary phase diagrams of Sn-Pb-Cu at various tempera-
tures. Below, we shall first discuss experimental studies of wetting reactions
of eutectic SnPb on Cu foils followed by interpretation using their ternary
phase diagrams. The morphology of scallop-type Cu6 Sn5 formation in the
wetting reaction is unique that it is a supply-limited reaction, in contrast to
the well-known diffusion-limited reaction or interfacial-reaction-limited reac-
tion. Then solid-state reactions shall be discussed, in which a layer-type mor-
phology of Cu6 Sn5 occurs and its growth obeys the diffusion-limited reaction.
SVNY339-Tu April 5, 2007 17:44

38 Chapter 2

A comparison between the wetting reaction and solid-state reaction will be

given [1].
While our interest in solder reactions concerns Pb-free solders, intermetal-
lic compound formation between Pb-free solders and Cu is similar to the SnPb
solder because the most important Pb-free solders are eutectic SnAgCu, eu-
tectic SnAg, eutectic SnCu, or pure Sn [2]. Hence, Cu-Sn reaction is the key
when these solders are joined to Cu. Since SnPb solder has been used on
Cu for a long time and much study has been done on the ternary system of
Sn-Pb-Cu, it is helpful to review the solder reactions of SnPb on Cu as the
background references for Pb-free solders. Many useful references are listed in
Refs. 1 and 2. In this chapter, thermodynamics and kinetics of Cu-Sn reaction
will be emphasized.

2.2 Wetting Reaction of Eutectic SnPb on Cu Foils

The samples of eutectic SnPb on Cu were prepared by melting a tiny bead of
eutectic SnPb alloy (63Sn37Pb in wt%) on a Cu foil in mildly activated rosin
flux (RMA). Solder beads 0.5 mm in diameter can be prepared by cutting tiny
pieces of solder, about 2 mg each, from a roll of commercial solder wire and
dropping the pieces into a disk containing RMA flux kept at 200◦ C on a hot
plate. The pieces melt and form beads owing to balance of surface tension [3].
Removing the disk from the hot plate and cooling it to room temperature, the
solid solder beads can be saved and stored in flux. Copper foils of 1 cm × 1 cm
in area and 0.5 mm in thickness were polished and cleaned before one of the
foils was fully immersed in the heated flux at 200◦ C with ± 3◦ C control.
A solder bead was dropped on the Cu foil. The bead melted and spread out
to form a cap on the Cu surface (see Fig. 2.1). After various periods of time
from 0.5 min to 40 min on the hot plate, the solder cap sample was cooled
to room temperature. To study the intermetallic compound (IMC) formation
at the interface between the cap and Cu, the samples was cross-sectioned,
polished, lightly etched, and examined by scanning electron microscopy (SEM)
(see Fig. 2.2). The wetting angle, from a sideview of the sample, was measured
as a function of wetting time. This is plotted in Fig. 2.3. The wetting angle of
molten eutectic SnPb on Cu is stable at 11◦ . In Fig. 2.3, the wetting angles of
pure Sn on Cu and SnBi on Cu, measured similarly, are shown too, and they
are much larger than that of eutectic SnPb [4,5].
To observe the three-dimensional morphology of the interfacial IMC, a
deep and selective Pb-etching was carried out. The etching removed the Pb
and also the Sn above the IMC but did not etch the IMC. SEM of the IMC
observed at a tilt angle reveals the three-dimensional scallop-type morphology
of the IMC in Fig. 2.4(a) and (b), which show the same area of IMC scallops
before and after the selective etching, and we can identify the same contour of
the IMC in both figures. Figure 2.5 shows a set of SEM of the IMC scallops as a
function of wetting time at 200◦ C. They were taken at the same magnification.
SVNY339-Tu April 5, 2007 17:44

Copper–Tin Reactions in Bulk Samples 39

Fig. 2.1. A solder bead melted and spread out to form a cap on the Cu surface.

Cu6 Sn5

Cross-sectional view(2D) Cu3 Sn

Solder side Cu side Cu

Solder

Cu

Fig. 2.2. Solder cap on Cu sample was cross-sectioned, polished, lightly etched, and
examined by scanning electron microscopy (SEM) for the study of IMC formation at
the interface between the cap and Cu.
SVNY339-Tu April 5, 2007 17:44

40

35 pure Sn/Cu

SnBi/Cu

Wetting Angle (deg.)

30

25

20

15
SnPb/Cu
10
0 10 20 30 40
Reflow Time (min)

Fig. 2.3. The wetting angle, from a side view of the sample, plotted as a function of
reflow time. The wetting angle of molten eutectic SnPb on Cu is stable at 11◦ . The
wetting angles of pure Sn and SnBi are shown too, and they are much larger than that
of eutectic SnPb.

Fig. 2.4. The same area of IMC scallops (a) before and (b) after etching. We can
identify the same contour of the IMC in both figures.
SVNY339-Tu April 5, 2007 17:44

Copper–Tin Reactions in Bulk Samples 41

Fig. 2.5. (a–d) Set of SEM images of IMC scallops as a function of reflow time.

We observe that the average size of the scallops increased with time. Since
the scallops are very close to each other, the growth is a parasite reaction. It
must have occurred by having the bigger ones grow at the expense of their
smaller neighbors. In other words, it is a ripening reaction. But the ripening
is nonconservative since the total volume of the scallops increases with time,
accompanied by the Cu-Sn chemical reaction between the Cu from the foil
and the Sn from the molten solder in order to grow the IMC. Kinetic analysis
of the nonconservative ripening will be presented in Chapter 5.
Since the manufacturing of a device involves several reflows, the total pe-
riod of wetting reaction in an actual solder joint in devices may add up to a few
minutes. Therefore, the study of wetting reaction up to 40 min is academic.
In wetting reactions over a long period of time, the scallops were found to
elongate, i.e., their height is much larger than their diameter. Often in cross-
sectional views of the solder joint, a few long and hollow Cu6 Sn5 tubes have
been observed. Their formation is not due to the interfacial reaction, rather it
has been attributed to the precipitation of supersaturated Cu from the molten
solder upon solidification, especially if the surface of the solder bump solidifies
before the Cu-Sn interface. These tubes were nucleated from the bump surface
rather than from the bump interface.
A more detailed discussion of the morphology, kinetics of growth, and
crystallographic orientation relationship between the scallop and Cu will be
given in Chapter 5.
SVNY339-Tu April 5, 2007 17:44

42 Chapter 2

Fig. 2.6. SEM images of a halo.

A rather unique behavior of the wetting of eutectic SnPb on Cu is the

side-band or halo formation on the Cu surface surrounding the solder cap.
While the cap diameter is stable after a minute or so in wetting, the halo
grows and enlarges its diameter with time. Figure 2.6 shows an SEM image
of a halo. Figure 2.7 plots the growth rate of a halo at 200◦ C. While the cap

160

140

120

100

80

60

40

20

0
0 1 2 3 4 5 6 7

Fig. 2.7. Growth rate of halo at 200◦ C.

SVNY339-Tu April 5, 2007 17:44

Copper–Tin Reactions in Bulk Samples 43

diameter is stable in the initial period of halo formation, the halo may change
the equilibrium condition of the wetting tip of the cap. It has been observed
that after several minutes of halo growth, the solder cap becomes unstable
and spreads out quickly to wet the entire area of the halo.

2.2.1 Crystallographic Relationship between Cu6 Sn5

Scallop and Cu
Fig. 2.8 is a top-view image of Cu6 Sn5 scallops on a Cu substrate, and the
crystallographic orientation relationship between the scallops and Cu has
been determined. Samples were prepared by reacting small beads (∼0.5mg) of
55Sn45Pb (in wt. %) solder with a mirror-like polished copper foil in flux at
200◦ C with different reaction times from 30 sec to 8 min. The remaining solder
was etched away to expose the scallops. The size of each scallop is about 1 to
3 μm as shown in Fig. 2.8, and each of them is a single crystal of Cu6 Sn5 . The
grain sizes of Cu in the Cu foil were in mm scale. By scanning the micro x-ray
beam in synchrotron radiation over an area of about 100 μm by 100 μm of
the sample, the crystallographic information of several thousands of scallops
and the Cu below them is obtained. The synchrotron radiation micro x-ray
beam is capable of penetrating through the Cu6 Sn5 layer since it is thin. Laue
patterns from both IMC scallops and Cu substrate can be obtained at the
same time.
Actually the strongest Laue spots were from the Cu, because the grain
size of the Cu is much larger (mm scale) than the Cu6 Sn5 (1∼2 μm in size)
and so the beam penetrated much deeper into the Cu. Grain orientation of
Cu was analyzed first. After finishing the analysis of the Cu, Laue spots from
the Cu were removed and the Laue pattern from the Cu6 Sn5 was analyzed.

Fig. 2.8. SEM image of top-view of Cu6Sn5 scallops on a Cu substrate.

SVNY339-Tu April 5, 2007 17:44

44 Chapter 2

In the early stage of reaction, the spots from Cu3 Sn were not detected; either
it did not form or its thickness was too thin to produce Laue spots strong
enough to be detected.
The crystal structure of η-Cu6 Sn5 has been thought to be hexagonal, but
a recent study using electron diffraction showed that it is actually mono-
clinic (Space group = P21 /c, a = c = 9.83Å, b = 7.27 Å, β = 62.5◦ ) [6].
From the Laue patterns, 3-dimensional computer models of crystal structures
were simulated to determine the orientation relationships between the Cu and
the monocline Cu6 Sn5 . Six types of preferred orientation relationships were
found:

(010)Cu6 Sn5 //(001)Cu [1̄01]Cu6 Sn5 //[110]Cu (2.1)

(343)Cu6 Sn5 //(001)Cu [1̄01]Cu6 Sn5 //[110]Cu (2.2)
(3̄43̄)Cu6 Sn5 //(001)Cu [1̄01]Cu6 Sn5 //[110]Cu (2.3)
(101)Cu6 Sn5 //(001)Cu [1̄01]Cu6 Sn5 //[110]Cu (2.4)
(141)Cu6 Sn5 //(001)Cu [1̄01]Cu6 Sn5 //[110]Cu (2.5)
(1̄41̄)Cu6 Sn5 //(001)Cu [1̄01]Cu6 Sn5 //[110]Cu (2.6)

In every cases, the [1̄01] direction of Cu6 Sn5 is aligned parallel to the [110]
direction of Cu. Fig. 2.9 (a) is a map representing the angle between the
[1̄01] direction of Cu6 Sn5 and the [110] direction of Cu, after 4 minutes of
wetting reaction. Scanned area was 100 μm × 100 μm, with a step size of
2 μm. Most of the spots have the angle close to 0◦ . Fig. 2.9 (b) is a histogram
of orientation distribution corresponding to Fig 2.9 (a). The angle is close
to zero for the majority of data points, indicating a very strong orientation
correlation between Cu6 Sn5 and Cu. The orientation relationship (2.1) to (2.6)

Fig. 2.9. (a) Map representing the angle between the direction of Cu6 Sn5 and the
[110] direction of Cu, after 4 minutes of wetting reaction. Most of the spots have the
angle close to 0◦ . (b) Histogram of orientation distribution corresponding to Fig 2.9(a).
SVNY339-Tu April 5, 2007 17:44

Copper–Tin Reactions in Bulk Samples 45

Fig. 2.10. (a) Structure of Cu6 Sn5 projected along [−101] direction. Copper atoms
in the Cu6 Sn5 are represented by circles. (b) Crystal planes for the six orientation
relationships labeled with respect to the Cu atom hexagon

shown above can be classified into two groups. This is due to the strong
pseudo-hexagonal symmetry around the Cu atoms in Cu6 Sn5 . Fig 2.10 (a)
shows the structure of Cu6 Sn5 projected along the [1̄01] direction. Copper
atoms in the Cu6 Sn5 are represented by circles in hexagonal arrangement.
In Fig. 2.10 (b), the crystal planes for the six relationships are labeled with
respect to the Cu atom hexagon. Relationships involving planes (010), (343)
and (3̄43̄) belong to the one group, and relationships with planes (101), (141)
and (1̄41̄) belong to another group.
The reason why the [1̄01] direction of Cu6 Sn5 prefers to be parallel to the
[110] direction of Cu is because of low misfit. Along the [1̄01] direction of
monoclinic Cu6 Sn5 , Cu atoms are located with intervals of 2.5573 Å. Since
the lattice constant of FCC Cu is a = 3.6078 Å, the distance between two
atoms along the diagonal of (001) plane is
√ √
2 2
aCu = × 3.6078 = 2.5511 Å
2 2

Thus the misfit between Cu and Cu6 Sn5 is

|2.5511 − 2.5573|
f= = 0.00244 = 0.24%
2.5511

The strong relation between the orientations of Cu6 Sn5 and Cu suggests
that at the earliest stage of IMC formation, the Cu6 Sn5 forms prior to the
Cu3 Sn. The lattice structure of Cu3 Sn is orthorhombic Cu3 Ti type [6], and it
does not have any low index plane or direction which can have a small misfit
with Cu or Cu6 Sn5 . If Cu3 Sn were formed first, before Cu6 Sn5 , it would have
been impossible for Cu6 Sn5 to have such a strong orientation relationship with
Cu.
Since the low misfit directions between Cu6 Sn5 and Cu lie on (001) plane
of Cu, (001) oriented single crystal Cu was used as substrate to verify the
orientation relationship. Fig.2.11 is morphology of Cu6 Sn5 formed on (001)
SVNY339-Tu April 5, 2007 17:44

46 Chapter 2

Fig. 2.11. Morphology of Cu6 Sn5 formed on (001) single crystal Cu.

single crystal Cu. It shows a dramatic change in morphology of Cu6 Sn5 on

(001) Cu. The Cu6 Sn5 scallops were elongated along two perpendicular direc-
tions and showed a roof-top morphology. Electron back scattered diffraction
(EBSD) analysis was performed on these scallops and the single crystal Cu
substrate. From the EBSD analysis, it was confirmed that the elongation di-
rections correspond to the two 110 directions on the (001) surface of Cu,
which are the low-misfit directions for the Cu6 Sn5 .
Fig. 2.12 shows the influence of the reflow or reaction time on the orienta-
tion relationship between the scallops and the Cu substrate shown in Fig. 2.8.
The histograms of the angle between the [110] direction of Cu and the [001]
direction of Cu6 Sn5 are given. We note that if the Cu6 Sn5 has a strong orien-
tation relationship with the Cu, the majority of data will have a value close
to 0◦ . At 30 sec which was the earliest reaction time of our measurement,
the distribution shown in Fig. 2.12(a) is more random compared to longer
annealing times. The distribution at 1 min (Fig. 2.12(b)) shows a strong rela-
tion between the Cu6 Sn5 scallops and the Cu grain. The correlation becomes
very strong at 4 min (Fig. 2.12(c)), but weakens at 8 min (Fig. 2.12(d)). This
change in distribution can be explained from the nucleation and growth of
Cu6 Sn5 and Cu3 Sn. The Cu6 Sn5 scallops should nucleate first with a certain
degree of randomness, resulting in a distribution such as that in Fig. 2.12(a).
During the growth and ripening of Cu6 Sn5 scallops, scallops with bad orienta-
tion relationship will be consumed due to their larger interfacial energy with
Cu. Hence, the fraction of scallops with good orientation relationship with Cu
will increase. This explains why the distribution gradually moves toward 0◦ ,
as shown in Fig. 2.12(a) to 2.12(c). After a certain amount of reaction time,
the orientation relationship between Cu6 Sn5 and Cu will be affected by the
SVNY339-Tu April 5, 2007 17:44

Copper–Tin Reactions in Bulk Samples 47

(a) (b)
140
300
120
250
100
# spots
200
80
40 150
40 100
20 50
0 0
0 20 40 60 80 0 20 40 60 80
(c) Angle (degree) (d) Angle (degree)
600
800 500
600 400
# spots

300
400
200
200 100

0 0
0 20 40 60 80 0 20 40 60 80
Angle (degree) Angle (degree)

Fig. 2.12. Influence of reflow or reaction time on the orientation relationship is shown
by histograms of the angle between the [110] direction of Cu and the [001] direction
of Cu6 Sn5 . (a) 30 min, (b) 1 min, (c) 4 min, and (d) 8 min.

nucleation and growth of Cu3 Sn between them. Formation of the Cu3 Sn by

solid state reaction between Cu6 Sn5 and Cu may not be uniform. After a long
reflow time, some of the Cu3 Sn grains will become thick and incoherent to
both Cu6 Sn5 and Cu. The presence of incoherent grains of Cu3 Sn may rotate
the Cu6 Sn5 scallops in order to release the misfit strain energy, so the orien-
tation distribution of Cu6 Sn5 changes. The effect of reflow on the orientation
of elongated scallops on (001) single crystal Co is much less.

2.2.2 Rate of Consumption of Cu in Soldering Reaction

with Eutectic SnPb
The rate of consumption of Cu in soldering reactions has been a critical ques-
tion in electronic packaging technology. This is because the Cu thickness in
most UBM is rather limited, except when a Cu column bump is used. The
consumption of Cu in multiple reflows must be under control so that not all
of the Cu will be consumed. Using cross-sectioned and top-polished samples,
the total volume of Cu-Sn IMC formed between eutectic SnPb and Cu as a
function of temperature from 200◦ C to 240◦ C and time up to 10 min has
been determined [7]. The consumed thickness of Cu at the three tempera-
tures is plotted in Fig. 2.13 versus time. From the slope of the curves, the
SVNY339-Tu April 5, 2007 17:44

48 Chapter 2

1.6 60

1.4 50
240 °C
Consumed Thickness

dh/dt×103 (μm/sec)
1.2
220 °C 40
1.0

Δh (μm)
200 °C
0.8 30

0.6 20
0.4
10
0.2 Consumption Rate
0
0.0
0 100 200 300 400 500 600 700

Fig. 2.13. Consumed thickness of Cu at temperatures of 200, 220, and 240◦ C versus
time.

consumption rate of Cu, dh/dt in μm/sec, was calculated for each temper-
ature. The consumption rate of Cu is relatively high at the initial stage of
wetting and decreases with time. After 1 min, the consumed Cu is about 0.36,
0.47, and 0.69 μm at 200, 200, and 240◦ C, respectively.

2.3 Wetting Reaction of SnPb on Cu Foil as a

Function of Solder Composition
To study the wetting behavior of SnPb solders on Cu as a function of composi-
tion, seven SnPb alloys (in wt%) were prepared: pure Pb, 5Sn95Pb, 10Sn90Pb,
40Sn60Pb, 63Sn37Pb, 80Sn20Pb, and pure Sn. Their melting temperatures
are 327, 305, 299, 245, 183, 205, and 232◦ C, respectively [8, 9].
Beads of about 0.5 mm in diameter of these solder alloys were reflowed on
Cu foils of 1 cm × 1 cm in area and 0.5 mm in thickness immersed in flux of
mildly activated resin (RMA) in a shallow beaker heated on a hot plate to
10◦ C above the melting temperature of the solder alloy. After about 1 min
when the solder cap was stable on the Cu surface, it was cooled and cleaned
for wetting angle measurement. The side view of a set of solder caps as a
function of SnPb composition is shown in Fig. 2.14. The wetting angles are
plotted against composition in Fig. 2.15. Wetting angle has been defined by
the classic Young’s equation,

γsl = γvs + γlv cos θ, (2.7)

where γ is the interfacial tension and the subscripts l, v, and s indicate the
liquid flux, molten solder, and Cu substrate, respectively, as shown in Fig. 2.16.
For the wetting of SnPb alloys on Cu, it is a reactive spreading process
since intermetallic compound forms at the interface. According to previous
studies of interfacial tension of molten SnPb solder in flux as a function of Sn
SVNY339-Tu April 5, 2007 17:44

Copper–Tin Reactions in Bulk Samples 49

Fig. 2.14. Side view of a set of

solder caps as a function of SnPb
composition. (a) Pb, (b) 5Sn95Pb,
(c) 10Sn90Pb, (d) 40Sn60Pb, (e)
63Sn37Pb, (f) 80Sn20Pb, and (g)
Pure Sn.

content, as redrawn in Fig. 2.17, the tension curve shows no minimum. The
interfacial tension in the high-Pb region is constant and it increases with the
Sn content. If we use this result and Young’s equation to calculate the wetting
angle, we expect no minimum wetting angle, but a minimum was found as
shown in Fig. 2.15. The disagreement indicates that if we consider only the
balance of interfacial tensions, we cannot account for the measured wetting
angle change of SnPb alloys as a function of composition.
As seen in Fig. 2.15, the pure Pb has a very large wetting angle on Cu, yet
by adding a small amount of Sn, the wetting angle is much reduced. On the
other hand, the addition of a small amount of Sn into Pb does not change the
interfacial tension as shown in Fig. 2.17. The strong effect of a small amount
of Sn on decreasing the wetting angle comes from the Sn-Cu reaction to form
interfacial intermetallic compound on Cu. The chemical reaction can provide
an additional driving force to change the equilibrium condition of the wetting
tip. To include the effect of the reaction on the wetting, Yost and Romig have
SVNY339-Tu April 5, 2007 17:44

50 Chapter 2

40

35

30
Wetting angle,

25

20

15

10

0
0 10 20 30 40 50 60 70 80 90 100
Pb content, wt%

Fig. 2.15. Wetting angle plotted as a function of composition.

estimated that [10, 11]

1 dE
= σ + Γ(θ), (2.8)
2πr dr

where σ corresponds to the driving force of compound formation and Γ(θ) is

the driving force of imbalance of interfacial tensions. On the other hand, if
we consider compound formation or chemical activity alone, it is difficult to
explain why pure Sn has a larger wetting angle than some of the other SnPb
solders, because pure Sn is expected to have a higher activity than other
SnPb solders and has formed a thicker compound. The average thickness of
scallop-type intermetallic compound formation, after 1 min at 10◦ C above the
melting point, is plotted in Fig. 2.18. The average IMC formation of pure Sn

γγlvlv

Fig. 2.16. Wetting tip configuration and

Young’s equation, where γ is the interfa-
θ cial tension and the subscripts l, v, and s
γvs
indicate the liquid flux, molten solder, and
γsl
Cu substrate, respectively.
SVNY339-Tu April 5, 2007 17:44

Copper–Tin Reactions in Bulk Samples 51

510

interfacial tension (mjm-2) 490

470

450

430

410

390

370

350
0 20 40 60 80 100
Sn content (weight percent)

Fig. 2.17. Interfacial tension of molten SnPb solder in flux plotted as a function of
Sn content. The curve shows no minimum.

4.5

4
Total compound thickness (μm)

3.5

2.5

1.5

0.5

0
0 10 20 30 40 50 60 70 80 90 100
Pb content, wt%

Fig. 2.18. Average thickness of scallop-type intermetallic compound formation, after 1

min at 10◦ C above the melting point, plotted as a function of solder alloy composition.
SVNY339-Tu April 5, 2007 17:44

52 Chapter 2

is the largest among the SnPb solders. But the surface tension of pure Sn
is also the largest as shown in Fig. 2.17. Combining the surface tension and
IMC formation, we can explain the behavior of a minimum wetting angle as
a function of alloy composition as shown in Fig. 2.15.
However, it is difficult to analyze the spreading mechanism and determine
the wetting angle at triple points without a thorough understanding of the
driving force and kinetics of wetting. One important question is how much
compound has formed during the wetting period, from the time the molten
solder bead touches the Cu surface to the time it forms the cap. It is difficult
to determine the interfacial reaction during the wetting period since it is a
very short period; the spreading rate of a molten solder on a Cu surface is
extremely high.

2.4 Wetting Reaction of Pure Sn on Cu Foils

Since the Cu-Sn reaction is a central theme of this book, we shall discuss
the reaction between molten Sn and Cu here. It is also crucial to the solder
reaction of Pb-free solders because the latter are high-Sn solders. The reac-
tions between thin-film Sn and thin-film Cu will be discussed in Chapter 3.
Both Cu6 Sn5 and Cu3 Sn form in the reaction between molten Sn and Cu,
and Cu6 Sn5 has the scallop-type morphology and Cu3 Sn has the layer-type
morphology. This is similar to the molten reaction between eutectic SnPb
and Cu. However, the scallop morphology of Cu6 Sn5 between pure Sn and
Cu is quite different from that formed between eutectic SnPb and Cu [4,
12]. Figure 2.19(a) is an SEM image of the rounded scallops formed between
40Sn60Pb on Cu. The image was taken after the unreacted SnPb was etched
away to expose the Cu6 Sn5 scallops. Figure 2.19(b) is an SEM image of the
faceted scallops formed between Sn and Cu. While both kinds of scallops are
Cu6 Sn5 , their morphology is very different. A rounded morphology occurs be-
cause the interfacial energy between molten SnPb and Cu6 Sn5 is isotropic,
and the faceted morphology occurs because the interfacial energy between
molten Sn and Cu6 Sn5 is highly anisotropic. The addition of Pb in the solder
has affected the interfacial energy or the bonding energy. In the rounded scal-
lops, the width-to-height ratio as measured from cross-sectional views can be
approximated by hemispheres, but the faceted scallops have a larger width-
to-height ratio.
On the other hand, the growth kinetics of these two kinds of scallops is
quite similar. Both obey t1/3 growth law. Figure 2.20 is a plot of the scallop
growth versus annealing time. In the plot, the lateral width (from plan view
images) and height (from cross-sectional views) were evaluated separately. Af-
ter 4 min of reflow at 245◦ C, the layer-type Cu3 Sn has a thickness of about
300 nm and the scallops of Cu6 Sn5 have a height of about 1 to 2 μm. A fun-
damental parameter of the growth of scallops is the narrow channel between
two of them. The channel will be analyzed in Chapter 5.
SVNY339-Tu April 5, 2007 17:44

Copper–Tin Reactions in Bulk Samples 53

Fig. 2.19. (a) SEM image of

the rounded scallops formed
between 40Sn60Pb on Cu. The
image was taken after the un-
reacted SnPb was etched away
to expose the Cu6 Sn5 scallops.
(b) SEM image of the faceted
scallops formed between Sn
and Cu.

12
11
Mean Grain Radius (μm)

10
9 240 °C
8 220 °C
7 200 °C
6
5
4
3
2
1
0
0 2 4 6 8 10 12 14
Reflow Time (sec)1/3

Fig. 2.20. Plot of scallop growth versus annealing time.

SVNY339-Tu April 5, 2007 17:44

54 Chapter 2

2.5 Ternary Phase Diagram of Sn-Pb-Cu

Since a solder–metal reaction involves initially a solid–liquid reaction with a
molten solder, followed by solidification and then a solid-state aging during
reliability tests and device operation, it is necessary to have the phase diagram
available in a wide temperature range, below and above the melting point of
the solder. With computers, the CALPHAD (Calculation of Phase Diagram)
technique has been developed to investigate the phase equilibrium of mul-
ticomponent systems [13]. In this technique, thermodynamic properties of a
system are analyzed by using thermodynamic models for the Gibbs free en-
ergy of individual phases. The Gibbs free energy of the SnPbCu liquid phase
in this formalism is
◦ liq ◦ liq ◦ liq
Glid
m = xCu GCu + xPb GPb + xSn GSn
+ RT (xCu ln xCu + xPb ln xPb + xSn ln xSn ) + E Gliq
m, (2.9)

where xi is the mole fraction of element i, ◦ Gliq

i is the mole Gibbs energy of
element i in the liquid state, and E Gliq
m is the excess Gibbs energy treated as
follows:
E liq liq liq liq
Gliq
m = xCu xPb LCu,Pb + xCu xSn LCu,Sn + xPb xSn LPb,Sn + xCu xPb xSn LCu,Pb,Sn ,
(2.10)

where Lliq
i,j are the binary interaction parameters of the i–j system, which
can be composition dependent and temperature dependent. The parameter
Lliq
Cu,Pb,Sn represents the ternary interaction, which is also composition depen-
dent according to the expression

Lliq 0 liq 1 liq 2 liq

Cu,Pb,Sn = xCu LCu,Pb,Sn + xPb LCu,Pb,Sn + xSn LCu,Pb,Sn . (2.11)

The coefficients n Lliq

Cu,Pb,Sn can also be temperature dependent.
Once the Gibbs free energy functions of all the phases in a system have
been obtained, in principle any kind of phase diagrams and thermodynamic
properties of interest may be calculated. When the experimental data needed
to determine the ternary interaction parameters are not available, this tech-
nique enables us to extrapolate the thermodynamic description of a mul-
ticomponent system from those of the subsystems that have already been
optimized. From Eq. (2.9) we see that when the content of any of the al-
loying elements is on the order of a few percent, the technique of ther-
modynamic extrapolation is a good approximation for a multicomponent
system.
Complete thermodynamic descriptions of the binary systems of Cu-Sn,
Cu-Pb, and Sn-Pb have been assessed in the literature [14–16] By extrapo-
lating them, i.e., by setting the ternary interaction parameter in Eq. (2.4) to
zero, the ternary SnPbCu phase diagrams are calculated as shown in Fig. 2.21.
SVNY339-Tu April 5, 2007 17:44

Copper–Tin Reactions in Bulk Samples 55

Fig. 2.21. Ternary SnPbCu phase diagrams calculated for (a) 170, (b) 200 and (c)
350◦ C. (d) Enlarged SnPb side of the SnPbCu phase diagram at 200◦ C (Courtesy of
Dr. Kejun Zeng, TI).

In the diagrams, L, η, and ε represent molten SnPb, Cu6 Sn5 , and Cu3 Sn,
respectively. To verify the diagrams, we compare the calculated eutectic equi-
librium temperature and composition of the SnPb binary eutectic at 181◦ C,
37.8 Pb, 62.12 Sn, and 0.08 Cu with the experimental data of Marcotte and
Schroeder [16] at 182◦ C, 38.1 Pb, 61.72 Sn, and 0.18 Cu.
An important assumption for interfacial reactions in bulk diffusion couples
is the condition of local equilibrium at interphase interfaces. It assumes that
every two neighboring phases (planar layers) are in equilibrium, which is rep-
resented by either a two-phase equilibrium region in a binary phase diagram
or by a tie-line in an isothermal section of a ternary phase diagram at the
temperature of interest. Since the assumption implies that the composition
of the phases coexisting at the interface are those given by the equilibrium
tie-line on the phase diagram, the effects of morphology and kinetics such as
the interfacial curvature, capillary tension, precipitate size, metastable phase
formation, stress gradient, and all external forces are totally neglected. Inves-
tigation of the theoretical nature of interfacial equilibrium has concluded that
whereas the effect of interfacial curvature on equilibrium may be small, the
SVNY339-Tu April 5, 2007 17:44

56 Chapter 2

effect of growth kinetics on interfacial equilibrium can be significant. As for

strain, in the early stage of precipitate growth, the precipitates may possess
a certain degree of coherency with the matrix, so the lattice strain must be
included in the description of equilibrium at the phase interface. However,
in the size scale of IMC growth in solder joints, the interfaces between
molten solder and IMC scallops are incoherent and the interfacial equilibrium
assumption tends to be valid. Currently, all data reported in the literature on
solder reaction concerning the growth of IMC several tenths of a micrometer
thick pertain to well-evolved microstructures. Thus, the assumption of local
equilibrium can be applied to the molten solder/metal reactions, and the
ternary phase diagrams can indeed provide a useful guide to the reactions.
For example, the equilibrium phase diagram may enable us to predict whether
Cu6 Sn5 or Cu3 Sn forms first in wetting reactions, which has been discussed in
Section 2.2.1. However, we must combine the phase diagrams together with
morphological and kinetic information in order to analyze the reactions at
lower temperatures, e.g., thin film Cu-Sn reaction at room temperature to be
considered in Chapter 3. In that case, the kinetic effect is more important.

2.5.1 Ternary SnPbCu Phase Diagrams at 200 and 170◦ C

The ternary phase diagrams of SnPbCu at 200 and 170◦ C are shown in
Fig. 2.21(b) and (a), respectively. The latter is for solid-state aging and the
former is for wetting reaction between the eutectic SnPb and Cu. In the dia-
grams, L, η, and ε represent molten SnPb, Cu6 Sn5 , and Cu3 Sn, respectively.
The eutectic composition is represented by the dot on the base connecting Sn
and Pb. If we draw a line to connect the dot and the Cu apex, it is a tie line.
The tie line cuts across several two-phase boundaries and three-phase zones in
the phase diagram, and these are the phases which can form in the reactions.
For the wetting reaction, we start from the eutectic dot in Fig. 2.21(b). The
molten eutectic SnPb dissolves a very small amount of Cu before η formation
[13]. Then the formation of η at the interface between the molten solder
and Cu affects the solubility of Cu in the molten solder, because the molten
solder is in contact with the η. However, the solubility may be modified by
the morphology of the η phase, i.e., whether it is a layer-type or a scallop-
type. The latter increases the solubility due to the curvature effect, but this
information is unavailable from the phase diagram. The η formation depletes
Sn (enriches Pb) in the boundary layer of molten solder next to the η. The
depletion will stop at about 45 wt% Pb before the solid phase of α-Pb(Sn)
forms [see Fig. 2.21(a)]. Since the η is thermodynamically unstable with Cu,
the ε phase tends to form between them. Experimentally, both η and ε have
been found. The latter is a very thin layer and it may have formed during
cooling, but the former is a much thicker scallop-type layer.
For the solid-state aging, we turn to Fig. 2.21(a). Again we start from the
eutectic dot. A very small amount of dissolution of Cu into the solid solder
SVNY339-Tu April 5, 2007 17:44

Copper–Tin Reactions in Bulk Samples 57

occurs first and is followed by the formation of η. The solubility of Sn in Cu

also will be modified by the formation and morphology of η. The depletion
of Sn can lead to a very high concentration of Pb in the solder next to the η,
up to about 85 wt% Pb. We note that it is not a solid solution of 85Pb15Sn
(wt%), but a high-Pb eutectic phase. Again because η is unstable with Cu,
the ε phase will form between them. Experimentally, both η and ε have been
found.
From the point of view of IMC formation, both the wetting reaction at
200◦ C and the solid-state aging at 170◦ C should have the same IMC products
as indicated by the phase diagrams. Yet no information about the morphology
of the IMC and their rate of formation is available from the phase diagrams.
Besides, the only difference in these two phase diagrams is that the one at
200◦ C, Fig. 2.21(b), has the triangular zone of liquid solder and η. The sig-
nificance of this zone on solder reaction is that it limits the amount of Sn to
be depleted from the molten solder to form IMC. What effect this zone has
on the wetting reaction as distinct from the solid-state aging is not obvious if
we only examine the phase diagrams. We shall show later that the morphol-
ogy and kinetics of reaction are actually very different between the wetting
reaction and solid-state aging.

2.5.2 5Sn95Pb/Cu Reaction and Ternary SnPbCu Phase

Diagrams at 350◦ C
At 350◦ C, Cu6 Sn5 cannot form between Cu and the high Pb solder because the
connection line between Cu and the solder crosses the liquidus line (Liq + ε)
[see Fig. 2.21(c)]. Only ε-Cu3 Sn can form at 350◦ C at the solder/Cu interface.
However, according to Fig. 2.21(c), η may form at 350◦ C if the solder is very
rich in Sn.
The availability or mass supply of the components that are involved in the
interfacial reaction plays an important role in phase evolution after the first
phase formation. In the SnPb/Cu system, if the solder volume is very small
compared to that of Cu, e.g., the supply of Sn is limited or the supply of
Cu is unlimited as in a Cu column bump, the Cu6 Sn5 layer that formed first
will transform into Cu3 Sn. The transformation can lead to the formation of a
large number of Kirkendall voids, which will be discussed in Section 2.6.1 and
Chapter 9. On the other hand, if the Cu is thin as in a solder-Cu thin-film
reaction, or the Cu supply is cut off, for instance by a crack between Cu and
Cu3 Sn, the Cu3 Sn will be converted back to Cu6 Sn5 .

2.6 Solid-State Reaction of Eutectic SnPb on Cu Foils

The wetting reaction of molten eutectic SnPb and Cu at temperatures above
200◦ C will be compared to the solid-state reaction of the same system below
SVNY339-Tu April 5, 2007 17:44

58 Chapter 2

170◦ C [15–17]. Since several reflows will be needed in package manufacturing,

solder is in molten state for several minutes. Therefore, the wetting reaction
at 200◦ C for periods from 0.5 to 10 min will be studied. Solid-state reaction
at 150◦ C for 1000 hr is a required reliability test. Thus, the solid-state re-
action between eutectic SnPb and Cu in the temperature range from 120◦ C
to 170◦ C up to 1000 hr will be studied. Another reliability test is thermal
cycling between −40◦ C and 125◦ C up to a few thousand cycles. We note that
while the temperature difference between wetting reaction and solid-state re-
action is small, perhaps only 30◦ C, the reaction time differs by four orders
of magnitude. Nevertheless, the very large difference in morphology and ki-
netics of intermetallic compound formation found between these two kinds
of reactions will be emphasized. The rate of wetting reaction is four orders
of magnitude faster than that in solid-state aging, although the temperature
difference between them is only about 30◦ C. The Cu6 Sn5 formed in the wet-
ting reaction has scallop-type morphology, but it becomes layer-type in the
solid-state aging.
In order to understand why the very large differences exist, we recall the
thermodynamic calculation of the ternary phase diagrams of SnPbCu at 170◦ C
and 200◦ C in Section 2.5.1. We shall conclude that the thermodynamic phase
diagram cannot explain the difference, it is the morphology that affects the
kinetics strongly. The wetting reaction is a high rate reaction because of its
unique morphology. The reaction is controlled by the rate of free energy change
rather than the free energy change.

2.6.1 Formation of Cu3 Sn and Kirkendall Voids

In solid-state aging, it grows a thicker layer of Cu3 Sn. The formation of Cu3 Sn
is accompanied by the formation of a large number of Kirkendall void in the
layer and especially in the interface between Cu3 Sn and Cu. Figure 2.22 is
a focused ion beam image of the cross section of eutectic SnPb solder on a
Cu foil after aging at 150◦ C for 3 days. Many Kirkendall voids are found in
the Cu3 Sn layer. Similar void formation has also been observed in aging of
eutectic Pb-free solder on Cu foils. This is because Cu is the dominant diffusing
species in the reaction, as identified by a marker motion experiment, which
will be covered in Section 3.2.3. The competition in growth between Cu6 Sn5
and Cu3 Sn tends to favor the latter when the ratio of Cu to Sn is large in
the sample, for example, in the case of a thin eutectic SnPb solder on a thick
Cu column bump, which will be covered in Section 9.6.1. The transformation
of 1 molecule of Cu6 Sn5 into 2 molecules of Cu3 Sn will leave behind 3 Sn
atoms, which will attract 9 atoms of Cu to form 3 more molecules of Cu3 Sn.
The vacancy flux needed to transport the Cu atoms will accumulate at the
Cu/Cu3 Sn interface to form Kirkendall voids. Voids are undesirable in device
applications, therefore it is of reliability interest to limit the growth of Cu3 Sn.
The growth of Cu3 Sn is not only controlled by time, temperature, impurities,
SVNY339-Tu April 5, 2007 17:44

Copper–Tin Reactions in Bulk Samples 59

Fig. 2.22. Focused ion beam image of cross section of eutectic SnPb solder on a Cu
foil after aging at 150◦ C for 3 days. Many Kirkendall voids are found in the Cu3 Sn
layer. (Courtesy of Dr. Kejun Zeng, Texas Instruments.)

and the ratio of Cu/Sn in the sample, but also by external forces such as
electromigration.

2.7 Comparison between Wetting and Solid-State

Reactions
To compare wetting and solid-state reactions, the samples were prepared by
reflowing eutectic SnPb solder paste on a thick Cu under-bump metallization
(UBM) that was electroplated on a Ti/W base [20] The size of the solder
bump was 125 μm in diameter. The electroplated Cu UBM had a thickness
of 20 μm. The samples were reflowed twice before the aging at 125, 150,
and 170◦ C for 500, 1000, and 1500 hr in an air ambient. The total reaction
product after two reflows is equivalent to a wetting reaction of 1 min at 200◦ C.
Samples were cross-sectioned, polished, and lightly etched for optical and SEM
observation both before and after aging. Wetting reaction and solid-state aging
were studied on the same set of samples.
SVNY339-Tu April 5, 2007 17:44

60 Chapter 2

Fig. 2.23. Cross-sectional optical mi-

croscopic images of the solder bump and
the Cu UBM after two reflows (a), two
reflows followed by 500 (b), 1000 (c),
and 1500 hr (d) at 170◦ C.

Figure 2.23 shows cross-sectional optical microscopic images of the solder

bump and the Cu UBM after two reflows (a), two reflows followed by 500 (b),
1000 (c), and 1500 hr (d) at 170◦ C. Figure 2.24 shows enlarged images of
the IMC in Fig. 2.23(a) to (c). The scallop-type Cu6 Sn5 IMC can be seen
SVNY339-Tu April 5, 2007 17:44

Copper–Tin Reactions in Bulk Samples 61

Fig. 2.24. Enlarged images of the IMC in Fig. 2.17(a) to (c).

in Figs. 2.23(a) and 2.24(a). The average diameter of the scallops is about
2 μm.
In Figs. 2.23(b), 2.23(c), 2.23(d), 2.24(b), and 2.24(c), a thick layer of IMC,
consisting of Cu6 Sn5 and Cu3 Sn, formed between the solder and the Cu in a
smoother layerlike morphology. Although the interface between the solder and
Cu6 Sn5 is not flat, there are no deep valleys between the scallops as shown in
Figs. 2.23(a) and 2.24(a). The layer of Cu3 Sn is quite uniform and is as thick
as the Cu6 Sn5 layer. An etching that preferentially removes Pb rendered a
deep groove between the IMC and the solder. This indicates that the solder
SVNY339-Tu April 5, 2007 17:44

62 Chapter 2

layer next to the IMC must contain a high concentration of Pb. In addition,
extensive grain growth occurred in the solder.
The total thickness of the IMC in Fig. 2.24( c) is only a few microme-
ters, which is not much bigger than the diameter of the scallops shown in
Fig. 2.23(a). In addition, the thickness of Cu3 Sn after two reflows [as shown
in Fig. 2.23(a)] is negligibly small, compared to that after aging [as shown in
Fig. 2.23(c)].
The IMC formed during the solid-state aging can be determined by sub-
tracting the amount of IMC formed during two reflows. The average thick-
ness of IMC formed during two reflows was obtained by dividing the total
cross-sectional area of the scallops by the total length. Figure 2.25 shows the
IMC thickness measured at 500, 1000, and 1500 hr for the 125, 150, and
170◦ C aging, respectively. The growth is diffusion-controlled and the activa-
tion energy of the solid-state aging is found to be 0.94 eV/atom, as shown in
Table 2.1.

2.7.1 Morphology of Wetting Reaction and Solid-State Aging

In the classical analysis of solid-state interfacial reactions in a binary bulk dif-
fusion couple, it is assumed that all the equilibrium IMCs form simultaneously
in a layered morphology. The kinetics of growth of each layer can be diffusion-
controlled or interfacial-reaction-controlled. For a bulk diffusion couple of suf-
ficient thickness at a high enough temperature, all the IMCs coexist and obey
a diffusion-controlled growth, so the ratio of thickness among the layers is
proportional to the ratio of the square root of the interdiffusion coefficient in
each layer [21, 22] The analysis has no consideration of surface and interfacial
energies, because the motion of a planar interface in a layered structure does
not change energy.
In wetting reaction between eutectic SnPb and Cu, the Cu6 Sn5 has a
scallop-type morphology. Furthermore, the growth kinetics of the scallops has
a t1/3 dependence on time, so it does not obey the diffusion-controlled or
interfacial-reaction-controlled kinetics. The scallop-type morphology of IMC
has also been observed in wetting reaction between eutectic SnPb and Ni, and
between most of the Sn-based Pb-free solders and Cu [2]. The scallops grow
bigger but fewer with time. This indicates a nonconservative ripening reaction
among the scallop-type grains.
In the morphology of solid-state aging as shown in Fig. 2.23(b), (c), and
(d), since the samples were reflowed twice before aging, they must possess
the scallops of Cu6 Sn5 before aging. Yet during the solid-state aging the
morphology of Cu6 Sn5 has changed from scallop-type to layer-type. Why does
it not keep the scallop-type growth in the solid-state reaction? More impor-
tantly, why has the change of morphology changed the kinetics of growth
significantly? The scallop-type morphology has been found to be stable in
wetting reaction, yet unstable in solid-state reactions. The issue of morpho-
logical stability will be discussed in Section 5.2. The scallop-type morphology
SVNY339-Tu April 5, 2007 17:44

Copper–Tin Reactions in Bulk Samples 63

Fig. 2.25. (a–c) IMC thickness (a) 5 Cu6Sn5

measured at 500, 1000, and 1500 hr Cu3Sn
for the 125, 150, and 170◦ C aging, 4 Total IMC
respectively. Consumed Cu

Thickness (μm)
3

0
0 500 1000 1500 2000 2500
Time1/2 (sec1/2)
(b) 10
Cu6Sn5
Cu3Sn
8 Total IMC
Thickness (μm) Consumed Cu
6

0
0 500 1000 1500 2000 2500
Time1/2 (sec1/2)
(c) 20
Cu6Sn5
Cu3Sn
Total IMC
Thickness (μm)

15 Consumed Cu

10

0
0 500 1000 1500 2000 2500
1/2
Time (sec1/2)

suggests that the scallops themselves are not a diffusion barrier to their own
growth, unlike that of a layer-type growth!

2.7.2 Kinetics of Wetting Reaction and Solid-State Aging

We will not review the kinetics of diffusion-controlled and interfacial-reaction-
controlled growth of layer-type IMC. Here we consider the scallop-type growth
SVNY339-Tu April 5, 2007 17:44

64 Chapter 2

Table 2.1. Activation energy of Cu consumption and compound growth.

Consumed Cu Cu3 Sn Total IMC

Q (eV) D0 (cm2 /s) Q (eV) D0 (cm2 /s) Q (eV) D0 (cm2 /s)

e-SnPb 0.94 0.0149 0.73 1.85 × 10−5 1.25 59.59

Sn–3.5Ag 1.03 0.115 0.95 0.00563 1.19 6.64
Sn–3.8Ag–0.7Cu 1.10 0.0560 1.05 0.0595 0.94 9.56 × 10−3
Sn–0.7Cu 1.05 0.128 1.08 0.109 1.00 0.109

in wetting reaction. A schematic diagram of the cross section of two of the

scallops is shown in Fig. 2.26. For simplicity, we ignore the thin Cu3 Sn between
the Cu6 Sn5 scallops and Cu. We assume that Cu will not diffuse through the
bulk of Cu6 Sn5 , instead Cu diffuses through the valley between the two Cu6 Sn5
scallops in order to reach the molten solder. This Cu flux is represented by the
vertical arrow. Once in the molten solder, the diffusivity of Cu is about 10−5
cm2 /sec, so it can quickly reach the front of the Cu6 Sn5 and react with Sn
to grow the compound. At the same time, there is a ripening reaction among
the scallops, so there is a Cu flux between the scallops, as represented by the
horizontal arrow. By combining these two fluxes, the growth equation of a
scallop has been given as [23, 24]

  
γΩ2 DC0 ρAΩυ(t)
r3 = + dt,
3NA LRT 4πmNP (t)

dr1 At to

r1
r2
dr2
Δ = Exposed area
Ti or Cr Si Substrate

Fig. 2.26. Schematic diagram of cross section of two neighboring scallops.

SVNY339-Tu April 5, 2007 17:44

Copper–Tin Reactions in Bulk Samples 65

where r is the radius of a scallop, γ is the surface energy of the scallop, Ω

is the average atomic volume, D is atomic diffusivity in the molten solder,
C0 is the solubility of Cu in the molten solder, NA is Avogadro’s number,
L is the numerical factor relating the mean separation between scallops and
the mean scallop radius, RT has the usual thermodynamic meaning, ρ is
the density of Cu, A is the total area of the solder/Cu interface, ν (=dh/dt,
where h and t are thickness of Cu and time, respectively) is the consump-
tion rate of Cu in the reaction, m is the atomic mass of Cu, and NP is the
total number of scallops at the interface. We note that this growth model
has included the surface energy γ of the scallops. But in a more detailed
analysis to be presented in Chapter 5, the growth is independent of sur-
face energy. In the equation, the first term on the right-hand side is the
ripening term, and the second term is the interfacial reaction term. It was
found that the flux of the first term (represented by the horizontal arrow in
Fig. 2.26) is about 10 times greater than that of the second term (represented
by the vertical arrow). Thus, the ripening process dominates the growth of
Cu6 Sn5 .
Since we have assumed that the diffusion of Cu in the molten solder is
very fast, on the order of 10−5 cm2 /sec, and since the diffusion distance be-
tween neighboring scallops is very short, diffusion is not the rate-limiting step.
However, the supply of Cu cannot come only from the regions just below the
valleys, they must also come from the regions below the scallops or the base of
a scallop, which requires a lateral diffusion of Cu along the interface between
the scallop and Cu. We have to assume that this interfacial diffusion is very
fast too. If this interfacial diffusion is rate-limiting, we should have observed a
1
t /2 dependence on time. On the other hand, we have assumed that scallops are
hemispherical. Therefore, the scallop surface must be full of atomic steps, so
we will not have an interfacial-reaction-controlled process. Since the growth of
a hemisphere is a three-dimensional growth, the rate of growth depends on the
supply of Cu through the channels to be discussed in Chapter 5, therefore it is
a supply-limited growth of the scallops. We define it to be a supply-controlled
growth, neither diffusion-controlled nor interfacial-reaction-controlled growth.
What is amazing is that the activation energy of the growth was found to
be about 0.2 to 0.3 eV/atom. It is a very low activation energy process, and is
comparable to the activation energy of dissolution of Cu into molten Sn. We
note that the activation energy of interfacial diffusion of Cu along the base of
the scallops may be higher, yet it is not the rate-limiting step!
In the solid-state aging of eutectic SnPb on Cu, we found that the morphol-
ogy of IMC is layer-type. The activation energy of the growth of the layer-type
Cu6 Sn5 , or Cu3 Sn (or both) is about 0.8 eV/atom. Thus, the solid-state aging
is a much slower kinetic process. Another way to compare the kinetics is to
compare the period of time needed to form the same amount of IMC. In the
wetting reaction at 200◦ C, it took only a few minutes to form IMC a few mi-
crometers thick, but in the solid-state aging at 170◦ C, it took 1000 hr. Thus,
the wetting reaction is four orders of magnitude (in terms of time) faster than
SVNY339-Tu April 5, 2007 17:44

66 Chapter 2

the solid-state aging, although the temperature difference between 200◦ C and
170◦ C is only 30◦ C.
The basic reason for the difference is atomic diffusivity. In the liquid state
it is about 10−5 cm2 /sec, but in a FCC solid near its melting point it is
about 10−8 cm2 /sec. Thus, across the melting point there is a difference of
three orders of magnitude in diffusivity. Specifically, for the solid-state aging
at 170◦ C, if we assume the diffusion of Cu across the IMC has an activation
energy of 0.8 eV/atom, the diffusivity is about 10−9 cm2 /cm. So there are four
orders of magnitude of difference in diffusivity between the wetting reaction
at 200◦ C and the solid-state aging at 170◦ C. This difference is correlated to
the time difference found on the basis of the relation, x2 ≈ Dt.
Knowing the activation energy of 0.8 eV/atom of the solid-state reaction,
we ask the question: if the growth of Cu6 Sn5 compound during the wetting
reaction at 200◦ C takes a layer-type morphology, what might happen? We
shall consider the formation of a layer of 1-μm-thick Cu6 Sn5 by consuming
about 0.5 μm of Cu. We find that it will take more than 1000 sec, this is
because the Cu6 Sn5 would have become a diffusion barrier layer to its own
growth. On the other hand, to consume 0.5 μm of Cu to form Cu6 Sn5 scallops
in the wetting reaction, it actually takes less than 1 min. In other words, the
rate of free energy gain in IMC formation will be much faster in the scallop-
type growth than the layer-type growth. Since the rapid diffusion of Cu in
the molten solder is not utilized in the layer-type growth, it becomes a slow
growth. Therefore, it is the rate of free energy change, rather than the free
energy change itself, which determines the morphology of scallop-type IMC
growth.
Clearly, the scallop-type morphology affects the kinetics strongly. The ra-
dius of the scallop cannot be constant because it must grow bigger with time.
If the scallops do not grow in radius, they must grow longer and become a
diffusion barrier layer because the valleys will be closed. However, not every
scallop can grow in radius, so some of them must shrink. Hence, ripening
occurs. But the ripening eventually must slow down as the scallops become
bigger and bigger. This is because the bigger the scallops, the less the number
of short circuit paths (the valleys) to reach the molten solder. It is of interest
to determine the distribution of the size of the scallops, which will be dis-
cussed in Chapter 5. Whether it obeys the LSW theory of ripening [25–27]
is of interest. Whether the distribution function is independent of time is un-
clear. In a typical wetting reaction in devices, the thickness of Cu consumed
is less than 1 μm, so the scallops are not big at all.
We should also question why the Cu6 Sn5 in solid-state aging does not keep
the morphology of scallops. This is because scallops have a larger interfacial
area than a flat interface. In the wetting reaction, the rapid gain in compound
formation energy can compensate the interfacial energy spent in growing the
scallops, but not in the solid-state reaction. In Chapter 5, we will show that the
total surface area of scallops is unchanged in growth while the total volume of
scallops increases. In solid-state aging, the rapid gain of free energy disappears,
SVNY339-Tu April 5, 2007 17:44

Copper–Tin Reactions in Bulk Samples 67

so the compound changes to a layer-type in order to reduce the interfacial

energy.

2.7.3 Reactions Controlled by Rate of Gibbs Free Energy Change

All chemical reactions at constant pressure and constant temperature occur
with a negative Gibbs free energy change. In interfacial reactions forming
IMCs, they should be governed by negative free energy change too. But in
order to have the largest negative free energy change in a short period of time
as in the wetting reaction, it is the rate of free energy change that is crucial.
To obtain the largest free energy change during a short period of time, e.g., 1
min in a wetting reaction, we have [28]
 τ
dG
ΔG = dt, (2.12)
0 dt

where dG/dt is the rate of free energy change of the reaction, and τ is a short.
period.
Thus, the system tends to choose the reaction path or product that can
give the largest dG/dt during the period of τ, resulting in the largest gain of
free energy change. Specifically, during a wetting reaction or reflow, a high
rate of reaction is possible with the formation of scallop-type IMC. On the
other hand, if we consider a reaction when τ is infinite, it is free energy change
that is important, not the rate anymore.
We have used the specific case of molten eutectic SnPb on Cu to illustrate
the importance of a high rate reaction. Actually it is a general phenomenon.
We have already mentioned that molten eutectic SnPb on Ni forms scallop-
type Ni3 Sn4 . In the case of molten eutectic SnPb on Pd, the formation of
PdSn3 has a lamellar-type morphology and has a growth rate over 1 μm/sec,
which is most likely the fastest rate of intermetalllic growth by interfacial
reaction, to be discussed in Chapter 7. The molten solder itself serves as a
matrix for fast atomic transport for the growth of the lamellae. Among the
Pd-Sn compounds, the formation energy of PdSn3 is much less than that of
Pd2 Sn and Pd3 Sn. The latter do not form because the former has a much
higher rate of growth. This is also true in the solid-phase amorphization in
reactions in the binary systems of Rh-Si, Ti-Si, and Ni-Zr [29–31] It is because
of the high rate of growth of the amorphous alloys that they can form first,
before the formation of the equilibrium IMC phases in those binary systems.

2.8 Wetting Reaction of Pb-Free Eutectic Solders on

Thick Cu UBM
The reaction of four different eutectic solders, SnPb, SnAg, SnAgCu, and
SnCu, on electroplated thick Cu UBM 15 μm in thickness were compared. A
SVNY339-Tu April 5, 2007 17:44

68 Chapter 2

photo-resist was applied to the Cu to define the UBM contact area. Solder
paste of the four eutectic solders was printed on the UBM and reflowed twice in
a belt furnace. The temperature profile had a peak of 240◦ C, and the duration
of time above the melting point of the solders is 60 sec. Following the reflows,
solid-state aging was performed in a furnace under atmospheric ambient at
three different temperatures of 125, 150, and 170◦ C for three different periods
of 500, 1000, and 1500 hr.
Figure 2.27 shows SEM images of the interface of the four solders on Cu af-
ter two reflows. All of them show scallop-type, rounded or faceted, morphology
of Cu6 Sn5 . The scallops in the Pb-free solders are larger than those in the
SnPb. The formation of Cu3 Sn is unclear because it is thin and below the
resolution of the imaging technique used. In the SnAg and SnAgCu, some
very large platelet-type Ag3 Sn IMC can be seen.
Figure 2.28 shows optical microscopic images of the interface of the four
solders on Cu after solid-state aging at 170◦ C for 1500 hr. The solid-state
aging has changed the scallop-type morphology of Cu6 Sn5 into layer-type.
Also, the formation of a layer of Cu3 Sn is clearly shown. In the SnPb solder,
the matrix has extensive grain growth and a Pb-rich layer forms next to the
Cu6 Sn5 . In the Pb-free solders, grain growth is not apparent. The thickness

Fig. 2.27. SEM images of the interface of four eutectic solders—(a) SnPb, (b) SnAg,
(c) SnAgCu, and (d) SnCu—on Cu after two reflows.
SVNY339-Tu April 5, 2007 17:44

Copper–Tin Reactions in Bulk Samples 69

Fig. 2.28. Optical microscopic images of the interface of the four solders on Cu
(Fig. 2.27) after solid-state aging at 170◦ C for 1500 hr.

of intermetallic compounds of Cu6 Sn5 and Cu3 Sn measured at 125, 150, and
170◦ C for 500, 1000, and 1500 hr has been reported in Ref. 20. The difference
between SnPb and Pb-free solders in terms of IMC formation during solid-
state aging is not large at all. From the IMC thickness, the consumption of Cu
during the solid-state aging can be calculated. It is rather surprising to find
that the amount of Cu consumed during solid-state aging for up to 1500 hr
is of the same order of magnitude as the wetting reaction for a couple of
minutes discussed in Section 2.7.1. If we just compare the amount of IMC
formation in Figs. 2.27 and 2.28, they are of the same order of magnitude, yet
the time difference between 2 min in reflow and 1500 hr (90,000 min) in aging
is a difference of four orders of magnitude. In other words, the rate of IMC
formation in wetting reaction is four orders of magnitude faster than that in
solid-state aging.
We note that thick Cu UBM is the trend in the electronic packaging indus-
try. The reason will be made clear after we have discussed two issues; the first
issue is the spalling of the scallop-type IMC in solder reaction on thin films
of Cu to be discussed in Chapter 3, and the second issue is current crowding
in electromigration in flip chip solder joints to be discussed in Chapter 9.
SVNY339-Tu April 5, 2007 17:44

70 Chapter 2

References
1. K. N. Tu and K. Zeng, “Tin-lead (SnPb) solder reaction in flip chip tech-
nology,” Materials Science and Engineering Reports, R34, 1–58 (2001).
(Review paper)
2. K. Zeng and K. N. Tu, “Six cases of reliability study of Pb-free solder
joints in electron packaging technology,” Materials Science and Engineer-
ing Reports, R38, 55–105 (2002). (Review paper)
3. T. Young, Philos. Trans. R. Soc. London, 95, 65 (1805).
4. H. K. Kim, H. K. Liou, and K. N. Tu, “Morphology of instability of wetting
tips of eutectic SnBi , eutectic SnPb, and pure Sn on Cu,” J. Mater. Res.,
10, 497–504 (1995).
5. H. K. Kim, H. K. Liou, and K. N. Tu, “Three-dimension morphology of
a very rough interface formed in the soldering reaction between eutectic
SnPb and Cu,” Appl. Phys. Lett., 66, 2337–2339 (1995).
6. A. K. Larsson, L. Stenberg, and S. Liden, “Crystal structure modulation
in η-Cu6 Sn5 ,” Z. Kristallogr., 210 (11), 832–837 (1995).
7. H. K. Kim and K. N. Tu, “Rate of consumption of Cu soldering accom-
panied by ripening,” Appl. Phys. Lett., 67, 2002–2004 (1995).
8. C. Y. Liu and K. N. Tu, “Morphology of wetting reactions of SnPb alloys
on Cu as a function of alloy composition,” J. Mater. Res., 13, 37–44 (1998).
9. C. Y. Liu and K. N. Tu, “Reactive flow of molten Pb(Sn) alloys in Si
grooves coated with Cu film,” Phys. Rev. E, 58, 6308–6311 (1998).
10. F. G. Yost and A. D. Romig, Jr., in “Electronic Packaging Materials
Science III,” R. Jaccodine, K. A. Jackson, and R. C. Subdahl (Eds.),
Materials Research Society Symp. Proc., 108, Pittsburgh, PA (1988).
11. W. J. Boettinger, C. A. Handwerker, and U. R. Kattner, “Reactive wetting
and intermetallic formation,” in “The Mechanics of Solder Alloy Wetting
and Spreading,” F. G. Yost, F. M. Hosking, and D. R. Frear (Eds.), Van
Nostrand Reinhold, New York (1993).
12. J. Gorlich, G. Schmidt, and K. N. Tu, “On the mechanism of the binary
Cu/Sn solder reaction,” Appl. Phys. Lett., 86, 053106–1 to –3 (2005).
13. L. Kaufman and H. Bernstein, “Computer Calculation of Phase Diagram,”
Academic Press, New York (1970).
14. J.-H. Shim, C.-S. Oh, B.-J. Lee, and D. N. Lee, “Thermodynamic assess-
ment of the Cu-Sn system,” Z. Metallkd., 87, 205–212 (1996).
15. A. Bolcavage, C. R. Kao, S. L. Chen, and Y. A. Chang, “Thermodynamic
calculation of phase stability between copper and lead-indium solder,” in
Proc. Applications of Thermodynamics in the Synthesis and Processing of
Materials, Oct. 2–6, 1994, Rosemont, IL, P. Nash and B. Sundman (Eds.),
TMS, Warrendale, PA, pp. 171–185 (1995).
16. V. C. Marcotte and K. Schroeder, “Cu-Sn-Pb phase diagram,” in Proc.
Thirteenth North American Thermal Analysis Society, A. R. McGhie
(Ed.), North American Thermal Analysis Society, 1984, pp. 294.
SVNY339-Tu April 5, 2007 17:44

Copper–Tin Reactions in Bulk Samples 71

17. H. Ohtani, K. Okuda, and K. Ishida, “Thermodynamic study of phase

equilibria in the Pb-Sn-Sb system,” J. Phase Equil., 16, 416–429 (1995).
18. K. N. Tu, T. Y. Lee, J. W. Jang, L. Li, D. R. Frear, K. Zeng, and J. K.
Kivilahti, “Wetting reaction vs. solid state aging of eutectic SnPb on Cu,”
J. Appl. Phys. 89, 4843–4849 (2001).
19. K. N. Tu, F. Ku, and T. Y. Lee, “Morphological stability of solder reac-
tion products in flip chip technology,” J. Electron. Mater., 30, 1129–1132
(2001).
20. T. Y. Lee, W. J. Choi, K. N. Tu, J. W. Jang, S. M. Kuo, J. K. Lin,
D. R. Frear, K. Zeng, and J. K. Kivilahti, “Morphology, kinetics, and
thermodynamics of solid state aging of eutectic SnPb and Pb-free solders
(SnAg, SnAgCu, and SnCu) on Cu,” J. Mater. Res., 17, 291–301 (2002).
21. G. V. Kidson, “Some aspects of the growth of different layers in binary
systems,” J. Nucl. Mater., 3, 21 (1961).
22. U. Gosele and K.N. Tu, “Growth kinetics of planar binary diffusion cou-
ples: Thin film case versus bulk cases,” J. Appl. Phys., 53, 3252 (1982).
23. H. K. Kim and K. N. Tu, “Kinetic analysis of the soldering reaction be-
tween eutectic SnPb alloy and Cu accompanied by ripening,” Phys. Rev.
B, 53, 16027–16034 (1996).
24. A. M. Gusak and K. N. Tu, “Kinetic theory of flux driven ripening,” Phys.
Rev. B, 66, 115403 (2002).
25. I. M. Lifshiz and V. V. Slezov, J. Phys. Chem. Solids, 19, 35 (1961).
26. C. Wagner, Z. Electrochem., 65, 581 (1961).
27. V. V. Slezov, “Theory of Diffusion Decomposition of Solid Solutions,”
Harwood Academic Publishers, pp. 99–112 (1995).
28. D. Turnbull, “Metastable structures in metallurgy,” Metall. Trans. A, 12,
695–708 (1981).
29. S. Herd, K.N. Tu, and K.Y. Ahn, “Formation of an amorphous Rh-Si
alloy by interfacial reaction between amorphous Si and crystalline Rh
thin films,” Appl. Phys. Lett., 42, 597 (1983).
30. R. B. Schwarz and W. L. Johnson, Phys. Rev. Lett., 51, 415 (1983).
SVNY339-Tu July 12, 2007 15:5

3
Copper–Tin Reactions in Thin-Film Samples

3.1 Introduction

On a silicon chip, thin-film under-bump metallization (UBM) is needed to join

the solder bump to the Al or Cu wiring on the chip and also to control the size
of the solder bump. This is because the oxide on a free Al surface prevents the
wetting of molten solder. On the other hand, Cu reacts extremely fast with
molten solder, therefore the Cu thin-film wiring cannot be wetted by molten
solder. Control of the size of the solder bump makes use of the concept of
ball-limiting metallization in C-4 technology; the so-called controlled-collapse-
chip-connection as discussed in Chapter 1. The most widely used thin-film
UBM, between solder bump and Al or Cu interconnect wiring, is a trilayer
film of Au/Cu/Cr or a trilayer film of Cu/Ni(V)/Al. In the trilayer thin film
of Au/Cu/Cr, the Cr is needed for adhesion to dielectric surface and to Al
wiring, the Cu is needed for soldering reaction, and the Au is needed for
passivation to prevent surface oxidation. Therefore, how solder reacts with
these thin films is crucial in device manufacturing, concerning both yield and
reliability.
In Chapter 1, we mentioned that several reflows are required in device
manufacturing. In each reflow, every solder joint on the chip surface must
be successfully joined and the UBM must survive several reflows, otherwise
a very unique phenomenon of spalling of thin-film intermetallic compound
(IMC) occurs, which is a major reliability subject to be covered in Sections
3.3 to 3.8.
To discuss solder reaction on UBM thin films, we begin with the room-
temperature reaction between thin films of Cu and Sn. It is of interest for
the following reasons. (1) The fast diffusion of noble metals in beta-Sn (white
Sn) and Pb has been interpreted by the mechanism of interstitial diffusion.
At 25◦ C, the diffusivity of Cu along the a- and c-axes of beta-Sn is about
0.5 × 10−8 and 2 × 10−6 cm2 /sec, respectively. This indicates that the mobil-
ity of Cu in Sn is high enough that the growth of Cu-Sn IMC can occur at
room temperature. (2) Spontaneous Sn whisker is known to occur at room
temperature on matte Sn plated Cu surfaces. In a spontaneous process, the
SVNY339-Tu July 12, 2007 15:5

74 Chapter 3

driving force must come from within the system. If Cu and Sn react at room
temperature, the free energy gain in interfacial chemical reaction might pro-
vide the driving force for whisker growth since the chemical energy per atom
is four to five orders of magnitude higher than the elastic strain energy per
atom. In other words, a chemical process can drive a mechanical process pro-
vided that it is a slow rate process such as creep, because the rate of solid
state chemical reaction by interdiffusion is slow. (3) Thin-film samples allow
the detection of the early stage of IMC formation. In this chapter, the kinetics
of Cu-Sn reaction will be emphasized first.

3.2 Room-Temperature Reaction in a Bilayer Thin

Film of Sn/Cu
A bilayer of polycrystalline thin film of Cu and Sn has been used to study
the room-temperature reaction. To detect the reaction in thin-film samples,
high-resolution glancing incidence x-ray diffraction was used to analyze the
interfacial IMC formation [1, 2]. The bilayer film was prepared by e-beam
deposition of Cu followed by Sn on 1-inch-diameter and 1/8-inch-thick fused
quartz disks kept at room temperature during the consecutive deposition in
one run without breaking the vacuum at better than 2 × 10−7 torr. The de-
position rate was about 0.5 nm/sec. Three sets of film having the followed
thicknesses were prepared: 350 nm Sn/180 nm Cu/quartz, 350 nm Sn/600 nm
Cu/quartz, and 2500 nm Sn/600 nm Cu/quartz. Since the thickness of the
quartz disk was 1/8 inch, no bending of the sample can occur. A single layer
of Sn film of thickness 350 nm and another one of 2500 nm were deposited
on the same kind of fused quartz substrate at room temperature and kept at
room temperature as references for lattice parameter measurements and for
spontaneous Sn whisker growth (to be discussed in Chapter 6).
Annealing of the bilayer films was carried out at four temperatures: −2◦ C
(in a refrigerator), 20◦ C (air-conditioned room), 60◦ C and 100◦ C (vacuum
furnace). Except for the room-temperature annealing where some temperature
fluctuation may have occurred, the temperature was controlled to within ±
1◦ C. The annealing time was up to 1 year.

3.2.1 Phase Identification by Glancing Incidence

X-ray Diffraction
Structural change of the bilayer films as a function of annealing due to inter-
diffusion and reaction was investigated by x-ray diffraction using a glancing
incidence Seeman-Bohlin diffractometer [1]. It has the sensitivity of detect-
ing a polycrystalline Au film of 10 nm by resolving the 111, 200, 220, and
311 reflections of Au. Beta-Sn has a body-centered-tetragonal lattice with
a = 0.58311 nm and c = 0.31817 nm. Copper has a face-centered-cubic lattice
with lattice parameter of a = 0.36149 nm.
SVNY339-Tu July 12, 2007 15:5

Copper–Tin Reactions in Thin-Film Samples 75

Figure 3.1 shows four diffraction spectra of the set of 350 nm Sn on 600
nm of Cu on fused quartz. Figure 3.1(a) was obtained from the as-deposited
state. It shows reflections of pure Cu and Sn, yet two reflections of Cu6 Sn5
(the η’ phase) were present, indicating that it forms during or right after the
deposition of Sn on Cu. Figure 3.1(b) was obtained after 15 days at room tem-
perature, and several more reflections of Cu6 Sn5 were detected. Comparing
Fig. 3.1(a) and 3.1(b), they show that Cu reacts with Sn at room tempera-
ture based on the growth of Cu6 Sn5 at room temperature. Figure 3.1(c) was
obtained after 1 year at room temperature. All reflections of Sn were gone.
While there are reflections of Cu, all the other reflections can be identified
to be those of Cu6 Sn5 . It is worthwhile noting that there is no reflection of
Cu3 Sn, even though there was excess Cu in the sample. Figure 3.1(d) was ob-
tained from a sample annealed at 100◦ C for 36 hr. The reflections of Cu6 Sn5
and Cu3 Sn were found, indicating the formation of both compounds.
Figure 3.2(a) and (b) are respectively Seeman-Bohlin x-ray diffraction
spectra from angles of 4θ from 40◦ to 190◦ taken from the sample annealed
at room temperature for 1 year and the sample annealed at 100◦ C for 60 hr.
In Fig. 3.2(a), only the reflections of Cu6 Sn5 can be detected, but reflections
of both Cu6 Sn5 and Cu3 Sn are present in Fig. 3.2(b). It was concluded that
Cu6 Sn5 , but not Cu3 Sn, forms at room temperature. The growth of Cu6 Sn5
was also detected in samples kept at −2◦ C. Both Cu6 Sn5 and Cu3 Sn were de-
tected in sample kept at 60◦ C, indicating that Cu3 Sn forms at temperatures
above 60◦ C.
According to the binary phase diagram of Cu-Sn, the phase Cu6 Sn5 under-
goes a phase transition of ordering around 170◦ C. The high-temperature phase
has the ordered hexagonal NiAs structure with a = 0.420 nm and c = 0.509
nm. The low-temperature phase is a long-period superlattice with a period
of 5 along both a- and c-directions. In Fig. 3.2(a), the superlattice reflections
are indexed with *. While Sn forms surface oxide at room temperature, no
oxide reflections were detected because the oxide was too thin. The reflections
with * are from the superlattice, not from oxide. Table 3.1 lists the indexed
reflections of Cu6 Sn5 shown in Fig. 3.2(a).
The Cu3 Sn phase is ordered and has an orthorhombic lattice with a =
0.5516 nm, b = 0.3816 nm, and c = 0.4329 nm. It is a long-period superlattice
of a smaller orthorhombic one with a = 0.5514 nm, b = 0.4765 nm, and c =
4329 nm. Table 3.2 lists the indexed reflections of Cu3 Sn shown in Fig. 3.2(b).

3.2.2 Growth Kinetics of Cu6 Sn5 and Cu3 Sn

The kinetics of growth of Cu6 Sn5 and Cu3 Sn at and above room temperature
is a reliability issue in solder joints because they consume Cu. Furthermore,
the Cu3 Sn growth is accompanied by Kirkendall void formation. The rate
of Cu consumption by solder reaction is of interest since the Cu thin-film
layer in UBM is not very thick. To study the kinetics, thin-film bilayers of
Sn/Cu were prepared on fused quartz substrate with the thickness of Cu at
SVNY339-Tu July 12, 2007 15:5

h¢ - Cu Sn
6 5
∋ - Cu3Sn
d
h¢ Cu

Cu
∋ h¢ ∋ h¢
022
∋ ∋ Cu
230 213
h¢ 222 ∋
h¢ 311 h¢ h¢ Cu

100°C 36 HOURS

Cu h¢ c
201
Cu
h
300
h¢ h¢ 212
202 211
Cu
h¢
112 h¢ h¢ h¢ h¢
103 203 213 220
Cu
COUNT (arbitrary units)

R.T. 1 YEAR
Cu

h¢
Cu b

h¢
h¢ Sn
Sn Sn h¢
Sn Sn Cu
Sn Sn
Sn h¢ h¢
h¢

R.T. 15 DAYS
Cu
a 220
Cu
200
Sn
301 Sn Sn
240
Sn 231 Cu
400 Sn Sn
141 132 311
Sn
Sn 341
112 Cu Sn
h¢ h¢ Sn
100 110 120 130 140 150 160 170 180 190 200
NO HEAT TREATMENT

Fig. 3.1. Four diffraction spectra of the set of 350 nm Sn on 600 nm Cu. (a) The as-
deposited state. It shows reflections of pure Cu and Sn, yet two reflections of Cu6 Sn5 (the
η  phase) are present, indicating that it forms during the deposition of Sn on Cu. (b) After
15 days at room temperature, several reflections of Cu6 Sn5 are detected. Comparing (a)
and (b), they show the Cu and Sn reaction and the growth of Cu6 Sn5 at room temperature.
(c) After 1 year at room temperature. All reflections of Sn are gone. While there are still
reflections of Cu, all the other reflections can be identified as being those of Cu6 Sn5 . It is
worth noting that there is no reflection of Cu3 Sn, even though there was excess Cu in the
sample. (d) Annealed at 100◦ C for 36 hr. The reflections of Cu3 Sn and Cu6 Sn5 are found,
indicating the formation of both compounds.
SVNY339-Tu July 12, 2007 15:5

Copper–Tin Reactions in Thin-Film Samples 77

(a) Cu111
h′
101 h′
' 110
210
COUNT (arbitrary units)

'∗ ∗
'
011 210 '∗
∗
' 002
h∗101
h′ Cu
h′ 201 220
'
002 Cu ' h′
200 200 h′ '
022 h′ 202
h∗ 202 230 211
' '
h∗ h∗
' h′ 213 222
Cu
311
112 112 h′ '
103
' ∗∗ ' ∗∗ 311 4q
40 50 60 70 80 90 100 110 120 130 140 150 160 170 180 190
100°C 60 HOURS
(b)
h′ Cu
101 111
h′
110
COUNT (arbitrary units)

Sn200

Cu
h∗ 200 h′
201 Cu
h∗ 220

h′
h′ h′ 300
h∗ Cu
h∗ 202 211
h∗ h′ 311
h∗ 112
h′ h′
103 203 4q

40 50 60 70 80 90 100 110 120 130 140 150 160 170 180 190
ROOM TEMPERATURE ONE YEAR

Fig. 3.2. Seeman-Bohlin x-ray diffraction spectra from angles of 4θ from 40◦ to 190◦
taken from (b) the sample annealed at room temperature for 1 year and (a) the sample
annealed at 100◦ C for 60 hr. In (a), only the reflections of Cu6 Sn5 can be detected, but
reflections of both Cu6 Sn5 and Cu3 Sn are present in (b). We conclude that Cu6 Sn5 ,
but not Cu3 Sn, can form at room temperature.

560 nm and that of Sn at either 200 or 500 nm. The bilayer thin films were de-
posited at liquid-nitrogen temperature in order to reduce as much as possible
the interfacial reaction during deposition. The change of thickness of Cu6 Sn5
on room-temperature aging was measured by Rutherford backscattering. No
formation of Cu3 Sn was detected on room-temperature aging [3].
In Fig. 3.3, three Rutherford backscattering spectra (RBS) of Sn/Cu sam-
ples are shown. The spectrum of curve (a) is that of a 200 nm Sn/560 nm
SVNY339-Tu July 12, 2007 15:5

78 Chapter 3

Table 3.1. Reflections of Cu6 Sn5 found in bimetallic Cu-Sn thin films annealed at
room temperature

Indexed as ordered Indexed as long

NiAs structure period superlattice
4θ d with a = 4.19 Å with a = 20.85 Å
(deg.) (Å) b = 5.09 Å b = 25.10 Å

49.10 3.60 501∗

52.20 3.41 503∗
60.65 2.95 101 505
63.75 2.74 515∗
70.30 2.55 002 0,0,10
73.05 2.45 525∗
79.20 2.27 535∗
86.55 2.09 110 550
90.15 2.01 554∗
107.50 1.72 201 10,0,5
114.15 1.61 112 5,5,10
120.85 1.54 103 5,0,15
126.15 1.47 202 10,0,10
142.65 1.32 211 10,5,5
154.35 1.23 203 10,0,15
153.85 1.21 300 15,0,0
∗
Long period superlattice line of η  .

Table 3.2. Reflections of Cu3 Sn found in bimetallic Cu-Sn thin films annealed at
100◦ C

Indexed as ordered Indexed as long

orthorhombic lattice with period superlattice with
a = 5.514 Å a = 5.514 Å
4θ d b = 4.765 Å b = 38.16 Å
(deg.) (Å) c = 4.329 Å c = 4.329 Å

52.15 3.40 101∗ 101

55.50 3.20 011∗ 081
64.89 2.75 200 181
67.00 2.67 191‡
75.25 2.38 210∗ 280
77.65 2.32 290‡
80.70 2.23 2,10,0‡
83.85 2.16 002∗ 002
96.70 1.90 112 182
115.30 1.59 022 0,16,2
136.05 1.37 230 2,24,0
155.25 1.23 213 283
166.35 1.16 222 2,16,2
168.25 1.15 311 381
∗ ‡
and indicates long period superlattice line.
SVNY339-Tu July 12, 2007 15:5

Copper–Tin Reactions in Thin-Film Samples 79

5
Cu Sn Cu Cu6 Sn5 Cu Cu3Sn

4 a. As Dep. b. Room Temp. c. 200° C - 10 min

84 Days
ARB. UNITS
3 a

Cu b
2 Sn

c
1

0
200 300 400
CHANNEL NUMBER (5 keV/CHANNEL)

Fig. 3.3. Three Rutherford backscattering spectra (RBS) of Sn/Cu samples. The
spectrum of curve (a) is that of 200 nm Sn/560 nm Cu right after the deposition at
liquid-nitrogen temperature. The two arrows labeled Sn and Cu indicate the backscat-
tered energy positions of Sn and Cu if they appear on the surface of the sample. As
can be seen, the Cu spectrum has been displaced to lower energy due to the top Sn
layer. Curve (b) is from the sample aged at room temperature for 84 days. The Sn
sign has lowered and extended backward. The front part of the Cu spectrum has also
lowered but extended forward. Together, they indicate a mixing of Sn and Cu.

Cu right after the deposition at liquid-nitrogen temperature. The two words

labeled Sn and Cu indicate the backscattered energy positions of Sn and Cu
if they were assumed to appear on the surface of the sample. As can be seen,
the Cu spectrum has been displaced to lower energy due to the absorption
of the top Sn layer. Curve (b) is from the sample aged at room temperature
for 84 days. The Sn sign has lowered and extended backward. The front part
of the Cu spectrum has also lowered but extended forward. Together, they
indicate a mixing of Sn and Cu. The ratio of the height of Sn to that of Cu
suggests a phase of Cu:Sn = 6:5, yet the confirmation of Cu6 Sn5 was obtained
by x-ray diffraction. Curve (c) is from a sample annealed at 200◦ C for 10 min.
The signal of Sn was reduced further, but that of Cu increased. Again the
phase was confirmed by x-ray diffraction to be Cu3 Sn.
During room-temperature aging, the measured thickness of Cu6 Sn5 is plot-
ted against time in Fig. 3.4 for two sets of samples. Both the thicker and
thinner Sn samples show a linear growth at a rate of 3.5 and 6 nm/day, re-
spectively.
When all the Sn was consumed by Cu6 Sn5 formation, the Cu6 Sn5 /Cu/SiO2
samples were annealed in the temperature range from 115◦ C to 150◦ C in a
purified He furnace. The growth of Cu3 Sn between the Cu6 Sn5 and Cu was
SVNY339-Tu July 12, 2007 15:5

80 Chapter 3

24

20

THICKNESS (× 10nm)
2000 Å Sn/ 5600 Å Cu TOTALLY
16 REACTED

12

4 5000 Å Sn/ 5600 Å Cu

0
0 5 10 15 20 25 30 35 40 45
TIME (DAYS)

Fig. 3.4. Measured thickness of Cu6 Sn5 plotted versus time for two sets of samples
after room-temperature aging. Both the thicker and thinner Sn samples show linear
growth at a rate of 3.5 and 6 nm/day, respectively.

measured from the reduction of Cu6 Sn5 by the Rutherford backscattering

technique. The phase of Cu3 Sn was confirmed by glancing x-ray diffraction.
Figure 3.5 plots the square of the measured remaining thickness of Cu6 Sn5
as a function of annealing time at temperatures of 115, 120, 130, 140, and
150◦ C. The linear relation indicates the reaction is diffusion-limited. The ac-
tivation energy of reduction of Cu6 Sn5 was obtained by plotting the thickness
of Cu6 Sn5 at a fixed annealing time on a logarithmic scale against inverse
temperature, as shown in Fig. 3.6. The slope of the curve gives an activation
energy of 0.99 eV/atom.

3.2.3 Copper Is the Dominant Diffusing Species

To determine whether Cu or Sn is the dominant diffusing species during the
room-temperature growth of Cu6 Sn5 , a flash of discontinuous W film about 1
nm thick was deposited between the Cu and Sn to serve as diffusion marker.
The in-depth positions of the W film in the Sn/Cu sample before and after
aging at room temperature for 60 hr were measured by Rutherford backscat-
tering to determine the marker displacement. Figure 3.7 shows two RBS of
the W marker signal and the bilayer thin films before and after the formation
of Cu6 Sn5 . In Fig. 3.7(a), before reaction the signal of W overlaps that of Sn
and appears as a small peak near the leading edge of the Sn signal. After
reaction, the W signal overlaps the middle part of the Sn signal, as shown in
Fig. 3.7(b). Together, they indicate that Cu is the dominant diffusing species.
This is because as Cu atoms move forward, the W will be displaced backward,
hence its signal will be shifted to lower energy. By simulation, the thickness
of Cu6 Sn5 formed was about 245 nm and the position of the W marker in the
sample was determined to be about 186 nm from the surface of the Cu6 Sn5 .
SVNY339-Tu July 12, 2007 15:5

Copper–Tin Reactions in Thin-Film Samples 81

4
(Cu6Sn5 THICKNESS)2 (× 106 Å)

Xo = 1500 Å

2 115°C

120°C

130°C
1

140°C

150°C
0
0 5 10 15 20 25 30 40 50 60 75 90
TIME (min)

Fig. 3.5. Square of the measured remaining thickness of Cu6 Sn5 plotted as a function
of annealing time at temperatures of 115, 120, 130, 140, and 150◦ C. The linear relation
indicates the reaction is diffusion-limited.

If we assume that the diffusion fluxes of Cu and Sn are equal, we expect the
marker would locate roughly in the middle of the Cu6 Sn5 layer. It is actually
much deeper into the Cu6 Sn5 layer, so the flux of Cu is larger than that of Sn.

3.2.4 Kinetic Analysis of Sequential Formation of Cu6 Sn5

and Cu3 Sn
In the thin-film reactions between Sn and Cu, the formation of Cu6 Sn5
and Cu3 Sn is sequential, i.e., Cu6 Sn5 forms first and alone at room tem-
perature, and Cu3 Sn will form only at temperatures above 60◦ C. We recall
that in the wetting reaction which occurs between molten Sn and Cu at
a much higher temperature, it seems that both Cu6 Sn5 and Cu3 Sn might
SVNY339-Tu July 12, 2007 15:5

82 Chapter 3

T (°C)
150 140 130 120 115
10
9

7
lnt (sec)

4
2.3 2.4 2.5 2.6
1000/T (K−1)

Fig. 3.6. Activation energy of reduction of Cu6 Sn5 obtained by plotting the thickness
of Cu6 Sn5 at a fixed annealing time on a logarithmic scale versus inverse temperature.
The slope of the curve shows an activation energy of 0.99 eV/atom.

form simultaneously. However, using synchrotron radiation, a strong crystal-

lographic orientation relationship between Cu6 Sn5 and Cu has been found on
studying the wetting reaction from 30 sec to 4 min, indicating that Cu6 Sn5
should have nucleated on Cu directly without Cu3 Sn, as discussed in Section
2.2.1 After a longer wetting time of several minutes, the latter forms and co-
exists with Cu6 Sn5 , yet the orientation relationship between Cu6 Sn5 and Cu
is affected by the growth of Cu3 Sn, as discussed in Section 2.2.1.
If we consider the binary phase diagram of Sn-Cu, both intermetallic
phases exist in the temperature range from room temperature to 250◦ C. Based
on the phase diagram or thermodynamics, we cannot explain why at room
temperature solid state reaction they form sequentially rather than simulta-
neously. Instead, we have to use kinetic reasoning to explain why Cu3 Sn does
not form at room temperature. In view of kinetics, either Cu3 Sn cannot nucle-
ate or it cannot grow even if it can nucleate. It cannot grow because it cannot
compete with the faster growth of Cu6 Sn5 . Since both Cu6 Sn5 and Cu3 Sn can
form together at temperatures above 60◦ C, Cu3 Sn must be able to nucleate
above 60◦ C. Nucleation depends on undercooling. Because the nucleation at
room temperature has a larger undercooling than the nucleation at 60◦ C, it
will be difficult to use no nucleation at room temperature to explain the ab-
sence of Cu3 Sn. We shall consider that it cannot compete in growth against
the rapid growth of Cu6 Sn5 .
Sequential phase formation in thin film reactions has been studied in the
reaction between Si and metallic films to form silicide IMC phases [4]. Single
phase formation of a specific silicide on Si to serve as ohmic contacts and
SVNY339-Tu July 12, 2007 15:5

Copper–Tin Reactions in Thin-Film Samples 83

w
5 (a)
w
5600Å 2000Å
4 Cu Sn

1
ARB. UNITS

0
w
5 4600 Å 1850 Å
(b) Cu6Sn5
Cu
4 200-600 Å w
Cu6Sn5 Cu Sn
3

0
200 300 400
CHANNEL NUMBER (5 KeV/CHANNEL)

Fig. 3.7. Two Rutherford backscattering spectra of the W marker signal and the bi-
layer thin films of Sn/Cu before and after the formation of Cu6 Sn5 . (a) Before reaction,
the signal of W overlaps that of Sn and appears as a small peak near the leading edge
of the Sn signal. (b) After reaction, the W signal overlaps the middle part of the Sn
signal. Together, they indicate that Cu is the dominant diffusing species.

gates in field-effect transistor devices has been a very important technological

issue. There are millions or even billions of silicide contacts and gates on a
Si chip having a very-large-scale integration of circuits. These contacts and
gates must have the same physical properties. For example, we cannot have
a contact that consists of a mixture of silicide phases. Therefore, the device
application demands single phase formation, which in principle is contrary to
thermodynamics. Thus, a kinetic rather than a thermodynamic reason has to
be given. The kinetics of single phase growth has been analyzed by Gosele and
Tu, assuming a layered model of competition of growth of coexisting phases
by combining diffusion-controlled growth and interfacial-reaction-controlled
growth [5].
SVNY339-Tu July 12, 2007 15:5

84 Chapter 3

eq
Cab
CA
Without interface reaction
with barriers
eq
Cba

Cbg
Cba

eq
AaB AbB Cbg Ag B
eq
Cgb

xb

xab xbg x

Fig. 3.8. Schematic diagram depicting the growth of a layered intermetallic compound
phase between two pure elements, for example, the growth of Cu6 Sn5 (Aβ B) between
Cu (Aα B) and Sn (Aγ B). The concentration change of Cu across the interfaces is
shown.

Figure 3.8 depicts the growth of a layered IMC phase between two pure
elements, for example, the growth of Cu6 Sn5 between Cu and Sn. We repre-
sent Cu, Cu6 Sn5 , and Sn by Aα B, Aβ B, and Aγ B, respectively. The thickness
of Cu6 Sn5 is xβ and the position of its interface with Cu and Sn is defined by
xαβ and xβγ , respectively. Across the interfaces, there is an abrupt change in
concentration. In Fig. 3.8, the concentration change of Cu across the inter-
faces is shown. In a diffusion-controlled growth of xβ , the concentrations at
its interface are assumed to have the equilibrium values, represented by the
broken curve in the xβ layer in Fig. 3.8. In an interfacial-reaction-controlled
growth, the concentrations at its interface are assumed to be nonequilibrium,
represented by the solid curve.
To consider a diffusion-controlled growth of a layered phase of xβ in Fig.
3.8, we shall use Fick’s first law of diffusion in one dimension. The correspond-
ing fluxes across the interface and in the layer are shown in Fig. 3.9.

dC
J = −D (3.1)
dx

and also the flux equation of

J = Cv, (3.2)
SVNY339-Tu July 12, 2007 15:5

Copper–Tin Reactions in Thin-Film Samples 85

xab xbg

A A
A A Jbg Jg b
Jab Jba
A
Jb

xb

Fig. 3.9. Corresponding fluxes across the interface and in the layer of xβ .

where J is atomic flux (number of atoms/cm2 -sec), D is atomic diffusivity

(cm2 /sec), C is concentration (number of atoms/cm3 ), x is length (cm), and
v is the velocity of the moving interface (cm/sec). For example, v = dxαβ /dt
of the interface xαβ . For the growth of this interface, by considering the con-
servation of flux that enters and leaves the interface, we have on the basis of
Eqs. (3.2) and (3.1),

dxαβ ∂C ∂C
(Cαβ − Cβα ) = Jαβ − Jβα = −D |αβ + D |βα (3.3)
dt ∂x ∂x

Rearranging, we obtain the expression of velocity of the xαβ interface,

    
dxαβ 1 ∂C ∂C
= −D − −D . (3.4)
dt Cαβ − Cβα ∂x αβ ∂x βα

To overcome the unknown of the concentration gradients in square brackets in

the above equation, we have to make a transformation by combining the two √
variables of x and t into one, i.e., we consider C(x, t) = C(η), where η = x/ t,
and so

∂C 1 dC
=√ . (3.5)
∂x t dη

The concentrations at the interface, i.e., Cαβ and Cβα , can be assumed to re-
main constant with respect to time and position, because we can take them as
the equilibrium values under the assumption of a diffusion-controlled growth
[6]. Hence, we have

dC(η)
= f (η), (3.6)
dη

where f (η) = constant if η is constant, independent of time and position,

at the interfaces for a diffusion-controlled process. Therefore, the equation of
SVNY339-Tu July 12, 2007 15:5

86 Chapter 3

velocity can be rewritten as

     
dxαβ 1 ∂C ∂C 1
= − D +D √ . (3.7)
dt Cαβ − Cβα ∂η αβ ∂η βα t

The quantity within square brackets is independent of time, after we take the
factor of time out of the square brackets. Integration of the above equation
gives
√
xαβ = Aaβ t, (3.8)

where
 
(DK)βα − (DK)αβ
Aαβ = 2 ,
Cαβ − Cβα
 
dC
Kij = .
dη ij

Following a similar approach, we can obtain at the other interface of xβγ ,

√
xβγ = Aβγ t. (3.9)

By combining the two interfaces, we have the width of the β phase as

√ √
Wβ = xβγ − xαβ = (Aβγ − Aαβ ) t = B t, (3.10)

which shows that the β phase has a parabolic rate or diffusion-controlled

growth. We note that the above is a very simple derivation of a diffusion-
controlled growth of a layered structure, or a relationship of x2 ∝ t for a
layered growth with abrupt change of composition at its interfaces.
A fundamental feature of a diffusion-controlled layer growth is that it will
not disappear or cannot be consumed in competition of growth in a multi-
layered structure since its velocity of growth is inversely proportional to its
thickness. As the thickness w approaches 0,

dw B
lim = → ∞. (3.11)
dt w
The growth rate will approach infinity, or the chemical potential gradient to
drive the growth will approach infinity. Therefore, in a multilayered structure,
for example, Cu/Cu3 Sn/Cu6 Sn5 /Sn, when both Cu3 Sn and Cu6 Sn5 exist and
have diffusion-controlled growth, they will coexist and grow together. For
this reason, in a sequential growth of Cu6 Sn5 followed by Cu3 Sn, we cannot
assume that both of them can nucleate and grow by a diffusion-controlled
process, then they will coexist.
SVNY339-Tu July 12, 2007 15:5

Copper–Tin Reactions in Thin-Film Samples 87

Next we shall consider the interfacial-reaction-controlled growth in which

the growth rate is linear with time, or the rate is constant and finite and
independent of thickness. We note that the linear growth rate cannot continue
forever; when the layer grows to a certain thickness, diffusion across the thick
layer will be rate-limiting and the growth will change to diffusion-controlled
or its time dependence will change from linear to parabolic [5].
In an interfacial-reaction-controlled growth, the concentration at the in-
terface will be nonequilibrium as shown in Fig. 3.8. The flux in the β-phase
will be given by
eq
Jβ = (Cβα − Cβα )Kβα , (3.12)

where Kβα is defined as the interfacial-reaction constant of the xαβ interface. It

has the unit of velocity (cm/sec), and it infers the rate of removal of Cu atoms
from the Aα B surface. We assume that there is a sluggishness in removing Cu
atoms from the surface of Aα B so that the concentration Cβα is less than the
equilibrium value. On the other hand, at the xβγ interface, there is sluggishness
in accepting the incoming Cu atoms, so there is a buildup of Cu atoms and
the concentration of Cβγ is greater than the equilibrium value. In Fig. 3.8, the
broken curve in the β-phase represents the equilibrium concentration gradient,
and the solid curve represents the nonequilibrium concentration gradient. The
rate of growth of the β-phase does not depend on diffusion across itself, but
rather it depends on the interfacial-reaction process at the two interfaces.
Details of the kinetic analysis of such a layer can be found in the literature
and will not be repeated here. The growth rate has been given as

dxβ Gβ ΔCβ Kβeff Gβ ΔCβ Kβeff

= K eff
= x , (3.13)
dt 1 + xβ Dββ 1 + xβ∗
β

1 1 1
where Kβeff
= Kβα + Kβγ is the effective interfacial-reaction constant of the β-
phase, Kβγ is defined as the interfacial-reaction constant at the xβγ interface,
Gβ is a constant, and ΔCβ is a concentration term, and Dβ is the interdiffusion
coefficient in the β-phase. We define a “changeover” thickness of

Dβ
x∗β = . (3.14)
Kβeff

For a large changeover thickness, or xβ /xβ ∗  1, i.e., under the condi-

tion that the interdiffusion coefficient is much larger than the effective
interfacial-reaction coefficient, we obtain

dxβ
= Gβ ΔCβ Kβeff . (3.15)
dt

The process is interfacial-reaction-controlled, and the growth rate is constant.

SVNY339-Tu July 12, 2007 15:5

88 Chapter 3

Fig. 3.10. Schematic diagram

depicting a four-layered thin-
film structure of Cu/Cu3 Sn/
CA
Cu6 Sn5 /Sn.

Aa B AbB Ag B Ad B

xb xg

To apply both diffusion-controlled growth and interfacial-reaction-

controlled growth to the thin-film Cu-Sn reaction, we depict in Fig. 3.10 a
four-layered thin-film structure of Cu/Cu3 Sn/Cu6 Sn5 /Sn. We assume that
the Cu3 Sn growth is interfacial-reaction-controlled and has velocity v1 , and
the Cu6 Sn5 growth is diffusion-controlled and has velocity v2 . When their
thickness is small, the magnitude of v2 can be quite large due to the inverse
dependence on layer thickness, so we can assume v2  v1 and the rapid growth
of Cu6 Sn5 can consume all of Cu3 Sn. We can also assume that both of them
have an interfacial-reaction-controlled growth, with v2  v1 , again the growth
of Cu6 Sn5 is dominant and we have a single phase growth. In Section 3.2.2 and
Fig. 3.4, it was stated that Cu6 Sn5 has a linear growth at room temperature.
Generally speaking, a drift velocity can be expressed as the product of
driving force and mobility. In thin film reaction, the driving force is chemical
affinity of IMC formation, e.g., the formation energy of Cu6 Sn5 . Then the
physical meaning of the interfacial-reaction constant K (velocity) is that of
interfacial mobility. An atomistic explanation of interfacial mobility will be
presented in Section 3.2.5.
Experimentally, it is of interest to prepare a bilayer of Cu/Sn thin-
film sample and age at 100◦ C for a short time to form a structure of
Cu/Cu3 Sn/Cu6 Sn5 /Sn, and then follow with a long aging at room temper-
ature to learn whether the Cu3 Sn will grow thicker or will be consumed by
the growth of Cu6 Sn5 . The aging at 100◦ C must be short so that both Cu3 Sn
and Cu6 Sn5 will not be too thick to have achieved diffusion-controlled growth.
If the room-temperature aging consumes the existing Cu3 Sn, it indicates that
even if Cu3 Sn can nucleate at room temperature, it cannot grow side-by-side
with Cu6 Sn5 , so the single phase formation of Cu6 Sn5 at room temperature is
due to growth selection. On the other hand, if existing Cu3 Sn can grow or can
coexist with Cu6 Sn5 , the finding of no Cu3 Sn on room-temperature reaction
between Cu and Sn means that it cannot nucleate.
SVNY339-Tu July 12, 2007 15:5

Copper–Tin Reactions in Thin-Film Samples 89

3.2.5 Atomistic Model of Interfacial-Reaction Coefficient

In the kinetic analysis discussed in the last section, the two most important
kinetic parameters are atomic diffusivity and interfacial-reaction-controlled
coefficient. Atomistic diffusion in face-centered-cubic metals via vacancy
mechanism has been well developed and presented in textbooks (see Appendix
A). For example, the diffusivity can be expressed as
     
ΔGf ΔGm ΔH
D = f nυ0 λ exp −
2
exp − = D0 exp − , (3.16)
kT kT kT

where the prefactor

 
2 ΔSf + ΔSm
D0 = f nυ0 λ exp (3.17)
k

and the activation enthalpy

ΔH = ΔHf + ΔHm (3.18)

and f is the correlation factor, n is the number of nearest neighbors and

n = 12 in face-centered-cubic lattices, υ0 is the Debye frequency of vibration,
λ is the atomic jump distance between an atom and its nearest neighbor
vacancy, ΔGm and ΔGf are the free energy of motion and formation of a
vacancy, respectively, and kT has the usual meaning of thermal energy. The
physical meaning of exp (−ΔGf /kT ) is the probability of having a vacancy
next to the jumping atom. The physical meaning of exp (−ΔGm /kT ) is the
probability of a successful exchange jump between an atom and a nearest
neighbor vacancy. Concerning the correlation factor, we note that f = 0.87
for vacancy mechanism in face-centered-cubic lattices.
For comparison, we present below a similar expression of the interfacial-
reaction-controlled coefficient. In Fig. 3.11, the energy of activation processes

F′
(Sn) (Cu6Sn5)
n + ΔGm
n−
ΔGf

Fig. 3.11. The energy of activation processes across the Cu6 Sn5 /Sn interface, where
ΔGm is the activation energy of motion across the interface and ΔG is the gain of free
energy (driving force) in the reaction or the growth of the compound per atom of the
Cu6 Sn5 molecule, and δ is the width of the interface.
SVNY339-Tu July 12, 2007 15:5

90 Chapter 3

across the Cu6 Sn5 /Sn interface is depicted, where ΔGm is the activation en-
ergy of motion across the interface and ΔG is the gain of free energy (driving
force) in the reaction or the growth of the compound per atom of the Cu6 Sn5
molecule, and δ is the width of the interface. If we consider a one-dimensional
growth and assume a unit area of the interface advancing a distance of dx, or
a volume of dx times 1, the number of atoms in the volume is dx/Ω and Ω is
atomic volume. The total free energy gained is ΔG(dx/Ω) and should equal
the work done,

dx
ΔG = pdV = pdx, (3.19)
Ω
ΔG
p= . (3.20)
Ω

p is pressure. To examine ΔG, we consider the chemical reaction of

6Cu + 5Sn → Cu6 Sn5 .

We have the chemical affinity

A = μη − 6μCu − 5μSn , (3.21)

where μη is the chemical potential of the Cu6 Sn5 compound molecule and
μCu and μSn are the chemical potential of the unreacted Cu and Sn atom,
respectively. The Gibbs free energy change of the reaction is

dG = −SdT + V dp − Adn, (3.22)

where n is the extent of the reaction (units of moles or molecules), and S, T , V ,

and p have their usual meaning in thermodynamics. At a constant temperature
and constant ambient pressure, the free energy gain of the reaction is

ΔG = AΔn. (3.23)

To relate the driving force to the kinetics of interfacial reaction, we consider

the atomic flux of Sn jumping from the Sn grain, across the interface, to the
Cu6 Sn5 grain, and we have [7]
 
ΔGm
J1→2 = A2 n1 υ1 exp − , (3.24)
kT

where A2 is the probability of accommodation of the atom per unit area on

the surface of Cu6 Sn5 , i.e., the probability of an atom that can attach to the
grain of Cu6 Sn5 , n1 is the number of atoms per unit area on the Sn grain
SVNY339-Tu July 12, 2007 15:5

Copper–Tin Reactions in Thin-Film Samples 91

ready to make a jump across the interface, and υ1 is the vibration frequency.
The reverse flux from the Cu6 Sn5 to the Sn is
 
ΔGm + ΔG
J2→1 = A1 n2 υ2 exp − . (3.25)
kT

If ΔG = 0, the two sides are at equilibrium; it means the growth stops. There
is no net motion, so the fluxes are equal. Thus, we have

A2 n1 υ1 = A1 n2 υ2 . (3.26)

With ΔG < 0, we have a net motion of growth,

   
ΔGm ΔG
Jnet = A2 n1 υ1 exp − 1 − exp −
kT kT
 
ΔGm ΔG
= A2 n1 υ1 exp − . (3.27)
kT kT

The linearization process assumes ΔG  kT . The velocity of growth is

 
A2 n 1 υ 1 Ω 2 ΔGm ΔG ΔG
v = Jnet Ω = exp − =M , (3.28)
kT kT Ω Ω

where M is the mobility, and ΔG/Ω = p as shown in Eq. (3.20).

 
A2 n1 υ1 Ω2 ΔGm
M= exp − . (3.29)
kT kT

To check the units of M , we note that on the basis of Einstein’s relationship,

M = D/kT , where D is diffusivity and has units of cm2 /sec. Since the units of
A2 n1 Ω2 and υ1 are cm2 and sec−1 , respectively, it is correct. For comparison,
we take the diffusivity,
 
ΔGm
D = A2 n1 υ1 Ω exp −
2
, (3.30)
kT

and compare it to the diffusivity of atomic diffusion via vacancy mechanism

in a face-centered-cubic lattice, for which we have
     
ΔGf ΔGm ΔH
D = f nυ0 λ exp −
2
exp − = D0 exp − . (3.31)
kT kT kT

The physical meaning of exp(−ΔGf /kT ) is the probability of having a vacancy

next to the jumping atom, so it is similar to the meaning of A2 in Eq. (3.24),
SVNY339-Tu July 12, 2007 15:5

92 Chapter 3

which is the probability of a site on the Cu6 Sn5 surface which is available to
accept a Cu atom or a Sn atom. If A2 = 1, the interfacial kinetic process is
fast, so the growth will be diffusion-controlled. If A2 < 1, the interfacial kinetic
process will be sluggish, so the growth will be interfacial-reaction-controlled.
Concerning the correlation factor, we note that f = 0.87 for vacancy mech-
anism in face-centered-cubic lattices, but we can take f ∼ 1 in the reactive
growth since the probability of reverse jump or dissociation jump of an atom
from the compound during the growth is small.
In the above, we have considered the growth atom by atom. Since there
are Cu atoms and Sn atoms, we should consider the growth per molecule,
but it will involve 6 Cu atoms and 5 Sn atoms. This is unlikely because the
interfacial reaction process will be extremely slow. On the other hand, when
we assume the growth occurs atom by atom, the gain in free energy per atom
is less than A/11, where A is chemical affinity of the formation of a molecule
of Cu6 Sn5 . This is because only when a molecule of Cu6 Sn5 is formed do we
gain the energy of A. If it is only partially formed, the average energy per
atom should be higher than A/11.

3.2.6 Measurement of Strain in Cu and Sn Thin Films

In Section 3.2.1, we discussed using the Seeman-Bohlin diffractometer to mea-
sure the lattice parameter of a thin film and to identify the phase. The di-
rection of the reciprocal lattice vector of each reflection in this diffractometer
makes a different inclination angle ϕ(= θ − γ) with respect to the normal of
the film surface, where θ is the Bragg angle and γ is the fixed incidence angle
of the x-ray beam. The x-ray measures the interplanar spacing, or strain, in
the direction of the reciprocal lattice vector. By extrapolation to ϕ = 90◦ ,
the strain in the direction of the principal stress, in the direction parallel to
the surface plane of the film, can be obtained [1]. The following reflections of
220, 311, 331, and 420 of Cu, and reflections of 400, 231, 420, 411, 440, 123,
303, 233, and 143 of Sn were used in the extrapolations. Since the lattice of
beta-Sn is body-centered tetragonal, the extrapolation to obtain both lattice
parameters a and c was carried out by successive iterations. It was found that
the remaining Cu film in the trilayer Cu/Cu6 Sn5 /Sn films when annealed at
room temperature was under tension and the remaining Sn film was under
compression. Table 3.3 lists the extrapolated lattice parameters, i.e., the lat-
tice parameter along the direction parallel to the film surface of Cu and Sn
annealed at room temperature. The lattice parameters of the single layer of Sn
deposited and annealed at room temperature on fused quartz are also listed
for reference. The strain in the Cu film in the direction parallel to the film
surface was found to be about +0.06% with uncertainty of ± 50%. The strain
in the Sn film along the same direction is about −0.16%, which is below the
elastic limit of 0.2%. The compressive stress will be related to spontaneous
Sn whisker growth to be discussed in Chapter 6.
SVNY339-Tu July 12, 2007 15:5

Copper–Tin Reactions in Thin-Film Samples 93

Table 3.3. Extrapolated lattice parameters of Cu and Sn after room-temperature

annealing

Lattice parameter Lattice parameter

Annealing of Cu (= 0.0010) of Sn (= 0.0020)
Specimen time (Å) (Å)
Sn/Cu Right after 3.6162 a = 5.8132
deposition c = 3.1701
15 days 3.6181 a = 5.8144
c = 3.1706
30 days 3.6176 a = 5.8063
c = 3.1692
1 yr 3.6180
Sn(3500 Å) a = 5.8205
c = 3.1747

3.3 Spalling in Wetting Reaction of Eutectic SnPb

on Cu Thin Films
When a thick Cu foil is replaced by a thin Cu film, a dramatic change in the
IMC morphology occurs in solder wetting reaction on the thin film. Figure 3.12
shows the cross-sectional SEM image of eutectic SnPb solder on an 870 nm
Cu thin film deposited on a 100 nm Ti film on an oxidized Si wafer after wet-
ting reaction for 10 min at 200◦ C. The scallop-type Cu6 Sn5 IMC no longer
exist, rather the IMC have become spheroids and some of them have left the
substrate and spalled into the molten solder [8–11]. This may be unexpected
because when the Cu film is completely consumed by the solder, we expect
that the interfacial reaction should stop since there is no more Cu. Yet the
ripening reaction among the scallops continues and transforms the hemispher-
ical scallops into spheroids. The spheroids have a 180◦ wetting angle on the
Ti surface. There is no adhesion between them, so the spheroids can detach
easily from the Ti surface and spall into the molten solder. In the molten

Fig. 3.12. Cross-sectional SEM

image of eutectic SnPb solder on
an 870 nm Cu thin film deposited
on a 100 nm Ti film on an oxi-
dized Si wafer after annealing for
10 min at 200◦ C. The scallop-type
Cu6 Sn5 IMC no longer exist; rather
the IMC have become spheroids
and some of them have left the sub-
strate and spalled into the molten 1μm
solder.
SVNY339-Tu July 12, 2007 15:5

94 Chapter 3

Fig. 3.13. Cross-sectional SEM image of a piece of eutectic SnPb solder sandwiched
between two Si chips having Au/Cu/Cr trilayer films. After 20 min at 200◦ C, the
Cu6 Sn5 spheroids have departed from the bottom surface (b) and moved to the upper
surface (a), assisted by gravity force. This is the phenomenon of “spalling” of IMC.

state of the solder, this can be illustrated with the help of gravity because the
density of the molten solder is greater than that of the Cu6 Sn5 compound.
Figure 3.13 shows the cross-sectional SEM image of a piece of eutectic SnPb
solder sandwiched between two Si chips having the Au/Cu/Cr trilayer films.
After 20 min at 200◦ C, the Cu6 Sn5 spheroids have departed from the bottom
surface and moved to the upper surface. This is the phenomenon of “spalling”
of IMC. When it takes place, the solder is in direct contact with the unwetted
substrate and dewetting occurs.
Figure 3.14 shows spalling of IMC in a solder cap on Au/Cu/Cr trilayer
thin films. Figure 3.15 shows an SEM image of a dewetted surface after
spalling. Figure 3.16 is a set of schematic diagrams of the sequence of ripening,
spalling, and dewetting in a molten solder–thin film reaction.
To explain the morphological transformation, we recall that Fig. 2.26
depicts the conservative or constant volume ripening between two neighboring
scallops and the opening of a large gap between them. When all of the Cu
SVNY339-Tu July 12, 2007 15:5

Copper–Tin Reactions in Thin-Film Samples 95

Fig. 3.14. Cross-sectional SEM images of a eutectic SnPb solder cap on Au/Cu/Cr
trilayer films. (a) Low-magnification image of the cap. (b)–(e) High-magnification im-
ages of various cross sections showing the spalling of IMC.

thin film has been reacted, the ripening among Cu6 Sn5 scallops becomes
conservative. The gap allows the molten solder to be in direct contact with
the Ti surface, which is unwetted by the molten solder. Figure 3.17 depicts
the transformation from a hemispherical-type scallop to a sphere driven by
SVNY339-Tu July 12, 2007 15:5

96 Chapter 3

Fig. 3.15. SEM image of a dewetted surface after spalling.

the lowering of the sum of surface and interfacial energies in a conservative

ripening.
In the following, we shall review a sequence of reactions between
molten solder and different thin-film under-bump metallizations in electronic
packaging technology, and we shall see that the spalling is a recurring phe-
nomenon and remains a very challenging reliability issue for Pb-free solders.

(a) Flux
Sn ripening
Cu

Cu-Sn compound Cr
Si

(b) Flux spalling

Sn
Cu
Cr
Si

Flux
γSn/Flux

(c) Sn dewetting
γCr/Flux θ γCr/Sn Cu
Cr
Si

Fig. 3.16. Schematic diagrams of the sequence of ripening, spalling, and dewetting in
a solder–thin film reaction.
SVNY339-Tu July 12, 2007 15:5

Copper–Tin Reactions in Thin-Film Samples 97

Molten Solder

rs
rh
Soild Soild

Cr Cr
Hemispherical Cu6Sn5 Cu6Sn5 Spheroid

γCu6Sn5/Cr is very large

Interfacial energy change for a fixed volume

πrh2γCu6Sn5/Cr + 2πrh2γsolder/Cu6Sn5 > πrh2γCr/solder + 4πrs2γsolder/Cu6Sn5

Fig. 3.17. Schematic diagram depicting the transformation from a hemispherical-type

particle to a sphere driven by lowering of the sum of surface and interfacial energies
in a conservative ripening.

3.4 No Spalling in High-Pb Solder on Au/Cu/Cu-Cr

Thin Films
First, we review the controlled-collapse-chip-connection (C-4) solder joint used
in mainframe computers [12]. Figure 1.9 is a schematic diagram of a 95Pb5Sn
solder ball joining a thin-film metallization which consists of 100 nm Au/ 500
nm Cu/ 300 nm co-deposited Cu-Cr to a ceramic module. Since Cu and Cr
have poor adhesion to each other, the co-deposited Cu-Cr or phased-in Cu-Cr
was developed to improve the adhesion between them. This is because Cr and
Cu are immiscible, so their grains form an interlocking microstructure when
they are co-deposited.
Figure 3.18 shows a selected area electron diffraction pattern of a phase-in
Cu-Cr thin film, in which the diffraction rings from Cu and Cr can be iden-
tified. Figure 3.19 shows a bright-field cross-sectional transmission electron
microscopy (TEM) image of the trilayer thin-film structure [13]. A selected
area diffraction pattern of the Cu-Cr layer can be indexed as a mixture of
reflection rings of Cu and Cr. A high-resolution TEM image of the mixed
Cu and Cr layer is shown in Fig. 3.20, in which the lattice of Cu or Cr can
be identified. When a molten high-Pb solder of 95Pb5Sn wets this trilayer
thin-film metallization, it dissolves the Au, forms AuSn4 compound particles
in the solder, and forms Cu3 Sn compound on the Cu-Cr layer, but there is no
Cu6 Sn5 formation. This is in agreement with the Sn-Pb-Cu phase diagram at
350◦ C, as discussed in Chapter 2. What is rather surprising is that there has
been no spalling of Cu3 Sn reported until recently [14]. The phase-in Cu-Cr
SVNY339-Tu July 12, 2007 15:5

98 Chapter 3

Fig. 3.18. Selected area electron diffraction pattern of a phase-in Cu-Cr. The diffrac-
tion rings of Cu and Cr are identified.

200 nm

Fig. 3.19. Bright-field cross-sectional transmission electron microscopy (TEM) image

of the trilayer thin-film structure.
SVNY339-Tu July 12, 2007 15:5

Copper–Tin Reactions in Thin-Film Samples 99

Fig. 3.20. High-resolution TEM image of the mixed Cu and Cr layer. The lattice of
Cu or Cr can be identified. (Courtesy of Prof. Ning Wang, Hong Kong University of
Science and Technology.)
SVNY339-Tu July 12, 2007 15:5

100 Chapter 3

thin film is quite stable with the high-Pb solder. Since the phase-in Cr-Cu
was developed not only to improve the adhesion between Cr and Cu, but
also to resist spalling, it indicates that spalling of Cu3 Sn might have occurred
if a layered structure of Cu/Cr is used instead of the phased-in Cu-Cr/Cr.
It is likely that the spalling of Cu3 Sn has been retarded by the interlocking
microstructure in the phase-in Cu-Cr.

3.5 Spalling in Eutectic SnPb Solder

on Au/Cu/Cu-Cr Thin Films
Due to the need for more functions in consumer electronic products and in
turn the need for more input/output (I/O) interconnections on a chip surface,
flip chip technology has gained wider application in electronic manufacturing.
In consumer products, chips are joined to polymer boards and cards to lower
the cost. The low-melting eutectic SnPb which can be reflowed at 220◦ C is
more suitable than the high-Pb solder in C-4 technology. When the eutectic
SnPb solder reacts with Cu at 220◦ C, the reaction product is Cu6 Sn5 rather
than Cu3 Sn. This is in agreement with the ternary phase diagrams of Sn-Pb-
Cu shown in Fig. 2.21. The Cu6 Sn5 , however, is morphologically unstable on
the Cu-Cr surface, leading to spalling [8–11]. Figure 3.13 showed the cross-
sectional SEM image of a eutectic SnPb solder bump sandwiched between two
Si chips having the Au/Cu/Cu-Cr films. After 20 min at 200◦ C, the Cu6 Sn5
spheroids have spalled.

3.6 No Spalling in Eutectic SnPb on Cu/Ni(V)/Al

Thin Films
Owing to the spalling behavior of eutectic SnPb on Au/Cu/Cu-Cr UBM dis-
cussed above, the Cu/Ni(V)/Al thin-film UBM was investigated for eutectic
SnPb solder bumping. Figure 3.21(a) shows a cross-sectional TEM image of
the solder–thin film interface after 1 reflow; the Si, SiO2 , a bilayer of Al and
Ni(V), and Cu6 Sn5 are seen [15]. Within the Cu6 Sn5 layer, isolated regions
of unreacted Cu surrounded by a cluster of small Cu3 Sn grains are found.
Between the Cu3 Sn and Cu, there are Kirkendall voids. Figure 3.21(b) shows
a cross-sectional TEM image of the interface after an additional annealing of
5 min at 200◦ C. Neither Kirkendall voids nor Cu3 Sn grains can be found any
more. The Cu6 Sn5 showed some grain growth but it attached very well to
the Ni(V). Even after 20 to 40 min annealing at 220◦ C, very little change was
found, and the Cu6 Sn5 and Ni(V) layers were stable. Figure 3.21(c) is a higher
magnification image of the Ni(V) layer and its interface with the Cu6 Sn5 layer.
When Ni contains more than 7 wt% V, it is diamagnetic and can be
sputtered at a high rate. The V seems to have lowered the stacking fault
energy of Ni, so a large number of twin boundaries in the Ni are seen in
Fig. 3.21(c). Why does the Cu6 Sn5 not transform into spheroids and spall
SVNY339-Tu July 12, 2007 15:5

Copper–Tin Reactions in Thin-Film Samples 101

(a)

Cu3Sn

Ni(V)
Cu6Sn5

Al
Al

SiO2
Si

1μm

(b)

Cu6Sn5
Ni(V)

Al

SiO2
Si
1μm

Fig. 3.21. (a) Cross-sectional TEM image of the solder–thin film interface after 1
reflow; the Si, SiO2 , a bilayer of Al and Ni(V), and Cu6 Sn5 are seen. Within the
Cu6 Sn5 layer, isolated regions of uneacted Cu surrounded by a cluster of small Cu3 Sn
grains are found. Between the Cu3 Sn and Cu, there are Kirkendall voids. (b) Cross-
sectional TEM image of the interface after an additional annealing of 5 min at 200◦ C.
Neither Kirkendall voids nor Cu3 Sn grains are seen. The Cu6 Sn5 showed some grain
growth but it attached very well to the Ni(V). (c) High-magnification image of the
Ni(V) layer and its interface with the Cu6 Sn5 layer.

into the solder? A plausible answer is that the interface between Cu6 Sn5 and
Ni(V) is a very low energy interface, hence it is stable against morphological
transformation.

3.7 Spalling in Eutectic SnAgCu Solder

on Cu/Ni(V)/Al Thin Films
Since the Cu/Ni(V)/Al UBM is stable with eutectic SnPb, it has been ex-
tended to Pb-free solder [16]. Figure 3.22 shows SEM backscattering images
SVNY339-Tu July 12, 2007 15:5

102 Chapter 3

solder (Cu,Ni)6Sn5 solder

Cu6Sn5

(a) (b)

(Cu,Ni)6Sn5 solder
(Cu,Ni)6Sn5

(c) (d)

Fig. 3.22. SEM backscattering images of cross sections of samples of eutectic SnAgCu
solder on Al/Ni(V)/Cu thin films after (a) 1 (i.e., the as-bonded condition), (b) 5, (c)
10, and (d) 20 reflows.

of cross sections of samples of eutectic SnAgCu solder on Al/Ni(V)/Cu thin

films after 1 (i.e., the as-bonded condition), 5, 10, and 20 reflows. Two types of
IMC, Cu6 Sn5 (also (Cu,Ni)6 Sn5 )) and Ag3 Sn, were found in the solder bump
after these reflows. The Cu6 Sn5 was present mainly at the interface, although
some large Cu6 Sn5 also existed inside the solder.
After 1 reflow, the Cu layer was consumed and converted to Cu6 Sn5 , while
the Ni(V) layer was intact as shown in Fig. 3.23(a). After 5 reflows, the mor-
phology of the IMC changed from a rounded scallop shape to an elongated
scallop or rod shape with an increase in the aspect ratio. The IMC was faceted
and some of them had broken away from the UBM. Since the 300 nm Cu layer
in UBM has been consumed after 1 reflow (the as-bumped condition), the vol-
ume of the Cu6 Sn5 IMC should not increase during the subsequent reflows.
However, the cross-sectional SEM image in Fig. 3.23(b) indicates that the
IMC volume does increase with the number of reflows, owing to the alloying
of Ni in the Cu6 Sn5 and its transformation to (Cu,Ni)6 Sn5 . EDX analyses
have confirmed the transformation. White patches were seen in the Ni(V)
SVNY339-Tu July 12, 2007 15:5

Copper–Tin Reactions in Thin-Film Samples 103

solder solder

Cu6Sn5 (Cu,Ni)6Sn5

(a) Ni (b)
Ni Sn

solder solder
(Cu,Ni)6Sn5 (Cu,Ni)6Sn5

(c) (d) Sn
Sn Ni

Fig. 3.23. (a) After 1 reflow, the Cu layer was consumed and converted to Cu6 Sn5 ,
while the Ni(V) layer was intact. (b) After 5 reflows, the cross-sectional SEM image
indicates that the IMC volume does increase with the number of reflows, owing to
the alloying of Ni in the Cu6 Sn5 and its transformation to (Cu,Ni)6 Sn5 . EDX analyses
have confirmed the transformation. White patches were seen in the Ni(V) layer, and
the patches were confirmed by EDX as mainly containing Sn and V. (c) After 10
reflows, the white patches dominated the Ni(V) layer. (d) After 20 reflows, the Ni(V)
layer disappeared and a layer of Sn separated the IMC from the Al layer. It is Sn
rather than Ni-Sn IMC that is found in place of the original Ni(V) layer. Some of the
intermetallic rods detached from the UBM and spalled into the solder.

layer shown in Fig. 3.23(b). The patches were confirmed by EDX as mainly
containing Sn and V. With the number of reflows increased to 10, the white
patches dominated the Ni(V) layer as shown in Fig. 3.23(c). After 20 reflows,
the Ni(V) layer disappeared and a layer of Sn separated the IMC from the
Al layer [Fig. 3.23(d)]. It is Sn rather than Ni-Sn IMC that is found to have
replaced the original Ni(V) layer. Some of the intermetallic rods detached
from the UBM and spalled into the solder. Clearly, the Cu/Ni(V)/Al UBM
becomes unstable with eutectic SnAgCu in multiple reflows.
The dissolution of Ni(V) by the molten Pb-free solder is nonuniform and
seems to have initiated on certain weak spots on the Ni(V) surface and spread
SVNY339-Tu July 12, 2007 15:5

104 Chapter 3

Fig. 3.24. SEM backscatter-

(Cu,Ni)6Sn5 ing images of the eutectic
SnAgCu solder on Al/Ni(V)/
Cu thin films annealed at
260◦ C for (a) 5 min, (b) 10
min, and (c) 20 min. The
nonuniform dissolution of the
Sn Ni Ni(V) layer and the formation
of white patches in the layer
(a) increased with annealing time.

(Cu,Ni)6Sn5

Ni Sn
(b)

(Cu,Ni)6Sn5

Sn
(c)

laterally to form patches. SEM backscattering images of the eutectic SnAgCu

solder on Al/Ni(V)/Cu thin films annealed at 260◦ C for 5, 10, and 20 min are
shown in Fig. 3.24(a) to (c), respectively. The nonuniform dissolution of the
Ni(V) layer and the formation of white patches in the layer increased with
annealing time.

3.8 Enhanced Spalling Due to Interaction across a

Solder Joint
The (Cu,Ni)6 Sn5 particles from the Cu/Ni(V)/Al thin-film UBM, as discussed
in the last section, can be induced to spall quickly by the interaction of the
metallization on the other interface of the solder joint. Figure 3.24 has already
shown such a case. Without metallization on the other side of a solder joint,
i.e., if we have just a bump of eutectic SnCuAg on Cu/Ni(V)/Al, spalling
SVNY339-Tu July 12, 2007 15:5

Copper–Tin Reactions in Thin-Film Samples 105

was observed after 20 reflows. With a metallization of Au/Ni(P) joined to the

other side of the bump, the spalling occurs after 5 reflows. Both SnPb and
Pb-free solder joints were studied and similar results were obtained.
In the molten solder, atomic diffusivity is about 10−5 cm2 /sec, hence it
takes only 10 sec for atoms to diffuse across a molten solder joint 100 μm in
diameter. Since there are Au, Ni, and P on the other side of the solder joint,
one of them could have enhanced the spalling. It turns out that if we replace
Au/Ni(P) by a piece of pure Ni on the other side of the solder joint, the
enhanced spalling of IMC occurs. The dissolution of pure Ni into the molten
solder enhances the dissolution of Cu6 Sn5 compound and exposes the Ni(V)
layer to the molten solder. The enhanced dissolution of Cu6 Sn5 is due to the
formation of (Cu,Ni)6 Sn5 . The increase in volume leads to a large compressive
stain.

3.9 Wetting Tip Reaction on Thin-Film-Coated

V-Grooves
The classic Young’s equation of an equilibrium wetting tip was derived by
minimizing the total surface and interfacial energies involved, but no free en-
ergy of formation of interfacial IMC was included. The wetting angle is defined
by the equilibrium condition among the surface and interfacial energies at the
wetting tip. Assuming that the wetting tip (or a wetting cap) configuration is
achieved instantaneously, the free energy of IMC formation may be ignored if
the rate of formation of the interfacial IMC is much slower than the spreading
rate for a drop of molten solder on a metal surface.
Here we shall present two phenomena in solder reactions wherein the wet-
ting tip is unstable. The first one is the wetting of molten SnPb on Pd and
Au. They show no stable wetting angle. Details of these wetting reactions will
be given in Chapter 7. In the case of eutectic SnPb on Pd, the tip advances
on the Pd surface unceasingly until the solder is consumed entirely [17]. In
the case of 95Pb5Sn solder on Au, the molten solder has a sunken interface
into Au which deepens with time [18]. In both cases, the wetting angle and
the tip configuration change with time. In the SnPb/Pd case, it is because of
rapid reaction to form IMC, and in the SnPb/Au case, it is because of rapid
dissolution of Au into the molten solder.
The second one concerns the wetting of molten eutectic SnPb cap on Cu.
While there is a stable wetting angle, the tip is unstable in the sense that it
grows a halo. The halo spreads out unceasingly due to the formation of a very
thin layer of IMC below the halo. The halo has also been found in front of a
molten tip of eutectic SnPb on Ni.
To study the effect of IMC formation on a reactive wetting tip, we must
study the very early stage of wetting reaction. It is possible to do so in
thin-film-coated V-grooves etched on a Si wafer surface. Using lithographic
SVNY339-Tu July 12, 2007 15:5

106 Chapter 3

Fig. 3.25. (a) Schematic diagram of the cross section of a V-groove and (b) corre-
sponding TEM image. (Courtesy of Prof. Ning Wang, Hong Kong University of Science
and Technology.)

technique, etched V-grooves along [110] direction on (001) surface of Si wafers

and coated with a bilayer of Cu/Cr film have been made. A schematic dia-
gram of the cross section of a V-groove and a corresponding TEM image are
shown in Fig. 3.25(a) and (b), respectively. Molten pure Pb will not run into
the V-groove, but the molten Pb(Sn) alloy having only 1 to 5% Sn will run
into it, driven by the horizontal capillary force, as shown in the lower part of
Fig. 3.26. The more Sn is present in the molten solder, the longer the length
of the run (or the faster the run). The length of the run shows a direct corre-
spondence to the wetting angle on Cu as a function of the Sn concentration
SVNY339-Tu July 12, 2007 15:5

Copper–Tin Reactions in Thin-Film Samples 107

Fig. 3.26. SEM images show that molten

pure Pb will not run into the V-groove, but
the molten Pb(Sn) alloy having only 1 to
5% Sn will run into it, driven by the hori-
zontal capillary force.

in the molten solder; see the SEM images in the upper part of Fig. 3.26.
While pure Pb unwets Cu, the Pb(Sn) alloy wets Cu and the wetting angle
decreases with increasing amount of Sn in Pb [19, 20]. Since the addition of
a few percent of Sn does not change the surface energy of the molten solder
[21], no change of wetting angle is expected on the basis of Young’s equation.
Hence, the change is due to Cu-Sn interfacial reaction in forming IMC. How
SVNY339-Tu July 12, 2007 15:5

108 Chapter 3

to calculate the change of wetting angle and in turn the wetting rate as a
function of solder composition is challenging. The wetting rate has been given
by the Washburn equation and can be measured using a CCD camera [22].
Knowing the rate, it is possible to estimate the rate of IMC formation in the
very early stage of the wetting reaction. However, for the melting solder to run
along the V-groove, the V-groove is typically kept at a temperature slightly
higher than the melting point of the solder. Then, it will take some time to
cool it to room temperature to measure the amount of IMC formation. Yet
during the cooling period, plenty of IMC can be formed at the wetting tip!
We cannot use the measured IMC to estimate how much IMC forms during
the initial wetting reaction because the cooling time is much longer than the
time of instantaneous wetting. However, it is difficult to run the molten solder
in a horizontal V-groove kept at room temperature. On the other hand, it can
be done if a room-temperature piece of Si with a coated V-groove is dipped
vertically into a pool of molten solder and the molten solder is allowed to rise
along the V-groove; after the piece touches the pool, it should be removed
from the pool quickly.

References
1. K. N. Tu, “Interdiffusion and reaction in bimetallic Cu-Sn thin films,”
Acta Metall., 21, 347 (1973).
2. J. W. Mayer, J. M. Poate, and K. N. Tu, “Thin films and solid-phase
reactions,” Science, 190, 228-234 (1975). J. M. Poate and K. N. Tu, “Thin
film and interfacial analysis,” Physics Today, May (1980), p.34.
3. K. N. Tu and R. D. Thompson, “Kinetics of interfacial reaction in bimetal-
lic Cu-Sn thin films,” Acta Metall., 30, 947 (1982).
4. K. N. Tu and J. W. Mayer, “Silicide formation,” in Thin Films—Inter-
diffusion and Reactions, J.M. Poate, K.N. Tu, and J.W. Mayer (Eds.),
John Wiley, New York (1978).
5. U. Gosele and K.N. Tu, “Growth kinetics of planar binary diffusion cou-
ples: Thin film case versus bulk cases,” J. Appl. Phys., 53, 3252 (1982).
6. G. V. Kidson, “Some aspects of the growth of different layers in binary
systems,” J. Nucl. Mater., 3, 21 (1961).
7. D. A. Porter and K. E. Easterling, “Phase Transformation in Metals and
Alloys,” Chapman & Hall, London (1992).
8. A. A. Liu, H. K. Kim, K. N. Tu, and P. A. Totta, “Spalling of Cu6Sn5
spheroids in the soldering reaction of eutectic SnPb on Cr/Cu/Au thin
films,” J. Appl. Phys., 80, 2774–2780 (1996).
9. C. Y. Liu, H. K. Kim, K. N. Tu, and P. A. Totta, “Dewetting of molten Sn
on Au/Cu/Cr thin film metallization,” Appl. Phys. Lett., 69, 4014–4016
(1996).
10. D. W. Zheng, Z. Y. Jia, C. Y. Liu, W. Wen, and K. N. Tu, “Size depen-
dent dewetting and sideband reaction of eutectic SnPb on Au/Cu/Cr thin
film,” J. Mater. Res., 13, 1103–1106 (1998).
SVNY339-Tu July 12, 2007 15:5

Copper–Tin Reactions in Thin-Film Samples 109

11. D. W. Zheng, W. Wen, K. N. Tu, and P. A. Totta, “In-situ scanning elec-

tron microscopy study of eutectic SnPb and pure Sn wetting on Au/Cu/Cr
multilayered thin films,” J. Mater. Res., 14, 745–749 (1999).
12. K. Puttlitz and P. Totta, “Area Array Technology Handbook for Micro-
electronic Packaging,” Kluwer Academic, Norwell, MA (2001).
13. G. Z. Pan, A. A. Liu, H. K. Kim, K. N. Tu, and P. A. Totta, “Microstruc-
ture of phased-in Cr-Cu/Cu/Au bump-limiting-metallization and its sol-
dering behavior with high Pb and eutectic SnPb solders,” Appl. Phys.
Lett., 71, 2946–2948 (1997).
14. J. W. Jang, L. N. Ramanathan, J. K. Lin, and D. R. Frear, “Spalling of
Cu3Sn intermetallics in high-Pb 95Pb5Sn solder bumps on Cu under-
bump-metallization during solid-state annealing,” J. Appl. Phys., 95,
8286–8289 (2004).
15. C. Y. Liu, K. N. Tu, T. T. Sheng, C. H. Tung, D. R. Frear, and P. Elenius,
“Electron microscopy study of interfacial reaction between eutectic SnPb
and Cu/Ni(V)/Al thin film metallization,” J. Appl. Phys., 87, 750–754
(2000).
16. M. Li, F. Zhang, W. T. Chen, K. Zeng, K. N. Tu, H. Balkan, and P.
Elenius, ”Interfacial microstructure evolution between eutectic SnAgCu
solder and Al/Ni(V)/Cu thin films,” J. Mater. Res., 17, 1612–1621 (2002).
17. Y. Wang and K. N. Tu, “Ultra-fast intermetallic compound formation
between eutectic SnPb and Pd where the intermetallic is not a diffusion
barrier,” Appl. Phys. Lett., 67, 1069–1071 (1995).
18. P. G. Kim and K. N. Tu, “Morphology of wetting reaction of eutectic
SnPb solder on Au foils,” J. Appl. Phys., 80, 3822–3827 (1996).
19. C. Y. Liu and K. N. Tu, “Morphology of wetting reactions of SnPb alloys
on Cu as a function of alloy composition,” J. Mater. Res., 13, 37–44 (1998).
20. C. Y. Liu and K. N. Tu, “Reactive flow of molten Pb(Sn) alloys in Si
grooves coated with Cu film,” Phys. Rev. E, 58, 6308–6311 (1998).
21. F. H. Howie and E. D. Hondros, J. Mater. Sci., 17, 1434 (1982).
22. J. A. Mann, Jr., L. Romero, R. R. Rye, and F. G. Yost, “Flow of simple
liquids down narrow V- grooves,” Phys. Rev. E, 52, 3967–3972 (1995).
SVNY339-Tu March 2, 2007 22:1

4
Copper–Tin Reactions in Flip Chip
Solder Joints

4.1 Introduction
In Chapters 2 and 3, we discussed solder reactions on bulk and thin-film Cu,
respectively. In those reactions, there is only one interface between the solder
and the Cu. However, in a solder joint there are two interfaces. Typically a
solder joint joins two pieces of Cu tube together as in plumbing, or it joins
two metallic contacts as in Si devices. These two interfaces in a flip chip solder
joint are not independent of each other from the point of view of interfacial
reactions.
In a flip chip solder joint, the solder bump joins a thin-film under-bump
metallization (UBM) on the Si chip side and a metallic bond-pad on the
substrate side. A schematic diagram of the cross section of a flip chip solder
joint was shown in Fig. 3.19. On the Si chip side, circuit lines are made of Al
or Cu interconnects. Between the interconnect line and solder bump, there is a
trilayer of Au/Cu/Cu-Cr thin films, and the contact area between the trilayer
thin film and the solder bump is defined by an opening in a dielectric film of
SiO2 . The trilayer film is called under-bump metallization (UBM). The shape
of the contact area between the UBM and solder bump is close to an octagon
rather than a circle because it is difficult to pattern a circle using lithographic
technique. On the substrate side, the circuit is typically made of Cu traces.
Between the Cu trace and solder bump, there is a bond pad, e.g., a very thick
Cu or electroless Ni(P) 10 to 20 μm in thickness and covered with a plated
Au layer. A solder mask is used to pattern the bond pad, which has a contact
area to the solder bump much larger than the contact area of the UBM on
the Si chip.
In this chapter, we introduce the concept that the reactions on the two
sides of a solder bump can influence each other. We shall illustrate that these
two interfaces, one between UBM and solder and the other between solder and
bond pad, are not independent of each other. This is because the chemical
diffusion of noble and near-noble metal atoms such as Cu or Ni in a molten
SVNY339-Tu March 2, 2007 22:1

112 Chapter 4

solder as well as in a solid-state solder are very fast. For example, it takes
just a few seconds for Cu atoms to diffuse from one side of a molten solder
joint to the other side of the joint, in other words, it is within the period
of a single reflow. Furthermore, in joining the chip to the substrate, thermal
stress is developed and then the joint will carry electric current in device
operation, so the effects of stress migration and electromigration will enhance
the interaction between the two interfaces across the solder joint. These effects
have become serious reliability issues, especially when they are combined, to
be discussed in later chapters. When a temperature gradient exists across a
solder joint due to joule heating, thermomigration tends to occur too. In this
chapter, the effect of interaction of Cu-Sn reactions across a solder joint on
its reliability will be emphasized.
Besides the two interfaces, the bulk of the solder bump is involved in the
interaction too. Eutectic solder in the solid state has a lamellar two-phase
microstructure. The two-phase eutectic structure has a very unique property,
namely, that the two phases are in equilibrium with each other independent of
their volume fraction. For example, the lamellar spacing in a eutectic struc-
ture is undefined. Thus, the lamellar spacing of each phase, or the volume
fraction of each phase, or the local composition of the eutectic solder, can
be changed without affecting the chemical potential of the eutectic structure.
In a eutectic two-phase mixture, a change of composition does not mean any
change of chemical potentials; it means instead just a change of local volume
fractions of the two phases. Thus, if we induce phase segregation in the eu-
tectic solder by an external driving force, for example, in thermomigration
or electromigration, it means a change of the gradient of volume fractions,
not a gradient of chemical potentials. Gradient of volume fractions is not a
driving force. Therefore, the segregation in two-phase mixtures such as eu-
tectic SnPb can be enormous, because there is no resistance to the change
since there is no change in chemical potential. Thus, when an external force
is applied to a solder joint, there is a microstructure instability in the bulk
of the solder. Typically, Pb is driven to the anode and Sn is driven to the
cathode in electromigration. Consequently, the Cu-Sn reaction at the cathode
is enhanced because of the continuing arrival of Sn. We shall emphasize this
unique reliability issue in Sections 4.4 and 4.5. We note that in eutectic SnPb
solder, the eutectic structure consists of Sn and Pb. But in Pb-free solders,
Sn and Cu6 Sn5 as well as Sn and Ni3 Sn4 also form eutectic structures too, so
the segregation of Cu6 Sn5 and/or Ni3 Sn4 in a Sn-based Pb-free solder can be
enormous.
Concerning the instability in the reactions, there is no doubt that the de-
sign of the dimensions of UBM, solder bump, and bond pad and the selection
of materials are very important in manufacturing flip chip technology. This
is not only because of the chip-to-substrate solder joint reaction to be dis-
cussed here, but also because the design and the selection will affect electric
current distribution in the joint, and in turn, will affect joule heating and elec-
tromigration and thermomigration. Hence, reliability concerns must be taken
SVNY339-Tu March 2, 2007 22:1

Copper–Tin Reactions in Flip Chip Solder Joints 113

into account in the design stage of the device. However, the optimal design
of a flip chip solder joint and intelligent selection of materials are beyond the
scope of this book. Nevertheless, a brief discussion will be given in Sections
4.8 and 9.6.2.

4.2 Processing a Flip Chip Solder Joint and

a Composite Solder Joint
The processing of flip chip solder joints starts with depositing the UBM fol-
lowed by electroplating a thick solder bump on the UBM. Figure 4.1 depicts
the steps (steps 1 to 9) of depositing and patterning the UBM and the plating
(steps 10 to 12) of a solder bump on a UBM and reflowing the cylindrical
bump to form a solder ball. We note that there is a TiW thin film below
the trilayer UBM. The continuous TiW film serves as electrodes during the
electroplating of the solder bump. After plating of the solder bump and re-
moval of the photoresist, the part of TiW which is uncovered by the plated
solder is etched away using the solder itself as etching mask so that the bumps
are insulated from each other electrically after the etching. We also note that
electroplating the solder bump requires a very thick photoresist of at least
50 μm. Since a thick photoresist is used to produce micro-electro-mechanical
systems (MEMS) devices on Si, it is available in many research laboratories.
After etching of the photo resist and the TiW, a first reflow will change the
plated cylindrical bump to a round solder ball. In the next steps the chip
having an array of solder balls is joined to a printed circuit board with an
array of bond pads by a second reflow, thus producing the flip chip samples.
The alignment between the array of solder balls on the chip and the array of
bond pads on the substrate is crucial. Typically a flip chip bonder is used.
However, the second reflow process has a built-in tolerance of misalignment.
When the solder ball melts, its liquid surface tension will pull and twist the
chip to achieve a near-perfect alignment in order to reduce the surface ten-
sion. This is a unique feature in “control-collapse-chip-connection” or C-4
process.
If the substrate is a ceramic module, the solder can be high-Pb having a
high melting point above 300◦ C so that a high reflow temperature is needed.
On the other hand, if the substrate is polymer-based such as FR4 board,
we have to use eutectic SnPb or Pb-free solder. Since eutectic solder will
cause spalling of intermetallic compound from thin-film UBM as discussed
in Chapter 3, either a thick UBM or a composite solder joint must be used
in order to overcome the spalling problem. In a composite solder joint, the
Au/Cu/Cu-Cr UBM and the high-Pb solder are kept on the chip side, but
a eutectic solder will be deposited on the bond pad on the polymer-based
substrate before joining the chip to the substrate. In Fig. 4.2, the processing
steps of a eutectic solder on the bond pad by stencil printing are shown. After
the reflow step to from a eutectic solder ball on all the bond pads, a step of
SVNY339-Tu March 2, 2007 22:1

114 Chapter 4

Fig. 4.1. The steps (steps 1 to 9) of depositing and patterning the UBM and plating
(steps 10 to 12) of a solder bump on a UBM and reflowing the bump to form a solder
ball. (Courtesy of Dr. J. W. Nah, UCLA.)
SVNY339-Tu March 2, 2007 22:1

Copper–Tin Reactions in Flip Chip Solder Joints 115

Fig. 4.2. The processing steps of a

eutectic solder on the bond pad by
stencil printing and caking.

caking will be followed. In caking, the eutectic solder balls will be pressed to
render a flat top surface. Then a flip chip bonder will be used to align the
chip and the substrate so that the high-Pb solder balls will sit on top of the
flattened eutectic solder plateaus; next, a low-temperature reflow will be able
to join the two solders together to form a composite solder joint. A schematic
diagram of the cross section of a pair of flip chip composite solder joint is
shown in Fig. 4.3.
The chip with 97Pb3Sn solder balls was flipped and assembled on the
substrate with 37Pb63Sn solder plateaus. The UBM on the chip side was
SVNY339-Tu March 2, 2007 22:1

116 Chapter 4

Fig. 4.3. Schematic diagram of the cross section of a pair of flip chip composite solder
joint.

sputtered TiW (0.2 μm)/Cu (0.4 μm)/electroplated Cu (5.4 μm) and the
bond pad on the substrate side was electroless Ni(P) (5 μm)/Au (0.1 μm).
The thickness of the Al metal line on the chip side was 1 μm and that of
the Cu metal line on the substrate side was 18 μm. For the flip chip assem-
bly, the typical reflow condition of 37Pb63Sn solder was applied with a peak
temperature of 220◦ and a dwell time of 90 sec in nitrogen atmosphere. The
composite solder joint was achieved by having the molten 37Pb63Sn solder
wet and coat the entire surface of the solid 97Pb3Sn solder ball. In Fig. 4.3,
a round coating of the eutectic on the high-Pb solder ball is depicted.

4.3 Chemical Interaction across A Flip Chip

Solder Joint
Typically, the Cu-Sn reaction on the Si chip side belongs to the class of solder
reactions with a thin film of Cu, as discussed in Chapter 3. But the Cu-Sn
reaction on the substrate side belongs to the class of solder reactions with
a bulk Cu, as discussed in Chapter 2. In flip chip solder joints, these two
kinds of reaction occur on the two sides of the joints, and they are about 100
μm apart. However, these reactions are not independent of each other. This
is because Cu and other noble and near-noble metals such as Au and Ni can
diffuse across a solder joint having a thickness of about 100 μm in a very short
time. For example, if we take the atomic diffusion in molten solder to be 10−5
cm2 /sec, we find that it will take only 10 sec for Cu atoms to diffuse across
the bump. In the reflow, the period within which the solder is in the molten
state is about 0.5 min, hence there is plenty of time for the Cu on both sides of
the solder bump to communicate with each other and can actually affect each
side’s reaction. In the solid state, noble and near-noble metal atoms diffuse
SVNY339-Tu March 2, 2007 22:1

Copper–Tin Reactions in Flip Chip Solder Joints 117

interstitially in Sn and Pb, with a diffusivity close to 10−8 cm2 /sec around
room temperature. Then it will take less than several hours for these atoms to
diffuse across a solder bump 100 μm in thickness. Yet the required reliability
test of solid state aging is 1000 hr at 150◦ C. Due to these fast diffusions in
the molten state as well as in the solid state, we cannot ignore the interaction
across a solder bump. Enhanced spalling of ternary intermetallic compound
of (Cu, Ni)6 Sn5 or (Ni, Cu)3 Sn4 occurs as shown in Fig. 1.13 and discussed in
Section 3.8.

4.4 Enhanced Dissolution of Cu-Sn IMC

by Electromigration
More serious interaction across a flip chip solder joint will come from elec-
tromigration. In a flip chip solder joint, electric current must go through the
chip and its packaging substrate; the contact of the UBM on the chip side
and the contact of the bond pad on the substrate side become respectively
the cathode and the anode, or vice versa, across a solder joint. Electromigra-
tion will dissolve Cu from the UBM and the Cu-Sn IMC at the cathode and
transport the dissolved Cu atoms to the anode and form Cu-Sn IMC at the
anode, or vice versa. Therefore, electromigration in the solder bump affects
chip–packaging interaction.
In Fig. 4.4, a set of five SEM images of the cathode contact in a composite
solder joint before and after electromigration are shown [1]. Figure 4.4(a) is
the image before electromigration, showing the TiW, Cu3 Sn and the matrix
of the high-Pb solder joint. The Cu3 Sn is found to be stable with the high-Pb
solder matrix in thermal aging without electric current stressing. However,
when a current density of 2.25 × 104 A/cm2 was applied and the direction
of electron current flow was from the chip side at the upper left corner into
the solder joint, some Cu6 Sn5 was formed below the Cu3 Sn after 3 hr, and
at the same time, some Cu above the Cu3 Sn was consumed and also in the
solder matrix, more Sn was found to have diffused from the anode side to
the cathode side [see Fig. 4.4(b)]. After 12 hr, more reaction has occurred at
the upper left corner and the Cu above the Cu3 Sn is completely gone and a
much thicker Cu6 Sn5 was formed below the Cu3 Sn [see Fig. 4.4(c)]. A small
void was found to have formed near the Cu6 Sn5 and Cu3 Sn interface. After
18 and 20 hr, a large void formed and it had extended all the way to the TiW
layer and the device failed with a very large increase in resistance [see Fig.
4.4(d) and (e), respectively]. Clearly, the chemical reaction at the cathode
was affected by electromigration. It is worth noting that at the upper right
corner of the contact in Fig. 4.4(d) and (e), neither dissolution of Cu nor
phase transformation similar to that at the upper left corner was found.
What is significant in the above observation is that a stable layered struc-
ture became unstable under electromigration. Typically, a layered morphology
SVNY339-Tu March 2, 2007 22:1

118 Chapter 4

(a) (b) e -Cu Cu

Cu3Sn Cu3Sn
Cu6Sn5
10 μm Sn

e- (d) e- Cu
(c) Cu

Cu3Sn Cu3Sn
Cu6Sn5
Sn Cu6Sn5 Sn

(e) e- Cu
A

Cu3Sn
A,
Cu6Sn5
Sn

Fig. 4.4. A set of five SEM images of the cathode contact in a composite solder joint
before and after electromigration. (a) The image before electromigration, showing the
TiW, Cu3 Sn, and high-Pb solder joint matrix.

formed in a diffusion couple, for example, is stable in constant-temperature

annealing. The layers may grow or shrink upon annealing, but the layered
morphology is maintained. Any perturbation on a flat interface is in prin-
ciple unstable and will be removed by ripening. On the other hand, such a
layered structure may become unstable under electromigration, particularly
when current crowding exists. The instability leads to the dissolution of the
Cu UBM as shown in Fig. 4.4 and in turn the failure of the solder joint.
In Chapters 8 and 9 we shall introduce the subject of electromigration,
the unique behavior of electromigraiton in flip chip solder joints, and the
interaction between chemical and electrical forces.
SVNY339-Tu March 2, 2007 22:1

Copper–Tin Reactions in Flip Chip Solder Joints 119

4.5 Enhanced Phase Separation in Solder Alloys by

Electromigration and Thermomigration
The so-called Soret effect refers to when a homogeneous single-phase alloy
becomes inhomogeneous under a temperature gradient [2]. We describe here
a similar but different effect in which a homogeneous eutectic two-phase alloy
becomes inhomogeneous under a temperature gradient, i.e., thermomigration.
The difference is that in the Soret effect the inhomogeneous alloy has created
a chemical potential gradient to resist the effect. But in a eutectic alloy un-
der thermomigration, there is no chemical potential gradient to resist phase
separation or redistribution of composition in the eutectic two-phase alloy.
We refer to this as the “eutectic effect on phase separation” induced by ther-
momigration or by electromigration. The subject of thermomigration will be
covered in Chapter 12.
Compositionally, what is unique in a eutectic system is that there is no
chemical potential gradient as a function of composition below the eutectic
temperature. In other words, the chemical potential is uniform and does not
depend on composition. For example, a schematic phase diagram of Sn-Pb
is shown in Fig. 4.5, and we consider a diffusion couple of 70Pb30Sn and
Temperature (°C)

A B

Fig. 4.5. Schematic phase diagram of a binary eutectic system of SnPb shows that
below the eutectic temperature, it will phase-separate into two primary phases having
a lamellar microstructure. The two lamellar phases are at equilibrium with each other,
independent of the lamellar spacing or their volume fraction. If we consider a diffusion
couple of “A” and “B” at 150◦ C, each of them has phase-separated into two lamellar
phases. Since they are at equilibrium, there is no chemical potential difference to drive
them to intermix under isothermal annealing at 150◦ C.
SVNY339-Tu March 2, 2007 22:1

120 Chapter 4

30Pb70Sn alloys annealed at 150◦ C. These two alloys are represented by the
two dots, A and B, on the constant-temperature line, the thick line, drawn
in Fig. 4.5. At ambient pressure and 150◦ C, it is a constant-pressure and
-temperature process, hence any composition along the broken line will de-
compose into the two end-phases of α and β at the two points indicated
by the arrows. Composition of each phase is determined by thermodynamic
equilibrium between them and is known from the phase diagram; they are the
primary or end phases. These two end phases are at equilibrium with each
other, independent of the amount of each phase. In the next section, it will
be shown that there is neither interdiffusion nor homogenization between the
diffusion couple annealed at ambient pressure and 150◦ C, except a minute
amount of ripening. This is because at 150◦ C, these two alloys in the diffusion
couple will phase-separate into the same two primary phases of Sn and Pb,
except that the amount of the two primary phases is different, which obeys
the level rule. They have the lamellar microstructure. However, the lamel-
lar spacing or lamellar thickness is not defined. In other words, the lamellar
thickness, or the amount of each phase, or the composition, can be changed
without affecting the equilibrium condition. To be precise, there is a change
of total lamellar interfacial energy.
Since these two primary phases are at equilibrium with each other, inde-
pendent of their amount, there is no driving force to homogenization upon
aging at a constant temperature, say 150◦ C.
However, if an external driving force is applied, such as a temperature
gradient in thermomigration or an electric field in electromigration, to a ho-
mogeneous two-phase eutectic alloy below the eutectic temperature, a very
large degree of phase separation or composition redistribution can be induced
because there will be no chemical potential change in phase separation, so
there is no chemical potential gradient to oppose the applied driving force as
in the Soret effect. In a eutectic two-phase mixture a change of concentration
does not mean any change of chemical potentials, but rather just a change of
local volume fractions of the two phases. Thus, if some segregation in solder
is induced by thermomigration or electromigration, it means a change of the
gradient of volume fractions, not of chemical potentials. Gradient of volume
fractions is not a driving force. Therefore, the segregation in two-phase mix-
tures such as eutectic SnPb can be enormous, compared with a single-phase
alloy such as PbIn, in which a change of composition will cause a counteract-
ing force due to concentration gradient. In principle, after a sufficiently long
duration of thermomigration or electromigration, the eutectic solder sample
could have achieved a complete phase separation into two parts. Actually
this has been observed in electromigration in flip chip solder joint of eutec-
tic SnPb. Figure 4.6(a) and (b) show respectively SEM images of the cross
section of a eutectic SnPb solder joint before and after electromigration at
5 × 103 A/cm2 at 160◦ C for 82 hr, and Sn has segregated to the anode side
and Pb to the cathode side. More discussion of the subject will be given in
Chapter 9. However, we note that when a large amount of Sn is driven to the
SVNY339-Tu March 2, 2007 22:1

Copper–Tin Reactions in Flip Chip Solder Joints 121

Pb

e- Sn

Fig. 4.6. SEM images of the cross section of a eutectic SnPb solder joint before (a)
and after (b) electromigration at 5 × 103 A/cm2 at 160◦ C for 82 hr. Sn has segregated
to the anode side and Pb to the cathode side. (Courtesy of Dr. Yi-Shao Lai, ASE,
Taiwan, ROC.)

cathode or the anode, it will react with Cu to form a large amount of IMC,
as shown in Fig. 4.4.
If we examine the binary phase diagram of Sn-Cu or Sn-Ni, we find that
Sn and Cu6 Sn5 as well as Sn and Ni3 Sn4 form eutectic couples. This means
that below their eutectic temperature, a large amount of these compounds
can be formed in a matrix of Sn, and the compounds will be in equilibrium
with the matrix. Indeed this has been observed in electromigration in solder
joints, as shown in Fig. 1.16. Since both Cu and Ni are being used as UBM in
solder joints, they can be dissolved by electromigration into the solder joint
and form a large amount of intermetallic compounds (IMC) near the anode,
especially the Pb-free solders which are Sn-based.
SVNY339-Tu March 2, 2007 22:1

122 Chapter 4

Fig. 4.7. SEM images of the in-

terface of the bulk couple before
and after annealing.

Before aging

After one week aging

SVNY339-Tu March 2, 2007 22:1

Copper–Tin Reactions in Flip Chip Solder Joints 123

4.6 Thermal Stability of Bulk Diffusion Couples

of SnPb Alloys
Experimentally, a bulk diffusion couple of alloys of 70Pb30Sn and 30Pb70Sn
was annealed at 150◦ C for 5 weeks. SEM images of the interface of the couple
before and after the annealing are shown in Fig. 4.7. Composition distribution
curves across the interface measured by electron micro-probe before and after
the annealing are shown in Fig. 4.8. Assuming an interdiffusion coefficient of
1 × 10−8 cm2 /sec, we expect to detect an interdiffusion zone 5 μm wide after
5 weeks. The detected width of the interface is much less than that, indicating
almost no interdiffusion.
Experimentally, in a composite solder joint consisting of 5Sn95Pb and
63Sn37Pb (eutectic), no interdiffusion or mixing of composition was found
after aging at 150◦ C at ambient pressure for several days. Figure 4.9 shows

Sn (wt %) composition distribution curves

100
Sn concentration (wt %)

80
60
40
20 Before Aging

0 After 5 weeks aging

0 500 1000 1500 2000
Distance starts from the middle of Pb-rich bulk sample to the
middle of Sn-rich bulk sample (μm)

Fig. 4.8. Composition distribution curves across the interface of the diffusion couple
measured by electron micro-probe before and after annealing.

a) b)

Fig. 4.9. SEM images of the cross sections of the 5Sn95Pb and eutectic SnPb com-
posite solder joint before (a) and after (b) aging. No substantial interdiffusion can be
detected.
SVNY339-Tu March 2, 2007 22:1

124 Chapter 4

SEM images of the cross sections of the composite solder joint before and after
aging. No substantial interdiffusion can be detected [3].
The results shown above are not unexpected. They were experiments under
constant temperature and constant pressure. When the Gibbs free energy
is minimized, the two phases have equal chemical potential, so there was
no driving force to homogenize them. Only when an electric current or a
temperature gradient is applied to the diffusion couple can interdiffusion occur
and lead to a change of volume fraction of the two phases.

4.7 Thermal Stress Due to Chip–Packaging

Interaction
To reduce resistance–capacitance delay in multilayered interconnect struc-
ture, ultralow-k materials with a value of k approaching 2 are being developed
for the integration with Cu conductors. The weak mechanical properties of
ultralow-k materials are of concern owing to thermal stress. The thermal stress
exists between Cu and ultralow k due to their difference in thermal expansion
coefficient, but also due to chip-to-packaging interaction. This is a relatively
new reliability issue. In Section 1.4.2, we discussed thermal-mechanical stress
across a flip chip due to the difference in thermal expansion coefficients of Si
and FR4 polymer substrate board. In flip chip technology, the Cu/ultralow-k
multilayered structure on a Si chip will be joined to a packaging substrate by
an area array of solder joints. Around the device operation temperature of
100◦ C, a thermal stress is developed between the Si chip and the substrate,
whether the latter is ceramic or polymer. The thermal stress will affect the
mechanical integrity of the solder bumps as well as the Cu/ultralow-k multi-
layered structure. When SiO2 was used as the interlayer dielectric, since SiO2
is mechanically rather strong, the load from the chip-to-packaging interaction
will be taken up by the solder bump which is softer. The thermal stress has led
to the well-known low cycle fatigue failure of flip chip solder bumps in device
applications. In the past, the microelectronics industry invented the epoxy
underfill to redistribute the thermal stress and reduce its effect on solder joint
failure. Moire interferometry has been developed to analyze the thermal stress
distribution in the flip chip solder joints [5–7].
However, when ultralow k is used as interlayer dielectric, the thermal stress
induced by the chip-to-packaging interaction will be shared between the solder
bump and the multilayer structure of Cu/ultralow k. The thermal stress may
crack the dielectric in the Cu/ultralow-k multilayered structure.

4.8 Design and Materials Selection of a Flip Chip

Solder Joint
In Chapter 3, Sections 3.3 to 3.7 presented a sequence of events of the selection
of UBM with respect to the change of solder alloy. Spalling of IMC has been
SVNY339-Tu March 2, 2007 22:1

Copper–Tin Reactions in Flip Chip Solder Joints 125

the issue. To overcome the spalling problem, very thick Cu UBM such as Cu
column bump has been studied, to be discussed in Section 9.6. Clearly, it is
better to take into account the reliability issues in the design stage of a device.
The design of the dimensions of a flip chip solder joint and the selection of
materials to form the joint depends on the application of the device and the
specifications from the device designer. For example, from the point of view
of electromigration, it is important to know the current carrying capacity
of a joint as required by the designer, then the current density distribution
in the circuit of the joint must be considered. Today, very powerful three-
dimensional simulation is capable of revealing any current crowding in the
circuit since it causes local joule heating and enhances electromigration. In
addition, the related temperature distribution due to joule heating and even
stress distribution can be obtained too. Nevertheless, the challenge to the
design is that reliability is a time-dependent event. The microstructure of a
joint will change in field use, so the current distribution will change with time.
Reliability is a dynamic problem!
Furthermore, up to now there has been no industrial standard of electro-
migration test of a flip chip solder joint. It will help greatly if a standard is
available. It is beyond the scope of this book to cover the details of the design
and selection rules of a flip chip solder joint. What is done here is to offer
some basic understanding of the reliability problems of solder joints, so that
designers may be aware of them in their circuit design.

References
1. J. W. Nah, K. W. Paik, J. O. Suh, and K. N. Tu, ”Mechanism of electromi-
gration induced failure in the 97Pb-3Sn and 37Pb-63Sn composite solder
joints,”J. Appl. Phys., 94, 7560–7566 (2003).
2. D. V. Ragone, “Thermodynamics of Materials,” Volume II, Chapter 8, John
Wiley, New York (1995).
3. A. Huang, Ph.D. dissertation, UCLA (2006).
4. J. W. Nah, UCLA, Personal communication.
5. Y. Gao, C. K. Liu, W. T. Chen, and C. G. Woychik, “Solder ball con-
nect assembles under thermal loading: 1. Deformation measurement via
Moire interferometry, and its interpretation,” IBM J. Res. Dev., 37, 635–
648 (1993).
6. D. Post, B. Han, and P. Ifju, “High Sensitivity Moire: Experimental Anal-
ysis for Mechanics and Materials,” Springer, Berlin (1994).
7. Z. Liu, H. Xie, D. Fang, H. Shang, and F. Dai, “A novel nano-Moire method
with scanning tunneling microscope,” J. Mater. Process. Technol., 148, 77–
82 (2004).
SVNY339-Tu March 3, 2007 7:44

5
Kinetic Analysis of Flux-Driven Ripening
of Copper–Tin Scallops

5.1 Introduction
In Chapter 2, we demonstrated that in the wetting reaction of a cap of
molten eutectic SnPb solder on a bulk Cu foil, the intermetallic compound
formation of Cu6 Sn5 takes a unique morphology of closely distributed scal-
lops. Morphology controls kinetics. The kinetics of growth of the scallops is
a supply-controlled reaction, rather than diffusion-controlled or interfacial-
reaction-controlled.
The scallops appear to have touched each other side-by-side and cover the
entire interface between the molten solder cap and the Cu. When the time of
wetting reaction is extended, the diameter of the cap does not increase, except
for the growth of a halo around the cap. Thus, the interfacial area between the
cap and Cu does not change. However, within the cap, the interfacial reaction
between the molten SnPb and Cu continues and the scallop grows, so the
average diameter of the scallop increases with time. Since the scallops are
touching each other, the growth of any one in diameter is a parasitic reaction;
it grows at the expense of its nearest neighbors, so the neighboring scallops
shrink. It is a ripening process. Figure 2.5 showed that on a fixed interfacial
area, while the average size of scallops has increased with time, the number of
scallops decreases with time. The ripening process is nonconservative, meaning
that the total volume of the scallops increases with time.
The volume increase is due to Cu-Sn reaction which grows the Cu6 Sn5
scallops by the diffusion of Cu into the molten solder. A very unique kinetic
behavior of the scallop-type morphology is that the scallop does not seem to
become a diffusion barrier to its own growth, unlike the growth of a layer-type
intermetallic compound. In a layered growth, it obeys a diffusion-controlled
growth; typically the thicker the layer, the better diffusion barrier it is to
its own growth. The layer thickness should have a square root dependence
on time, as discussed in Chapter 3. In scallop growth, the diameter of the
scallop was measured to have a cubic root dependence on time, quite similar
SVNY339-Tu March 3, 2007 7:44

128 Chapter 5

to a conservative ripening described by the LSW theory [1–3]. However, in the

classic ripening process described by the LSW theory, it is diffusion-controlled.
While the total volume of all the scallops increases with time, the base area
or the interfacial area between the scallops (or the cap) and the Cu is fixed.
This is a major constraint of the reaction; a constraint of constant interfacial
or base area. It is very interesting to note that if the scallops are assumed
to be hemispheres, it follows that the total surface area of the hemispherical
scallops is also fixed at twice the base area, independent of the size distri-
bution of the hemispheres. Therefore, we have a ripening process which has
a constant-surface area, rather than a constant volume. In the classic LSW
ripening, it proceeds under a constant-volume constraint and is driven by the
reduction of the total surface area. The scallop-type ripening proceeds under
a constant-surface-area constraint and is driven by the increase of the total
volume of the scallops, in other words, by the gain in the formation energy of
intermetallic compound.
In the scallop-type ripening process, there are two important constraints.
The first is that the reaction interface or the total surface of scallops is con-
stant. The second is conservation of mass, in which all the flux of Cu diffusing
into the molten solder is consumed by scallop growth to increase the aver-
age diameter of scallops, but at the same time it will reduce the number of
channels between scallops under the first constraint. This is the reason for
supply-controlled reaction or flux-driven ripening since the reaction rate de-
pends on the supply of the Cu flux. Here we define the “channel” between
two neighboring scallops to be a gap, not a grain boundary. In other words,
the width of the gap is several nanometers, not 0.5 nm of a grain boundary in
the solid state. A reduction of the number of channels will reduce the flux of
Cu needed for the growth. In experiments, scallops of Cu6 Sn5 were observed
to elongate in the growth direction after approximately 10 min of reaction
between molten eutectic SnPb and Cu at 200◦ C.
In the following, we shall first consider the stability of scallop-type mor-
phology in wetting reactions before we present a kinetic analysis of the distri-
bution and growth of scallops. On kinetics, first a simple model of growth of
mono-size scallops is presented to illustrate the basic idea. Second, a general
model of the distributional growth of different sized scallops in ripening will
be analyzed.

5.2 Morphological Stability of Scallop-Type IMC

Growth in Wetting Reactions
Why is morphology important in phase transformations? It affects the ki-
netic path in the transformation. In wetting reactions of molten solder on
Cu films, it is the morphological change that leads to “spalling” of IMC,
as discussed in Chapter 3. In wetting tip reactions, when surface energy or
SVNY339-Tu March 3, 2007 7:44

Kinetic Analysis of Flux-Driven Ripening of Copper–Tin Scallops 129

interfacial energy is considered, morphological stability or equilibrium of the

triple point at the wetting tip is conserved. Only when we are sure that the
scallop-type morphology is stable in wetting reactions will the kinetic analysis
of its ripening make sense [4]. In the following, we shall compare the morphol-
ogy and the kinetics of solid-state reactions to those of wetting reactions. In
solid-state reactions, a layer-type rather than a scallop-type morphology is
stable and kinetic analysis of the layer-type growth will be reviewed briefly
below [5].
In the classical analysis of solid-state interfacial reactions in a binary bulk
diffusion couple, the kinetics of growth of each IMC layer can be diffusion-
controlled or interfacial-reaction-controlled, as discussed in Chapter 3. For
a bulk diffusion couple of sufficient thickness at a high enough temperature
and after a very long time, we can have all the layered IMCs coexisting with
diffusion-controlled growth. The ratio of thickness among the layers is pro-
portional to the ratio of the square root of the interdiffusion coefficient in the
layers.
The analysis of reaction in bulk diffusion couples has been extended to
reactions between two metallic thin films and between a metallic thin film and
a silicon wafer. With the availability of modern analytical techniques, which
can detect IMC formation with atomic resolution, it was found that not all
the equilibrium IMCs would form in thin-film reactions [6, 7]. Actually, only
one of them forms if the reaction takes place at a low temperature. In the
case of a Ni film on a silicon wafer at 200◦ C, the phase of Ni2 Si will form
alone [8, 9]. When it consumes all the Ni, then NiSi will form in between
the Ni2 Si and Si. After the growth of NiSi has consumed all the Ni2 Si, the
phase NiSi2 will form in between the NiSi and Si, and the growth of NiSi2 will
consume all the NiSi finally. The reaction ends with a film of NiSi2 on Si. It is a
sequential formation of silicide phases; they form one-by-one. Using a model of
competing growth between two thin layers, e.g., Ni2 Si and NiSi between Ni and
Si, and employing both diffusion-controlled and interfacial-reaction-controlled
kinetics, Gosele and Tu were able to explain the phenomenon of “single phase
growth” in thin film reactions [10]. Their model defines a critical thickness or
a thin film thickness below which only one IMC can form. To verify the single-
phase growth phenomenon, a Ni film was evaporated on NiSi/Si to obtain a
sample of Ni/NiSi/Si and the sample was annealed. Instead of the growth of
NiSi, the Ni was found to react with NiSi to form Ni2 Si. The latter grew and
consumed all the NiSi and only after that was the sequential growth as stated
above repeated. How to predict which of the IMCs will be the first phase to
form, i.e., the Ni2 Si in the Ni/Si reaction, and what is the sequence of the
ensuing phase formations have been actively studied.
Many models exist to predict the first phase formation of solid-state inter-
facial reactions in thin films. Actually they are quite successful in predicting
the phase using equilibrium phase diagrams and applying the criterion of the
largest Gibbs free energy change or the largest driving force in IMC forma-
tion. However, a metastable phase, such as an amorphous alloy, which does
SVNY339-Tu March 3, 2007 7:44

130 Chapter 5

not exist in equilibrium phase diagrams, can be the first phase formation, as
in Rh/Si and Nb/Zr reactions [11, 12]. The metastable phase has a relatively
small Gibbs free energy of formation compared to that of the equilibrium
IMCs, so amorphous phase formation does not have the largest gain in Gibbs
free energy. Also, the first phase formation tends to depend on temperature.
For example, an amorphous phase cannot be the first phase formation if the
formation temperature is very high. At the moment, a plausible explanation
of the metastable phase formation or the selection of the first phase formation
and its temperature dependence is that it has the highest “rate” of Gibbs
free energy change [13]. It is the largest Gibbs free energy change in a short
time or the fastest rate of growth that determines the first phase formation.
Or, the phase which has the largest flux of interdiffusion tends to become
the first phase of formation. In other words, the first phase selection is based
on kinetics, not thermodynamics. However, we need kinetic data in order to
predict which phase has the highest rate of growth, but unfortunately kinetic
data are not available for most reactions.
In wetting reaction between eutectic SnPb and Cu, the morphology of
the Cu6 Sn5 is not layer-type, rather it has a scallop-type morphology. The
scallop-type morphology is stable as long as there is unreacted Cu. When the
thickness of Cu changes from a foil to a thin film, the nonconservative ripening
changes to a conservative ripening after the Cu film is completely consumed.
It becomes unstable and leads to spalling of IMC as we have discussed in
Section 3.3.
Now we turn to solid-state aging of the same system. Figure 2.23 showed
four SEM images of eutectic SnPb on a thick Cu substrate before and af-
ter aging at 170◦ C for 500, 1000, and 1500 hr. These samples were reflowed
twice at 200◦ C before the solid-state aging. In these figures, we observed that
both Cu6 Sn5 and Cu3 Sn compounds have layer-type morphology; their inter-
faces are rather flat. Specifically, the Cu6 Sn5 no longer has the channels in
the scallop-type morphology and the Cu3 Sn is very thick. Since the samples
were reflowed twice before aging, they must possess the scallops of Cu6 Sn5
in the initial stage of aging. Yet during the solid-state aging the morphology
of Cu6 Sn5 has changed from the scallop-type to layer-type. Naturally, we ask
why it does not maintain the scallop-type growth in the solid-state aging.
Figure 5.1 shows a top view of the surface of Cu6 Sn5 of a sample of eutectic
SnPb on Cu aged at 150◦ C for 2 months. The solder over the IMC was etched
away. The Cu6 Sn5 has a rather flat surface, and grain boundaries were formed
between grains.
Interestingly, if we wet the layered IMC by molten eutectic SnPb solder,
the scallop-type morphology of Cu6 Sn5 returns. Figure 5.2(a) shows a SEM
image of the cross section of a sample of eutectic SnPb on a Cu foil aged at
170◦ C for 960 hr. In the layered IMC, a few of the grains with vertical and flat
grain boundaries are seen, and a rather thick layer of Cu3 Sn formed below
the Cu6 Sn5 . Figure 5.2(b) shows a cross section of the sample after it was
reflowed at 200◦ C for 40 min. The scallop-type grains of Cu6 Sn5 reappeared
SVNY339-Tu March 3, 2007 7:44

Kinetic Analysis of Flux-Driven Ripening of Copper–Tin Scallops 131

Fig. 5.1. SEM image of the top view of the surface of Cu6 Sn5 of a sample of eutectic
SnPb on Cu aged at 150◦ C for 2 months. The solder over the IMC was etched away.
The Cu6 Sn5 has a rather flat surface, and grain boundaries were formed between grains.
(Courtesy of Jong-ook Suh, UCLA.)

with a curved surface. In Fig. 5.3(a), the morphology of a layer-type IMC of

another sample after a solid-state aging at 130◦ C for 480 hr is shown. While
its top surface is rough, the grains are columnar, not scallop-type, and no
channels exist between the grains. When this sample was reflowed at 200◦ C
for only 1 min, the columnar grains transformed back to scallops, as shown
in Fig. 5.3(b). Both Figs. 5.2 and 5.3 reveal that the scallop-type grains are
stable in contact with molten solder, but the layer-type grains are stable in
contact with solid solder.

5.2.1 Analysis of Morphological Stability of Scallops in

Wetting Reactions
The formation of layer-type IMC in solid-state reactions is common. What is
uncommon here is why the layer-type IMC transforms to scallop-type IMC
when it is wetted by molten solder. The transformation indicates that the
scallop-type morphology is thermodynamically stable in wetting reactions.
We consider such a transformation in Fig. 5.4, where the solid lines represent
a schematic diagram of the cross section of a layer-type Cu6 Sn5 . The broken
lines represent the wetting of the top surface and grain boundaries of Cu6 Sn5
grains by molten solder. The change in interfacial and grain boundary energies
SVNY339-Tu March 3, 2007 7:44

132 Chapter 5

Fig. 5.2. (a) SEM image of

(a) the cross section of a sam-
ple of eutectic SnPb on a Cu
foil aged at 170◦ C for 960
hr. In the layered IMC, a
few of the grains with ver-
tical and flat grain bound-
aries are shown, and a rather
thick layer of Cu3 Sn formed
below the Cu6 Sn5 . (b) Cross-
sectional image of the sam-
ple after reflowing at 200◦ C
for 40 min. The scallop-type
grains of Cu6 Sn5 reappeared
and the IMC grains had a
curved surface.
(b)

is given as

1
2πrhσGB + πr2 σSS ≥ 2πrhσLS + πr2 σLS ,
2
where σGB , σSS , and σLS represent grain boundary energy in Cu6 Sn5 , interfa-
cial energy between solid solder and Cu6 Sn5 , and interfacial energy between
molten solder and Cu6 Sn5 , respectively, and r and h are respectively the ra-
dius and height of the solid Cu6 Sn5 grains. The factor 1/2 appears in the first
term on the left-hand side of the inequality equation because a grain bound-
ary is shared by two grains. The sum of energies on the left-hand side of the
equation is greater than that on the right-hand side.
We compare morphological stability in wetting reaction to that in solid-
state aging. In wetting reaction, Fig. 5.5(a), we represent the stable scallop-
type morphology by the solid lines and the unstable layer-type morphology
by the broken lines, and we have

1
2πR2 σLS ≤ 2πrhσGB + πr2 σLS ,
2
where R is the radius of the scallop. For the hemispherical scallop, we assume
SVNY339-Tu March 3, 2007 7:44

Kinetic Analysis of Flux-Driven Ripening of Copper–Tin Scallops 133

Fig. 5.3. (a) The morphology of a layer-type IMC of a eutectic SnPb on a Cu foil after
solid-state aging at 130◦ C for 480 hr. The grains are not scallop-type and no channels
between the grains can be found. (b) When this sample was reflowed at 200◦ C for only
1 min, the grains transformed back to scallops.

that the interfacial energy σLS is isotropic. In solid-state aging, Fig. 5.5(b), we
represent the stable layer-type morphology by the solid lines and the unstable
scallop-type morphology by the broken lines and we have

1
2πR2 σSS ≥ 2πrhσGB + πr2 σSS .
2

sSS Molten Solder

Cu6Sn5 sLS Cu6Sn5

sGB

Fig. 5.4. Schematic diagram of the transformation of a layer to a scallop-type Cu6 Sn5 .
The cross section of a layer-type Cu6 Sn5 is depicted by the solid lines. The broken
lines represent the wetting of the top surface and grain boundaries of Cu6 Sn5 grains
by molten solder.
SVNY339-Tu March 3, 2007 7:44

134 Chapter 5

(a) sLS Fig. 5.5. (a) In wetting re-

action, the transformation of
a layered to a scalloped IMC.
R sGB The stable scallop-type mor-
h
phology is represented by the
solid lines and the unstable
r layer-type morphology by the
broken lines. (b) In solid-
state aging, the transforma-
sSS tion of a scalloped to a layered
(b) IMC. The stable layer-type
morphology is represented by
R the solid lines and the un-
h sGB
stable scallop-type morphol-
ogy by the broken lines.
r
In the above two inequalities, we have assumed a constant-volume transfor-
mation between a cylindrical grain and a hemispherical grain, so that we need
not consider volume energy but only interfacial and grain boundary energies.
To simplify all of the above inequality equations, we assume that the radius
of the cylindrical grain is equal to the radius of the hemispherical grain (i.e.,
r = R). Alternatively, we can also assume that the height of the cylindrical
grain is equal to its radius, r = h, and the result is similar. We obtain

2
h= r.
3

Substituting this relation into the last three inequality equations, we have the
following two inequalities:

7
σSS ≥ σLS ,
6
2
σSS ≥ σGB ≥ σLS .
3

The first inequality equation shows that the interfacial energy between a
molten solder and Cu6 Sn5 is less than that between a solid solder and Cu6 Sn5 .
The second inequality equation implies that the energy of large-angle grain
boundaries in Cu6 Sn5 must be quite high. When a molten solder reacts with
the layer-type Cu6 Sn5 , the reaction wets the large-angle grain boundaries in
the layer-type Cu6 Sn5 , as depicted in Fig. 5.4. The interfaces between the solid
solder and Cu6 Sn5 and the grain boundaries in Cu6 Sn5 can be replaced by the
low-energy interfaces of the molten solder and Cu6 Sn5 . Therefore, the scallop-
type morphology persists in wetting reactions and the layer-type morphology
persists in solid-state aging.
SVNY339-Tu March 3, 2007 7:44

Kinetic Analysis of Flux-Driven Ripening of Copper–Tin Scallops 135

The stability is due to minimization of interfacial and grain boundary

energies. The interfacial energy between a molten solder and Cu6 Sn5 is less
than that between a solid solder and Cu6 Sn5 , so the scallop-type morphology
of grains persists in wetting reactions and the layer-type morphology persists
in solid-state aging. When the molten solder reacts with the layer-type Cu6 Sn5
and transforms it back to scallop-type Cu6 Sn5 , the reaction must begin by
wetting the large-angle grain boundaries in the layer-type Cu6 Sn5 , as depicted
in Fig. 5.4. It implies that the energy of large-angle grain boundaries in Cu6 Sn5
must be quite high, so they can be replaced by the low-energy interfaces of
the molten solder and Cu6 Sn5 .
During the wetting reaction, the neighboring scallops will not join together
to form grain boundaries. Therefore, they can only grow bigger by ripening. In
ripening, the total surface area of the hemispherical scallops is conserved. On
the other hand, the scallops cannot keep on growing to become very big. This
is because the bigger the scallops, the fewer the channels in between them.
Since channels are the short circuit paths for Cu to reach the molten solder,
the scallop-type growth has to slow down when the channels are reduced. In
order to keep the channels, the scallops will elongate, and very long scallops
have been observed in extended wetting reactions.

5.3 A Simple Model for the Growth of Mono-Size

Hemispheres
Figure 5.6(a) is a schematic diagram of the cross section of an array of hemi-
spherical Cu6 Sn5 scallops grown on Cu, represented by the solid curves. The
following assumptions are made to analyze the kinetics of scallop growth [14–
16].

(a) The presence of Cu3 Sn and Pb in the reaction is ignored for convenience.
(b) A liquid channel exists between two scallops, the depth of which reaches
the Cu surface. The width of the channel “δ” is assumed to be small
compared to the radius of scallops. The morphology of scallops and chan-
nels is assumed to be thermodynamically stable in the presence of molten
solder. The channels serve as rapid diffusion paths for Cu to go into the
molten solder to grow the scallops. Although the scallops are separated
by channels, they remain in close contact with each other. Figure 5.6(b) is
a cross-sectional transmission electron microscopic image of Cu6 Sn5 scal-
lops, channels, and a thin layer of Cu3 Sn on Cu. The channel, as indicated
by an arrow, has a width less than 50 nm.
(c) The shape of the scallops is represented by hemispheres. On a given in-
terfacial area of “S total ” between the scallops and the Cu, the total
surface area between all hemispherical scallops and molten solder is just
twice “S total .” In Fig. 5.6(a), if we represent the cross section of a single
large hemispherical scallop by the broken half circle, its surface is 2 S total ,
SVNY339-Tu March 3, 2007 7:44

136 Chapter 5

(a)

Jout

Jin

R1 R2 R4
R3

N
2πR2 ≅ Σ 2πRi2 ≅
i=1
2Stotal

(b)

1 μm

Fig. 5.6. (a) Schematic diagram of the cross section of an array of hemispherical
Cu6 Sn5 scallops grown on Cu, represented by the solid curves. (b) Cross-sectional
transmission electron microscopic image of Cu6 Sn5 scallops, channels, and a thin layer
of Cu3 Sn on Cu. The channel, indicated by the arrow, has a width less than 50 nm.
SVNY339-Tu March 3, 2007 7:44

Kinetic Analysis of Flux-Driven Ripening of Copper–Tin Scallops 137

the same as the sum of the surfaces of the smaller scallops represented by
the solid curves. Hence, while the growth increases the total volume, it
does not change the total surface area of the scallops.
(d) By conservation of mass, all the influx of Cu from the Cu substrate is
consumed by the growth of the scallops. Hence, the outflux of Cu from
the ripening zone into the bulk of the molten solder is assumed to be
negligible.
In this ripening process, there are two important constraints. The first
constraint is that the interface of reaction is constant. When the scallops are
assumed to be hemispherical, it also means that the total surface area of scal-
lops is constant. Therefore, we have a ripening process which proceeds under
constant surface with increasing volume. The second constraint is conserva-
tion of mass, in which all the influx of Cu is consumed by scallop growth.
In the classical LSW ripening, the process proceeds under almost a constant
volume, so the decrease of surface (and surface energy) provides the driving
force.
Strictly speaking, mono-size distribution (size distribution as Dirac’s
δ-function) is not compatible with scallop growth since scallop growth is par-
asitic and dependent upon the shrinkage of neighboring smaller scallops. It is
an unrealistic model and at the best, it approximates a narrow distribution
function of scallops. The big advantage of this model is that we change a
many-body problem to a one-body problem. We shall show below that the
mono-size approximation is good for a rough estimate of average values of the
kinetics.
According to the first constraint that the interface between the scallops
and Cu is occupied completely by scallops except the thin channels, we have

N πR2 ∼
= S total = const, (5.1)

where N is the number of scallops and R is radius. The free surface (the
cross-sectional area of channels at the bottom) for the supply of Cu from the
substrate is

δ δ
S free = N 2πR = S total , (5.2)
2 R

where δ is the channel width. Thus, the free surface decreases during the
scallop growth as 1/R. The volume of the reaction product of IMC scallops
is equal to

2π 3 2
Vi = N R = S total R. (5.3)
3 3

According to the second constraint that all influx of Cu atoms diffusing

into the molten solder from the substrate is consumed by the growth of IMC
SVNY339-Tu March 3, 2007 7:44

138 Chapter 5

Fig. 5.7. The meaning of C e and

C b in Eq. (5.5).

scallops due to conservation of mass, we have

dVi
ni Ci = J in S free . (5.4)
dt

Here ni is atomic density in IMC, i.e., number of atoms per unit volume, and
Ci is atomic fraction of Cu in IMC, which is 6/11 in Cu6 Sn5 . The influx is
taken approximately as
 
Ce + α
− Cb
J in = −nD R
, (5.5)
R

where α = (2γΩ/RG T )C e , γ is isotropic surface tension at IMC/melt interface,

Ω is molar volume, RG is gas constant, and T is temperature. The meaning
of C e and C b is defined in Fig. 5.7. In our case α ∼= 4.4 × 10−7 cm.
First, we consider the case α/R  C b − C e , so that

Cb − Ce
J in ∼
= nD , (5.6)
R

where n is atomic density or number of atoms per unit volume in the melt or
molten solder. Then, substituting Eqs. (5.6), (5.2), and (5.3) into the balance
Eq. (5.4), we obtain
e  
2 dR Cb − C δ total
ni Ci S total = nD S , (5.7)
3 dt R R

which immediately gives

R3 = kt, (5.8a)
9 n D(C b − C e )δ
k= . (5.8b)
2 ni Ci
SVNY339-Tu March 3, 2007 7:44

Kinetic Analysis of Flux-Driven Ripening of Copper–Tin Scallops 139

Note that the surface tension is absent in the expression for the rate constant,
despite the “ripening-like” time law.
If we take n/ni ≈ 1, Ci = 6/11, D ≈ 10−5 cm2 /sec, δ ≈ 5 ∗ 10−6 cm, C b −
C ≈ 0.001, where the concentration C b is taken for equilibrium of melt with
e

the Cu3 Sn phase, the rate constant k ≈ 4 ∗ 10−13 cm3 /sec. For example, for
annealing time t = 300 sec, it gives R ≈ 5 ∗ 10−4 cm, which agrees very well
with experimental data.

5.4 Theory of Nonconservative Ripening with a

Constant Surface Area
For scallops having a distribution of size, we let f (t, R) be the size distribution
function of scallops, so that the total number of scallops is equal to [15]

∞
N (t) = f (t, R)dR, (5.9)
0

and the average values are

∞
m 1
<R >= Rm f (t, R)dR. (5.10)
N
0

The first constraint of constant interface takes the form

∞
πR2 f (t, R)dR = S total − S free ∼
= S total = const. (5.11)
0

The cross-sectional area of all channels for copper supply is

∞
free δ
S = 2πRf (t, R)dR. (5.12)
2
0

The total volume of the growing scallops is

∞
2 3
Vi = πR f (t, R)dR. (5.13)
3
0

According to the second constraint, we have

dVi
ni Ci = J in S free . (5.14)
dt
SVNY339-Tu March 3, 2007 7:44

140 Chapter 5

Here ni is atomic density in IMC, i.e., number of atoms per unit volume, and
Ci is atomic fraction of Cu in IMC, which is 6/11 in Cu6 Sn5 . The influx is
taken approximately as
 
Ce + α
− Cb
J in
= −nD R
, (5.15)
R

where n is atomic density in solder, α = (2γΩ/kG T )C e , γ is isotropic surface

tension at IMC/melt interface, Ω is molar volume, kG is Boltzmann’s constant,
and T is temperature. C e and C b are defined as the equilibrium concentration
(the atomic fraction of Cu in molten solder) on a flat surface and concentration
at the entrance of the channels (corresponding to the concentration of Cu in
the molten solder on the substrate surface).
Since scallops must grow and shrink atom by atom, the distribution func-
tion should satisfy the usual continuity equation in the size space:

∂f ∂
=− (f uR ), (5.16)
∂t ∂R

where the velocity in the size space, uR , is simply the growth rate of scallops
with radius R and is determined by the flux density, j(R), on each individual
scallop. The expression for j(R) is usually found as a quasi-stationary solution
of diffusion problem in infinite space around a spherical grain with fixed super-
saturation < C > −C e at infinity. Then

dR j(R) n D < C > −(C e + α/R)

uR = =− = . (5.17)
dt ni Ci ni Ci R

While this expression is good for LSW theory, it is not good for the present
case because the scallops are on an interface and the diffusion distance between
them is of the same order of magnitude of the size of scallops. On the other
hand, the diffusion in the melt is very fast, so we suggest that the expression of
j(R) can be obtained by assuming that the flux on (or out of) each individual
scallop should be proportional to the difference between the average chemical
potential of copper μ in reaction zone (we take it to be the same everywhere—
mean-field approximation) and the chemical potential at the curved scallop–
β
melt interface, μ∞ + R , and β = 2γΩ:

   
β dR L β
−j(R) = L μ − μ∞ − , = μ − μ∞ − , (5.18)
R dt ni Ci R

where the parameters L, β, μ − μ∞ are determined self-consistently from the

above-mentioned two constraints of constant surface, Eq. (5.1), and mass
SVNY339-Tu March 3, 2007 7:44

Kinetic Analysis of Flux-Driven Ripening of Copper–Tin Scallops 141

conservation, Eq. (5.4). We obtain

 
dR k 1 <R>
uR = = 1− , (5.19)
dt 9 < R2 > − < R >2 R
9 n
k= D(C b − C e )δ. (5.20)
2 ni Ci

During the nonconservative flux-driven ripening, the rate of growth/shrinkage

of each scallop is determined not only by diffusivity and the average size of
all scallops, < R >, but also by the capacity of channels to supply Cu for the
reaction.
Thus, in the mean-field approximation, the basic equation for the distri-
bution function has the following form:
  
∂f k <R> ∂ 1 1
=− f − , (5.21)
∂t 9 < R2 > − < R >2 ∂R <R> R

where the rate coefficient k is determined by the incoming flux conditions,

which in turn are determined by the channels.
The formal solution of the distribution function is
⎧ ⎫
⎪
⎨ 
R/(bt)1/3
⎪
⎬
B R 3 − 4ξ
f (t, R) = exp dξ
bt (bt)1/3 ⎪
⎩ ξ − 3ξ + 9/4 ⎪
2
⎭
0

B R
= ϕ(η), τ = bt, η = 1/3
, (5.22)
τ (bt)
S total
B = ∞ . (5.23)
π ξ 2 ϕ(η)dη
0

The parameter b should be found self-consistently.

A standard integration gives
 
3
ϕ(η) = 0, η > ,
2
   
η 3 3
ϕ(η) =  4 · exp − , 0 < η < . (5.24)
2 −η
3 2
3
2 −η

A plot of ϕ(η) versus η is shown in Fig. 5.8. Thus, we have a unique asymptotic
solution which satisfies the universal scaling expression in Eq. (5.15). Also we
have

< η 2 > − < η >2 ∼

= 0.0615, < η > = 3/4, < η 3 > ∼
= 0.5535.
SVNY339-Tu March 3, 2007 7:44

142 Chapter 5

. ϕ (η )versus h
.
.

Fig. 5.8. Plot of ϕ(η) versus η.

k
The parameter b = 9(<η2 >−<η> ∼ k ∼ k
2 ) = 0.5535 = <η 3 > .

Hence, average cube of grain size is equal to

< R3 >=< ξ 3 > bt ∼

= kt.

The averaged size will be

 1/3
1/3 ∼ 3 k ∼ 1/3
< R > = < ξ > (bt) = t = 0.913 (kt) . (5.25)
4 0.5535

If we take n/ni ≈ 1, Ci = 6/11, D ≈ 10−5 cm2 /sec, δ ≈ 5 ∗ 10−6 cm, C b − C e ≈

0.001, where the concentration C b is taken for equilibrium of melt with the
Cu3 Sn phase, the rate constant k ≈ 4 ∗ 10−13 cm3 /sec. For example, for an-
nealing time t = 300 sec, it gives R ≈ 5 ∗ 10−4 cm, which agrees very well with
experimental data. Rate of ripening and growth of R is determined by the
incoming flux condition.

5.5 Size Distribution of Scallops

The morphology of Cu6 Sn5 IMC was found to be highly faceted when the
solder is pure tin, and rounded when the solder has a composition near
SVNY339-Tu March 3, 2007 7:44

Kinetic Analysis of Flux-Driven Ripening of Copper–Tin Scallops 143

the eutectic. The dependence of IMC morphology on solder composition

and reaction temperature has been investigated systematically, prior to mea-
surement of size distribution and growth rate according to reaction time.
This is because morphology of IMC will affect the kinetic path of ripening
among IMC scallops, and the FDR theory assumed that IMC scallop shape is
hemispherical.
To investigate the morphology of IMC formation in SnPb/Cu reactions,
samples with different solder composition were prepared: pure Sn, 90Sn10Pb,
80Sn20Pb, 70Sn30Pb, 60Sn40Pb, 50Sn50Pb, 40Sn60Pb, 30Sn70Pb, and
20Sn80Pb [17]. Solder with Pb concentration higher than 90 wt% Pb was
not investigated since Cu3 Sn will form instead of Cu6 Sn5 . These solder alloys
were cut into small pieces weighing about 0.5 mg, and melted in mildly ac-
tivated flux (197 RMA) to form a spherical bead. Cu foil (99.999%) was cut
into 1 cm × 1 cm square pieces, 1 mm thick. Each Cu piece was mechanically
polished down to colloidal silica to reduce surface roughness, cleaned ultra-
sonically with acetone, followed by methanol and DI water to remove organic
contaminants on the surface, etched with 5% HNO3 +95% H2 O for 15 seconds
to remove native oxide, and rinsed with DI water and followed by drying with
nitrogen gas. Then these Cu pieces were quickly immersed into hot 197 RMA
flux.
Solder wetting on Cu was prepared by dropping the small solder bead on
the polished Cu foil at 20 ◦ C above melting temperature of each alloy for
2 min. The 55Sn45Pb solder was selected for measurement of size distribution
and growth rate of IMC for 30 sec, 1 min, 2 min, 4 min, and 8 min, at
200◦ C. Reaction time over 10 min was not investigated because of elongation
of scallops.
To observe IMC scallops in plan view, the unreacted solder was removed
by mechanical polishing, followed by selective chemical etching. The selec-
tive etching was performed by using 1 part nitric acid, 1 part acetic acid,
and 4 parts glycerol at 80◦ C. Since overetching or underetching of solder can
change the IMC morphology, different etching times were applied repeatedly
to confirm the correct morphology.

5.5.1 Dependence of Cu6 Sn5 Morphology on Solder Composition

Figure 5.9 shows the change of IMC morphology as a result of reaction between
copper and SnPb solder of different compositions. The observed Cu6 Sn5 IMC
morphology is summarized in Table 5.1. When the Pb content of SnPb solder
composition was higher than 70 wt%, faceted scallops were observed all over
the sample together with some round scallops. When the Pb concentration of
solder was within the range from 60 wt% to eutectic composition (34 wt%),
only round scallops were observed. When the Pb concentration was further
decreased down below 30 wt%, again faceted scallops were observed together
with some round scallops. Finally, when the solder became pure tin, only
faceted IMC was observed.
SVNY339-Tu March 3, 2007 7:44

144 Chapter 5

(a) (b) (c)

(d) (e) (f)

(g) (h)

Fig. 5.9. SEM images of Cu6 Sn5 IMC formed by wetting reaction with copper and sol-
ders having compositions of (a) 20Sn80Pb, (b) 30Sn70Pb, (c) 40Sn60Pb, (d) 50Sn50Pb,
(e) 70Sn30Pb, (f) 80Sn20Pb, (g) 90Sn10Pb, and (h) pure Sn.

For eutectic SnPb solder, some samples showed only round scallops. But
in other samples, clusters of faceted scallops were observed at the center of
solder cap. This also took place when the solder composition was 60Sn40Pb.
When a small piece (0.5 mg) was taken from a solder chunk (∼3 g), small
inhomogeneity of the solder chunk may cause a 1–2 wt% change of composition
in the small piece, in extreme cases. Since the eutectic SnPb solder cap on
Cu has a wetting angle of 11o , the amount of solder covering IMC is much
thinner at the edge. At the edge of solder cap, the consumption of Sn from
the molten solder due to IMC growth will result in higher concentration of
Pb. This effect may change IMC morphology across the solder cap, from the
center to the edge.
To accurately determine the IMC morphology of eutectic SnPb solder, sol-
ders with different compositions near the eutectic temperature were melted
to control the liquidus phase composition of solder. Solders of 80Sn20Pb,
50Sn50Pb, 30Sn70Pb, and eutectic (63Sn37Pb) composition were reacted with
copper at 0.5◦ C above eutectic temperature (183.5◦ C). The temperature con-
trol was within ±1◦ C. The solders went through partial melting, and the com-
position of liquidus phase became very close to the eutectic composition. As
 SVNY339-Tu
March 3, 2007
7:44

Table 5.1. Summary of the observed Cu6 Sn5 IMC morphology.

Solder Composition 20Sn80Pb 30Sn70Pb 40Sn60Pb 50Sn50Pb 60Sn40Pb 63Sn37Pb 70Sn30Pb 80Sn20Pb 90Sn10Pb 100Sn

Reaction 295 275 255 235 200 200 210 225 240 250
Temperature (◦ C)
Morphology Facet Facet Round Round Round Round Facet Facet Facet Facet
Kinetic Analysis of Flux-Driven Ripening of Copper–Tin Scallops
145
SVNY339-Tu March 3, 2007 7:44

146 Chapter 5

(a) (b)

(c) (d)

Fig. 5.10. SEM images of Cu6 Sn5 IMC formed by wetting reaction between copper
and solders of (a) 80Sn20Pb, (b) 63Sn37Pb, (c) 50Sn50Pb, and (d) 30Sn70Pb compo-
sition at 0.5◦ C above eutectic temperature (183.5◦ C). The temperature control was
within ±1◦ C. The solders went through partial melting, and the composition of liq-
uidus phase came very close to the eutectic composition.

shown in Fig. 5.10, the morphology of scallops had round shape, regardless of
solder composition. Thus the true morphology of Cu6 Sn5 when SnPb solder
composition is eutectic is smooth round morphology.
The general idea of classic theories on the formation of faceted or rounded
liquid-solid interface is that if the interface is faceted, the adatoms at the
surface of solid phase tend to fill nearly all the available surface sites before
advancing to the next atomic layer, resulting in atomically flat interface with
small number of kink sites. [18, 19] If the crystal surface is round, the interface
is more or less atomically rough, possessing a large number of kink sites. To
have round shaped scallops, they should have many atomic steps and kinks
at the surface. Since step and kink have many broken bonds, scallops tend to
have more of them at the surface when the broken bond energy is low. The
broken bond energy is low when the IMC/solder interfacial energy is low. The
IMC/solder interfacial energy can be estimated by considering the wetting
angle between solder and copper.
When solder wets copper, copper surface is replaced by IMC/solder inter-
face. Thus IMC/solder interfacial energy is low when solder/copper wetting
angle is low, because the molten solder will prefer to spread out to increase
the IMC/solder interfacial area. C.Y. Liu investigated effect of SnPb solder
composition on the wetting angle of molten SnPb on Cu and found the lowest
SVNY339-Tu March 3, 2007 7:44

Kinetic Analysis of Flux-Driven Ripening of Copper–Tin Scallops 147

Fig. 5.11. Plot of ln(mean height (μm)) versus ln(time (sec)) along with ln(mean
radius) versus ln(time).

wetting angle when solder composition was somewhat higher in Pb than the
eutectic composition (around 55 wt% Pb). [20] This is in agreement with the
finding shown in Table 1 that scallops had round shape near the eutectic
composition.

5.5.2 Size Distribution and Average Radius of Scallops

Since scallop morphology showed a dependence on solder composition, the
solder with 55Sn45Pb composition was selected to ensure round morphology
of scallops for size distribution analysis. The plot of the data of radius versus
time to check the agreement with Eq. (5.25) is shown in Fig. 5.11. The growth
exponent was found to be 0.35 to 0.33, which is close to 1/3. The magnitude
of k in Eq. (5.25) was found to be 1.65 × 10−2 μm3 /sec. The particle size
distribution (PSD) is shown in Fig. 5.12. The theoretical curve is the curve
of f (r/ < r >), where < r > is average radius, normalized to f (y)dy = 1.
The heights of histogram bars were also normalized for comparison with the
theoretical curve. The height indicates frequency density, and the total area of
bars is 1. The experimental data showed a much better agreement with FDR
theory than LSW theory [see Fig. 5.12(a)]. The widths and peak positions
of bars are in good agreement with the theoretical curve from FDR theory,
but heights are slightly lower than expected, which is found quite often when
experimental data are compared to theoretical curves.
In the very early stage of the wetting reaction, nucleation of scallops will
have a greater effect on the size distribution. Thus, the PSD of short reaction
SVNY339-Tu March 3, 2007 7:44

148 Chapter 5

Fig. 5.12. Normalized particle size distributions of Cu6 Sn5 scallops. The reaction time
is shown in each histogram. Theoretical curves from LSW theory and FDR theory are
shown in (a) for comparison.

time was expected to be less ideal. However, the size distribution of 30 sec
of reaction already showed very good agreement with the FDR theory [see
Fig. 5.12(a)]. Thus, 30 sec is enough time for scallops to reach statistically
stable size distribution, and ripening is already dominant over nucleation and
growth. The standard deviations of PSD showed very little variation with
reaction time (around 0.4).
The average scallop height was also measured from cross-sectional SEM
images, as shown in Fig. 5.13. The growth exponent of the height was 0.35 and
k was 2.40 × 10−2 μm3 /sec. Because cross-sectional images were used to mea-
sure the height of scallops, the number of scallops measured is much less than
the previous case of measuring the radius of scallops from top-view images.
Thus, height measurement is not statistically as reliable as radius measure-
ment, although the growth exponent of height versus time was also close to
SVNY339-Tu March 3, 2007 7:44

Kinetic Analysis of Flux-Driven Ripening of Copper–Tin Scallops 149

Fig. 5.13. Morphology of cross-sectional scallop at the interface between 55Sn45Pb

and Cu.

1/3. The aspect ratio between height and radius remained almost constant.
The average value of aspect ratio was 1.05.
The constant k in Eq. (5.25) is composed of several thermodynamic pa-
rameters. k is given as

9 n D(C b − C e )δ
k= ,
2 ni Ci

where Ci is Cu concentration in the IMC scallop, C e is Cu concentration

in solder melt with stable equilibrium with planar Cu6 Sn5 , and C b is Cu
quasi-equilibrium concentration in the vicinity of substrate. n is atomic den-
sity in molten solder, ni is atomic density in IMC, and D is diffusivity of
Cu in molten solder. We take Ci ≈ 6/11, C b − C e ≈ 0.001, n/ni ≈ 1, D ≈
10−5 cm2 /sec. Since k = 2.10 × 10−14 cm3 /sec, the channel width is calculated
to be δ = 2.54 nm.
Since FDR assumed hemispherical scallops, it is of interest to see if there
is any change of ripening behavior when scallop morphology deviates a lot
from hemispherical shape. Ghosh investigated Ni3 Sn4 formation by the reac-
tion between various eutectic solders and Cu/Ni/Pd metallization [21]. Ni3 Sn4
scallops had extremely faceted morphology, and their size distribution devi-
ated a lot from the FDR theoretical curve. But their growth rate also followed
t1/3 . Görlich et al. investigated ripening of Cu6 Sn5 when pure Sn reacted with
SVNY339-Tu March 3, 2007 7:44

150 Chapter 5

Cu [22]. As mentioned in the last section, the Cu6 Sn5 formed by reaction be-
tween pure Sn and Cu has faceted morphology. However, the faceted scallops
in top-view micrographs of Görlich’s work do not reveal edges as sharp as those
in Fig. 5.9. This could be due to the use of too strong metallographic etchant.
The fit of size distribution of the faceted Cu6 Sn5 scallops with the FDR the-
oretical curve is not as good as for the round scallops, but still a moderately
good agreement is obtained [22]. The growth exponent was 0.34 for scallop
diameter and 0.40 for height. This deviation from the theory can be exam-
ined in terms of geometric height-to-radius aspect ratio (height/(width/2)). In
Görlich’s work, the scallop aspect ratio was about 0.71 when pure Sn reacted
with Cu, and 1.67 for eutectic PbSn reacted with Cu.

5.6 Nano Channels between Scallops

One of the most important kinetic parameters in the FDR theory of interfacial
ripening is width of the channel between scallops. In the analysis presented in
the last section, the channel width was calculated to be about 2.54 nm. The
width is much larger than the effective width of a large-angle grain boundary
in metals and alloys. The channel is assumed to be wetted by molten solder
during the reactive ripening. The experimental value of the width and whether
the width can be measured in situ during the reactive ripening are of interest.

References
1. I. M. Lifshiz and V. V. Slezov, “The kinetics of precipitation from super-
saturated solid solutions,” J. Phys. Chem. Solids, 19 (1/2), 35–50 (1961).
2. C. Wagner, Z. Electrochem., 65, 581 (1961).
3. V. V. Slezov, “Theory of Diffusion Decomposition of Solid Solution,”
Harwood Academic Publishers, Newark, NJ, pp. 99–112 (1995).
4. K. N. Tu, F. Ku, and T. Y. Lee, “Morphology stability of solder reac-
tion products in flip chip technology,” J. Electron. Mater., 30, 1129–1132
(2001).
5. K. N. Tu, T. Y. Lee, J. W. Jang, L. Li, D. R. Frear, K. Zeng, and J. K.
Kivilahti, “Wetting reaction vs. solid state aging of eutectic SnPb on Cu,”
J. Appl. Phys., 89, 4843–4849 (2001).
6. H. Foell, P. S. Ho, and K. N. Tu, “Cross-sectional TEM of silicon-silicide
interfaces,” J. Appl. Phys, 52, 250 (1981).
7. H. Foell, P. S. Ho, and K. N. Tu, “Transmission electron microscopy of
the formation of nickel silicides,” Philos. Mag. A, 45, 32 (1982).
8. K. N. Tu, W. K. Chu, and J. W. Mayer, “Structure and growth kinetics
of Ni2 Si on Si,” Thin Solid Films, 25, 403 (1975).
9. K.N. Tu, “Selective growth of metal-rich silicide of near noble metals,”
Appl. Phys. Lett., 27, 221 (1975).
SVNY339-Tu March 3, 2007 7:44

Kinetic Analysis of Flux-Driven Ripening of Copper–Tin Scallops 151

10. U. Gosele and K. N. Tu, “Growth kinetics of planar binary diffusion cou-
ples: Thin film case versus bulk cases,” J. Appl. Phys., 53, 3252 (1982).
11. S. Herd, K. N. Tu, and K. Y. Ahn, “Formation of an amorphous Rh-Si
alloy by interfacial reaction between amorphous Si and crystalline Rh thin
films,” Appl. Phys. Lett., 42, 597 (1983).
12. S. Newcomb and K. N. Tu, “TEM study of formation of amorphous NiZr
alloy by solid state reaction,” Appl. Phys. Lett., 48, 1436–1438 (1986).
13. D. Turnbull, “Meta-stable structure in metallurgy,” Metall. Trans. A, 12,
695–708 (1981).
14. H. K. Kim and K. N. Tu, “Kinetic analysis of the soldering reaction be-
tween eutectic SnPb alloy and Cu accompanied by ripening,” Phys. Rev.
B, 53, 16027–16034 (1996).
15. A. M. Gusak and K. N. Tu, “Kinetic theory of flux-driven ripening,” Phys.
Rev. B, 66, 115403-1 to -14 (2002).
16. K. N. Tu, A. M. Gusak, and M. Li, “Physics and materials challenges for
Pb-free solders,” J. Appl. Phys., 93, 1335–1353 (2003). (Review paper)
17. J. O. Suh, Ph.D. dissertation, UCLA (2006).
18. K. A. Jackson, “Current concepts in crystal growth from the melt,” in
“Progress in Solid State Chemistry,” H. Reiss (Ed.), Pergamon Press,
New York, pp. 53–80 (1967).
19. D. P. Woodruff, “The Solid–Liquid Interface,” Cambridge University
Press, London (1973).
20. C. Y. Liu and K. N. Tu, “Morphology of wetting reactions of SnPb alloys
on Cu as a function of alloy composition,” J. Mater. Res., 13, 37–44 (1998).
21. G. Ghosh, “Coarsening kinetics of Ni3 Sn4 scallops during interfacial reac-
tion between liquid eutectic solders and Cu/Ni/Pd metallization,” J. Appl.
Phys., 88, 6887–6896 (2000).
22. J. Görlich, G. Schmitz, and K. N. Tu, “On the mechanism of the binary
Cu/Sn solder reaction,” Appl. Phys. Lett., 86, 053106 (2005).
SVNY339-Tu April 5, 2007 17:18

6
Spontaneous Tin Whisker Growth:
Mechanism and Prevention

6.1 Introduction
Whisker growth on beta-tin (β-Sn) is a surface relief phenomenon of creep
[1–16]. It is driven by a compressive stress gradient and occurs at room tem-
perature. Spontaneous Sn whiskers are known to grow on matte Sn finish on
Cu. Today, due to the wide application of Pb-free solders on Cu conductors
used in the packaging of consumer electronic products, Sn whisker growth has
become a serious reliability issue because the Sn-based Pb-free solders are very
rich in Sn. The matrix of most Sn-based Pb-free solders is almost pure Sn.
The well-known phenomena of tin such as tin-cry, tin-pest, and tin-whisker
are receiving attention again.
The Cu leadframes in surface mount technology of electronic packaging are
finished with a layer of solder for surface passivation and for enhancing wetting
during the joining of the leadframes to printed circuit boards. When the solder
finish is eutectic SnCu or matte Sn, whiskers are often observed. Some whiskers
can grow to several hundred micrometers in length, which are long enough to
become electrical shorts between neighboring legs of a leadframe. The trend
in consumer electronic products is to integrate systems in packaging, so that
elements of devices and parts of components are getting closer and closer
together and the probability of shorting by whiskers is becoming much greater.
A broken whisker can fall between two electrodes and become a short.
How to suppress Sn whisker growth, and how to perform systematic tests
of Sn whisker growth in order to understand the driving force, the kinetics, and
the mechanism of growth are challenging tasks for the electronic packaging
industry today. Due to the very limited temperature range of Sn whisker
growth, from room temperature to about 60◦ C, accelerated tests are difficult.
This is because if the temperature is lower, the kinetics is insufficient due
to slow atomic diffusion, and if the temperature is higher, the driving force
is insufficient because of stress relief by lattice diffusion owing to the high
homologous temperature of Sn.
SVNY339-Tu April 5, 2007 17:18

154 Chapter 6

The whisker growth is spontaneous, indicating that the compressive stress

needed for the growth is self-generated; no externally applied stress is re-
quired. Otherwise, we expect the growth to slow down and stop when the
applied stress is exhausted, if it is not applied continuously. Therefore, it is
of interest to ask where is the self-generated compressive stress coming from,
how can the driving force maintain itself to sustain the spontaneous whisker
growth, and also how large must the compressive stress gradient be to grow a
whisker?
Spontaneous whisker growth is a unique process of creep in which both
stress generation and stress relaxation occur simultaneously at room tempera-
ture. The three indispensable conditions of Sn whisker growth are: (1) the fast
room-temperature diffusion in Sn, (2) the room-temperature reaction between
Sn and Cu or another element to form IMC which generates the compressive
stress in Sn, and (3) the breaking of the protective surface oxide on Sn. The
last condition is needed in order to produce a compressive stress gradient for
creep. When the oxide is broken at a weak spot, the exposed free surface is
stress-free, so a compressive stress gradient is developed, and creep or the
growth of a whisker can occur to relax the stress.
While whisker growth occurs at a constant temperature, it does not occur
under a constant pressure, therefore we cannot use minimum Gibbs free en-
ergy change to describe the growth. Rather an irreversible process of whisker
growth will be presented in Section 6.7.
Cross-sectional scanning and transmission electron microscopy have been
used to examine Sn whiskers, with samples prepared by focused ion beam
thinning [14]. Also, x-ray micro-diffraction in synchrotron radiation has been
used to study the structure and stress distribution around the root and vicinity
of a whisker grown on eutectic SnCu [15].
The growth of Sn whiskers is from the bottom, not from the top, since
the morphology of the tip does not change with whisker growth. Many Sn
whiskers are long enough to short two neighboring legs of the leadframe shown
in Fig. 1.10. It is possible that when there is a high electrical field across the
narrow gap between the tip of a whisker and the point of contact on the other
leg, just before the tip of the whisker touches the other leg, a spark may cause
a fire. The fire may result in failure of the device or a satellite [17–20].

6.2 Morphology of Spontaneous Sn Whisker Growth

In Fig. 6.1(a), an enlarged SEM image of a long whisker on the eutectic
SnCu finish is shown. The whisker in Fig. 6.1(a) is straight and its surface
is fluted. The crystal structure of Sn is body-centered tetragonal with the
lattice constant “a” = 0.58311 nm and “c” = 0.31817 nm. The whisker growth
direction, or the axis along the length of the whisker, has been found mostly
to be the “c” axis, but growth along other axes such as [100] and [311] has
also been found.
SVNY339-Tu April 5, 2007 17:18

Spontaneous Sn Whisker Growth 155

(a)

10 μm

(b)

10 μm

Fig. 6.1. (a) Enlarged SEM image of a long whisker on the eutectic SnCu finish. (b)
Short whiskers or hillocks observed on the pure or matte Sn finish surface.

On the pure or matte Sn finish surface, short whiskers or hillocks were

observed as shown in Fig. 6.1(b). The surface of the whisker in Fig. 6.1(b) is
faceted. Besides the difference in morphology, the rate of whisker growth on
the pure Sn finish is much slower than that on the SnCu finish. The direction
of growth is more random too.
Comparing the whiskers formed on eutectic SnCu and pure Sn, it seems
that the Cu in eutectic SnCu enhances Sn whisker growth. Although the
composition of eutectic SnCu consists of 98.7 at % Sn and 1.3 at % Cu, the
small amount of Cu seems to have a very large effect on whisker growth on
the eutectic SnCu finish.
In Fig. 6.2(a), a cross-sectional SEM image of a leadframe leg with SnCu
finish is shown. The rectangular core of Cu leadframe is surrounded by an
approximately 15-μm-thick SnCu finish. A higher-magnification image of the
interface between the SnCu and the Cu, prepared by focused ion beam, is
shown in Fig. 6.2(b). An irregular layer of Cu6 Sn5 compound can be seen
between the Cu and SnCu. No Cu3 Sn was detected at the interface. The grain
size in the SnCu finish is about several micrometers. More importantly, there
are Cu6 Sn5 precipitates in the grain boundaries of SnCu. The grain boundary
SVNY339-Tu April 5, 2007 17:18

156 Chapter 6

(a)

Cu

SnCu

(b)
(c)

Fig. 6.2. (a) Cross-sectional SEM image of a leadframe leg with SnCu finish. The
rectangular core of Cu is surrounded by an approximately 15-μm-thick SnCu finish.
(b) Higher-magnification image of the interface between the SnCu and Cu layers,
prepared by focused ion beam. An irregular layer of Cu6 Sn5 compound can be seen
between the Cu and SnCu. No Cu3 Sn was detected at the interface. The grain size
in the SnCu finish is about several micrometers. More importantly, there are Cu6 Sn5
precipitates in the grain boundaries of SnCu. (c) Cross-sectional SEM image, prepared
by focused ion beam, of matte Sn finish on Cu leadframe. While the layer of Cu6 Sn5
compound can be seen between the Cu and Sn, there is much less Cu6 Sn5 precipitate
in the grain boundaries of Sn.

precipitation of Cu6 Sn5 is the source of stress generation in the CuSn finish. It
provides the driving force of spontaneous Sn whisker growth. We shall address
this critical issue of stress generation later.
In Fig. 6.2(c), a cross-sectional SEM image of matte Sn finish on Cu lead-
frame is shown, prepared by focused ion beam. While the layer of Cu6 Sn5
compound can be seen between the Cu and the Sn, there is much less Cu6 Sn5
precipitate in the grain boundaries of Sn. The grain size in the Sn finish is also
about several micrometers. The lack of grain boundary Cu6 Sn5 precipitates
SVNY339-Tu April 5, 2007 17:18

Spontaneous Sn Whisker Growth 157

Fig. 6.3. TEM images of the crosssection of whiskers, normal to their length, together
with [001] electron diffraction patterns.

is the most important difference between the eutectic SnCu and the pure Sn
finish with respect to whisker growth.
TEM images of the cross section of whiskers, normal to their length, are
shown in Fig. 6.3(a) and (b) together with electron diffraction patterns. The
growth direction is the c-axis. There are a few spots in the images which might
be dislocations.

6.3 Stress Generation (Driving Force) in Sn Whisker

Growth by Cu-Sn Reaction
The origin of the compressive stress can be mechanical, thermal, and chem-
ical. The mechanical and thermal stresses tend to be finite in magnitude, so
they cannot sustain a spontaneous or continuous growth of whiskers for a
SVNY339-Tu April 5, 2007 17:18

158 Chapter 6

Sn
Whisker

Sn Oxide
Crack

(210)
(321) (321) (321) (321) (321)
σ σ
Sn
Cu6Sn5

Cu

Fig. 6.4. A fixed volume “V ” in the Sn finish is indicated by the dotted square. It
contains an IMC precipitate, and the growth of the precipitate due to the diffusion of
a Cu atom into this volume to react with Sn will produce a compressive stress in the
volume.

long time. The chemical force is essential for spontaneous Sn whisker growth,
but not obvious. The chemical force is due to the room-temperature reaction
between Sn and Cu to form the intermetallic compound (IMC) of Cu6 Sn5 .
The chemical reaction provides a sustained driving force for spontaneous
growth of whiskers as long as the reaction continues with unreacted Sn and
Cu [13].
Compressive stress is generated by interstitial diffusion of Cu into Sn and
the formation of Cu6 Sn5 in the grain boundary of Sn. When the Cu atoms
from the leadframe diffuse into the finish to grow the grain boundary Cu6 Sn5 ,
as shown in Fig. 6.2(b), the volume increase due to the IMC growth will exert
a compressive stress to the grains on both sides of the grain boundary. In
Fig. 6.4, we consider a fixed volume “V ” in the Sn finish that contains an
IMC precipitate, as shown by the dotted square. The growth of the IMC due
to the diffusion of a Cu atom into this volume to react with Sn will produce
a stress,

Ω
σ = −B , (6.1)
V

where σ is the stress produced, B is the bulk modulus, and Ω is the partial
molecular volume of a Cu atom in Cu6 Sn5 (we ignore the molar volume change
of Sn atoms in the reaction for simplicity). The negative sign indicates that
the stress is compressive. In other words, we are adding an atomic volume into
the fixed volume. The assumption of a fixed volume means the constraint of a
constant volume. If the fixed volume cannot expand, a compressive stress will
occur within the volume. When more and more Cu atoms, say “n” Cu atoms,
SVNY339-Tu April 5, 2007 17:18

Spontaneous Sn Whisker Growth 159

diffuse into the volume, V, to form Cu6 Sn5 , the stress in the above equation
increases by changing Ω to n Ω.
In diffusional processes, such as the classic Kirkendall effect of interdiffu-
sion in a bulk couple of A and B, the atomic flux of A diffusing into B is not
equal to the opposite flux of B diffusing into A. If we assume that A diffuses
into B faster than B diffuses into A, we might expect that there will be a
compressive stress in B since there are more A atoms diffusing into it than
B atoms diffusing out of it. However, in Darken’s analysis of interdiffusion,
there is no stress generated in either A or B or no consideration of stress was
given. Why? Darken made a key assumption that vacancy concentration is
in equilibrium everywhere in the sample [21, 22]. To achieve vacancy equilib-
rium, we must assume that vacancies (or vacant lattice sites) can be created
and/or annihilated in both A and B as needed. Hence, provided that the lat-
tice sites in B can be added to accommodate the incoming A atoms, there is
no stress. The addition of a large number of lattice sites implies an increase
in lattice planes if we assume that the mechanism of vacancy creation and/or
annihilation is by dislocation climb mechanism. It further implies that lattice
plane can migrate, which means “Kirkendall shift,” in turn implying marker
motion if markers are embedded in the moving lattice planes in the sample.
Hence, we have the marker motion equation in Darken’s analysis. However,
we must recall that in some cases of interdiffusion in bulk diffusion couples,
vacancy may not be in equilibrium everywhere in the sample, so very often
Kirkendall void formation has been found due to the existence of excess va-
cancies [23].
To absorb the added atomic volume by the fixed volume of V in the finish
due to the in-diffusion of Cu as considered in Fig. 6.4, we must add lattice sites
in the fixed volume. Furthermore, we must allow Kirkendall shift or allow the
added lattice plane to migrate, otherwise compressive stress will be generated.
Since Sn has a native and protective oxide on the surface, the interface between
the oxide and Sn is a poor source and sink for vacancies. Furthermore, the
protective oxide ties down the lattice planes in Sn and prevents them from
moving. This is the basic mechanism of stress generation in spontaneous Sn
whisker growth.
For the oxide to be effective in tying down lattice plane migration, the
finish cannot be too thick. In a very thick finish, say over 100 μm, there are
more sinks in the bulk of the finish to absorb the added volume of Cu. We
note that a whisker is a surface relieve phenomenon. When bulk relaxation
mechanism occurs, whiskers will not grow. There is a dependence of whisker
formation on the thickness of finish. Since the average diameter of whiskers
is about a few micrometers, whiskers will grow more frequently on a finish
having a thickness of a few micrometers to a few times its diameter.
Sometimes it is puzzling to find that Sn whiskers seem to grow on a tensile
region of a Sn finish. For example, when a Cu leadframe surface was plated
with SnCu, the initial stress state of the SnCu layer right after plating was
tensile, yet whisker growth was observed. If we consider the cross section of a
SVNY339-Tu April 5, 2007 17:18

160 Chapter 6

Cu leadframe leg coated with a layer of Sn as shown in Fig. 6.2(a), the lead-
frame experienced a heat treatment of reflow from room temperature to 250◦ C
and back to room temperature. Since Sn has a higher thermal expansion coef-
ficient than Cu, the Sn should be under tension at room temperature after the
reflow cycle. Yet with time, a Sn whisker grows, so it seems that a Sn whisker
grows under tension. Furthermore, if a leg is bent, one side of it will be in
tension and the other side in compression. It is surprising to find that whiskers
grow on both sides, whether the side is under compression or tension. These
phenomena are hard to understand until we recognize that the thermal stress
or the mechanical stress, whether it is tensile or compressive, is finite. It can
be relaxed or overcome quickly by atomic diffusion at room temperature. Af-
ter that, the continuing chemical reaction will develop the compressive stress
needed to grow whiskers. So the chemical force is dominant and persistent.
When we consider the driving force of spontaneous whisker growth on Sn or
SnCu solder finish on Cu, the compressive stress induced by chemical reaction
at room temperature is essential. Room-temperature reaction between Sn and
Cu was studied by using thin film samples; see Chapter 3.
The idea of compressive stress induced by the growth of a grain boundary
precipitate of Cu6 Sn5 has a few variations. One is the wedge model proposed
by Lee and Lee [24] that the Cu6 Sn5 phase between the Cu and Sn has a
wedge shape in growing into the grain boundaries of Sn. The growth of the
wedge will exert a compressive stress to the two neighboring Sn grains, same
as splitting a piece of wood with a wedge. So far, very few wedge-shaped IMCs
have been observed in XTEM; for example, see Fig. 6.2(b).

6.4 Effect of Surface Sn Oxide on Stress Gradient

Generation and Whisker Growth
To discuss the effect of surface oxide on Sn whisker growth, we shall refer to the
effect on Al hillock growth. In an ultrahigh vacuum, no surface hillocks were
found on Al surfaces under compression [25]. Hillocks grow on Al surfaces
only when the Al surface is oxidized, and Al surface oxide is known to be
protective. Without surface oxide in an ultrahigh vacuum, the free surface of
Al is a good source and sink of vacancies, so a compressive stress can be
relieved uniformly on the entire surface or on the surface of every grain of
the Al based on the Nabarro–Herring model of lattice creep or Coble model
of grain boundary creep.
To apply these models to a Sn finish without oxide, as depicted in the top
diagram in Fig. 6.5, the relaxation can occur in each of the grains by diffu-
sion to the free surface of each grain since there is a stress gradient. The free
surfaces are stress-free and are effective source and sink of vacancies. There-
fore, the relaxation is uniform over the entire Sn film surface; all the grains
just become slightly thicker. Consequently, no localized growth of hillocks or
whiskers will take place.
SVNY339-Tu April 5, 2007 17:18

Spontaneous Sn Whisker Growth 161

Surface with no oxide

Vacancy flow Uniform relaxation

No hillocks
No whiskers

Fig. 6.5. Nabarro–Herring model of lattice creep or Coble model of grain boundary
creep in a Sn layer without surface oxide. The schematic diagram shows that when
the surface has no oxide, relaxation of stress can occur in each of the grains by atomic
diffusion to the free surface of each grain.

We note that a whisker or hillock is a localized growth on a surface. To

have a localized growth, the surface cannot be free of oxide, and the oxide
must be a protective oxide so that it effectively blocks all the vacancy sources
and sinks on the surface. Furthermore, a protective oxide also means that it
pins down the lattice planes in the matrix of Sn (or Al), so that no lattice
plane migration can occur to relax the stress in the volume, V, considered in
Fig. 6.4. Only those metals which grow protective oxides, such as Al and Sn,
are known to have serious hillock or whisker growth. When they are in thin
film or thin layer form, the surface oxide can pin down the lattice planes near
the surface easily. On the other hand, it is obvious that if the surface oxide
is very thick, it will physically block the growth of any hillock and whisker.
No hillocks or whiskers can penetrate a very thick oxide or a thick coating.
No break means no free surface and no stress gradient. Thus, a necessary
condition of whisker growth is that the protective surface oxide must not be
too thick so that it can be broken at certain weak spots on the surface to form
free surfaces, and from these spots whiskers grow to relieve the stress.
In Fig. 6.6(a), a focused ion beam image of a group of whiskers on the SnCu
finish is shown. In Fig. 6.6(b), the oxide on a rectangular area of the surface of
the finish was sputtered away by using a glancing incidence ion beam to expose
the microstructure beneath the oxide. In Fig. 6.6(c), a higher-magnification
image of the sputtered area is shown, in which the microstructure of Sn grains
and grain boundary precipitates of Cu6 Sn5 are clear. Due to the ion channeling
effect, some of the Sn grains appear darker than the others. The Cu6 Sn5
particles distribute mainly along grain boundaries in the Sn matrix, and they
are brighter than the Sn grains due to less ion channeling. The diameter of
the whiskers is about a few micrometers, comparable to the grain size in the
SnCu finish.
In ambient, we assume that the surface of the finish and the surface of
every whisker are covered with oxide. The growth of a hillock or whisker
is an eruption from the oxidized surface. It has to break the oxide. When
the Sn matrix is under compression, its oxide is under tension, so the oxide
breaks under tension. The stress that is needed to break the oxide may be the
SVNY339-Tu April 5, 2007 17:18

Fig. 6.6. (a) Focused ion beam image of a group of whiskers on the SnCu finish. (b)
The oxide on a rectangular area of the surface of the finish was sputtered away by
using a glancing incidence ion beam to expose the microstructure beneath the oxide.
(c) Higher-magnification image of the sputtered area, in which the microstructure of
Sn grains and grain boundary precipitates of Cu6 Sn5 are clear.
SVNY339-Tu April 5, 2007 17:18

Spontaneous Sn Whisker Growth 163

minimum stress needed to grow whiskers. It seems that the easiest place to
break the oxide is at the base of the whisker. Then to maintain the growth,
the break must remain open so that it behaves like a stress-free free surface
and vacancies can be supplied continuously from the break and can diffuse
into the Sn layer to sustain the long-range diffusion of the Sn atoms needed to
grow the whisker. In case a part of the break is healed by oxide, the growth of
the whisker will lead to a turn in whisker growth direction toward the healed
side, and as a consequence, a bent whisker is formed.
In Fig. 6.4, we have depicted that the surface of the whisker is oxidized,
except for the base. The surface oxide of the whisker serves the very im-
portant purpose of confinement so that the whisker growth is essentially a
one-dimensional growth. The surface oxide of the whisker prevents it from
growing in the lateral direction, thus it grows with a constant cross section
and has the shape of a pencil. Also, the oxidized surface may explain why
the diameter of a Sn whisker is just a few micrometers. This is because the
gain in strain energy reduction in whisker growth is balanced by formation of
the surface of the whisker. By balancing the strain energy against the surface
energy in a unit length of the whisker, πR2 ε = 2πRγ, we find that

2γ
R= , (6.2)
ε

where R is radius of the whisker, γ is surface energy per unit area, and ε is
strain energy per unit volume. Since strain energy per atom is about four to
five orders of magnitude smaller than the chemical bond energy or surface
energy per atom of the oxide, the diameter of a whisker is found to be several
micrometers, which is about four orders of magnitude larger than the atomic
diameter of Sn. For this reason, it is very difficult to have spontaneous growth
of nano-diameter Sn whiskers.
The broken oxide at the base is a key assumption in the model of spon-
taneous Sn whisker growth to be discussed in Section 6.6. The free surface
exposed by the broken oxide produces the stress gradient to drive the whisker
growth.

6.5 Measurement of Stress Distribution by

Synchrotron Radiation Micro-diffraction
The micro-diffraction apparatus in the Advanced Light Source (ALS), at
Lawrence Berkeley National Laboratory, was used to study Sn whiskers grown
on SnCu finish on Cu leadframe at room temperature [15]. The white radia-
tion beam was 0.8 to 1 μm in diameter and the beam step-scanned over an
area of 100 μm by 100 μm in steps of 1 μm. Several areas of the SnCu finish
were scanned and those areas were chosen so that in each of them there was
a whisker, especially the areas that contained the root of a whisker. During
SVNY339-Tu April 5, 2007 17:18

164 Chapter 6

Fig. 6.7. Low-magnification SEM image of an area of finish in which a whisker is

circled and scanned.

the scan, the whisker, and each grain in the scanned area, can be treated as
a single crystal to the beam. This is because the grain size is larger than the
beam diameter. At each step of the scan, a Laue pattern of a single crystal is
obtained. The crystal orientation and the lattice parameters of the Sn whisker
and the grains in the SnCu matrix surrounding the root of the whisker were
measured by the Laue patterns. The software in ALS is capable of determin-
ing the orientation of each of the grains, and displaying the distribution of
the major axis of these grains. Using the lattice parameters of the whisker
as stress-free internal reference, the strain or stress in the grains in the SnCu
matrix can be determined and displayed. Figure 6.7 shows a low-magnification
picture of an area of finish wherein a whisker is circled and scanned.
Figure 6.8 shows an in-plane orientation map of the angle between the
(100) axis of Sn grains and the x-axis of the laboratory frame. An image of
the whisker is seen. The x-ray micro-diffraction study shows that in a local area
of 100 μm × 100 μm the stress is highly inhomogeneous with variations from
grain to grain. The finish is therefore under a biaxial stress only on the average.
This is because each whisker has relaxed the stress in the region surrounding
it. But the stress gradient around the root of a whisker does not have a radial
symmetry. The numerical value, and the distribution of stress, are shown in
Fig. 6.9, where the root of the whisker is at the coordinates of x = −0.8415
and y = −0.5475 as shown in Fig. 6.8. Overall, the compressive stress is quite
low, of the order of several mega pascals, but we can still see the slight stress
gradient going from the whisker root area to the surroundings. This means
that the stress level just below the whisker is slightly less compressive than
the surrounding area. This is because the stress near the whisker has been
SVNY339-Tu April 5, 2007 17:18

Spontaneous Sn Whisker Growth 165

In-plane Orientation Map

[Angle between (100) and X]

90.00
-0.82 85.50
81.00
76.50
72.00
67.50
-0.84 63.00
58.50
X (mm)

54.00
49.50
45.00
-0.86 40.50
36.00
31.50
27.00
-0.88 22.50
18.00
13.50
9.000
4.500
-0.90 0

-0.60 -0.58 -0.56 -0.54 -0.52 -0.50

Y (mm)

Fig. 6.8. Plot of in-plane orientation map of the angle between the (100) axis of
Sn grains surrounding the whisker (shown in Fig. 6.7) and the x-axis of laboratory
reference frame.

-0.830
35.00
32.38
29.75
-0.835 27.13
24.50
21.88
19.25
-0.840
X (mm)

16.63
14.00
11.38
8.750
-0.845 6.125
3.500
0.8750
-1.750
-0.850 -4.375
-7.000
-9.625
-12.25
-0.855 -14.88
-17.50

-0.56 - 0.55 -0.54

Y (mm)
Fig. 6.9. Stress distribution around the root of a whisker, which is at the coordinates
of x = −0.8415 and y = −0.5475.
SVNY339-Tu April 5, 2007 17:18

166 Chapter 6

1.5 μm
(Unit : Mpa)
-0.5400 -0.5415 -0.5430 -0.5445 -0.5460 -0.5475 -0.5490 -0.5505 -0.5520 -0.5535 -0.5550

-0.8340 -2.82 -3.21 -2.26 0.93 0.93 -0.23 -8.17 2.22 1.49 1.6 -0.03
-0.8355 -2.26 -2.64 -2.64 -1.04 1.37 1.37 -1.31 0.87 0.87 0.87 -0.7
-0.8370 -2.53 -3.21 -3.21 -2.64 -1.04 3.61 0.75 0.87 0.7 0.7 -0.19
-0.8385 -7.37 -9.62 -6.57 -2.64 3.61 4.52 3.61 0.29 -1.31 0 -4.79
-0.8400 -7.37 -8.22 -6.57 -1.18 0.75 4.23 0.75 -2.25 -2.27 -2.91 -6.91
-0.8415 -4.17 -4.84 -4.17 -1.81 -0.67 .000 -1.96 -1.96 -3.74 -5.08 -5.08
-0.8430 -4.17 -4.17 -3.63 -1.81 -1.81 -2.29 -2.29 -1.96 -1.96 -3.27 -3.27
-0.8445 -4.14 -4.17 -3.86 -3.63 -2.79 -4.64 -4.78 -0.84 -1.4 -1.49 -3.27
-0.8460 -3.14 -3.63 -3.86 -3.63 -3.13 -4.78 -4.78 0.04 0.04 -1.41 -2.33
-0.8475 -4.14 -4.49 -4.49 -4.64 -3.86 -6.64 -1.72 3.55 3.55 -0.41 -2.33
-0.8490 -3.33 -5.67 -6.20 -6.29 -2.66 -2.08 -1.72 -1.79 0 -1.79 -3.73

Whisker

Fig. 6.10. Plot of −σzz , which is the deviatoric component of the stress along the
surface normal. The total strain tensor is equal to the sum of the deviatoric strain
tensor and the dilatational strain tensor. In the figure, the whisker part is removed in
order to observe the stress around the whisker root more clearly. The absolute value
of stress in the whisker is higher than that in the surrounding grains. If we assume the
whisker to be stress-free, the surface of SnCu finish is under compressive stress.

relaxed by whisker growth. In Fig. 6.10 the light-colored arrows indicate the
directions of local stress gradient. Some circles next to each other in Fig. 6.10
show a similar stress level, which most likely means that they belong to the
same grain.
Figure 6.10 shows a plot of −σzz , which is the deviatoric component of
the stress along the surface normal. The total strain tensor is equal to the
sum of the deviatoric strain tensor and the dilatational strain tensor. The
latter is measured from energy of Laue spot using monochromatic beam and
the former is measured from deviation in crystal Laue pattern using white
radiation beam.

εij = εdeviatoric + εdilatational

    
 ε11 ε12 ε13   δ 0 0 
  
=  ε21 ε22 ε23  +  0 0  ,

δ (6.3)
ε ε32 ε33   0

0 δ
31

where the dilatational strain δ = 13 (ε11 + ε22 + ε33 ) and εii = εii + δ.
We explain in the following the measurements of these two strain tensors.
The deviatoric strain tensor is calculated from the deviation of spot positions
SVNY339-Tu April 5, 2007 17:18

Spontaneous Sn Whisker Growth 167

in the Laue pattern with respect to their “unstrained” positions. The latter
is obtained from an “unstrained” reference. By assuming that the whisker
is strain-free, we used the Sn whisker itself as the unstrained reference and
calibrated the sample–detector distance and the tilt of detector with respect
to the beam. The geometry is fixed. From the Laue spot positions of the
strained sample, we can then measure any deviation of their positions from
the calculated positions if the sample has zero strain. The transformation
matrix which relates the unstrained to the strained Laue spot positions is
then calculated and the rotational part is taken out. The deviatoric strain
can then be computed from this transformation matrix. The more spots we
have in the Laue pattern, the more accurate will be the deviatoric strain tensor
determined. We note that the deviatoric strain is related to the change in the
shape of the unit cell, but the unit cell volume is assumed to be constant and
it consists of five independent components. The sum of the three diagonal
components should be equal to zero.
To obtain the total strain tensor, we must add the dilatational strain ten-
sor to the deviatoric strain tensor. The dilatational component is related to
the change in volume of the unit cell and it consists of a single component of
expansion or shrinkage, δ, in the last equation. In principle, if the deviatoric
strain tensor is known, only one additional measurement is needed, i.e., the
energy of a single reflection is required to obtain this single dilatational com-
ponent. We can use the monochromatic beam to do so. From the orientation of
the crystal and the deviatoric strain we can calculate for each reflection what
the energy of E0 for zero dilatational strain would be. We scan the energy by
rotating the monochromator around this energy E0 and watch the intensity
of the peak of interest on the CCD camera. The energy which maximizes the
intensity of the reflection is the actual energy of the reflection. The difference
in the observed energy and the E0 gives the dilatational strain.
Since σxx + σyy + σzz = 0 by definition, −σ zz is a measure of the in-plane
stress (note that for a blanket film, with free or passivated surface, on the
average the total normal stress σzz = 0), from that σb (biaxial stress) = (σxx +
σyy )/2 = (σxx + σyy )/2 − σzz = −3σzz /2 . This relation is always true on the
average. A positive value of −σzz indicates an overall tensile stress whereas a
negative value indicates an overall compressive stress. However, the measured
stress values, corresponding to a strain of less than 0.01%, are only slightly
larger than the strain/stress sensitivity of the white beam Laue technique
(sensitivity of the technique is 0.005% strain).
No very long range stress gradient has been observed around the root of a
whisker, indicating that the growth of a whisker has released most of the local
compressive stress in the distance of several surrounding grains. In Fig. 6.9,
the whisker part is removed in order to observe the stress around the whisker
root more clearly. The absolute value of stress in the whisker is higher than
that in the surrounding grains. If we assume the whisker to be stress-free, the
surface of SnCu finish is under compressive stress.
SVNY339-Tu April 5, 2007 17:18

168 Chapter 6

6.6 Stress Relaxation (Kinetic Process) in Sn Whisker

Growth by Creep: Broken Oxide Model
Whisker growth is a unique creep phenomenon in which stress generation
and stress relaxation occur simultaneously. Therefore, we must consider two
kinetic processes of stress generation and stress relaxation and their coupling
by irreversible processes [13]. About the two processes in whisker growth,
the first is the diffusion of Cu from the leadframe into the Sn finish to form
grain boundary precipitates of Cu6 Sn5 . This kinetic process generates the
compressive stress in the finish. The second is the diffusion of Sn from the
stressed region to the stress-free region at the root of a whisker to relieve the
stress. The distance of diffusion in the second process is much longer than the
first and also the diffusivity in the second process is slower too, so the second
process tends to control the rate in the growth of whiskers.
Since the reaction of Sn and Cu occurs at room temperature, the reaction
continues as long as there are unreacted Sn and Cu. The stress in the Sn
will increase with the growth of Cu6 Sn5 in it. Yet the stress cannot build up
forever; it must be relaxed. Either the added lattice planes in the volume, V,
in Fig. 6.4 must migrate out of the volume, or some Sn atoms will have to
diffuse out from the volume to a stress-free region.
Room temperature is a relatively high homologous temperature for Sn,
which melts at 232◦ C, hence the self–diffusion of Sn along Sn grain bound-
aries is fast at room temperature. Therefore, the compressive stress in the Sn
induced by the chemical reaction at room temperature can also be relaxed at
room temperature by atomic rearrangement via self grain boundary diffusion.
The relaxation occurs by the removal of atomic layers of Sn normal to the
stress, and these Sn atoms can diffuse along grain boundaries to the root of
a stress-free whisker to feed its growth. This is a creep process driven by a
stress gradient.
It is worth noting that creep at a slow rate is driven by a stress gradient,
not by a stress. Typically creep is defined as a time-dependent deformation
under a constant load. If we take a rectangular bar and apply a constant
load or stress at its two ends, it will creep. However, in the normal direction
of the free surface of the four sides of the bar, there is no stress. Hence,
there is a stress gradient between the end surfaces and the side surfaces. The
driving force of atomic migration in creep is stress gradient as given in the
Nabarro–Herring model. Under a hydrostatic tension or compression, there
may be random walk of atoms, but no creep. The diffusion of Cu into the Sn
finish generates a compressive stress. However, a uniform compressive stress
will not cause creep; we need a mechanism to produce a compressive stress
gradient in the Sn finish. In Section 6.4, we discussed that a broken surface
oxide can produce a free surface which is stress-free by definition. Hence, a
stress gradient is developed in the broken oxide model and creep or whisker
growth can occur.
SVNY339-Tu April 5, 2007 17:18

Spontaneous Sn Whisker Growth 169

Furthermore, because of the stress gradient, creep is a process which does

not occur under the condition of constant pressure, therefore we cannot min-
imize Gibbs free energy change to describe the creep process. It is an irre-
versible process.

6.7 Irreversible Processes

To formulate the irreversible process, we have the linear phenomenological

relation between the fluxes Ji and the forces Xj ,

Ji = Lij Xj ,
j

where Lij are the phenomenological coefficients. In pairing the forces of chem-
ical reaction and stress, the fluxes of Cu and Sn atoms can be given as

J1 = L11 X1 + L12 X2 ,
(6.4)
J2 = L21 X1 + L22 X2 ,

where J1 = JCu is atomic flux of Cu in units of atoms/cm2 sec, J2 = JSn is

atomic flux of Sn in units of atoms/cm2 sec, X1 = −∇μ1 is chemical potential
gradient, and X2 = −∇μ2 is stress potential gradient. We note that stress
potential is a chemical potential too.
To examine X1 , we consider the chemical reaction of

6Cu + 5Sn → Cu6 Sn5

We have the chemical affinity as given in Section 3.2.5,

A = μη − 6 μCu − 5 μSn (6.5)

where μη is the chemical potential of the Cu6 Sn5 compound molecule and
μCu and μSn are the chemical potentials of the unreacted Cu and Sn,
respectively.

X1 = −∇G |T,p = A∇n (6.6)

under constant temperature and pressure, where n is number of moles or

molecules of Cu6 Sn5 formed. Since our model of the growth of Cu6 Sn5 is by
interstitial diffusion of Cu into the finish and reaction with Sn at the interface
of Cu6 Sn5 and Sn, we assume it is an interfacial-reaction-controlled reaction.
We recall that in Section 3.2.2, the growth of Cu6 Sn5 at room temperature
was measured to be linear with time.
On X2 , we recall that the stress induced by the diffusion of Cu into Sn can
be represented by σ = −B(Ω/V ), so the stress is assumed to be hydrostatic.
SVNY339-Tu April 5, 2007 17:18

170 Chapter 6

Since the stressed region is near a free surface, the stress state may not be
isotropic, nevertheless it is a vector. We define

X2 = −∇σΩ, (6.7)

where σ and Ω are the stress and atomic volume (or partial molar volume
in a binary system), respectively. The driving force is the stress gradient. We
emphasize that J1 , J2 , X1 , and X2 are vectors, so Lij are tensors. The pair of
flux equations is
Cu Cu
Dij Dij
JiCu = CCu (A∇n)i + CCu (−∇σΩ)i ,
kT kT (6.8)
Sn
Dij
JiSn = CSn M21 (A∇n)i + CSn (−∇σΩ)i .
kT
In the second equation of Eq. (6.4), the meaning of the first term after the
equals sign, i.e., the L21 X1 term, is to describe the flux of Sn driven by chemical
reaction to form Cu6 Sn5 . Because we are considering the formation of Cu6 Sn5
within Sn, so Sn is everywhere and no long-distance diffusion of Sn is needed in
the formation of Cu6 Sn5 . Actually, we have assumed that the reaction occurs
by interstitial diffusion of Cu to a grain boundary precipitate of Cu6 Sn5 and
by the reaction with Sn at the interface of the Cu6 Sn5 precipitate. We assume
that the growth of the precipitate is interfacial-reaction-controlled. Hence, we
have

L21 X1 = CSn K21 = CSn M21 X1 = CSn M21 (A∇n)i , (6.9)

where K21 = M21 X1 is the interfacial reaction constant and it has the dimen-
sion of velocity, and M21 is the mobility of the interface between Cu6 Sn5 and
Sn. The choice of M21 must fulfill Onsager’s reciprocity relation of L12 = L21 .
How to choose M21 requires an analysis of the interfacial reaction process,
which has been discussed in Section 3.2.5.
On the other cross term or the last term in the first equation of Eq. (6.4),
it means the Cu flux in the Sn finish driven by stress gradient. Since the
chemical potential is much larger than the stress potential, this cross term
is negligible. Therefore, in modeling the growth of a Sn whisker in the next
section, we shall assume a very simple model of creep.

6.8 Kinetics of Grain Boundary Diffusion-Controlled

Whisker Growth
When the surface oxide is broken around a grain whose surface normal is near
a close-packed direction and surrounded by high angle grain boundaries, this
grain can be pushed out by pressure in the Sn layer. If the deposited Sn finish
has a texture, this grain should not be a part of the texture because it should
SVNY339-Tu April 5, 2007 17:18

Spontaneous Sn Whisker Growth 171

be surrounded by high-angle boundaries to facilitate pressure transmission

through the grain boundary structure. This may explain why there are only
a few places from where whiskers nucleate and grow.
The growth of a whisker occurs at the root; it is being pushing out. What
is the growth mechanism? When Sn atoms diffuse to the root of a whisker,
how can they be incorporated into the root of a whisker? The growth can
be regarded as grain growth because the whisker is a single crystal and it
grows longer with time. In the classical model of normal grain growth, the
basic process is grain boundary migration against its curvature by atoms
jumping from one grain across a grain boundary to the grain on the other
side of the grain boundary. Yet in whisker growth, it is unclear if there is
grain boundary migration at the root. Using a series of cross-sectional SEM
images, the microstructure of the root of a whisker and its surrounding grains
have been observed [15]. It suggests that most likely there is no migration of
the grain boundaries between the whisker and the surrounding grains during
the growth of the whisker. Whisker growth is a grain growth with very little
grain boundary migration at the root of the whisker. It seems that Sn atoms
arrive at the root region along grain boundaries and they can be incorporated
into the root of a whisker without jump across a grain boundary as in normal
grain growth. This is because the Sn atoms are already diffusing in grain
boundaries. Hence, no grain boundary migration is needed. The atomistic
model of incorporation of atoms into the whisker for its growth requires more
study; it may take place at the kink sites on the bottom side of the whisker,
similar to stepwise growth on a free crystal surface in epitaxial growth of thin
films. We must mention that there are vacancies coming in from the surface
crack at the root area to assist the growth.
To study the grain boundary structure around the root of a whisker, cross-
sectional TEM samples for direct observation of the root of a whisker were
prepared. Figure 6.11 is a SEM image of the preparation of cross-sectional
TEM samples by focused ion beam etching. Focused ion beam (FIB) was
used to etch two rectangular holes into the finish separated by a thin wall as
shown in Fig. 6.11. The location of the two holes was selected to have a whisker
on top of the wall between them. After etching, the thickness of the wall is
less than 100 nm so that it is transparent to 100-keV electrons. The thin wall
contains a thin vertical section of the whisker, the root of the whisker, and
a couple of grains surrounding the whisker. Figure 6.12 is a FIB image and
the corresponding bright-field TEM image of the thin slice. Figure 6.13 is a
higher-magnification bright-field TEM image of a grain boundary between a
whisker and a neighboring grain in the root region of the whisker. The grain
boundary plane appears straight.
Sometimes the surface of a whisker along its length has the appearance
of very fine sawtooth steps. It indicates that the growth of the whisker may
have the ratcheted motion instead of a smooth motion. The ratcheted motion
may be due to a repetitive breaking of the oxide at the root of the whisker.
The growth of the whisker has to break the oxide and expose a free surface.
SVNY339-Tu April 5, 2007 17:18

172 Chapter 6

Fig. 6.11. SEM image of the preparation of cross-sectional TEM samples by focused
ion beam etching. Focused ion beam was used to etch two rectangular holes into
the finish separated by a thin wall. The thin wall will be cut out and examined by
TEM.

Yet the free surface in ambient will form oxide right away, so the oxide has
to be broken repetitively and may be in ratcheted steps. The atomistic mech-
anism of the growth of a whisker will require more experimental study and
analysis.

Fig. 6.12. FIB image and the corresponding bright-field TEM image of the thin wall
containing the crosssection of a whisker and grains around its root.
SVNY339-Tu April 5, 2007 17:18

Spontaneous Sn Whisker Growth 173

Fig. 6.13. Bright-field cross-sectional TEM image of a grain boundary between a Sn

whisker and a neighboring grain in the root region of the whisker. The grain boundary
plane appears straight.

To analyze the growth kinetics of a whisker, we assume a two-dimensional

model in cylindrical coordination. The distribution of whiskers is assumed
to have a regular arrangement so that each occupies a diffusional field of
diameter of 2b, as shown in Fig. 6.14. We assume that the whisker has a
constant diameter of 2a and a separation of 2b and it has a steady-state
growth in the diffusional field which can be described by a two-dimensional
continuity equation in cylindrical coordinates. We recall that stress can be
regarded as an energy density and a density function obeys the continuity
equation [13, 26].

∂ 2 σ 1 ∂σ
∇2 σ = + = 0. (6.10)
∂r2 r ∂r
The boundary conditions are

σ = σ0 at r = b,
σ = 0 at r = a.

The solution is σ = Bσ0 ln(r/a), where B = [ln(b/a)]−1 and σ0 is the stress

SVNY339-Tu April 5, 2007 17:18

174 Chapter 6

Fig. 6.14. The whiskers are assumed to have a regular arrangement so that each
occupies a diffusional field of diameter of 2b.

in the Sn film. Knowing the stress distribution, we can evaluate the stress
gradient,

∂σΩ
Xr = − . (6.11)
∂r

Then the flux to grow the whisker is calculated at r = a,

D Bσ0 D
J =C Xr = . (6.12)
kT kT a

We note that in a pure metal, C = 1/Ω. The volume of materials transported

to the root of the whisker in a period of dt is

JAdtΩ = πa2 dh, (6.13)

where A = 2πas is the peripheral area of the growth step at the root, s is the
step height, and dh is the increment of height of the whisker in dt. Therefore,
the growth rate of the whisker is

dh 2 σ0 ΩsD
= . (6.14)
dt ln(b/a) kT a2
SVNY339-Tu April 5, 2007 17:18

Spontaneous Sn Whisker Growth 175

To evaluate the whisker growth rate, we need to know the self-diffusivity of

Sn. The self-lattice diffusivities in the direction parallel and normal to the
c-axis are slightly different and are given as

D// = 7.7 × exp(−25.6 kcal/RT ) cm2 /sec,

D⊥ = 10.7 × exp(−25.2 kcal/RT ) cm2 /sec.

The lattice diffusivities at room temperature are about 10−17 cm2 /sec. This
means that in 1 year, or t = 108 sec, the diffusion distance calculated by
using x2 ∼= Dt is about 1 μm. Thus, the lattice diffusivities are too slow to
be responsible for whisker growth at room temperature. Self-grain-boundary
diffusion of Sn has not been determined. If we assume that the large-angle
grain boundary diffusivity requires one-half of the activation energy of lattice
diffusion given above, we obtain a self-grain-boundary diffusivity of about
10−8 cm2 /sec.
Using a = 3 μm, b = 0.1 mm, σ0 Ω = 0.01 eV (at σ0 = 0.7 × 109 dyne/cm2 ),
kT = 0.025 eV at room temperature, s = 0.3 nm, and D = 10−8 cm2 /sec (the
self-grain-boundary diffusivity of Sn at room temperature), we obtain a growth
rate of 0.1 × 10−8 cm/sec. At this rate, we expect a whisker of 0.3 mm after
1 year, which agrees well with the observed result. Since we assume grain
boundary diffusion, we note that there are only several grain boundaries con-
necting the base of a whisker to the rest of the Sn matrix. Hence, in taking
the total atomic flux which supplies the growth of a whisker to be JAdtΩ,
where A = 2πas, we have assumed that the flux goes to the entire periph-
ery of the whisker “2πa” but only for a step height of “s” for its growth.
The values of b and σ0 used in the above calculation were taken from Ref.
12. These values differ from what we found in Ref. 15, where the stress is
about 10 MPa or 108 dyne/cm2 but the diffusion distance is only a few grain
diameters. Using the latter values, the calculated growth rate is about the
same.

6.9 Accelerated Test of Sn Whisker Growth

One of the most annoying behaviors of a Sn whisker is that it does not grow
when we want it to grow, yet it grows when we do not want it to . The growth of
just one whisker is a threat of reliability. In order to predict the lifetime of Pb-
free solder finish without whisker growth, we should conduct accelerated tests
as in most reliability problems. An accelerated test can be conducted at larger
driving force or faster kinetics, provided that the mechanism of failure remains
the same. Typically, tests at higher temperatures are performed to obtain the
activation energy of the rate controlling process, which then enables us to
extrapolate the lifetime at the device operation temperature. For Sn whisker
growth, while it is possible to conduct the tests up to 60◦ C, the rate of whisker
growth is still quite slow due to slow atomic diffusion. When the temperature
SVNY339-Tu April 5, 2007 17:18

176 Chapter 6

approaches 100◦ C, the diffusion is fast enough to relieve the stress. Hence,
we encounter a situation of competition between driving force and kinetics.
Although we can add Cu to Sn to have a faster whisker growth as in eutectic
SnCu solder, the rate is still not fast enough. Besides, we need to isolate the
effect of Cu on whisker growth.
We consider here the use of electromigration to conduct accelerated tests of
whisker growth. The subject of electromigration will be covered in Chapter 8.
In the classic Blech test structure of electromigration in Al short strips, atoms
of Al are being driven from the cathode to the anode and a compressive stress
is built up at the anode end of a stripe and hillocks grow there. The advantage
of using electromigration to study whisker growth is that not only can we
vary the applied current density (larger driving force), we can also use higher
temperatures (faster kinetics). Hence, we can control both the driving force
as well as the kinetics.
Figure 6.15 shows the growth of a Sn whisker at the anode of a test sample
of pure Sn under electromigration [27, 28]. Measuring the growth rate and the
diameter of the whisker, we obtain the volume change per unit time of the
whisker, V = JAdtΩ, where J is the electromigration flux in units of number

Fig. 6.15. The sequential growth of a

Sn whisker at the anode of a test sample
of pure Sn under electromigration.
SVNY339-Tu April 5, 2007 17:18

Spontaneous Sn Whisker Growth 177

of atoms/cm2 -sec, A is the cross-section of the whisker, dt is unit time, and Ω

is atomic volume. Knowing J, we have
 
D dσΩ ∗
J =C + Z ejρ (6.15)
kT dx

where C = 1/Ω in pure Sn, D is diffusivity, kT is thermal energy, σ is stress

at the anode and we may assume the stress at the cathode is zero, dσ/dx is
the stress gradient along the short stripe of Sn of length dx, Z ∗ is the effective
charge number of the diffusing Sn atoms in electromigration, e is electron
charge, j is current density, and ρ is resistivity of Sn at the test temperature.
We can determine σ from the last equation.
To determine the stress gradient for the whisker growth, we note that the
electrons flow out of the stripe at the bottom corner of the anode. It has
the highest stress since Sn atoms are being pushed to this point. When the
oxide on the upper corner of the anode is broken, a vertical stress gradient is
generated and it is this stress gradient which drives whisker or hillock growth
at the anode end.
Figure 6.16 shows a SEM image of whiskers at the anode end of a set of
stripes of eutectic SnPb driven by electromimgration. The stripes were cut
by focused ion beam, and it was found that whiskers grow at the anode ends.

Fig. 6.16. SEM image of whiskers at the anode end of a set of stripes of eutectic
SnPb driven by electromigration. (Courtesy of Prof. Chih Chen, National Chiao Tung
University.)
SVNY339-Tu April 5, 2007 17:18

178 Chapter 6

If we keep the stripe dimension and the applied current density unchanged,
we may determine the activation energy of whisker growth when the growth
as a function of temperature is obtained. However, the accelerated test may
not be meaningful until we can confirm that the whisker driven by electromi-
gration has the same growth behavior and mechanism as the whisker grown
spontaneously on the Pb-free finish.

6.10 Prevention of Spontaneous Sn Whisker Growth

On the basis of the analysis presented in this chapter, we have three indis-
pensable conditions of spontaneous whisker growth: (1) the room-temperature
grain boundary diffusion of Sn in Sn, (2) the room-temperature reaction be-
tween Sn and Cu to form Cu6 Sn5 , which provides the compressive stress or
the driving force for whisker growth, and (3) the breaking of the protective
surface Sn oxide. If we remove any one of them, we will have in principle
no whisker growth. However, we have found from the synchrotron radiation
study that it takes only a very small stress gradient to grow Sn whiskers,
hence it is difficult to prevent whisker growth. National Electronics Manu-
facturing Initiative (NEMI) has recommended a solution to remove condition
(2) by introducing a diffusion barrier of Ni between the Cu leadframe and the
Pb-free finish to prevent Cu from reacting with Sn. To remove condition (3) is
unrealistic since we must have no oxide on the finish, and to do so we must
keep the sample in ultrahigh vacuum. We propose here to remove condition
(1) by blocking the grain boundary diffusion of Sn. Furthermore, if we can
remove both conditions (1) and (2), it is even better.
Since Sn whisker growth is an irreversible process which couples stress
generation and stress relaxation, it is essential to uncouple them in order
to prevent Sn whisker growth. In other words, we must remove both stress
generation and stress relaxation. Stress generation can be removed by using a
diffusion barrier to block the diffusion of Cu into Sn. Besides Ni, we can also
use Cu3 Sn intermetallic compound. In Chapter 3, we have shown that above
60◦ C, Cu3 Sn will form between Cu and Sn. A heat treatment of the finish on
Cu leadframe above 60◦ C will form Cu6 Sn5 and Cu3 Sn between them to serve
as diffusion barrier.
Up to now, no solution to remove stress relaxation has been given. In other
words, how to prevent the creep process or the diffusion of Sn atoms to the
whiskers is unknown. We can do so by using another diffusion barrier. Since
we have to block the diffusion of Sn atoms from every grain of Sn in the finish,
it is nontrivial. We may succeed by adding several percent of Cu or another
element into the matte Sn or the eutectic SnCu solder. We recall that the Cu
concentration in the eutectic SnCu is only 1.3 at.% or 0.7 wt%. We shall add
about several (3 to 7) wt% of Cu. The reason for doing so is to have enough
precipitation of Cu6 Sn5 in all the grain boundaries in the finish, so that every
grain of Sn in the finish will be coated by a grain boundary layer of Cu6 Sn5 .
SVNY339-Tu April 5, 2007 17:18

Spontaneous Sn Whisker Growth 179

Fig. 6.17. A layered structure of a Sn-Cu finish on a Ni diffusion barrier on a Cu

leadframe. Each of the Sn grains is surrounded by Sn-Cu precipitates.

Thus, the grain boundary coating becomes a diffusion barrier to prevent the
Sn atoms from leaving each of the grains in the Sn. When there is no diffusion
of Sn, there is no growth of Sn whisker since the supply of Sn is cut. Figure 6.17
depicts a layered structure of a Sn-Cu finish on a Ni diffusion barrier on a Cu
leadframe. The optimal concentration of Cu in the finish requires more study.
Cross-sectional SEM and FIB images of the samples should be obtained to
investigate the microstructure of electroplated Sn-Cu finish, having a high
concentration of Cu.
There are two key reasons for the selection of Cu (or another element) to
form grain boundary precipitate in Sn. The first is that when the Sn has so
much supersaturated Cu, it will not take more Cu from the leadframe. The
second is that the addition of Cu will not affect strongly the wetting property
of the surface of the finish. Without a good property of wetting, it cannot be
used as finish on the leadframe since the most important property of a surface
finish is that it should be wetted easily by molten solder with the help of flux.
For low-reliability devices, it will be sufficient to plate the Sn-3 to 7 wt% Cu
finish directly on Cu leadframe without the Ni diffusion barrier. As previously
mentioned, when the Sn is supersaturated with Cu, it will not take more Cu
from the leadframe. The advantage is that it is a low-cost process without the
additional deposition of Ni. For high-reliability devices, we should keep the
Ni diffusion barrier and deposit the Sn-Cu finish on the Ni. The combination
has diffusion barriers to prevent the diffusion of both Cu and Sn. It will be
much more effective than just using one of them.
Whether the addition of several percent of Cu is effective or not may
depend or how the Cu is added, e.g., by using nanosize Cu particles. Whether
there are other problems must be studied, such as the brittleness of the grain
boundary precipitates. Whether there are other elements better than Cu for
the purpose of whisker prevention also remains to be studied. It is known
that the addition of several percent of Pb will prevent Sn whisker growth
since Pb is soft and it tends to reduce the local stress gradients in Sn. Also,
because Sn-Pb is a eutectic system, the eutectic microstructure consists of
two separated and intermixing phases, and they block each other in terms of
long-range diffusion. Thus, adding several percent of the other soft elements
SVNY339-Tu April 5, 2007 17:18

180 Chapter 6

that have eutectic phase diagram with Sn, such as Bi, In, and Zn, may be
a good choice. If there is no Sn diffusion, we expect no Sn whisker growth.
Finally, the simplest method to avoid whisker problem may be to spray the
entire packaging structure with a thick coating.

References
1. C. Herring and J. K. Galt, Phys. Rev. 85, 1060 (1952).
2. J. D. Eshelby, Phys. Rev., 91, 755 (1953).
3. F. C. Frank, Philos. Mag., 44, 854 (1953).
4. G. W. Sears, Acta Metall., 3, 367 (1955).
5. S. Amelinckx, W. Bontinck, W. Dekeyser, and F. Seitz, Philos. Mag., 2,
355 (1957).
6. W. C. Ellis, D. F. Gibbons, and R. C. Treuting, “Growth of metal whiskers
from the solid,” in “Growth and Perfection of Crystals,” R. H. Doremus,
B. W. Roberts, and D. Turnbull (Eds.), John Wiley, New York, pp. 102–
120 (1958).
7. A. P. Levitt, in “Whisker Technology,” Wiley–Interscience, New York
(1970).
8. U. Lindborg, “Observations on the growth of whisker crystals from zinc
electroplate,” Metall. Trans. A, 6, 1581–1586 (1975).
9. I. A. Blech, P. M. Petroff, K. L. Tai, and V. Kumar, “Whisker growth in
Al thin-films,” J. Cryst. Growth, 32, 161–169 (1975).
10. N. Furuta and K. Hamamura, “Growth mechanism of proper tin-whisker,”
Jpn. J. Appl. Phys., 8, 1404–1410 (1969).
11. R. Kawanaka, K. Fujiwara, S. Nango, and T. Hasegawa, “Influence of
impurities on the growth of tin whiskers,” Jpn. J. Appl. Phys. Part I, 22,
917–922 (1983).
12. K. N. Tu, “Interdiffusion and reaction in bimetallic Cu-Sn thin films,”
Acta Metall., 21, 347–354 (1973).
13. K. N. Tu, “Irreversible processes of spontaneous whisker growth in
bimetallic Cu-Sn thin film reactions,” Phys. Rev. B, 49, 2030–2034
(1994).
14. G. T. T. Sheng, C. F. Hu, W. J. Choi, K. N. Tu, Y. Y. Bong, and L.
Nguyen,“Tin whiskers studied by focused ion beam imaging and trans-
mission electron microscopy,” J. Appl. Phys., 92, 64–69 (2002).
15. W. J. Choi, T. Y. Lee, K. N. Tu, N. Tamura, R. S. Celestre, A. A. Mac-
Dowell, Y. Y. Bong, and L. Nguyen, “Tin whiskers studied by synchrotron
radiation micro-diffraction,” Acta Mater., 51, 6253–6261 (2003).
16. W.J. Boettinger, C.E. Johnson, L. A. Bendersky, K.-W. Moon, M.E.
Williams, and G.R. Stafford, Whisker and hillock formation in Sn, Sn-
Cu, and Sn-Pb lectrodeposists; Acta Mater., 53, 5033–5050 (2005).
17. I. Amato, “Tin whiskers: The next Y2K problem?” Fortune magazine,
vol. 151, issue 1, p.27 (2005).
SVNY339-Tu April 5, 2007 17:18

Spontaneous Sn Whisker Growth 181

18. R. Spiegel, “Threat of tin whiskers haunts rush to lead-free,” Electronic

News, 03/17/2005.
19. http://www.nemi.org/projects/ese/tin whisker.html
20. W. J. Choi, G. Galyon, K. N. Tu, and T. Y. Lee, “The structure and ki-
netics of tin whisker formation and growth on high tin content finishes,”
in “Handbook of Lead-Free Solder Technology for Microelectronic Assem-
blies,” K. J. Puttlitz and K. A. Stalter (Eds.), Marcel Dekker, New York
(2004).
21. P. G. Shewmon, “Diffusion in Solids,” McGraw–Hill, New York (1963).
22. D. A Porter and K. E. Easterling, “Phase Transformations in Metals and
Alloys,“ Chapman & Hall, London (1992).
23. K. Zeng, R. Stierman, T.-C. Chiu, D. Edwards, K. Ano, and K. N. Tu,
“Kirkendall void formation in SnPb solder joints on bare Cu and its effect
on joint reliability,” J. Appl. Phys., 97, 024508–1 to –8 (2005).
24. B.-Z. Lee and D. N. Lee, “Spontaneous growth mechanism of tin
whiskers,” Acta Mater., 46, 3701–3714 (1998).
25. C. Y. Chang and R. W. Vook, “The effect of surface aluminum oxide film
on thermally induced hillock formation,” Thin Solid Films, 228, 205–209
(1993).
26. K. N. Tu and J. C. M. Li, “Spontaneous whisker growth on lead-free solder
finishes,” Mater. Sci. and Eng. A, 409, 131–139 (2005).
27. C. Y. Liu, C. Chen, and K. N. Tu, “Electromigration of thin stripes of
SnPb solder as a function of composition,” J. Appl. Phys., 80, 5703–5709
(2000).
28. S. H. Liu, C. Chen, P. C. Liu, and T. Chou, “Tin whisker growth driven
by electrical currents,” J. Appl. Phys., 95, 7742 (2004).
SVNY339-Tu April 5, 2007 17:13

7
Solder Reactions on Nickel, Palladium,
and Gold

7.1 Introduction
In this chapter we shall discuss solder reaction with Ni, Pd, and Au. These
metals and Cu are being used in under-bump metallization (UBM) or in bond
pad, yet the role of Cu and Ni differs from that of Pd and Au. For Cu and Ni,
the formation of intermetallic compound (IMC) of Cu-Sn or Ni-Sn is chosen
so as to achieve metallic bonds in a solder joint. For Pd and Au, they have
been used as surface coating to passivate the surface of Cu and Ni as well as
to enhance wetting reaction. Typically, the surface of Cu is protected by a
thin film of Au and that of Ni is protected by a film of Pd. Often Au is used
on Ni too.
The reaction between solder and Ni has received much attention because
the reaction rate is about two orders of magnitude slower than that of Cu,
so the effect of spalling of IMC on thin film Ni is less serious and Ni can
also serve as a diffusion barrier of Cu, as discussed in Section 6.10. Why the
reaction rate between Ni and solder is much slower than that between Cu
and solder has been an interesting kinetic question. The answer is unclear at
the moment; mostly it is because the supply of Ni to the reaction may be
much slower than Cu. The supply may depend on the diffusion of Ni along
the interface between Ni3 Sn4 and Ni and also on the solubility of Ni in the
molten solder [1–3].
The reaction between Au and Sn forms Sn4 Au and consumes a large frac-
tion of the solder and it is well-known that if a solder joint has more than 5%
Au, it will have the problem of “cold joint” or brittle joint due to the presence
of a large volume fraction, over 25%, of the Sn4 Au intermetallic in the joint.
The reaction between Pd and eutectic SnPb solder has the fastest rate of
IMC formation in wetting reaction as well as in solid-state aging. The growth
rate is about 1 μm/sec in the wetting reaction, so the reaction can consume
an entire solder joint in 1 minute when the joint diameter is below 50 μm.
This fast reaction has the potential to transform a solder joint completely into
SVNY339-Tu April 5, 2007 17:13

184 Chapter 7

an intermetallic joint. This will become a critical issue in flip chip technology
when the solder bump size is below 25 μm.
A common nature of reactions of molten solder on Pd and Au is the very
fast rate of IMC formation. It is due to the effect of IMC morphology on the
kinetics of growth. In the following, we shall discuss solder reaction with Ni
first, followed by the reactions with Au and Pd.

7.2 Solder Reactions on Bulk and Thin-Film Ni

The reaction rate of molten eutectic SnPb on Ni is about 100 times slower
than that of molten eutectic SnPb on Cu [1–32]. The growth of Ni3 Sn4 between
the molten solder and Ni has the scallop-type morphology, and the diffusivity
of Ni in molten solder should be nearly the same as that of Cu. But why
the formation rate is slow is unclear. Because of this slow wetting reaction,
electronic packaging companies have attempted to replace Cu-based thin-film
UBM by Ni-based thin-film UBM.
Ghosh [7, 8] has recently optimized the thermodynamic descriptions of the
Ni-Sn and Ni-Pb binary systems and calculated several isothermal sections of
the Ni-Sn-Pb ternary phase diagram by extrapolation. The calculation ob-
tained phase diagrams of SnPbNi at 170◦ C, and 240◦ C in weight percentage
[see Fig. 7.1(a) and (b)]. They are rather similar to each other except the
zones of the liquid solder + Ni3 Sn4 and liquid solder + Ni3 Sn4 + (Pb) in the
diagram at 240◦ C. The solubility of Ni in the molten solder is expected to
be higher than that in the solid solder. The first compound to form in the
wetting reaction will be Ni3 Sn4 , and the molten solder that is in equilibrium
with the compound may contain up to 65 wt% Pb, but the solid solder may
contain up to 88 wt% Pb. Since Ni3 Sn4 is unstable on Ni, the other compounds
such as Ni3 Sn2 and Ni2 Sn may form between them, provided that temperature
is high enough and time is long enough.
Figure 7.2(a) to (c) show SEM images of the three-dimensional morphology
of Ni3 Sn4 formation between eutectic SnPb and Ni at 240◦ C for 1, 10 and 40
min, respectively. It is a tilt view of the interface after a preferential etching
of Pb. The scallops of Ni3 Sn4 can be seen. The rate of consumption of Ni by
the wetting reaction in the temperature range of 200 to 240◦ C is shown in
Fig. 7.3. Compared it to the rate of consumption of Cu by the same solder
as shown in Fig. 2.13, the Ni is much slower. For example, after 40 min at
240◦ C, the size (linear dimension) of the Ni3 Sn4 scallops is about 2 μm. In
Fig. 2.5, the size of Cu6 Sn5 scallops after 40 min at 200◦ C is already 10 μm,
so in three-dimensional growth, the latter is at least 100 times faster.
The slow reaction rate of Ni with molten eutectic SnPb solder has been
of keen interest in UBM applications. Nevertheless, when a Ni/Ti thin film
was reacted by the molten solder, the spalling phenomenon was observed.
Figure 7.4(a) to (c) show cross-sectional SEM images of eutectic SnPb on 200
nm Ni/50 nm Ti at 220◦ C for 1, 5, and 40 min, respectively. The absence
SVNY339-Tu April 5, 2007 17:13

Solder Reactions on Ni, Pd, and Au 185

Fig. 7.1. Calculated phase diagrams of SnPbNi at 170◦ C (a), 240◦ C (b), 400◦ (c),
and enlargement of the phase diagram near the SnPb eutectic point (d) in weight
percentage. (Courtesy of Dr. K. Zeng, TI.)

of Ni3 Sn4 at the interface can be seen in Fig. 7.4(b) and (c). Although the
spalling process is slower than that of Cu6 Sn5 , it does occur.
The above discussions are for the wetting reaction when solder is in the
molten state. When solder is in the solid state, the dissolution of Ni into the
solder should result in the formation of Ni3 Sn4 [see Fig. 7.1(a)]. However, if
the temperature is too low (<160◦ C), a metastable phase that has an approx-
imate composition of NiSn3 may form instead of the stable Ni3 Sn4. The NiSn3
grows extremely fast as platelets and can quickly erupt at the surface of solder
coatings on nickel, degrading their solderability. Although the exact mecha-
nism of wettability deterioration due to the presence of NiSn3 is unclear, it
was proposed that the oxidation of NiSn3 near or at the surface of finish was
most likely the cause. One of the solutions to this NiSn3 oxidation problem
would be to decompose it into stable compounds.
Recalling the C-4 technology as discussed in Chapter 1, Ni was used on
the substrate side, but not on the chip side. This is because of the concern for
SVNY339-Tu April 5, 2007 17:13

186 Chapter 7

Fig. 7.2. SEM images of the three-dimensional morphology of Ni3 Sn4 formation be-
tween eutectic SnPb and Ni at 240◦ C for (a) 1 min, (b) 10 min, and (c) 40 min.

intrinsic stress in Ni thin films. To use Ni as UBM on the chip side, one must
have a cushion layer below the Ni film to absorb the stress. In Sections 3.6
and 3.7, we discussed the thin-film Al/Ni(V)/Cu UBM. Otherwise, one must
deposit a low-stress Ni. This leads to the discussion of the thick electroless
Ni(P) UBM below.
SVNY339-Tu April 5, 2007 17:13

Fig. 7.3. Rate of consumption of Ni by the wetting reaction in the temperature range
from 200 to 240◦ C. After 40 min at 240◦ C, the size (linear dimension) of the Ni3 Sn4
scallops is about 2 μm.

Fig. 7.4. Cross-sectional SEM images of eutectic SnPb on 200 nm Ni/50 nm Ti at

220◦ C for (a) 1 min, (b) 5 min, and (c) 40 min.
SVNY339-Tu April 5, 2007 17:13

188 Chapter 7

7.2.1 Reaction between Eutectic SnPb and Electroless Ni(P)

Electroless Ni(P) contains 15 to 20 at.% P and can be deposited by a mask-
less plating process on a patterned and zincated Al surface. Its growth is
isotropic and has a mushroomlike appearance. The binary phase diagram of
Ni-P has a deep eutectic point at 19 at.% P, so it is very easy to form amor-
phous Ni-P alloys near this composition without rapid quenching. Electroless
Ni(P), near the deep eutectic composition, is typically amorphous in the as-
plated state and has low stress. For UBM applications, it has a thickness more
than 10 μm.
Figure 7.5(a) shows a schematic diagram of the cross section of a eutectic
SnPb solder ball on an electroless Ni(P) UBM. The dielectric which defined
the contact opening on the zincated Al surface is SiON. Figure 7.5(b) shows an
SEM cross-sectional image of a eutectic SnPb solder ball of 100-μm diameter
on an electroless Ni(P) UBM. On the Ni(P), a layer of Ni3 Sn4 is seen. The
morphology of the Ni3 Sn4 is faceted, and the grains are either chunky-type or
needle-type, unlike round scallops of Cu6 Sn5 . Nevertheless, there are valleys in
between the chunky and needle-type grains. Figure 7.5(c) shows an enlarged
SEM image of a corner of the Ni(P) UBM. Between the Ni3 Sn4 and electroless
Ni(P), there is a layer of Ni3 P compound. Between the electroless Ni(P) and
SiON, there is an interfacial layer of Ni3 Sn4 .
The Ni3 P has a layer-type morphology. Figure 7.6(a) to (d) show respec-
tively the x-ray mapping of elements of Sn, Ni, Pb, and P across an interfacial
region. The distribution of Sn matches that of Ni in the Ni3 Sn4 , and a layer
of P corresponds to the Ni3 P layer. During the growth of Ni3 Sn4 , it depletes
Ni from the amorphous Ni(P) and enriches the concentration of P toward
that of 75Ni25P, which then crystallizes into the Ni3 P compound. This is an
enhanced crystallization of the amorphous Ni(P) by the soldering reaction. It
is a conservative reaction, in which the Ni3 P consumes nearly all the P in the
amorphous Ni(P) and very little P goes into the solder. The formation of a
thick Ni3 P layer will lead to brittle fracture since Ni3 P cracks easily.
The growth of the Ni3 P is a diffusion-controlled growth with an activation
energy of 0.33 eV/atom. At the moment it is unclear whether it is the Ni dif-
fusion from the Ni(P)/Ni3 P interface or the P diffusion from the Ni3 Sn4 /Ni3 P
interface that controls the growth of Ni3 P;. No doubt the valleys or chan-
nels in the Ni3 Sn4 enable Ni to dissolve into the molten solder quickly. If P
is the dominant diffusing species, it implies that Ni3 P would decompose at
the Ni3 Sn4 /Ni3 P interface. Then as the Ni diffuse into the molten solder, the
P will diffuse back to the Ni3 P/Ni(P) interface to enrich and crystallize the
Ni(P). On the other hand, if Ni is the dominant diffusing species, it will depart
from the Ni3 P/Ni(P) interface and diffuse through the Ni3 P layer to reach the
Ni3 Sn4 /Ni3 P interface. However, what is of interest here is that overall it is a
low-activation energy process, even with the Ni3 P acting as a diffusion barrier!
Again, in the wetting reaction, several micrometers of Ni(P) can be consumed
in a few minutes at 200 to 240◦ C.
SVNY339-Tu April 5, 2007 17:13

Solder Reactions on Ni, Pd, and Au 189

Fig. 7.5. Schematic diagram of the cross section of a eutectic SnPb solder ball on
an electroless Ni(P) UBM. The dielectric which defined the contact opening on the
zincated Al surface is SiON. (b) SEM cross-sectional image of a eutectic SnPb solder
ball of 100 μm diameter on an electroless Ni(P) UBM. On the UBM, a layer of Ni3 Sn4
is seen. (c) SEM cross-sectional image of the penetration of Ni3 Sn4 along the interface
between SiON and Ni(P).

However, in solid-state aging of eutectic SnPb on electroless Ni(P), the

rate of Ni3 Sn4 formation is much slower. Figure 7.7 shows a cross-sectional
SEM image of the interface after 500 hr at 150◦ C. The thickness of Ni3 Sn4 is
only a few micrometers, and it is unclear whether or not Ni3 P forms at such
a low temperature. However, the Ni3 Sn4 now has a layer-type morphology
SVNY339-Tu April 5, 2007 17:13

190 Chapter 7

(b) Ni P (d)

Ni3P

Ni3Sn4

Electroless Ni-P Electroless Ni-P

Sn Pb
(a) (c)
Sn-rich solder

Pb-rich solder
Ni3Sn4

5 μm

Fig. 7.6. X-ray mapping of elements of Sn (a), Ni (b), Pb (c) and P (d) across an
interfacial region. The distribution of Sn matches that of Ni in the Ni3 Sn4 , and a layer
of P corresponds to the Ni3 P layer.

rather than the chunky-type or needle-type morphology. By comparing the

wetting reaction and solid-state reaction, we again conclude that it is similar
to the SnPb/Cu reactions where the wetting reaction is much faster and has
a high rate of free energy gain.
While the electroless Ni(P) is thick enough that no spalling of IMC
takes place, it nevertheless has a weak interface with the dielectric SiON.
In Fig. 7.5(c) where the cross section of a solder ball on electroless Ni(P) is
shown, the molten solder can penetrate the interface, starting from the triple
point where the electroless Ni(P), the SiON, and the molten solder meet to-
gether. The penetration forms Ni3 Sn4 compound between the electroless Ni(P)
and SiON and it extends all the way to the Al interface. Besides solder, corro-
sion reagent can also penetrate the interface, so it becomes a reliability issue.
The kinetics of penetration is interesting since it involves interfacial diffusion
and IMC formation. The solution of the penetration has been given as [16]

 1/2  1/4
12δDi 2(bC 0 − C34 )
y0 = t1/4 , (7.1)
b C34 Dc
SVNY339-Tu April 5, 2007 17:13

Solder Reactions on Ni, Pd, and Au 191

Fig. 7.7. Cross-sectional SEM image of the interface between eutectic SnPb and elec-
troless Ni(P) after 500 hr at 150◦ C. The rate of Ni3 Sn4 formation is very slow.

where y is the penetration depth, δ is the effective width of the interface, Di

is the interfacial diffusivity, b(= C34 /Ci ) is the ratio of concentration in the
IMC (C34 ) and in the interface (Ci ), C0 is the concentration at the triple point
or at the origin of the interface, and Dc is the diffusivity in the compound.
From the solution, a t1/4 dependence of penetration is found, which is similar
to the classic Fisher’s analysis of grain boundary penetration by a diffusional
process.

7.2.2 Reaction between Eutectic Pb-Free Solders

and Electroless Ni(P)
The enhanced crystallization of amorphous electroless Ni(P) into Ni3 P and
Ni3 Sn4 when it is reflowed with eutectic SnPb at 200◦ C is a concern when
Pb-free solder is used. Eutectic SnAg solder has a higher concentration of
Sn and a higher reflow temperature, about 240◦ C, which is very close to the
self-crystallization temperature of amorphous Ni(P) of 250◦ C.
In a eutectic SnAg solder joint on Ni(P), the interfacial IMC is Ni3 Sn4 .
Figure 7.8 shows the interfacial region of a sample after a reflow at 250◦ C for
SVNY339-Tu April 5, 2007 17:13

192 Chapter 7

Fig. 7.8. SEM images of the interfacial region of a sample after a reflow at 250◦ C for
1 hr followed by an aging at 215◦ C for 225 hr. A large number of voids can be seen in
the Ni3 P layer. Between Ni3 Sn4 and Ni3 P, there exists a layer of NiSnP. (Courtesy of
Professor Zhong Chen, Nanyang Technological University, Singapore.)

1 hr followed by an aging at 215◦ C for 225 hr. A large amount of voids can
be seen in the Ni3 P layer. Between Ni3 Sn4 and Ni3 P, there exists a layer of
NiSnP. Figure 7.9 is a SEM image of the voids in Ni3 P layer when the sample
was aged at 190◦ C for 400 hr. Figure 7.10 is a schematic diagram of plausible
fluxes of Ni and Sn during the reactions.

Fig. 7.9. SEM image of the voids in Ni3 P layer when the sample was aged at 190◦ C
for 400 hr. (Courtesy of Professor Zhong Chen, Nanyang Technological University,
Singapore.)
SVNY339-Tu April 5, 2007 17:13

Solder Reactions on Ni, Pd, and Au 193

Inter-diffusion in Sn-3.5Ag/Ni-P

Fig. 7.10. Schematic diagram of plausible fluxes of Ni and Sn during the reaction
between SnAg solder and Ni(P).

For eutectic SnAgCu solder joint on Ni(P), the interfacial IMC is found
to be (Cu,Ni)6 Sn5 rather than Ni3 Sn4 . The Ni3 P layer has a well-developed
columnar structure. The Sn has penetrated into the Ni3 P layer. Between Ni3 P
and (Cu,Ni)6 Sn5 there is a very thin (less than 0.2 μm) dark layer of NiSnP
layer, and there are voids in this NiSnP layer. When the sample was aged at
170◦ C for 64 hr, the interfacial structure is similar, but there are more voids
in the NiSnP layer, and also more Sn has entered into the Ni3 P layer. The
remaining amorphous Ni(P) layer was not crystallized by solid-state aging.

7.2.3 Formation of (Cu, Ni)6 Sn5 versus (Ni, Cu)3 Sn4

Without Cu, the reaction in the binary system of Sn on Ni or in the ternary
system of SnPb on Ni leads to the formation of Ni3 Sn4 . With the presence of
Cu in the SnAgCu solder, the formation of Ni3 Sn4 is suppressed. Instead, the
(Cu, Ni)6 Sn5 forms with some solution of Ni, as reported in the literature. This
can be explained by using the thermodynamic–kinetic considerations. Since
no Ag was detected in the interfacial IMC layer, it is assumed that Ag was
not directly involved in the interfacial reactions. It is found that the amount
of Ni in (Cu, Ni)6 Sn5 increased with the number of reflows. The maximum
solubility of Ni in (Cu, Ni)6 Sn5 in the solder joint, i.e., without the formation
of (Ni, Cu)3 Sn4 , has been calculated to be 8 wt% at 260◦ C, which is in good
agreement with the experimental finding. The distribution of Ni within a
particle of (Cu, Ni)6 Sn5 is quite uniform. This uniform distribution is expected
in a precipitation upon solidification, yet it is unexpected in a solid-state
SVNY339-Tu April 5, 2007 17:13

194 Chapter 7

diffusion process, in which Ni atoms diffuse from the Ni(V) layer into a particle
of Cu6 Sn5 .
The ternary phase of (Cu, Ni)6 Sn5 is more stable than (Ni, Cu)3 Sn4 , hence
the latter forms by dissolution of Cu into an existing layer Ni3 Sn4 , for example,
by interaction across a solder joint as discussed in Section 4.3.

7.2.4 Formation of Kirkendall Voids

The formation of Kirkendall voids in the Ni3 P and NiSnP layer, as shown in
Fig. 7.9, is a reliability issue since the voids will enhance cracking in a brittle
phase. In the reaction, Ni atoms diffuse from the Ni(P) layer to the Ni3 Sn4
layer, and in the reverse direction, a flux of Sn diffuses to the Ni3 P since Sn
has been detected there. A marker motion experiment is needed to determine
whether Ni or Sn is the dominant diffusing species. Void formation is not a
unique phenomenon of the SnAg/Ni(P) reaction. It has also been observed in
the eutectic SnPb/Ni(P) as well as SnAgCu/Ni(P) reactions.

7.3 Solder Reactions on Bulk and Thin-Film Pd

7.3.1 Reaction between Eutectic SnPb and Pd Foil
The unique property of Pd in soldering reaction is that it is highly resistive
to oxidation and very easily wetted by solder to form Pd-Sn IMC [33–41].
Frequently it is used together with Ni to passivate the surface and to enhance
wetting of Ni. In the wetting reaction of molten eutectic SnPb on a Pd foil,
no stable wetting angle between the solder cap and the Pd was found [34].
This behavior is unlike that on Cu and Ni surfaces, where the wetting angle is
stable. The molten solder cap on Pd spreads out unceasingly until all the Sn in
the solder cap transforms completely into Pd-Sn compounds. The instability
of the wetting tip is due to the extremely fast IMC formation between the
molten solder and Pd. The gain in free energy of IMC formation provides
the driving force for the tip motion. The IMC formation at the wetting tip
does not become a diffusion barrier to the wetting reaction; the molten solder
tip is able to advance and wet the Pd in front of it unceasingly. Thus, the
morphology of the IMC at the wetting tip is of interest; it must have channels
to allow the molten solder to pass through.
To confirm the ultrafast reaction, strips of pure Pd foil (0.5 mm in thickness
and 0.5 cm in width) were rolled into rings 3 mm in diameter, and a eutectic
SnPb solder wire was inserted into each of the rings. Then they were immersed
in flux at 250◦ C for 1 to 20 min. The ring cross sections were polished, etched,
and examined by optical microscopy, SEM, and EDX. Formation of a very
thick layer of PdSn3 compound between the Pd and solder was observed [35].
Figure 7.11(a) and (b) show optical microscopic images of a very thick layer
of PdSn3 formation, as indicated by the arrow, of 170 and 360 μm in thickness
SVNY339-Tu April 5, 2007 17:13

Solder Reactions on Ni, Pd, and Au 195

Fig. 7.11. (a,b) Optical mi-

croscopic images of a very
thick layer of PdSn3 forma-
tion (indicated by the arrow)
170 and 360 μm thick after 2
and 5 min of wetting reaction,
respectively.

after 2 and 5 min, respectively. The growth rate within this period is linear
with time and exceeds 1 μm/sec, which is extraordinarily fast and it may be
the fastest IMC growth reported. Figure 7.12(a) and (b) show SEM images
of the PdSn3 layer and its interface with Pd, respectively. The layer has a
lamellar structure, wherein the bright phase is PdSn3 and the dark areas in
between PdSn3 are solder which has been etched away. The lamellar structure
is unique in that the molten solder is able to contact the unreacted Pd all the
time during the reaction. These molten solder channels serve as fast diffusion
paths so that a very high rate of reaction can be achieved. The diffusivity in
the molten solder is about 10−5 cm2 /sec, which is more than enough for the
measured linear growth rate. But beyond 5 min, we found that the growth
rate slowed down when the thickness was more than 500 μm. This is expected
because the growth will eventually become diffusion-controlled even if it has
a diffusivity as high as 10−5 cm2 /sec.
The binary phase diagrams of Pd-Sn and Pd-Pb show that Pd forms sev-
eral compounds with Sn as well as with Pb, such as Pd2 Sn and PdPb2 .
Nevertheless, during the wetting reaction at 250◦ C, the only compound
formed is PdSn3 . A similar result was obtained at 260◦ C. Among the Pd-Sn
SVNY339-Tu April 5, 2007 17:13

196 Chapter 7

Fig. 7.12. SEM images of (a)

the PdSn3 layer and (b) its in-
terface with Pd. The layer has
a lamellar structure, wherein
the bright phase is PdSn3 and
the dark areas in between
PdSn3 are solder which has
been etched away.

compounds, PdSn3 has a very low melting point, about 345◦ C, which is
much lower than that of Pd2 Sn and Pd3 Sn, which have a melting point
around 1300◦ C. If maximum free energy change were the criterion of reac-
tion, those Pd-rich compounds should have formed. Yet, PdSn3 wins the race
and becomes the first phase of formation because the morphology enables
it to have a high rate of growth or a high rate of free energy gain in the
reaction.
The first phase formation has a strong dependence on temperature in the
ternary system of SnPbPd. At or above 250◦ C the first phase formation in
the reaction between eutectic SnPb and Pd is PdSn3 , but at a lower tem-
perature of 220◦ C, PdSn4 forms instead as reported by Ghosh [39–41]. This
finding is in agreement with the ternary phase diagrams of SnPbPd at 250◦ C
and 220◦ C, shown in Fig. 7.13(c) and (b), respectively. The phase diagrams
are calculated using the set of thermodynamic data optimized by Ghosh [40].
While many compounds exist in these phase diagrams, only one or two of the
Sn compounds (PdSn4 and PdSn3 ) have been detected in the wetting reac-
tions, depending on the temperature and time of reaction. Thermodynamic
SVNY339-Tu April 5, 2007 17:13

Solder Reactions on Ni, Pd, and Au 197

Fig. 7.13. Ternary phase diagrams of SnPbPd at (a) 125◦ C, (b) 220◦ C, (c) 250◦ C,
and (d) 303◦ C. (Courtesy of Dr. K. Zeng, TI.)

calculation indicates that below 245◦ C the wetting reaction of eutectic SnPb
with Pd produces PdSn4 [see Fig. 7.13(d)], between 245 and 303◦ C PdSn3
forms, but above 303◦ C PdSn2 may form as the first reaction product.
The solid-state reaction between eutectic SnPb and Pd was found to be
much slower. After reaction at 220◦ C for 50 sec, the same sample was aged
at 125◦ C for 30 days. In addition to a thicker (130 μm) PdSn4 layer, there
were a 50-μm-thick PdSn3 layer and a 40-μm-thick Pb layer. These phase
formations are consistent with the calculated phase diagram [Fig. 7.13(a)].
Yet the thickness of the compounds formed in 30 days at 125◦ C is comparable
to that formed in a few minutes at 220◦ C.

7.3.2 Reaction between Eutectic SnPb and Pd Thin Film

To passivate Ni, the Pd film used is typically about 50 to 100 nm thick.
When the thin-film Pd is in contact with molten solder, it dissolves very
SVNY339-Tu April 5, 2007 17:13

198 Chapter 7

quickly into the solder and allows the solder/Ni reaction to take place. But
a slightly thicker Pd may lead to Pd-Sn compound formation, which could
become a diffusion barrier in the subsequent reflow. The wetting behavior
of eutectic SnPb on Pd/Ni plated Cu leadframe has been studied [36–38].
The wetting reaction forms a Pd-Ni-Sn ternary compound and Ni3 Sn4 . The
ternary compound grains were broken off from the interface and scattered
into the molten solder. The Ni3 Sn4 , consisting of small scallops, remains as a
rather uniform layer on the unreacted Ni.

7.4 Solder Reactions on Bulk and Thin-Film Au

Both Au and Pd serve the same purpose for surface passivation and wetting
enhancement, yet a major difference between them is that Au has a very high
solubility in molten eutectic SnPb solder, about 7.8 wt% at 220◦ C. When
a molten solder joint has dissolved this amount of Au, the precipitation of
Au-Sn compound upon solidification will form a brittle solder joint, or “cold
joint.” On the other hand, a bump of Au may be jointed by a thin layer of
solder or Sn. While Au bumps have better dimensional stability or better
creep resistance than soft solder bumps, their interfaces are poorer in fracture
and fatigue resistance. When the solder layer on the Au bump is thin, it can
be transformed completely to Au-Sn IMC during reflow. If the IMC is thin
and brittle, it will not survive the very large thermal strain we have discussed
in direct chip attachment in Chapter 1 [42–53].
Isothermal sections of the SnPbAu ternary phase diagram at varied tem-
peratures are needed for a comparison of wetting reaction and solid-state
aging. These diagrams have been calculated on the basis of the optimized bi-
nary systems. The ternary phase diagrams of SnPbAu at 160, 200, 225, and
330◦ C are shown in Fig. 7.14(a), (b), (c), and (d), respectively. At 200◦ C, the
molten solder will dissolve about 4.5 wt% Au before the formation of AuSn4 .
Since AuSn4 is unstable on Au, we expect the other compounds such as AuSn2
and AuSn will form between AuSn4 and Au.

7.4.1 Reaction between Eutectic SnPb and Au Foil

Experimentally, in the wetting reaction of a eutectic SnPb solder cap on a Au
foil at 200◦ C, the wetting angle is unstable and it decreases with time, but
it differs from that on a Pd surface [45, 46]. The cap begins with a wetting
angle of 20◦ , decreases to 6◦ in 2 to 3 min, and then stops because all the
molten solder in the cap has reacted with Au to form IMC. The surface of
the cap becomes very rough and the faceted surface of IMC can be seen.
Cross-sectional SEM images of the cap after only 5 and 60 sec at 200◦ C are
shown in Fig. 7.15, where the AuSn4 compound has extended all the way to
the surface of the cap. The cross-sectional profile of the top surface of the cap
SVNY339-Tu April 5, 2007 17:13

Solder Reactions on Ni, Pd, and Au 199

Fig. 7.14. Ternary phase diagrams of SnPbAu at (a) 160◦ C, (b) 200◦ C, (c) 225◦ C,
and (d) 330◦ C. (Courtesy of Dr. K. Zeng, TI.)

is not smooth and is no longer circular. Also, the bottom surface of the cap
has sunken into the Au. A large fraction of the AuSn4 compounds was formed
during solidification because of the high solubility of Au in the molten solder.
With a diffusivity of 10−5 cm2 /sec, Au atoms can diffuse a distance of 100 μm
in 5 sec and reach the top surface of the solder cap, so Au can easily saturate
the molten solder during the reflow. However, the interesting question here is
why the formation of AuSn4 does not become a diffusion barrier to stop the
interface from sunking into Au. We recall that in the cases of SnPb/Cu and
SnPb/Ni, there are no sunken interfaces.
We may also ask how the high solubility of Au in molten eutectic SnPb
affects IMC formation. According to thermodynamic principles, dissolution
of Au into the molten solder must occur first when the molten solder is in
contact with the Au, and it is assumed that only when the molten solder has
reached the solubility limit of Au can the IMC start to form at the interface.
However, it is unnecessary for the entire molten solder in the sample to reach
SVNY339-Tu April 5, 2007 17:13

200 Chapter 7

Fig. 7.15. Cross-sectional SEM images of eutectic SnPb solder cap on Au foil after
only 5 sec (a) and 60 sec (b) at 200◦ C. It is seen that the AuSn4 compound has
extended all the way to the surface of the cap.

the solubility limit. Rather, it is only necessary to have a boundary layer of

molten solder next to the Au become saturated or slightly supersaturated,
thereafter the IMC can nucleate heterogeneously and grow. Thus, we may
assume that a boundary layer of molten solder next to Au can dissolve Au
to supersaturation, the IMC then nucleates and precipitates heterogeneously
on the Au surface. The nucleation of IMC may only require a very small
supersaturation in the molten solder since the IMC is supercooled at the
wetting temperature. When the IMC grows into a continuous layer, it becomes
a diffusion barrier to the subsequent dissolution.
To study such dissolution and reaction issue, a strip of Au foil 0.5 mm
thick was rolled into a ring of 0.4-mm diameter and a eutectic SnPb wire
was inserted into the ring and immersed in flux for reaction at 200◦ C. Figure
7.16(a) to (c) show the interface after 10, 90, and 210 sec at 200◦ C. To arrest
the fast reaction during cooling, the ring sample was quenched in ethanol.
The cross section was polished and etched for SEM observation. Both AuSn4
and AuSn2 were found. The latter is a thin continuous layer between the Au
and AuSn4 and it has a constant thickness, independent of the annealing time.
Because the thickness of the AuSn2 layer is constant, it is believed that it forms
during quench, not during the annealing at 200◦ C. The AuSn4 has a chunky-
type morphology, has a three-dimensional growth, and becomes longer, wider,
and thicker with time. However, there are solder phases or channels in between
the chunky grains. The initial growth of AuSn4 is extremely fast, about 1
μm/sec, which is similar to that of PdSn3 . However, unlike PdSn3 which can
grow to several hundred micrometers thick, the layer of AuSn4 at the interface
SVNY339-Tu April 5, 2007 17:13

Solder Reactions on Ni, Pd, and Au 201

Fig. 7.16. Cross-sectional images of a ring of Au foil containing eutectic SnPb after
10 sec (a), 90 sec (b), and 210 sec (c) at 200◦ C.

does not grow very thick. The morphology of AuSn4 is different from that of
PdSn3 . The AuSn4 tends to disperse all over the solder within the ring and
occupied over one quarter of the total volume of the solder. These dispersed
AuSn4 crystals come from two processes; one is from the breakage of grains
growing from the interface, and the other is from precipitation of the dissolved
Au in the molten solder during cooling. Because the AuSn4 contains channels,
it is not a diffusion barrier and will not prevent the continuous dissolution of
Au into the molten solder. The porous morphology of AuSn4 enables it to
SVNY339-Tu April 5, 2007 17:13

202 Chapter 7

grow fast and at the same time to be accompanied by a fast dissolution of Au

into the molten solder. The channels, when they are filled with molten solder,
serve as short-circuit paths for the transport of Au needed to grow the AuSn4
as well as to maintain the dissolution of Au into the pool of molten solder.
The dissolution has also been studied by the reaction between Pb-5 wt%
Sn on Au at 330◦ C for 10 sec. Figure 7.17(a) and (b) show cross-sectional
SEM images of the solder cap and an enlarged image at one end of the cap,
respectively. Two striking appearances are to be noted in these figures. The
first is a very deep sunken interface, which shows that the molten solder has
dissolved about 70 μm of Au in depth. It indicates a dissolution rate of Au of
about 7 μm/sec, assuming a linear dissolution rate. This is an extremely fast
rate of dissolution. The second is that the solder cap is full of Au2 Pb compound
and there is no other IMC at the interface. According to the SnPbAu ternary
phase diagram at 330◦ C, shown in Fig. 7.14(d), the molten high-Pb solder
can dissolve up to 50 wt% Au and the IMC that forms at 330◦ C is Au2 Pb.
On the basis of the phase diagram and the SEM image shown in Fig. 7.17(a),
they suggest that in the 10-sec reflow, the solder cap has not dissolved all
the Au to reach the solubility limit, not even in the boundary layer next to

Fig. 7.17. (a) Cross-sectional SEM image of the solder cap of Pb-5 wt% Sn on Au
and (b) an enlarged image at one end of the cap.
SVNY339-Tu April 5, 2007 17:13

Solder Reactions on Ni, Pd, and Au 203

the Au. Thus, no IMC is formed at the interface. On the other hand, if the
sample were annealed more than 10 sec, an interfacial layer of Au2 Pb might
have been found. Upon cooling, the dissolved Au comes out in the form of
Au2 Pb dispersed all over the solder. What has happened to the 5 wt% Sn in
the solder is unclear. According to the liquidus surface projection of the phase
diagram, a very small amount of the compound phase AuSn will form after
the precipitation of the primary Au2 Pb.
Since the sunken interface changes the wetting tip angle, no constant wet-
ting angle exists during the reflow. This is another case of a chemical reaction
(dissolution) at the wetting interface defeating the Young’s equilibrium condi-
tion of the wetting tip. The wetting angle, when the sunken part is included,
is not constant but increases with time.
Solid-state aging of near-eutectic SnPb on Au has been studied in the
temperature range from 80 to 160◦ C. Electro-deposited Au 12 μm thick was
patterned into 2-mm-diameter disks and a drop of molten solder of Pb-72 at.%
Sn was applied to wet the Au surface with flux in about 1 sec. Solid-state aging
was followed after cooling to room temperature. After aging for 7 hr at 160◦ C,
a very thin but very Au-rich δ phase was found next to Au and followed by
1.5 μm of AuSn, 10 μm of AuSn2 , and 30 μm of AuSn4 . All three Au-Sn
compounds contain a certain amount of Pb. The morphology of AuSn4 is very
much like that of PdSn3 shown in Fig. 7.11. The microstructure of AuSn4 was
described as having lamellar (columnar) grains with a thickness (diameter)
of about 3 μm, lying parallel to the growth direction, with sheets or rods of
solder alloy between them along the same direction. The growth of AuSn4 was
found to be diffusion-controlled with an activation energy of 0.84 eV/atom.
No precipitation of Au-Sn and Au-Pb compounds in the remaining solder was
reported. According to the phase diagram of SnPbAu at 160◦ C, shown in
Fig. 7.14(a), we expect the formation of AuSn4 , AuSn2 , AuSn and perhaps
other Au-rich compounds between the solder and the Au. No formation of Au-
Pb compound is expected. The phase formation is in good agreement with the
experimental finding mentioned.
Again, the major difference between the wetting reaction and the solid-
state reaction is in the rate of reaction. The wetting reaction at 200◦ C has
a rate of IMC formation of about 1 μm/sec, and it takes only a few minutes
to form a layer of AuSn4 of 30μm. On the other hand, it takes 7 hr for
solid-state reaction at 160◦ C to form the same amount of AuSn4 . Thus, there
is a difference of at least two orders of magnitude in rate between the two
reactions. Furthermore, the wetting reaction involves a substantial amount of
dissolution of the Au, but not the solid-state reaction.
The difference in rates of IMC formation is not as large as what we have
found in the cases of Cu and Ni. We note that this is because the AuSn4 does
not form a continuous layer and Au can diffuse interstitially in the solid solder
lamellae in between the AuSn4 . While the diffusivity of Au in molten solder is
about 10−5 cm2 /sec, the solid-state diffusivity of Au in Pb and Sn at 160◦ C
is about 10−7 cm2 /sec, which is quite fast.
SVNY339-Tu April 5, 2007 17:13

204 Chapter 7

7.4.2 Reaction between Eutectic SnPb and Au Thin Film

When the Au film is about 100 nm as in the case of a trilayer of Cr/Cu/Au,
the molten solder dissolves all the Au very quickly before reacting with the Cu.
Tiny Au-Sn particles were segregated out on the surface of Cu6 Sn5 spheroids
upon cooling. When the Au is thick and is about 1 μm as in the case of
Cu/Ni/Au trilayer used in ball-grid-array (BGA) packaging, the reflow at
225◦ C for 10 sec has converted all the Au into a porous layer of AuSn2 and
AuSn4 . The grains in this layer then began to separate themselves from the
Ni layer and to spall into the molten solder. In the cross-sectional SEM image,
a few very large AuSn4 grains can be seen to have dispersed in the solder ball.
When Ni(P) is used as UBM in flip chip, a thin coating of Au is deposited
on Ni(P) by electroplating. During reflow, the Au dissolves into the molten
solder and allows the Ni(P) to react with solder. The Au layer is so thin that
the overall content of Au in the molten solder bump is below the saturation
solubility. The compound AuSn4 will precipitate throughout the bulk of the
solder during cooling. However, after a solid-state aging at high temperatures
for several hours, some of the AuSn4 crystals that dispersed in the bulk of
BGA solder joints before aging have redeposited at the solder/Ni3 Sn4 interface
as a continuous layer. The aged joints were found to be significantly weaker
than those before aging, and they failed by brittle fracture along the interface
between the AuSn4 layer and the Ni3 Sn4 layer. To prevent the redeposition
problem, one percent of Ni particles was added to the solder so as to keep the
dispersion of Au remaining in the bulk of the solder.

References
1. P. W. Dehaven, “The reaction kinetics of liquid 60/40 Sn/Pb solder with
copper and nickel: a high temperature X-ray diffraction study,” in Proc.
Electronic Packaging Materials Science, Materials Research Society, USA,
1984, pp. 123–128.
2. W. G. Bader, “Dissolution of Au, Ag, Pd, Pt, Cu, and Ni in a molten
tin-lead solder,” Weld. J. Res. Suppl., 28, 551s–557s (1969).
3. K. N. Tu and R. Rosenberg, “Room temperature interaction in bimetallic
thin films,” Jpn J. Appl. Phys., Suppl. 2 (Part 1), 633 (1974).
4. C.-Y. Lee and K.-L. Lin, “The interaction kinetics and compound forma-
tion between electroless Ni-P and solder,” Thin Solid Films, 249, 201–206
(1994).
5. C.-Y. Lee and K.-L. Lin, “Preparation of solder bumps incorporating elec-
troless nickel-boron deposit and investigation on the interfacial interaction
behavior and wetting kinetics,” J. Mater. Sci. Mater. Electron., 8, 377–383
(1997).
6. C.-J. Chen and K.-L. Lin, “The reactions between electroless Ni-Cu-P
deposit and 63Sn-37Pb flip chip solder bumps during reflow,” J. Electron.
Mater. 29, 1007–1014 (2000).
SVNY339-Tu April 5, 2007 17:13

Solder Reactions on Ni, Pd, and Au 205

7. G. Ghosh, “Thermodynamic modeling of the Ni-Pb-Sn system,” Metall.

Mater. Trans. A, 30, 1481–1494 (1999).
8. G. Ghosh, “Kinetics of interfacial reaction between eutectic Sn-Pb solder
and Cu/Ni/Pd metallizations,” J. Electron. Mater., 28, 1238–1250, 1999.
9. D. Frear, F. Hosking, and P. Vianco, “Mechanical behavior of solder
joint interfacial intermetallics,” in Proc. Materials Developments in Mi-
croelectronic Packaging: Performance and Reliability, 19–22 Aug. 1991,
Montreal, Canada, ASM International, Materials Park, Ohio, pp. 229–240
(1991).
10. A. C. Harman, “Rapid tin-nickel intermetallic growth: Some effects on
solderability,” in Proc. InterNepcon, Brighton, UK, pp. 42–49 (1978).
11. J. Haimovich, “Intermetallic compound growth in tin and tin-lead platings
over nickel and its effects on solderability,” Weld. J., 68, 102s–111s (1989).
12. P. J. Kay and C. A. MacKay, “Growth of intermetalic compounds on
common basis material coated with tin or tin-lead alloys,” Trans. Inst.
Met. Finish., 54, 68–74 (1976).
13. J. W. Jang, P. G. Kim, K. N. Tu, D. R. Frear, and P. Thompson, “Sol-
der reaction-assisted crystallization of electroless Ni-P under bump met-
allization in low cost flip chip technology,” J. Appl. Phys., 85, 8456–8463
(1999).
14. J. W. Jang, D. R. Frear, T. Y. Lee, and K. N. Tu, “Morphology of inter-
facial reaction between Pb-free solders and electroless Ni(P) under-bump-
metallization,” J. Appl. Phys., 88, 6359–6363 (2000).
15. P. G. Kim, J. W. Jang, T. Y. Lee, and K. N. Tu, “Interfacial reaction
and wetting behavior in eutectic SnPb solder on Ni/Ti thin films and Ni
foils,” J. Appl. Phys. 86, 6746–6751 (1999).
16. P. G. Kim, J. W. Jang, K. N. Tu, and D. Frear, “Kinetic analysis of in-
terfacial penetration accompanied by intermetallic compound formation,”
J. Appl. Phys. 86, 1266–1272 (1999).
17. Y.-D. Jeon, K.-W. Paik, K.-S. Bok, W.-S. Choi, and C.-L. Cho, “Studies
on Ni-Sn intermetallic compound and P-rich Ni layer at the electroless
nickel UBM–solder interface and their effects on flip chip solder joint reli-
ability,” in Proc. 51st Electronic Components & Technology Conference,
May 29-June 31, 2001, Orlando, FL, IEEE, Piscataway, NJ, pp. 1326–1332
(2001).
18. K. C. Hung, Y. C. Chan, C. W. Tang, and H. C. Ong, “Correlation
between Ni3 Sn4 intermetallics and Ni3 P due to solder reaction-assisted
crystallization of electroless Ni-P metallization in advanced packages,” J.
Mater. Res. 15, 2534–2539 (2000).
19. K. C. Hung and Y. C. Chan, “Study of Ni3 P growth due to solder reaction-
assisted crystallization of electroless Ni-P metallization,” J. Mater. Sci.
Lett., 19, 1755–1757 (2000).
20. P. Liu, Z. Xu, and J. K. Shang, “Thermal stability of electroless-Ni/solder
interfaces: Part A. Interfacial chemistry and microstructure,” Metall.
Mater. A, Trans. 31, 2857–2866 (2000).
SVNY339-Tu April 5, 2007 17:13

206 Chapter 7

21. K. Zeng, V. Vuorinen, and J. K. Kivilahti, “Intermetallic reactions be-

tween lead-free SnAgCu solder and Ni(P)/Au surface finish on PWBs,”
in Proc. 51st Electronic Components & Technology Conference, May 29-
June 31, 2001, Orlando, FL, IEEE, Piscataway, NJ, pp. 685–690 (2001).
22. M. Li, F. Zhang, W. T. Chen, K. Zeng, K. N. Tu, H. Balkan, and P.
Elenius, “Interfacial microstructure evolution between eutectic SnAgCu
solder and Al/Ni(V)/Cu thin films,” J. Mater. Res., 17, 1612–1621
(2002).
23. C. Y. Liu, K. N. Tu, T. T. Sheng, C. H. Tung, D. R. Frear, and P. Elenius,
“Electron microscopy study of interfacial reaction between eutectic SnPb
and Cu/Ni(V)/Al thin film metallization,” J. Appl. Phys., 87, 750–754
(2000).
24. K. Y. Lee, M. Li, and K. N. Tu, “Growth and ripening of (Au,Ni)Sn4
phase in Pb-free and Pb containing solder on Ni/Au metallization,” J.
Mater. Res., 18, 2562–2570 (2003)
25. M. O. Alam, Y. C. Chan, and K. N. Tu, “Effect of reaction time and P-
content on mechanical strength of the interfaces formed between eutectic
Sn-Ag solder and Au/electroless Ni(P)/Cu bond pad,” J. Appl. Phys., 94,
4108–4115 (2003)
26. M. O. Alam, Y. C. Chan, and K. N. Tu, “Effect of 0.5 wt% Cu addition in
the Sn-3.5%Ag solder on the interfacial reaction with Au/Ni metallizaion,”
Chem. Mater., 15, 4340–4342 (2003).
27. C. E. Ho, Y. W. Lin, S. C. Yang, C. R. Kao, and D. S. Jiang, “Effect
of limited Cu supply on soldering reactions between SnAgCu and Ni,” J.
Electron. Mater., 35, 1017–1024 (2006).
28. Y. D. Jeon, S. Nieland, A. Ostmann, H. Reichl, and K. W. Paik, “A study
on interfacial reactions between electroless Ni-P under bump metallization
and 95.5Sn4Ag 0.5 Cu alloy,” J. Electron. Mater., 32, 548–557 (2006).
29. M. L. Huang, T. Loeher, D. Manessis, L. Boettcher, A. Ostmann, and H.
Reichl, “Morphology and growth kinetics of intermetallic compounds in
solid-state interfacial reaction of electroless Ni-P with Sn based Pb-free
solders,” J. Electron. Mater., 35, 181–188 (2006).
30. K. S. Kim, S. H. Huh, and K. Suganuma, “Effects of intermetallic com-
pounds on properties of SnAgCu Pb-free soldered joints,” J. Alloys Com-
pounds, 352, 226–236 (2006).
31. L. Y. Hsiao, S. T. Kao, and J. G. Duh, “Characterizing metallurgical re-
action of Sn3Ag0.5Cu composite solder by mechanical alloying with elec-
troless Ni-P/Cu under bump metallization after various reflow cycles,” J.
Electron. Mater., 35, 81–88 (2006).
32. C. Lin, J. G. Duh, and B. S. Chiou, “Wettability of electroplated Ni-P
in under bump metallization wuth SnAgCu solders,” J. Electron. Mater.,
35, 7–14 (2006).
33. K. N. Tu, “Single intermetallic compound formation in Pd-Pb and Pd-
Sn thin-film couples studied by X-ray diffraction”, Mater. Lett., 1, 6–10
(1982).
SVNY339-Tu April 5, 2007 17:13

Solder Reactions on Ni, Pd, and Au 207

34. Y. Wang, H. K. Kim, H. K. Liou, and K. N. Tu, “Rapid soldering reactions

of eutectic SnBi and eutectic SnPb solder on Pd surfaces,” Scr. Metall.
Mater., 32, 2087–2092 (1995).
35. Y. Wang and K. N. Tu, “Ultra-fast intermetallic compound formation
between eutectic SnPb and Pd where the intermetallic is not a diffusion
barrier,” Appl. Phys. Lett., 67, 1069–071 (1995).
36. P. G. Kim, K. N. Tu, and D. C. Abbott, “Effect of Pd thickness on sol-
dering reaction between eutectic SnPb and plated Pd/Ni thin films on Cu
leadframe,” Appl. Phys. Lett., 71, 61–63 (1997).
37. P. G. Kim, K. N. Tu, and D. C. Abbott, “Soldering reaction between
eutectic SnPb and plated Pd/Ni thin films on Cu leadframe”, Appl. Phys.
Lett., 71, 61–63 (1997).
38. P. G. Kim, K. N. Tu, and D. C. Abbott, “Time and temperature de-
pendent wetting behavior of eutectic SnPb on Cu lead-frame plated
with Pd/Ni and Au/Pd/Ni thin films,” J. Appl. Phys. 84, 770–775
(1998).
39. G. Ghosh, “Diffusion and phase transformations during interfacial reac-
tion between lead-tin solders and palladium,” J. Electron. Mater., 27,
1154–1160 (1998).
40. G. Ghosh, “Thermodynamic modeling of the Pd-Pb-Sn system,” Metall.
Mater. Trans. A, 30, 5–18 (1999).
41. G. Ghosh, “Interfacial microstructure and the kinetics of interfacial reac-
tion in diffusion couples between Sn-Pb solder and Cu/Ni/Pd metalliza-
tion,” Acta. Mater., 48, 3719–3738 (2000).
42. A. Prince, “The Au-Pb-Sn ternary system,” J. Less-Common Met., 12,
107–116 (1967).
43. G. Humpston and D. S. Evans, “Constitution of Au-AuSn-Pb partial
ternary system,” Mater. Sci. Technol., 3, 621–627 (1987).
44. E.-B. Hannech and C. R. Hall, “Diffusion controlled reactions in Au/Pb-
Sn solder system,” Mater. Sci. Technol., 8, 817–824 (1992).
45. P. G. Kim and K. N. Tu, “Morphology of wetting reaction of eutectic
SnPb solder on Au foils,” J. Appl. Phys., 80, 3822–3827 (1996).
46. P. G. Kim and K. N. Tu, “Fast dissolution and soldering reactions on Au
foils,” Mater. Chem. Phys., 53, 165–171 (1998).
47. Z. Mei, M. Kaufmann, A. Eslambolchi, and P. Johnson, “Brittle interfacial
fracture of PBGA packages soldered on electroless nickel/immersion gold,”
in Proc. 48th Electronic Components and Technology Conference, 25–28
May 1998, Seattle, WA, IEEE, Piscataway, NJ, pp. 952–961 (1998).
48. Z. Mei, P. Callery, D. Fisher, F. Hua, and J. Glazer, “Interfacial frac-
ture mechanism of BGA packages on electroless Ni/Au,” in Proc. Pacific
Rim/ASME International Intersociety Electronic and Photonic Packaging
Conf., Advances in Electronic Packaging 1997, ASME, New York, Vol. 2,
pp. 1543–1550 (1997).
49. S. C. Hung, P. J. Zheng, S. C. Lee, and J. J. Lee, “The effect of Au plat-
ing thickness of BGA substrates on ball shear strength under reliability
SVNY339-Tu April 5, 2007 17:13

208 Chapter 7

tests,” in Proc. 24th IEEE/CPMT International Electronics Manufactur-

ing Technology Symp., Oct. 18–19, 1999, Austin, TX, IEEE, Piscataway,
NJ, pp. 7–15 (1999).
50. C. E. Ho, R. Zheng, G. L. Luo, A. H. Lin, and C. R. Kao, “Formation and
resettlement of (Aux Sn1−x )Sn4 in solder joints of ball-grid-array packages
with the Au/Ni surface finish,” J. Electron. Mater., 29, 1175–1181 (2000).
51. A. M. Minor and J. W. Morris, “Growth of a Au-Ni-Sn intermetallic
compound on the solder-substrate interface after aging,” Metall. Mater.
Trans. A, 31, 798–800 (2000).
52. A. M. Minor and J. W. Morris, “Inhibiting growth of the Au0.5Ni0.5Sn4
intermetallic layer in Pb-Sn solder joints reflowed on Au/Ni metalliza-
tion,” J. Electron. Mater., 29, 1170–1174 (2000).
53. S. Anhock, H. Oppermann, C. Kallmayer, R. Aschenbrenner, L.
Thomas, and H. Reichl, “Investigations of Au-Sn alloys on different
end-metallizations for high temperature applications,” in Proc. 22nd
IEEE/CPMT International Symp. Electronics Manuf. Technol., 27–29
April 1998, Berlin, IEEE, pp. 156–165 (1998).
SVNY339-Tu April 5, 2007 17:25

II

Electromigration and Thermomigration

SVNY339-Tu April 5, 2007 17:25

8
Fundamentals of Electromigration

8.1 Introduction

An ordinary household extension cord conducts electricity without electromi-

gration in the cord because the electric current density in the cord is low,
about 102 A/cm2 , and also the ambient temperature is too low for atomic
diffusion to occur in copper. The free electron model of conductivity of metals
assumes that the conduction electrons are free to move in the metal, uncon-
strained by the perfect lattice of atoms except for scattering interactions due
to phonon vibration. The scattering is the cause of electrical resistance and
joule heating. When an atom is out of its equilibrium position, for example, a
diffusing atom at the activated state, it possesses a very large scattering cross
section. Nevertheless, when the electric current density is low, the scattering
or the momentum exchange between the electrons and the diffusing atom does
not enhance displacement of the latter and it has no net effect on atomic diffu-
sion. However, the scattering by electrons in a high current density, above 104
A/cm2 , enhances atomic displacement in the direction of electron flow. The
enhanced atomic displacement and the accumulated effect of mass transport
under the influence of electric field (mainly a high-density electric current)
are called electromigration.
It is worth noting that a household cord is allowed to carry only a very
low current density, otherwise joule heating will burn the fuse. Yet, a thin-
film interconnect in a Si device can carry a much higher current density, which
facilitates electromigration. This is because the Si chip on which interconnects
are built is a very good heat conductor, hence the interconnect on Si can
carry a very high current density without overheating. On the other hand, in
a device having a very dense integration of circuits, the heat management is
a serious issue. Typically, a device is cooled by a fan or other means in order
to maintain the working temperature around 100◦ C.
In very-large-scale integration (VLSI) of circuits on a Si device, assuming
an Al or Cu thin-film line 0.5 μm wide and 0.2 μm thick carrying a current of 1
mA, for example, the current density will be 106 A/cm2 . Such current density
can cause electromigration in the line at the device working temperature of
SVNY339-Tu April 5, 2007 17:25

212 Chapter 8

Fig. 8.1. (a) SEM image of Al short strips on TiN baseline after electromigration. (b)
High-magnification image of one of the Al strips. (Courtesy of Dr. Alexander Straub,
MPI Stuttgart, Germany.)

100◦ C and lead to void formation at the cathode and extrusion at the anode.
These defects are the most persistent and most serious reliability failures in
thin-film integrated circuits. As device miniaturization demands smaller and
smaller interconnects, the current density goes up, as does the probability
of circuit failure induced by electromigration. This is a subject which has
demanded and attracted much attention [1–25].
The phenomenon of electromigration can be observed directly from the re-
sponse of a set of short Al strips on a baseline of TiN as shown in Fig. 8.1(a).
This structure is called the Blech structure for electromigration tests [5, 6].
The Al strips have a line width of 10 μm and thickness of 100 nm. The ap-
plied electric current in the TiN baseline took a detour to go along the strips
because the latter are paths of low resistance. When the current density and
temperature are high enough, atomic transport occurs and void and extru-
sion formations can be observed directly. Under applied current density of
106 A/cm2 at 225◦ C for 24 hr, depletion at the cathode end and extrusion
formation at the anode end of several of the strips can be seen in Fig. 8.1(a).
It is worth noting that no electromigration damage can be seen in the shorter
strips in the upper right corner. Figure 8.1(b) is a higher-magnification SEM
image of one of the strips. It is important to note that the atomic displace-
ment and mass transport are in the same direction as the electron flow from
the lower left corner to the upper right corner.
Figure 8.2(a) shows the morphology of a Cu strip in electromigration with
a current density of 5 × 105 A/cm2 at 350◦ C for 99 hr [12]. At the cathode end
of the strip, a depleted region can be seen, but at the anode end, an extrusion
is seen. By conservation of mass, the depletion (void) equals the extrusion in
the same strip. The rate of depletion at the cathode can be measured so the
drift velocity can be calculated. Figure 8.2(b) is a set of SEM images of the
depletion at the cathode of a Cu strip taken at different intervals at 400◦ C
with a current density of 2.1 × 106 A/cm2 . The drift velocity is about 2 μm/hr.
SVNY339-Tu April 5, 2007 17:25

Fundamentals of Electromigration 213

(a)

(b)

Fig. 8.2. (a) SEM image of a Cu strip on W baseline after electromigration. (b) SEM
images of the depletion at cathode of a Cu strip taken at different intervals at 400◦ C
with current density of 2.1 × 106 A/cm2 . The drift velocity is about 2 μm/hr.
SVNY339-Tu April 5, 2007 17:25

214 Chapter 8

(a) Cu/dielectric cap

interface

Current Crowding

(b) Cu/dielectric cap

interface

Current Crowding

Cu/dielectric cap
interface

Fig. 8.3. (a) SEM images of void formation at the upper end and lower end (b) of a via
in a dual damascene structure of Cu interconnect due to electromigration. (Courtesy
of Professor S. G. Mhaisalkar, Nanyang Technological University, Singapore.)

Figure 8.3 shows images of void formation in the upper (a) and lower
end (b) of a via at the cathode end in a dual damascene structure of Cu
interconnect due to electromigration. Each of them had caused open circuit
and was detected by a very large resistance increase. The kinetic process of
void formation was the propagation and accumulation of a sequence of small
voids on the upper surface of the Cu interconnect. Clearly, electromigration
in Cu interconnect is dominated by surface diffusion [21–25].

8.2 Electromigration in Metallic Interconnects

Electromigration is the result of a combination of thermal and electrical effects
on mass transport. If the conducting line is kept at a very low temperature
SVNY339-Tu April 5, 2007 17:25

Fundamentals of Electromigration 215

Table 8.1. Diffusivity of Al, Cu, and SnPb

Melting Temperature Diffusivities at Diffusivities at

point (K) ratio 373K/Tm 100◦ C(cm2 /sec) 350◦ C (cm2 /sec)

Lattice D l = 7 × 10−28 D l = 5 × 10−17

Cu 1356 0.275 Grain boundary D gb = 1.2 × 10−9
D gb = 3 × 10−15
Surface D s = 10−12 D s = 10−8
Al 933 0.4 Lattice D l = 1.5 × 10−19 D l = 10−11
Grain boundary D gb = 5 × 10−7
D gb = 6 × 10−11
Eutectic 456 0.82 Lattice D l = 2 × 10−9 Molten state
SnPb to 2 × 10−10 D l > 10−5

(e.g., liquid nitrogen temperature), electromigration cannot occur because

there is no atomic mobility of diffusion, even though there is a driving force.
The contribution of thermal effect can be recognized by the fact that electro-
migration in a bulk eutectic solder bump occurs at about three-quarters of its
melting point in absolute temperature, electromigration in a polycrystalline
Al thin-film line occurs at less than one-half of its melting point in absolute
temperature, and electromigration in a Cu thin-film line having bamboo-type
grain structure occurs at about one-quarter of its melting point in absolute
temperature. At these homologous temperatures, there are atoms which un-
dergo random walks in the bulk of the solder bump, in the grain boundaries of
the Al thin-film line, and on the free surface of Cu damascene interconnect, re-
spectively, and these are the atoms which take part in electromigration under
the applied current. Indeed, we assume the Si device working temperature to
be 100◦ C, which is about three-quarters of the melting point of solders, about
slightly less than half of the melting point of Al, and about one-quarter of the
melting point of Cu. In this homologous temperature scale, lattice diffusion,
grain boundary diffusion, and surface diffusion occur predominantly at three-
quarters, one-half, and one-quarter of the absolute temperature of a metal,
respectively. [16]
Table 8.1 lists the melting points and diffusivities which are relevant to
the electromigration behaviors in Cu, Al, and eutectic SnPb. The diffusivities
for Cu and Al were calculated on the basis of the following equations from
the master log D versus Tm /T plot for face-centered cubic metals (see Fig. 15
in Ref. 26):

Dl = 0.5 exp(−34Tm /RT ),

Dgb = 0.3 exp(−17.8Tm /RT ), (8.1)
Ds = 0.014 exp(−13Tm /RT ),

where Dl , Dgb , and Ds are respectively lattice diffusivity, grain boundary

SVNY339-Tu April 5, 2007 17:25

216 Chapter 8

diffusivity, and surface diffusivity [25–27]. Tm is the melting point, and the
units of 34 Tm , 17.8 Tm , and 13 Tm are in cal/mole. As shown in Table 8.1,
at 100◦ C the lattice diffusivity of Cu and Al are insignificantly small, and
the grain boundary diffusivity of Cu is three orders of magnitude smaller
than the surface diffusivity of Cu. At 350◦ C the difference between surface
diffusivity and grain boundary diffusivity of Cu is much less, indicating that
we cannot ignore the latter. The lattice diffusivity of eutectic SnPb (not a
face-centered cubic metal) at 100◦ C given in Table 8.1 is an average value of
tracer diffusivity of Pb and Sn in the alloy [29]. It depends strongly on the
lamellar microstructure of the eutectic sample. Since a solder joint 100 μm in
diameter has typically a few large grains, the smaller diffusivity is better for
our consideration. The surface diffusivity of Cu, grain boundary diffusivity
of Al, and lattice diffusivity of the solder are actually rather close at 100◦ C.
To compare atomic fluxes transported by these three kinds of diffusion in a
metal, we should have multiplied the diffusivity by their corresponding cross-
sectional area of path of diffusion. But the outcome is the same.
Table 8.1 also shows the homologous temperature of Cu, Al, and solder
at the device operation temperature of 100◦ C; they are 0.25, 0.5, and 0.82,
respectively. The homologous temperature of solder is very high. It means
that the application of solder joint in devices will be affected by the high-
temperature properties of the solder, or controlled by thermally activated
processes such as diffusion. For example, the mechanical properties of solder
joints will be influenced greatly by creep. This is a very important point to
remember when we study the mechanical properties of solder joints.
In face-centered cubic metals such as Al and Cu, atomic diffusion is medi-
ated by vacancies. A flux of Al atoms driven by electromigration to go to the
anode, requires a flux of vacancies to go to the cathode in the opposite direc-
tion. If we can stop the vacancy flux, we stop electromigration. To maintain
a vacancy flux we must supply vacancies continuously. Hence, we can stop a
vacancy flux by removing the sources or supplies of vacancies. Within a metal
interconnect, dislocations and grain boundaries are sources of vacancies, but
the free surface is generally the most important and effective source of va-
cancies. For Al, its native oxide is protective, which means that the interface
between the metal and its oxide is not a good source or sink of vacancies.
This is also true for Sn. When vacancies are removed without replenishment
or to be added without an effective sink, equilibrium vacancy concentration
cannot be maintained, so back stress will be generated. This topic is discussed
in Section 8.5.
If the atomic or vacancy flux is continuous in the interconnect, i.e., the
anode can supply vacancies and the cathode can accept them continuously,
and if there is no flux divergence in between, vacancy concentration is in equi-
librium everywhere, then there will be no electromigration-induced damage
such as void and extrusion formation. In other words, without mass flux diver-
gence no electromigration damage will occur in an interconnect when fluxes
of atoms and vacancies can pass through it uniformly. Hence, atomic or mass
SVNY339-Tu April 5, 2007 17:25

Fundamentals of Electromigration 217

flux divergence is a necessary condition concerning electromigration failure in

real devices. The most common mass flux divergences are the triple points of
grain boundaries and interfaces between dissimilar materials. Since a solder
joint has two interfaces, one at the cathode and one at the anode, they are
the common failure sites, especially the cathode interface where accumulation
of vacancies to form voids occurs.
In summary, electromigration involves both atomic and electron fluxes.
Their distribution in interconnects is the most important concern in electro-
migration damage. In a region where both distributions are uniform, there
will be electromigration, but no electromigration-induced damage because of
lack of flux divergence.
Concerning atomic or vacancy fluxes, the most important factor is the
temperature scale shown in Table 8.1. Atomic diffusion must be thermally ac-
tivated. The second is the design and processing of the interconnect structure.
Nonuniform distribution or divergence in interconnects occurs at microstruc-
ture irregularities such as grain boundary triple points and interphase inter-
faces, and they are the sites of failure initiation. In the following sections, the
effects of microstructure, solute, and stress on electromigration in solder joints
will be discussed in turn. The mean-time-to-failure analysis on the basis of
void and hillock formation due to flux divergence will also be discussed.
Concerning the electron flux, the current density must be high enough
for electromigration to occur. Because transistors in devices are turned on by
pulsed direct current, for transistor based devices such as computers we con-
sider only electromigration under direct current. A brief review of electromi-
gration by pulsed direct current can be found in Ref. 18. While a uniform cur-
rent distribution is expected in straight lines, nonuniform current distribution
occurs at corners where a conducting line turns, at interfaces where conductiv-
ity changes, and also around voids or precipitates in a matrix. One of the most
important factors that affects electromigration in a flip chip solder joint is the
unique line-to-bump geometry, where a very large change of current density
takes place from the line to the bump, which in turn leads to a large current
crowding at the contact between the line and the bump. The effect of cur-
rent crowding on electromigration-induced damage in the solder bump will be
discussed below. Furthermore, because of current crowding and rectified inter-
faces, AC electromigration can occur in flip chip solder joints. In a uniform cur-
rent distribution region, only DC electromigration occurs, but in a nonuniform
current distribution region, both DC and AC electromigration may occur.

8.3 Electron Wind Force of Electromigration

The electrical force acting on a diffusing atom (ion) proposed by Huntington
and Grone is taken to be [1]

Fem = Z ∗ eE = (Zel∗ + Zwd

∗
)eE, (8.2)
SVNY339-Tu April 5, 2007 17:25

218 Chapter 8

(a) (b)

electron flow

Fig. 8.4. Schematic diagram of electromigration of a diffusing atom (a) before, and
(b) at the activated state, where it possesses a very large scattering cross section.

where e is the charge of an electron and E is the electric field (E = ρj, where ρ
is resistivity and j is current density) and Z ∗ is the effective charge number of
electromigration. Zel∗ can be regarded as the nominal valence of the diffusing
ion in the metal when the dynamic screening effect is ignored; it is responsible
∗
for the field effect and Z∗el eE is called the direct force. Zwd is an assumed
charge number representing the momentum exchange effect between electrons
∗
and the diffusing ion, and Zwd eE is called the electron wind force, and it is
generally found to be of the order of 10 for a good conductor, so the electron
wind force is much greater than the direct force for electromigration in metals.
Hence, in electromigration, the enhanced flux of atomic diffusion is in the same
direction as electron flux.
To appreciate the electron wind force, we depict in Fig. 8.4(a) the con-
figuration of a shaded Al atom and a neighboring vacancy in a face-centered
cubic lattice structure before they exchange position along a <110> direction.
They have four nearest neighbors in common, including the two shown by the
broken circles, one on top and one on the bottom of the close-packed atomic
plane. When the shaded atom is diffusing halfway toward the vacancy as de-
picted in Fig. 8.4(b), it is at the activated state, sitting at a saddle point while
displacing the four nearest-neighbor atom. Since the saddle point is not part
of the lattice periodicity, the atom at the saddle position is out of its equilib-
rium position and will make a much larger contribution to the resistance to
electrical current than a normal lattice atom. In other words, it experiences a
greater electron scattering and hence a greater electron wind force which will
push it to an equilibrium position, the vacant site. The diffusion of the atom
is enhanced in the direction of the electron flow. We note that the diffusing
atom will experience the electron wind force, not just at the saddle point, but
all the way from the beginning to the end in the entire jumping path of the
diffusion.
To estimate the electron wind force, the ballistic approach to the scattering
process was developed by Huntington and Grone [1]. The model postulates
a transition probability of free electrons per unit time from one free electron
state to another free electron state due to the scattering by the diffusing atom.
SVNY339-Tu April 5, 2007 17:25

Fundamentals of Electromigration 219

The force, the momentum transfer per unit time, is calculated by summing
over the initial and final states of the scattered electrons. The step-by-step
derivation of the model is presented in Appendix C. Below, a simple derivation
is given.
During elastic scattering of electrons by a diffusing atom, the system mo-
mentum is conserved. The average change in electron momentum in the trans-
port direction is equal to 2me < υ >, where me is the electron mass and < υ >
is the mean velocity of electrons in the direction of current flow. The force on
the ion induced by the scattering is

−→ 2me < υ >

Fwd = , (8.3)
τcol

where τcol is the mean time interval between two successive collisions. The net
momentum lost per second per unit volume of electrons to the diffusing ions
is then 2nme < υ > /τcol , and the force on a single diffusing ion is

−→ 2nme < υ >

Fwd = , (8.4)
τcol Nd

where n is the electron density and Nd is the density of diffusing ions. The
electron current density can be written as

j = −ne < υ > . (8.5)

Substituting < υ > in Eq. (8.5) into Eq. (8.4), we obtain

−→ 2me j
Fwd = −
eτcol Nd
  
ρd n
=− eE, (8.6)
Nd ρ

where ρ = E/j is the total resistivity of a conductor, ρd = m/ne2 τcol is the

metal resistivity due to the diffusing atoms, and E is the applied electrical
field.
Aside from the electron “wind” force, the electric field E will produce a
direct force on the diffusing ion to be given by

−
→
Fd = Zel∗ eE, (8.7)

where Zel∗ can be regarded as the nominal valance of the metal ion when the
dynamical scattering effect around the ion is ignored. Thus, the total force
SVNY339-Tu April 5, 2007 17:25

220 Chapter 8

will be
    
−−→ ρd N
FEM = Zel∗ − Z eE, (8.8)
Nd ρ

where N is the atomic density of the conductor and n = N Z is used. Equation

(8.8) can be written as

−−→
FEM = Z ∗ eE (8.9)

and
    
ρd N
Z ∗ = Zel∗ − Z . (8.10)
Nd ρ

Z ∗ is called the effective charge number of the ion in electromigration.

Electromigration model based on the ballistic scattering of electrons is the
first and simplest model of the phenomenon of electromigration. Theoretical
understanding is further developed by contributions of numerous researchers
to date. In spite of these theoretical developments, the model developed by
Huntington and Grone, especially the drift velocity of electromigration, is
employed as the theoretical basis in nearly all experimental studies of electro-
migration. For example, the drift velocity is taken as

D ∗
vd = M F = Z ejρ. (8.11)
kT

This indicates that if we measure the drift velocity using short stripes (to be
discussed in Sections 8.5 and 8.6) and know the diffusivity, D, we will be able
to calculate Z ∗ .
The above model shows that the effective charge number can be given
in terms of specific resistivities of a diffusing atom and a normal lattice
atom,

ρd
∗ N d m0
Zwd = −Z ρ , (8.12)
m∗
N

where ρ = m0 /ne2 τ and ρd = m∗ /ne2 τd are the resistivity of the equilibrium

lattice atoms and the diffusing atoms, respectively. m0 and m∗ are the free
electron mass and effective electron mass, respectively, and we can assume
that they are equal, and τ and τd are relaxation times. In an fcc lattice, there
are 12 equivalent jump paths along the < 110 > directions. For a given current
direction, the average specific resistivity of a diffusing atom must be corrected
SVNY339-Tu April 5, 2007 17:25

Fundamentals of Electromigration 221

by a factor of one-half. By rewriting Eq. (8.10), we have

⎡ ρd ⎤
⎢1 N d m0 ⎥
Z ∗ = −Z ⎣ ρ m∗ − 1 ⎦ , (8.13)
2
N

where Zel has been taken as Z, the nominal valence of the metal atom. This
is the equation of Huntington and Grone for the effective charge number of
electromigration. To calculate Z ∗ , we need to know the specific resistivity of
a diffusing atom, or its ratio to that of a lattice atom.

8.4 Calculation of the Effective Charge Number

If the specific resistivity of an atom in a metal is assumed to be proportional
to the elastic cross section of scattering, which in turn is assumed to be pro-
portional to the average square displacement from equilibrium, or < x2 >, the
cross section of a normal lattice atom can be estimated from the Einstein
model of atomic vibration in which the energy of each mode is

1 1
mω 2 < x2 >= kT, (8.14)
2 2

where the product m ω2 is the force constant of the vibration, and m and ω
are atomic mass and angular vibrational frequency, respectively.
To obtain the cross section of scattering of a diffusing atom, < x2d >, we as-
sume that the atom and its surrounding as shown in Fig. 8.4(b) have acquired
the motion energy of diffusion, ΔHm , which is independent of temperature,

1
mω 2 < x2d >= ΔHm . (8.15)
2

Then, the ratio of the last two equations gives the ratio of cross section of
scattering,

< x2d > 2ΔHm

= . (8.16)
< x2 > kT

This shows that the ratio varies inversely with temperature. This dependence
comes from the well-known fact that the resistivity of normal metals varies
linearly with temperature above the Debye temperature. Substituting the last
equation into the equation of Z*, we obtain [9]
 
ΔHm m0
Z ∗ = −Z − 1 . (8.17)
kT m∗
SVNY339-Tu April 5, 2007 17:25

222 Chapter 8

Table 8.2. Comparison of the measured and calculated values of Z*

Metal a
Measured Z ∗ Temp (◦ C) ΔHm (eV)b c
Calculated Z ∗

Monovalent
Au −9.5 to −7.5 850 to 1000 0.83 −7.6 to −6.6
Ag −8.3 ± 1.8 795 to 900 0.66 −6.2 to −5.5
Cu −4.8 ± 1.5 870 to 1005 0.71 −6.3 to −5.4
Trivalent
Al −30 to −12 480 to 640 0.62 −25.6 to −20.6
Quadrivalent
Pb −47 250 0.54 −44
a
Data of measured Z* taken from Huntington (1974), where the correlation factor is ignored.
b
Data of ΔHm taken from Table 3.1.
c
Data of calculated Z* are obtained by using Eq. (14.12).

In the above equation, the numerical factor of 1/2 is canceled when the prob-
ability of averaging jumps in a given direction (i.e., the direction of electron
flow) from among the 12 <110> paths in an fcc metal is taken into account.
Now the value of Z* can be calculated at a given temperature by using the
last equation. The calculated values of Z* agree quite well with those mea-
sured for Au, Ag, Cu, Al, and Pb (see Table 8.2). For example, at 480◦ C,
the measured and calculated Z* for Al (taking ΔHm = 0.62 eV/atom) are
about −30 and −26, respectively. The temperature dependence of Z* cal-
culated for Au is also found to agree well with the measured values (see
Fig. 8.5).
Roughly speaking, we can see from Fig. 8.4(b) that the diffusing atom
at the activated state possesses a scattering cross section of about 10 atoms,
therefore its effective charge number will be roughly equal to 10Z, where Z is
its nominal valence number, so we have the order of magnitude of Z* of −10,
−30, and −40 for Cu (noble metal), Al, and Pb (or Sn), respectively.

8.5 Effect of Back Stress on Electromigration

and Vice Versa
Figure 8.6 is a schematic diagram of a short Al strip patterned on a baseline of
TiN. In electromigration, a high current density of electrons moves from left
to right and it transports Al atoms from the cathode to the anode, leading to
depletion or void formation at the cathode and pileup or hillock formation at
the anode. Hence, the damage of electromigration can be recognized directly.
The depletion rate at the cathode can be measured and the drift velocity of
electromigration can be deduced. Furthermore, it was found that the longer
the strip, the more the depletion at the cathode side in electromigration.
SVNY339-Tu April 5, 2007 17:25

Fundamentals of Electromigration 223

Fig. 8.5. The temperature dependence of Z* calculated for Au is found to agree well
with the measured values.

But below a “critical length,” there was no observable depletion as shown in

Fig. 8.1.
The dependence of depletion on strip length was explained by the effect
of back stress. In essence, when electromigration transports Al atoms in a
strip from cathode to anode, the latter will be in compression and the for-
mer in tension. On the basis of the Nabarro–Herring model of equilibrium
vacancy concentration in a stressed solid, the tensile region has more and the
compressive region has less vacancies than the unstressed region, so there is
a vacancy concentration gradient decreasing from the cathode to the anode.

Fig. 8.6. Schematic diagram of a set of short Al strips patterned on a baseline of TiN.
SVNY339-Tu April 5, 2007 17:25

224 Chapter 8

The gradient induces an atomic flux of Al diffusing from the anode to the
cathode, and it opposes the Al flux driven by electromigration from the cath-
ode to the anode. The vacancy concentration gradient depends on the length
of the strip; the shorter the strip, the greater the gradient. At a certain short
length defined as the “critical length,” the gradient is large enough to balance
electromigration so no depletion at the cathode and no extrusion at the anode
occur [10–13].
In analyzing this stress effect, irreversible processes have been proposed by
combining electrical and mechanical forces on atomic diffusion. The electrical
force proposed by Huntington and Grone is taken to be

Fem = Z ∗ eE. (8.2)

The mechanical force is taken as the gradient of chemical potential in a

stressed solid,

dσΩ
Fme = −∇μ = − , (8.18)
dx

where σ is hydrostatic stress in the metal and Ω is atomic volume. In essence,

it is a creep process, driven by a stress gradient. Thus, we have a pair of
phenomenological equations for atomic flux and electron flux [9]:

D dσΩ D ∗
Jem = −C +C Z eE, (8.19a)
kT dx kT
dσΩ
Je = −L21 + nμe eE, (8.19b)
dx

where Jem is atomic flux in units of atoms/cm2 -sec, and Je is electron flux in
units of coulomb/cm2 -sec. C is the concentration of atoms per unit volume,
and n is the concentration of conduction electrons per unit volume. D/kT
is atomic mobility and μe is electron mobility. L21 is the phenomenological
coefficient of irreversible processes and it contains the deformation potential.
In Eq. (8.19a), if we take Jem = 0, there is no net electromigration flux or
no damage. In other words, it reaches a steady state in an irreversible process.
The expression for the “critical length” is obtained as

ΔσΩ
Δx = . (8.20)
Z ∗ eE

Since the resistance of the conductor can be taken to be constant at a constant

temperature, we have instead the “critical product” or “threshold product”
of “jΔx” by rearranging the current density from the right-hand side to the
SVNY339-Tu April 5, 2007 17:25

Fundamentals of Electromigration 225

left-hand side of the equation:

ΔσΩ
jΔx = . (8.21)
Z ∗ eρ

Under a constant applied current density, a bigger value of critical product in

Eq. (8.21) means a longer critical length, in turn a larger back stress in Eq.
(8.20). For Al and Cu interconnects, we take j = 106 A/cm2 and Δx = 10 μm,
and we have a typical value of critical product about 1000 A/cm. We note that
while the back stress in the above is induced by electromigration, the interac-
tion between an applied stress and electromigration is the same; for example,
an applied compressive stress at the anode will retard electromigration.
If we consider a short strip deposited on an insulating substrate and take
Je = 0 in Eq. (8.19b), we have

dφ

N∗ = − at Je = 0, (8.22)
Ω
dσ

where dφ/dσ is deformation potential defined as the electrical potential per

unit stress difference at zero current. By using Onsager’s reciprocity relation,
L12 = L21 , we obtain

dφ Z ∗ Dρe
=− . (8.23)
dσ kT
−1
The dimensions of dφ/dσ and N * are cm3 /C and C , respectively.

8.6 Measurement of Critical Length, Critical Product,

Effective Charge Number
To calculate the critical length in Eq. (8.20) for Al strips, we take the stress
at the elastic limit of Al, σ = −1.2 × 109 dyn/cm2 ; Ω = 16 × 10−24 cm3 ; e =
1.6 × 10−19 C; Ex = jρ = 1.54 V/cm, where j = 3.7 × 105 A/cm2 and ρ =
4.15 × 10−6 Ω cm at 350◦ C. Substituting these values into Eq. (8.20), we obtain

−78μm
ΔxAl = .
Z∗

By taking Z ∗ = −26 for bulk Al, we have the critical length of 3 μm, which is of
the right order of magnitude, but shorter than the experimental value found in
between 10 and 20 μm. Since the Al strips are polycrystalline thin films, grain
boundary diffusion might have played a dominant role in electromigration, and
Z* for atomic diffusion in grain boundaries might be smaller than in the bulk.
Note that the critical length can be measured experimentally in a long strip
SVNY339-Tu April 5, 2007 17:25

226 Chapter 8

by extending the time of electromigration to a sufficiently long period until

the mass transport or depression at the cathode end ceases.
The temperature dependence of the critical length can be examined by
substituting Z* into Eq. (8.20), obtaining

ΔσΩ
Δx =   . (8.24)
ΔHm m0
−Z − 1 ejρ
kT m∗

For normal metals whose electrical resistivity increases linearly with temper-
ature above the Debye temperature, the last equation shows that the critical
length is rather insensitive to temperature, provided that ΔHm  kT so that
the unity in the denominator can be dropped.
To calculate the critical product in Eq. (8.21), we take j = 106 A/cm2 and
Δx = 10 μm for Al stripes, and we have a typical value of critical product of
about 1000 A/cm.
To calculate the effective charge number, we can use Eq. (8.20) provided
that we have measured Δx and Δσ. On the other hand, if we use a very long
strip and ignore the back stress effect and measure the drift velocity, we have
 
D ∗ D0 Q
vd = M F = Z ejρ = exp − Z ∗ ejρ. (8.25)
kT kT kT

This indicates that if we know the diffusivity, D, we will be able to calculate

Z*. Furthermore, if we measured the drift velocity at several temperatures,
we can take the natural logarithm of the last equation and obtain
 
D0 ∗ Q
ln(vd T ) = ln eZ jρ − . (8.26)
k kT

Thus, by plotting ln (vd T ) versus 1/kT , we can determine the activation

energy of the diffusion process in electromigration.

8.7 Why Is There Back Stress in Electromigration?

Although the Blech structure has been used very often in experimental studies
of electromigration in Al strips, there has been a question about the origin of
the back stress. If we confine a short strip by rigid walls as shown in Fig. 8.7,

Fig. 8.7. A short strip confined by rigid walls.

SVNY339-Tu April 5, 2007 17:25

Fundamentals of Electromigration 227

we can envisage easily the compressive stress at the anode induced by electro-
migration. In a fixed or constant volume of V at the anode, the stress change
in the volume by adding atoms or adding ΔV into it by electromigration
is
ΔV
ΔV ΔC
Δσ = −B = −B VΩ − B , (8.27)
V Ω
C

where B is bulk modulus and Ω is atomic volume. The negative sign indicates
that the stress is compressive. In other words, we are adding atomic volume
into the fixed volume. If the fixed volume cannot expand, a compressive stress
will occur. Theoretically, a fixed volume means the constraint of a constant
volume. Thus, the implicit assumption in the origin of back stress is the as-
sumption of a constant volume constraint. Why we can have such a constraint
in Al short strips will be explained below.
In a fixed volume confined within rigid walls, the compressive stress in-
creases with the addition of atoms. However, in short strip experiments,
there are no rigid walls to cover the Al strips, except native oxide. How
can the back stress build up at the anode if the native oxide is not a rigid
wall?
In Section 6.3, we mentioned that in diffusional processes, such as the clas-
sic Kirkendall effect of interdiffusion in a bulk diffusion couple of A and B,
while the atomic flux of A is not equal to the opposite flux of B, no stress
was assumed in the analysis of Darken’s model of interdiffsuion. If it is as-
sumed that more A atoms are diffusing into B, we might expect that there
will be a compressive stress in B, on the basis of the argument given above.
However, Darken has made a key assumption that vacancy concentration is
in equilibrium everywhere, hence vacancies (or lattice sites) can be created
and/or annihilated as needed in the sample. Hence, there is no compressive
stress in B, provided that the lattice sites in B can be added readily to accom-
modate the incoming A atoms. The addition of a large number of lattice sites
implies an increase in lattice planes and an expansion in volume if we assume
that the mechanism of vacancy creation and/or annihilation is by dislocation
climb mechanism. However, if we assume a constraint of constant volume, we
must allow the excess lattice planes to migrate in the reverse direction and
out of the volume, in turn implying marker motion if markers are embedded
in the frame of the moving lattice in the sample. Therefore, in the ideal case
of Darken’s model of interdiffusion, lattice shift occurs due to an effective va-
cancy generation and annihilation under the constant volume constraint. If
not, strain or stress will be generated.
Thus, a plausible explanation of the back stress in Al short strips is that
the Al native oxide has removed the sources and sinks of vacancies from the
surface, therefore when electromigration drives atoms into the anode region,
the out-diffusion of vacancies will reduce the vacancy concentration in the
anode region if there is no source to replenish it. Furthermore, we must allow
SVNY339-Tu April 5, 2007 17:25

228 Chapter 8

the added lattice planes to move, otherwise a compressive stress will be gen-
erated. Since the Al thin film has a native and protective oxide on the sur-
face, the oxide tends to tie down the lattice planes in Al and prevent them
from moving. The oxide is effectively tying down the lattice planes because
the film is thin. This is the basic mechanism of back stress generation in Al
interconnects.
In Chapter 6, we discussed spontaneous Sn whisker growth at room tem-
perature; the mechanism of compressive stress generation is similar to that
presented in the above since Sn also has a protective native oxide. However,
since Sn has a low melting point, no spontaneous whisker growth occurs near
or above 100◦ C because of fast stress relaxation. For the same reason, back
stress induced by electromigration in solder joints is not as strong as that in
Al because of the high homologous temperature of solder. The back stress can
be relaxed quickly over a certain distance from the anode.
On the basis of the above discussion, it is clear that the origin of back
stress, in turn the existence of a critical product or critical length of electro-
migration in Al short strips, depends on the effectiveness of sources and sinks
of vacancies in the samples. If the sources and sinks are as effective as in the
assumption of Darken’s model of interdiffusion, there will be no back stress,
no critical product, and no threshold current density of electromigration. Elec-
tron wind force can be regarded as a driving force of atomic diffusion, and the
latter is a thermally activated process. Theoretically, even at 1 K, atomic dif-
fusion can take place except that the probability or the frequency of exchange
jumps will be infinitely small, so as electromigration. However, in real devices,
it is not electromigration itself but rather electromigration-induced damage
that is of concern, and the damage should not occur within the lifetime of the
device. The Blech short strip structure has enabled us to see electromigration
induced damage of void formation at the cathode and hillock formation at the
anode very conveniently. Indeed there is a threshold current density, a back
stress, and a critical length of Al short strip, but they are unique because of
the surface oxide on the thin-film Al short strips. For Cu interconnect, the
situation is different since it has no protective oxide.
The time dependence of stress buildup in a short strip by electromigration
can be obtained by solving the continuity equation since stress is energy den-
sity and a density function obeys the continuity equation, and we can convert
ΔC to Δσ from the last equation [11, 12]:

 2
C ∂σ D ∂2σ D ∂σ CDZ∗ eE ∂σ
=− 2
− − . (8.28)
B ∂t kT ∂x BkT ∂x BkT ∂x

The solution for a finite line and the manner of stress buildup as a function
of time is shown in Fig. 8.8. Clearly, in the beginning of electromigration
the back stress is nonlinear along the length of the stripe, represented by
the curved lines. In reality, the buildup is asymmetrical since the hydrostatic
tensile stress at the cathode can hardly be developed.
SVNY339-Tu April 5, 2007 17:25

Fundamentals of Electromigration 229

600

400

STRESS (MPa)
200

–200

–400

–600
0.0 0.2 0.4 0.6 0.8 1.0
REDUCED DISTANCE (x/L)

Fig. 8.8. The solution for a finite line and the manner of stress buildup as a function
of time.

8.8 Measurement of the Back Stress Induced

by Electromigration
A serious effort has been dedicated to measuring the back stress in Al
strips during electromigration. It is not an easy task since the strip is thin
and narrow; typically it is only a few hundred nanometers thick and a few
micrometers wide, so a very high intensity and focused x-ray beam is needed
in order to determine the strain in Al grains by precision lattice parameter
measurement. Micro-diffraction x-ray beams using synchrotron radiation have
been employed to study the back stress. White x-rays of 10μm × 10μm beam
from the National Synchrotron Light Source (NSLS) at Brookhaven National
Laboratory were used to study electromigration-induced stress distribution
in pure Al lines [13]. The line was 200 μm long, 10 μm wide, and 0.5 μm
thick with 1.5 μm SiO2 passivation layer on top, 10 nm Ti/60 nm TiN shunt
layer at the bottom, and 0.2-μm-thick W pads at both ends which connect
the line to contact pads. The electromigration tests were performed at 260◦ C.
Results of the steady-state rate of resistance increase, δ(ΔR/R)/δt, and the
electromigration-induced steady-state compressive stress gradient, δσEM /δx,
versus current density are shown in Fig. 8.9. No electromigration occurred
below the threshold current density “jth ” of 1.6 × 105 A/cm2 . Below the
threshold current density, the electromigration-induced steady-state stress
gradient increased linearly with current density, wherein the electron wind
force was counterbalanced by the mechanical force, so no electromigration
drift was observed. However, the basic reason for the existence of the
threshold stress is unclear.
SVNY339-Tu April 5, 2007 17:25

230 Chapter 8

Fig. 8.9. Steady-state rate of

0.010
resistance increase, δ(ΔR/R)/
2
δt, and electromigration-
dsEM /dx [MPa/μm]

induced steady-state compres-

d(ΔR/R)/dt [hr-1]
sive stress gradient, δσEM /δx,
0.005 plotted versus current density.
1
(Courtesy of Prof. G. S.
Cargill, Lehigh University.)

0 0.000
jth
0.0 2.0×105 4.0×105 6.0×105
Current density [A/cm2]

The x-ray micro-diffraction apparatus at the Advanced Light Source (ALS)

of Lawrence Berkeley National Laboratory is capable of delivering white x-
ray beams (6–15 keV) focused to 0.8 to 1 μm by a pair of elliptically bent
Kirkpatrick–Baez mirrors [34, 35]. In the apparatus, the beam can be scanned
over an area of 100 μm by 100 μm in steps of 1 μm. Since the diameter of the
grains in the strip is about 1 μm, each grain can be treated as a single crystal
with respect to the micro-beam. Structural information such as stress/strain
and orientation can be obtained by using white beam Laue diffraction. Laue
patterns were collected with a large-area (9 × 9 cm2 ) charge-coupled device
(CCD) detector with an exposure time of 1 sec or longer, from which the ori-
entation and strain tensor of each illuminated grain can be deduced and dis-
played by software. The resolution of the white beam Laue technique is 0.005%
strain. In addition, a four-crystal monochromator can be inserted into the
beam to produce monochromatic light for diffraction. The combined white and
monochromatic beam diffractions are capable of determining the total strain-
stress tensor in each grain. The technique and applications of Scanning X-ray
Microdiffraction (μSXRD) have been described by MacDowell et al. [30, 31].
Whether back stress exists in electromigration in Cu damascene structure
at the device working temperature has not been confirmed. If electromigration
occurs by surface diffusion in a Cu damascene structure, we need a mechanism
of back stress generation in the bulk of the structure induced by surface dif-
fusion. Without a protective surface oxide, surface diffusion occurs, and thus
the surface is a good source and sink of vacancies.

8.9 Current Crowding and Current Density

Gradient Force
Interconnect in VLSI technology is a three-dimensional multilevel structure.
When electrical current turns or converges, e.g., at vias in the 3D struc-
ture, where the current passes from one level of interconnect to another level,
SVNY339-Tu April 5, 2007 17:25

Fundamentals of Electromigration 231

e-

(a) Cathode AI
Anode
TiN

106 0.6

105 0.4
(b)
104 Cathode 0.2
103 0.0
0.0 0.5 1.0 1.5 2.0 μm

(c) Cathode AI
Anode
TiN

Fig. 8.10. Sketch of the cross section of the well-known Blech–Herring short strip
test structure of electromigration. (b) Simulation of the current crowding picture. (c)
If a void nucleates and grows in the high current density region, it will not be able to
deplete the cathode completely.

current crowding occurs and affects electromigration significantly. We postu-

late that defects such as vacancies and solute atoms will have a higher poten-
tial in the high current density region than in the low current density region.
The potential gradient in the current crowding area provides a driving force
to push these defects from the high current density region to the low current
density region. As a consequence, the voids tend to form in the low current
density region rather than in the high current density region. In other words,
failure tends to occur in a low current density region in 3D interconnects
[32, 33].
Figure 8.10(a) is a sketch of the crosssection of the well-known Blech short
strip test structure of electromigration. We assume electrons go from the left
(cathode) to the right (anode). The W baseline has a higher resistance than
the short Al strip so the electrons will make a detour from the W into the Al
because the latter is a better conductor. In the region where the current enters
the Al, current crowding occurs. The current crowding has been simulated and
shown in Fig. 8.10(b), where the arrow indicates that the upper left corner
and its neighborhood are the low current density region. Clearly, the lower
left corner of Al/TiN interface, where current crowding occurs, is the high
current density region [34–36].
SVNY339-Tu April 5, 2007 17:25

232 Chapter 8

Over-hang
e-
AI

Cathode W via W Anode

AI AI

Fig. 8.11. Sketch of the cross section of a two-level Al interconnect structure con-
nected by W vias. One way to delay the wear-out failure is to add an overhang of the
Al interconnect above the W via, as shown by the dotted line.

Many SEM images have shown that void formation occurs at the cathode
of the strip as a result of electromigration. If a void nucleates and grows in
the high current density region, as depicted in Fig. 8.10(c), it will not be able
to extend to the low current density region in the upper corner because the
void is an open, and the current will be pushed back toward the anode. In
order to deplete the entire cathode, the vacancies must go to the low current
density region, so the void must start from the upper left corner of the left
end of the strip.
Figure 8.11 is a sketch of the crosssection of a two-level Al interconnect
structure connected by W vias. Again it is assumed that electrons go from left
to right and current crowding occurs in passing through the vias. We consider
the via on the left and indicate by an arrow that the upper left corner and its
neighborhood over-hang region are the low current regions where a void tends
to form first. Since atomic diffusion in W is much slower than that in Al, the
W/Al interface is a flux divergence plane of diffusion, where more Al atoms
are leaving. The reverse flux of vacancies would lead to vacancy condensation
near the interface. But the void formation will not begin at the right-hand
edge of the W/Al interface where the current density is the highest; rather it
tends to occur at the upper left corner or the neighboring regions. As the void
grows, it leads to circuit failure when it covers the entire via so the via is open.
In the microelectronic industry, this has been called the wear-out mechanism
of failure. One way to delay the wear-out failure is to add an overhang of the
Al interconnect above the W via, as shown by the dotted line in Fig. 8.11.
The overhang provides an additional volume or reservoir for void growth, so
it can lengthen the mean time to failure. However, it is implicitly assumed
in this remedy that vacancies will go to the low current density region of the
overhang.
Often it is assumed that there is a stress gradient to drive the vacancies
to the low current density region. A stress gradient is developed after a void
is formed because of its free surface. The nucleation of a void requires su-
persaturation of vacancies, hence vacancies have to diffuse to the low current
density region first before the void formation. Following the electron wind
SVNY339-Tu April 5, 2007 17:25

Fundamentals of Electromigration 233

force, we expect vacancies to go to the high current density region, in turn

void formation there, but this is not true since void formation occurs in the
low current density region.
To envisage the driving force that enables vacancies to go from the high
current density region to the low current density region, we consider a single-
crystal Al strip and assume that a vacancy in the Al crystalline lattice has a
specific resistivity of ρv . The specific resistivity may depend on current den-
sity due to joule heating since resistivity depends on temperature. However,
for simplicity, we ignore the temperature effect here and assume that it is in-
dependent of current density for the simple analysis to be given below. Since
vacancy is a lattice defect, we can regard its specific resistivity as the excess
resistivity over that of a lattice atom. Under electromigration in a current
density of je , a voltage drop of je ρv occurs around the vacancy. From the
energetic viewpoint, we can regard that the vacancy has a potential of je ρv
above the surrounding lattice atoms. Knowing the charge of the vacancy, we
have the potential energy of the vacancy in the current density of je . Let the
charge of the vacancy be Z ∗∗ e, where Z ∗∗ is the effective charge number of
the vacancy and e is the charge of an electron, we have the potential energy
of Pv = Z ∗∗ eje ρv for a vacancy stressed by the current density je .
If we assume the equilibrium vacancy concentration in the crystal without
any electrical current (je = 0) to be Cv ,

Cv = C0 exp (−ΔGf /kT ), (8.29)

where C0 is the atomic concentration of the crystal and ΔGf is the formation
energy of a vacancy. When we stress the crystal by a current density of je , the
vacancy concentration will be reduced to

Cve = C0 exp − (ΔGf + Z ∗∗ eje ρv )/kT. (8.30)

Under a uniform high current density, the equilibrium vacancy concentra-

tion in the crystal decreases. In other words, the electric current dislikes any
excess high-resistance obstacles (or defects) and prefers to get rid of them
until the equilibrium is reached. When there exists a current density gradient
as in current crowding, a driving force exists to do so,

F = −dPv /dx. (8.31)

This force drives the excess vacancies to diffuse in the direction normal to
the direction of current flow. Now if we go back to Fig. 8.10(a) and consider
the electromigration flux in the short strip, in the middle section of the strip,
the current density is constant so there is a constant flux of Al atoms moving
from left to right and a balance flux of vacancies moving from right to left.
A few of the vacancies near the surface or the substrate may escape to the
surface or the substrate interface, but the concentration as given by Eq. (8.30)
SVNY339-Tu April 5, 2007 17:25

234 Chapter 8

is maintained. When the vacancy flux approaches the cathode and enters the
current crowding region, most of them become excess and a force to divert
them to the low current density region comes into play. Consequently, a com-
ponent of the vacancy flux is moving in the direction normal to the current
flow,

Jcc = Cve (Dv /kT )(−dPv /dx) (8.32)

where Dv /kT is the mobility and Dv is the diffusivity of vacancies in the

crystal. Since a constant flux of vacancies keeps moving from anode to cathode
due to electromigration, the total flux of vacancies moving toward the cathode
is given by the sum of two components,

Jsum = Jem + Jcc = Cve (Dv /kT )(−Z ∗ eE − dPv /dx), (8.33)

where the first term is due to electromigration driven by the current density
(electron wind force) and the second term is due to current crowding driven
by the current density gradient. In the first term, Z ∗ is the effective charge
number of the diffusing Al atom, and E = je ρ (where ρ is the resistivity of
the Al). We note here that we assume the vacancy flux is opposite but equal
to the Al flux. Also, it is important to note that the sum in brackets is a
vector sum; the first term is directed along the current and the second term
is directed normal to the current. In other words, the vacancies are driven
by two forces in the current crowding region. They are depicted in Fig. 8.12.
Since the current turns in the current crowding region, the direction of Jsum
changes with position. Clearly, a detailed simulation is needed in order to
unravel the magnitude and distribution of the forces in the current crowding
region.
How large is the gradient force is of interest. If we apply a current density
of 105 A/cm2 through the strip and assume that the current density will drop
to zero across the thickness of the strip of 1 μm, the gradient can be as high
as 109 A/cm3 . The gradient force is of the same magnitude as the electron
wind force. Under such a large gradient, high-order effect might exist, but we
will ignore it at this moment.
Current crowding in flip chip solder joints and its effect on
electromigration-induced failure will be presented in Section 9.3.

AI
Cathode

TiN
Fig. 8.12. The vacancies are driven by two forces in the current crowding region.
SVNY339-Tu April 5, 2007 17:25

Fundamentals of Electromigration 235

8.10 Electromigration in Anisotropic Conductor

of Beta-Sn
White tin (or beta-Sn) has a body-centered tetragonal crystal structure, and
its lattice parameters are a = b = 0.583 nm and c = 0.318 nm. Its electrical
conductivity is anisotropic; the resistivity along the a and b axes is 13.25 μΩ-
cm and that along the c-axis is 20.27 μΩ-cm. In electromigration under an
applied current density of 6.25 × 103 A/cm2 at 150◦ C for 1 day, the beta-Sn
stripes were found to show a voltage drop of up to 10%, reported by Lloyd [37].
The electromigration in white tin (Tm = 232◦ C) is of interest because most
of the Pb-free solders are Sn-based and the electromigration at the device
working temperature occurs primarily by lattice diffusion which may lead to
a noticeable microstructural change affected by its anisotropic conductivity.
Figure 8.13 shows SEM images of the top view of a Sn strip before (a) and
after (b) electromigration at 2 × 104 A/cm2 at 100◦ C for 500 hr. The lower
image shows that a few of the grains rotated after electromigration.
In Fig. 8.14, the cross section of a beta-Sn grain is shown. It has a body-
centered tetragonal structure and we assume that its c-axis is making an angle

(a) 0 h

(b) 500 h

Fig. 8.13. (a) SEM images of

the top view of a Sn strip be-
fore and (b) after electromi-
gration at 2 × 104 A/cm2 at
100◦ C for 500 hr. The lower
image shows that a few of the
grains rotated after electromi-
gration. (Courtesy of Profes-
sor C. R. Kao, National Cen-
tral University, Jhongli, Tai-
wan, ROC.)
SVNY339-Tu April 5, 2007 17:25

236 Chapter 8

Fig. 8.14. Cross section of a

beta-Sn grain. E versus j.

θ with the x-axis, its a-axis is on the plane of the figure and is making an
angle of (90◦ − θ) with the x-axis, and its b-axes is normal to the plane of
the figure. The applied current density of electrons, j, is from left to right,
along the x-axis. The resistivity along the a- and b-axis is the same and
smaller than that along the c-axis. Due to the anisotropy of resistivity, the
electrical field, E, can be written in two components, Ea and Ec , along the a-
and c-axes, respectively. The electrical field along the a-axis is Ea = ρa ja ; the
field along the c-axis is Ea = ρc jc , where ja and jc are respectively the two
components of the electrical (electron) current density, j, along the a- and
c-axes.
In isotropic materials, such as Cu or Al, they have the same resistivity
along all the axes; therefore, the magnitude of ρc jc is the same as ρa ja . This
also means Ea = Ec , the overall electrical field within the grain will coincide
with the current flow direction, j. However, due to the difference of the mag-
nitude between Ea and Ec in anisotropic materials, such as beta-Sn, there will
be an angle ϕ, as shown in Fig. 8.14, between the combined electric field E
and the direction of j inside the grain. This is a unique property of anisotropic
conducting material, and it is important to examine analytically how the angle
ϕ would affect the interaction between the electrical current and the electrical
force exerted on the grain [38].
In Fig. 8.14, the two components of current density j along the a- and
c-axes would be

ja = j sin θ, jc = j cos θ. (8.34)

Therefore, the resulting electrical fields along these two axes are

Ea = ρa j sin θ, Ec = ρc j cos θ. (8.35)

SVNY339-Tu April 5, 2007 17:25

Fundamentals of Electromigration 237

The combined electrical field, E, would be

 
2 2 2 2
E= Ea2 + Ec2 = (ρa j sin θ) + (ρc j cos θ) = j (ρa sin θ) + (ρc cos θ) .
(8.36)

To evaluate the magnitude of the angle ϕ, we consider the components of

E, Ea and Ec , on the y-axis. From Fig. 8.14,

E sin ϕ = Ec sin θ − Ea cos θ. (8.37)

By rearrangement, we have

Ec sin θ − Ea cos θ (ρc j cos θ) sin θ − (ρa j sin θ) cos θ

sin ϕ = = . (8.38)
E E

By substituting E from Eq. (8.36), we have

j[(ρc cos θ) sin θ − (ρa sin θ) cos θ]

sin ϕ =  . (8.39)
2 2
j (ρa sin θ) + (ρc cos θ)

The current term, j, can be canceled out and by substituting 2 sin θ cos θ =
sin 2θ, the last equation becomes

(ρc − ρa ) sin 2θ
sin ϕ =  . (8.40)
2 2
2 (ρa sin θ) + (ρc cos θ)

Similarly, if we consider the components of E, Ea and Ec , on the x-axis, we

obtain

ρc cos2 θ + ρa sin2 θ
cos ϕ =  . (8.41)
2 2
(ρa sin θ) + (ρc cos θ)

Either Eq. (8.40) or (8.41) defines the magnitude of the angle ϕ between
the electrical-field and the applied current density from the data of resistivity
and the orientation of the grain. Equation (8.40) shows that if θ = 0◦ and
θ = 90◦ , then ϕ = 0◦ or in these cases E will be parallel to j, which will be
considered later.
Since the direction of the electrical field deviates from that of the current
density, it follows that the force originating from this field will also deviate
from the current flow direction. The effect on force can have two significant
consequences. The first is a torque and the second is that the force has a
component parallel to the grain boundary plane or the y-axis as shown in
Fig. 8.14. It can be seen that the force generated from momentum exchange
SVNY339-Tu April 5, 2007 17:25

238 Chapter 8

between electrons and atoms exerted on the boundaries of the grain is also
making an angle to the current density. The existence of a pair of totally
opposite forces provides a torque. Note that the boundaries need not be all
grain boundaries, they could also be the interface between the sample and the
substrate.

8.11 Electromigration of a Grain Boundary

in Anisotropic Conductor
We consider electromigration of a grain boundary. The electron flux is normal
to the plane of the grain boundary. In the case of grain boundaries in Al
thin films, electromigration-induced migration of the grain boundaries has
been observed. We shall consider a grain boundary in beta-Sn, which is an
anisotropic conductor, and we shall demonstrate that electromigraion will lead
to an atomic flux along the plane of the grain boundary. In other words, the
induced atomic flux along the grain boundary is moving in a direction normal
to the electron flux or the electron wind force [39].
In Fig. 8.15, a simple and geometrically ideal situation of a grain boundary
between two beta-Sn grains is depicted; grain 3 on the right and grain 2 on
the left of the grain boundary. We assume that grain 2, on the left, has its
crystallographic c-axis directed along the flow direction of electron current, j,
which is directed from left to right as indicated by a long arrow. We further
assume that grain 3, on the right, has its crystallographic a-axis also directed
along the current flow direction. Since both resistivity and diffusivity along
a- and c-axes in beta Sn are different, the electron wind force and the corre-
sponding vacancy fluxes in the c-axis (grain 2) and a-axis (grain 3) grains will

δ e- d

Grain 1 Grain 2 Grain 3

V

c a c
jv More jv More V jv
Compressive Tensile h

Grain Boundary I Grain Boundary II

Fig. 8.15. Schematic diagram of a simple and geometrically ideal situation of a sand-
wiched grain structure.
SVNY339-Tu April 5, 2007 17:25

Fundamentals of Electromigration 239

be different:

Cvbulk Dvc,bulk ∗ c Cvbulk Dva,bulk ∗ a

Jvc = Z eρ j, Jva = Z eρ j. (8.42)
kT kT

Jvc and Jva are the vacancy fluxes in grain 2 along the c-axis and in grain
1 along the a-axis, respectively. We have the reference data below for the
diffusivity and resistivity of Sn atoms along these two directions:

Dc = 5.0 × 10−13 cm2 /sec, ρc = 20.3 × 10−6 Ω-cm;

Da = 1.3 × 10−12 cm2 /sec, ρa = 13.3 × 10−6 Ω-cm.

The atomic flux under electromigration should be in the same direction

of electron flow. Therefore, a counterflux of vacancy flows from right to left,
as indicated by the two short arrows. The effective charge, Z ∗ , is considered
to be the same in both directions. Since Dc ρc < Da ρa , from Eq. (8.42), we
find a larger vacancy flux reaches the grain boundary from grain 3 to grain
2, yet a smaller vacancy flux leaves the grain boundary going into grain 2.
In grain 2, supersaturation of vacancies occurs at the grain boundary and a
corresponding tensile stress occurs near the grain boundary.
If we consider a small volume within the grain boundary of d × h × δ,
where d is the width of the grain, h is the height of the grain, and δ is the
effective grain boundary width, we have in the steady state, the balance of
the flux or conservation of mass,

Cvbulk Z ∗ ej a,bulk a C ∞ − CvL

(Dv ρ − Dvc,bulk ρc ) × d × h ∼
= DvGB v × d × δ. (8.43)
kT h

The first term of this equation states the difference of the flux through bulk
diffusion; the second term means that the difference of the vacancy flux goes to
the surface via grain boundary diffusion, since the surface is a good sink/source
of vacancy.
From Eq. (8.43), the difference of vacancy concentrations can be evaluated
as

ΔCv ∼ Cvbulk Z ∗ ej a,bulk a

= (Dv ρ − Dvc,bulk ρc ) × h, (8.44)
h kT δDvGB

where ΔCv = Cv∞ − CvL , where Cv∞ is the equilibrium vacancy concentration
of the free surface and CvL is the vacancy concentration in the grain bound-
ary. Thus, we have obtained a flux of vacancies (or atoms) along the grain
boundary, provided that a sink for vacancies exists at the end of the grain
boundary, which can be a free surface or a void. Again we note that this
flux is moving in a direction normal to the electron flux. We recall that this
SVNY339-Tu April 5, 2007 17:25

240 Chapter 8

is the second case where atomic or vacancy flux is moving normal to elec-
tron flux, and the first case is due to current density gradient presented in
Section 8.9.
If we extend the above analysis to a three-grain structure of one c-axis
grain sandwiched between two a-axis grains, it will lead to grain rotation
of the c-axis grain in the sandwiched structure, as shown in Fig. 8.15 [39].
Furthermore, the analysis presented above can also be applied to interphase
interfaces. For example, if we consider the interface between solder and Cu6 Sn5
in a flip chip solder joint, since the resistivity and diffusivity in these two
phases are different, there will be a vacancy flux along the interface under
electromigration with electron flow normal to the plane of the interface. This
interfacial flux could lead to void formation and morphological change of the
interface, to be discussed in Section 9.4.6.

8.12 AC Electromigration
Electromigration in interconnects is commonly a DC behavior. In devices
based on field-effect transistors used in computers, such as dynamic random
assess memory (DRAM) devices, the gate of the transistor is turned on and
off by pulsed DC current. On the other hand, in most communication devices,
AC current is used. Especially in power switching devices, and radio frequency
and audio power amplifiers, a large AC swing occurs during operation. The
question whether AC can induce electromigration is often asked. Typically it
is believed that AC has no effect on electromigration.
We follow Huntington and Grone’s model that the driving force of elec-
tromigration is due to momentum exchange in the scattering of electrons by
diffusing atoms. A diffusing atom will not be in equilibrium and will have
a large scattering cross section. If we consider AC of frequency 60 Hz or 60
cycles/sec, it means in a period of 1/120 sec, the scattering will be reversed
once. For lattice diffusion in Pb assuming a vacancy mechanism of diffusion
at 100◦ C, the fraction of equilibrium vacancy in 1 cm3 of Pb is given by
 
nV ΔGf
= exp − ,
n kT

where ΔGf is the formation free energy of a lattice vacancy. Taking ΔGf =
0.55 eV/atom, we have nV /n = 10−7 . If we take n = 1022 atoms/cm3 , we have
nV = 1015 vacancies/cm3 , meaning that these many vacancies are attempting
to jump. The successive jumps are limited by the frequency factor
 
ΔGm
υ = υ0 exp − ,
kT

where υ0 is the Debye frequency or attempt frequency of jumping of the

diffusing atom and is about 1013 Hz for metals above their Debye temperature.
SVNY339-Tu April 5, 2007 17:25

Fundamentals of Electromigration 241

Taking ΔGm = 0.55 eV/atom, which is assumed to be the motion free energy
of a vacancy in Pb, we obtain υ = 106 jumps/sec. Then in 1/120 sec, there
are about 104 successive jumps of each of the 1015 vacancies/cm3 . Implicitly,
we have assumed that the lifetime of the transition state or the activated
state is very short. In other words, in each cycle of the 60 Hz AC, a large
number of vacancies (or atoms) jump in one direction in the first half cycle
driven by electromigration and then an equal number of vacancies will jump
in the opposite direction in the second half cycle. They cancel out each other
statistically, so there is no net atomic flux driven by AC current. Nevertheless,
AC will generate joule heating and joule heating may develop a temperature
gradient that induces atomic diffusion.
In the above analysis, an implicit assumption is that the electric field or
electric current is uniform. However, when the current distribution is nonuni-
form, it is unclear whether AC electromigration can occur or not. Nonuniform
current distribution occurs when electric current turns as in a flip chip solder
joint, or in a metal interconnect between a Cu line and via, or in a two-phase
alloy where a precipitate and its matrix have different resistivities, or at a
reactive interphase interface such as Sn/Cu. In the last case, if the interface is
not at equilibrium, the atomic jumps across the interface in one direction are
not the same as the jumps in the reverse direction. Since it is irreversible, the
AC effect of electromigration may enhance the jumping in one direction. At
a Schottky barrier across a metal/n-type semiconductor interface, the carrier
flow is oneway from the semiconductor to the metal, so a high current density
of AC may enhance the diffusion of the semiconductor into the metal.

References
1. H. B. Huntington and A. R. Grone, “Current-induced marker motion in
gold wires,” J. Phys. Chem. Solids, 20, 76 (1961).
2. H. B. Huntington, in “Diffusion,” H. I. Aaronson (Ed.), American Society
for Metals, Metals Park, OH, p. 155 (1973).
3. H. B. Huntington, in “Diffusion in Solids: Recent Development,” A. S.
Nowick and J. J. Burton (Eds.), Academic Press, New York, p. 303 (1974).
4. I. Ames, F. M. d’Heurle, and R. Horstman, IBM J. Res. Dev., 4, 461
(1970).
5. I. A. Blech, “Electromigration in thin aluminum films on titanium ni-
tride,” J. Appl. Phys., 47, 1203–1208 (1976).
6. I. A. Blech and C. Herring, “Stress generation by electromigration,” Appl.
Phys. Lett., 29, 131–133 (1976).
7. F. M. d’Heurle and P. S. Ho, in “Thin Films: Interdiffusion and Reac-
tions,” J. M. Poate, K. N. Tu, and J. W. Mayer (eds.), Wiley–Interscience,
New York, p. 243 (1978).
8. P. S. Ho and T. Kwok, Rep. Prog. Phys., 52, 301 (1989).
9. K. N. Tu, “Electromigration in stressed thin films,” Phys. Rev. B, 45,
1409–1413 (1992).
SVNY339-Tu April 5, 2007 17:25

242 Chapter 8

10. R. Kircheim, Acta Metall. Mater., 40, 309 (1992).

11. M. A. Korhonen, P. Borgesen, K. N. Tu, and C. Y. Li, J. Appl. Phys., 73,
3790 (1993).
12. J. J. Clement and C. V. Thompson, J. Appl. Phys., 78, 900 (1995).
13. P. C. Wang, G. S. Cargill III, I. C. Noyan, and C. K. Hu, Appl. Phys.
Lett., 72, 1296 (1998).
14. K. L. Lee, C. K. Hu, and K. N. Tu, J. Appl. Phys., 78, 4428 (1995).
15. R. S. Sorbello, in “Solid State Physics,” H. Ehrenreich and F. Spaepen
(Eds.), Academic Press, New York, Vol. 51, pp. 159–231 (1997).
16. C. K. Hu and J. M. E. Harper, Mater. Chem. Phys., 52, 5 (1998).
17. R. Rosenberg, D. C. Edelstein, C. K. Hu, and K. P. Rodbell, Annu. Rev.
Mater. Sci., 30, 229 (2000).
18. E. T. Ogawa, K. D. Li, V. A. Blaschke, and P. S. Ho, IEEE Trans. Reliab.,
51, 403 (2002).
19. K. N. Tu, “Recent advances on electromigration in very-large-scale-
integration of interconnects,” J. Appl. Phys., 94, 5451–5473 (2003).
20. C. L. Gan, C. V. Thompson, K. L. Pey, W. K. Choi, H. L. Tay, B. Yu,
and M. K. Radhakrishnan, Appl. Phys. Lett., 79, 4592 (2001).
21. A. V. Vairagar, S. G. Mhaisalkar, A. Krishnamoorthy, K.N. Tu, A.M.
Gusak, M. A. Mayer, and E. Zschech, “In-situ observation of electromi-
gration induced void migration in dual-damascene Cu interconnect struc-
tures,” Appl. Phys. Lett., 85, 2502–2504 (2004).
22. A. V. Vairagar, S. G. Mhaisalkar, M. A. Meyer, E. Zschech, A. Krish-
namoorthy, K. N. Tu, and A. M. Gusak, “Direct evidence of electromi-
gration failure mechanism in dual-damascene Cu interconnect tree struc-
tures,” Appl. Phys. Lett., 87, 081909 (2005).
23. M. Y. Yan, J. O. Suh, F. Ren, K. N. Tu, A. V. Vairagar, S. G. Mhaisalkar,
and A. Krishnamoorthy, “Effect of Cu3 Sn coatings on electromigration
lifetime improvement of Cu dual-damascene interconnects,” Appl. Phys.
Lett., 87, 211103 (2005).
24. M. Y. Yan, K. N. Tu, A. V. Vairagar, S. G. Mhaisalkar, and A. Krish-
namoorthy, “Confinement of electromigration induced void propagation
in Cu interconnect by a buried Ta diffusion barrier layer,” Appl. Phys.
Lett., 87, 261906 (2005).
25. T. V. Zaporozhets, A. M. Gusak, K. N. Tu, and S. G. Mhaisalkar, “Three-
dimensional simulation of void migration at the interface between thin
metallic film and dielectric under electromigration,” J. Appl. Phys., 98,
103508 (2005).
26. P. G. Shewman, “Diffusion in Solids,” 2nd ed., The Minerals, Metals, and
Materials Society, Warrendale, PA (1989).
27. N. A. Gjostein, in “Diffusion,” H. I. Aaronson (Ed.), American Society
for Metals, Metals Park, OH, p. 241 (1973).
28. P. Wynblatt and N. A. Gjostein, Surf. Sci., 12, 109 (1968).
29. D. Gupta, K. Vieregge, and W. Gust, Acta Mater., 47, 5 (1999).
SVNY339-Tu April 5, 2007 17:25

Fundamentals of Electromigration 243

30. A. A. MacDowell, R. S. Celestre, N. Tamura, R. Spolenak, B. Valek, W.

L. Brown, J. C. Bravman, H. A. Padmore, B. W. Batterman, and J. R.
Patel, Nucl. Instrum. Methods., A 467, 936 (2001).
31. N. Tamura, A. A. MacDowell, R. S. Celestre, H. A. Padmore, B. Valek,
J. C. Bravman, R. Spolenak, , W. L. Brown, T. Marieb, H. Fujimoto, B.
W. Batterman, and J. R. Patel, Appl. Phys. Lett., 80, 3724 (2002).
32. H. Okabayashi, H. Kitamura, M. Komatsu, and H. Mori, AIP Conf. Proc.,
373, 214 (1996). (See Figs. 2 and 4)
33. S. Shingubara, T. Osaka, S. Abdeslam, H. Sakue, and T. Takahagi, AIP
Conf. Proc., 418, 159 (1998). (See Table I).
34. K. N. Tu, C. C. Yeh, C. Y. Liu, and C. Chen, “Effect of current crowding
on vacancy diffusion and void formation in electromigration,” Appl. Phys.
Lett., 76, 988–990 (2000).
35. C. C. Yeh and K. N. Tu, “Numerical simulation of current crowding phe-
nomena and their effects on electromigration in VLSI interconnects,” J.
Appl. Phys., 88, 5680–5686 (2000).
36. E. C. C. Yeh and K. N. Tu, J. Appl. Phys., 89, 3203 (2001).
37. J. Lloyld, J. Appl. Phys., 94, 6483 (2003).
38. A. T. Wu, K. N. Tu, J. R. Lloyd, N. Tamura, B. C. Valek, and C. R. Kao,
“Electromigration induced microstructure evolution in tin studied by syn-
chrotron x-ray microdiffraction,” Appl. Phys. Lett., 85, 2490–2492(2004).
39. A. T. Wu, A. M. Gusak, K. N. Tu, and C. R. Kao, “Electromigration
induced grain rotation in anisotropic conduction beta-Sn,” Appl. Phys.
Lett., 86, 241902(2005).
SVNY339-Tu April 5, 2007 17:31

9
Electromigration in Flip Chip Solder Joints

9.1 Introduction

In 1998, Brandenburg and Yeh reported electromigration failure in flip chip

eutectic SnPb solder joints [1]. Using a current density of 8 × 103 A/cm2 at
150◦ C for about 100 hr, they detected failure by the formation of a pancake
type of void across the entire cathode contact to the Si chip. In addition,
a significant amount of phase separation in the bulk of the eutectic solder
bump was observed; the Pb has moved to the anode side and the Sn to the
cathode side. On the other hand, in the pair of bumps tested, while the one in
which electrons flowed from the Si to the substrate failed as described by the
pancake-type void formation, the other one in which electrons flowed from the
substrate to the Si did not fail. On the basis of their data, an attempt was made
by using Black’s equation of mean-time-to-failure to extrapolate lifetime of flip
chip solder joint. Taking the exponential factor of current density n = 1.8 and
activation energy of diffusion Q = 0.8 eV, they showed that electromigration
in flip chip solder joint would become a serious reliability issue in flip chip
technology. Since then, the subject has been included in the International
Technological Roadmap for Semiconductors for study.
The four interesting findings in their work are (1) the low current den-
sity, which is about two orders of magnitude less than that required to cause
electromigration failure in Al or Cu interconnects, (2) the failure mode of
pancake-type void formation, (3) the failure mode is asymmetrical in the pair
of bumps tested; the failure occurs only in the one bump with the cathode
contact to the Si side, and (4) the redistribution of Pb and Sn; they moved in
opposite directions. Since Sn moves against the electron flow, it might mean
that the effective charge of Sn is positive, which is unreasonable.
It is generally accepted that the basic mechanism of electromigration in
solder is the same as that in Al or Cu, and the basic concept of electromigration
in a metallic conductor remains true that it enhances atomic diffusion in the
electron flow direction by electron wind force. Hence, there is no reason to
expect that Pb and Sn behave differently in electromigration.
SVNY339-Tu April 5, 2007 17:31

246 Chapter 9

Furthermore, in eutectic SnPb solder joints, it is found that at tempera-

tures above 100◦ C, while atoms of Pb move in the same direction as electrons,
Sn atoms move in the opposite direction. However, in electromigration in eu-
tectic SnPb conducted at room temperature, Sn and Pb reversed their diffu-
sion direction. On the other hand, in pure Sn or Pb-free solders, Sn moves
in the same direction as electrons, evidenced by the growth of Sn hillocks
and whiskers at the anode [2–6]. An attempt to explain this behavior of re-
verse flow of Sn by the constraint of constant volume will be presented in
Section 9.7.
Flip chip solder joint technology was discussed in Section 1.3.3. A
schematic diagram of the cross section of such a joint was depicted in Fig. 1.9.
In today’s circuit design, each power solder joint in a flip chip may carry 0.2 A
and it will be doubled in the near future. At present, the diameter of a solder
joint is about 100 μm and it will be reduced to 50 μm and even 25 μm soon.
If the diameter of solder bumps and the spacing between them is 50 μm, we
can place 100 × 100 = 10, 000 solder joints on a 1 cm × 1 cm chip surface. In
the most advanced devices today, there are already over 7000 solder joints on
a chip. When a current of 0.2 A is applied to a bump, the average current
density in a 50-μm solder bump is about 104 A/cm2 . This current density is
about two orders of magnitude smaller than that in Al and Cu interconnects.
Nevertheless, electromigration does occur in flip chip solder joints at such low
current density, and it occurs by lattice diffusion.
Often the easy electromigration in solder joints is explained by the low
melting point of solder or fast atomic diffusion in solder. However, Table 8.1
shows that at the device working temperature of 100◦ C, lattice diffusion in
solder is not much slower than grain boundary diffusion in Al, nor slower than
surface diffusion in Cu. While the total atomic flux in lattice diffusion is much
greater than that in grain boundary diffusion or in surface diffusion, so is the
larger volume of a void required to cause failure of a solder joint. Therefore,
low melting point or fast diffusion is not the key answer. Why electromigra-
tion can occur in flip chip solder joints at such low current densities will be
explained below; it is due to the low critical product of electromigration of
solder alloys [6]. Furthermore, the line-to-bump geometry of a flip chip solder
joint leads to a large current crowding at the cathode contact where acceler-
ated electromigration occurs. Currently, there are attempts to use Cu column

Table 9.1. Resistance of Al and Cu interconnects and solder bumps

Al or Cu
interconnects Solder bumps

Cross section 0.5 × 0.2 μm2 100 × 100 μm2

Resistance 1–10 Ω 10−3 Ω
Current 10−3 A 1A
Current density 106 A/cm2 103 –104 A/cm2
SVNY339-Tu April 5, 2007 17:31

Electromigration in Flip Chip Solder Joints 247

bumps to reduce the effect of current crowding; electromigration in Cu column

bumps will be covered in Section 9.6.

9.2 Unique Behaviors of Electromigration

in Flip Chip Solder Joints
9.2.1 Low Critical Product of Solder Alloys
We recall the “critical product” in Eq. (8.21). If we replace Δσ by Y Δε, where
Y is Young’s modulus and Δε = 0.2% is the elastic limit, we see that the
“critical product” is a function of Young’s modulus, resistivity, and effective
charge number of the interconnect material:

Y ΔεΩ
jΔx = . (9.1)
Z ∗ eρ

We have used bulk modulus instead of Young’s modulus to explain the back
stress generation in Eq. (8.27). Since bulk modulus and Young’s modulus are
related, for convenience we use Young’s modulus here.
To compare the value of “critical product” among Cu, Al, and eutectic
SnPb, we recall that eutectic SnPb has a resistivity that is one order of mag-
nitude larger than those of Al and Cu, see Table 9.1. The Young’s modulus
of eutectic SnPb (30 GPa) is a factor of two to four smaller that those of Al
(69 GPa) and Cu (110 GPa) [7]. The effective charge number of eutectic SnPb
(Z ∗ of lattice diffusion) is about one order of magnitude larger than those of
Al (Z ∗ of grain boundary diffusion) and Cu (Z ∗ of surface diffusion). There-
fore, in Eq. (9.1), if Δx is kept constant for comparison, the current density
needed to cause electromigration damage in eutectic SnPb solder is two orders
of magnitude smaller than that needed for Al and Cu interconnects. If Al or
Cu interconnect fails in electromigration by a current density of 105 to 106
A/cm2 , solder joint will fail by 103 to 104 A/cm2 . This is the major reason
why electromigration in flip chip solder joints can be serious.

9.2.2 Current Crowding in Flip Chip Solder Joints

The failure mode in flip chip solder joint by the pancake-type void forma-
tion at the cathode contact interface can be explained by current distribution
in the unique geometry of a flip chip joint [8–11]. Figure 9.1 is a schematic
diagram depicting the geometry of a flip chip solder bump between an in-
terconnect line on the chip side (top) and a conducting trace on the board
or module side (bottom). Current crowding occurs at the contact interface
between the solder bump and interconnect line (the simulation will be dis-
cussed in Section 9.3). The high current density due to current crowding is
about one order of magnitude higher than the average current density in the
SVNY339-Tu April 5, 2007 17:31

248 Chapter 9

e- Al line
2 μm
Solder
J = I/A
AAl<<Asolder

Fig. 9.1. Schematic diagram depicting the geometry of a flip chip solder bump joining
an interconnect line on the chip side (top) and a conducting trace on the board or
module side (bottom).

bulk of solder joint. The low threshold of the current density needed to cause
electromigration in solder and the high current density induced by current
crowding are the key reasons why electromigration in flip chip solder joints
can compete with electromigration in Al and Cu interconnects as the major
reliability problem in microelectronic devices. Because of the current crowd-
ing, the related failure mode of pancake-type void formation in flip chip solder
joints is unique (see Section 9.3).

9.2.3 Phase Separation in Eutectic Solder Joints

In electromigration, the microstructure change in the bulk of a flip chip solder
joint as well as at the cathode and anode interfaces is very different from that
in Al and Cu interconnects. The reverse flow of Sn at high temperatures or
the reverse flow of Pb near room temperature is due to the two-phase eutectic
structure of solder alloys. The reverse flow behavior can be explained by as-
suming a constant volume constraint in segregation in a two-phase structure.
Concerning the unique behavior of phase segregation in a two-phase eutectic
structure driven by an applied force, a detailed discussion will be given in
Sections 9.7 and 12.3.

9.2.4 Narrow Range of Current Density

The trend of device miniaturization has reduced the diameter of solder bump
below 100 μm, so the average current density in the bump is approaching 104
A/cm2 . While this current density is not high, the joule heating associated
with current crowding and pancake-type void formation will cause a large
increase in temperature of the solder bump. In a systematic experimental
study of melting of certain flip chip eutectic SnPb solder joints, it was found
that a threshold current density around 1.6 × 104 A/cm2 exists, above which
melting occurs. The melting is time-dependent and partial, meaning it takes
time to melt a part of the solder joint. More discussion on time-dependent
melting of flip chip solder joint will be given in Section 9.4.7. Since the ap-
plied current density in real devices is approaching 1 × 104 A/cm2 and since
it is very close to the threshold current density of melting, melting of flip
SVNY339-Tu April 5, 2007 17:31

Electromigration in Flip Chip Solder Joints 249

chip solder joints has become a new reliability concern because it is time-
dependent.
The reason for melting is unclear. Nevertheless, it is worth mentioning that
when the current density in the bump is 1 × 104 A/cm2 , the current density
in the interconnect will be about 1 × 106 A/cm2 since the interconnect cross
section is two orders of magnitude smaller. Such current density will cause
electromigration in the Al interconnect. Hence, from the point of view of reli-
ability, there is a competition between the failure in the solder bump and in the
Al interconnect. Especially in the flip chip bump where the electrons flow from
the bump to the interconnect, there is an electromigration-induced atomic flux
divergence in the Al interconnect, similar to the Al line on a via of W. At the
point where current enters the Al interconnect, more Al atoms are driven away
due to the very large change in current density. Then the condensation of the
reverse flux of vacancies may lead to the formation of a void in the Al above
the solder bump. It will increase the resistance and joule heating of the Al.
Knowing the upper bound of current density in melting, we may ask what
is the lower bound or the threshold current density below which very little
electromigration damage occurs in a flip chip solder joint. It is about 1 ×
103 A/cm2 . Combining the two bounds, we see that there is only a narrow
range of one order of magnitude of difference in current density, in which
electromigration in flip chip solder joints can be studied. The upper bound
may depend on the design of UBM and bond pad, and an optimal design
may increase the upper bound to 5 × 104 A/cm2 to avoid melting. The lower
bound may depend on the composition of the solder, for example, if Pb-free
solder is used, it may change to 5 × 103 A/cm2 . Anyway, the range between
the two bounds is narrow.

9.2.5 Effect of Under-Bump Metallization on Electromigration

Because of the use of under-bump metallization (UBM) to form the solder
joint, electromigration has enhanced the dissolution of UBM on the cathode
side into solder bump and the transportation of the solute to the anode side,
followed by a large amount of intermetallic compound formation near the
anode [12–14]. This behavior will be discussed later. The effect of thick Cu
UBM will be discussed in Section 9.6.

9.3 Failure Mode of Electromigration in

Flip Chip Solder Joints
Figure 9.1 depicted the line-to-bump geometry of a flip chip solder joint.
Because the cross section of the line on the chip side is at least two orders
of magnitude smaller than that of the solder bump, there is a very large
current density change at the contact between the bump and the line because
the same current is passing between them. Since electric current will take
SVNY339-Tu April 5, 2007 17:31

250 Chapter 9

the lowest resistance path, electrons will jam at the entrance into the solder
bump, resulting in current crowding. There are two significant effects due to
the current crowding. First, there is an abrupt change in current density in
the interconnect, before and beyond the point where current turns into the
bump. Second, the current density in the solder bump near the entrance point
will be about one order of magnitude higher than the average current density
in the middle of the bump. It will be 105 A/cm2 near the entrance when the
average current density in the middle of the bump is 104 A/cm2 .
Figure 9.2(a) is a two-dimensional simulation of current distribution in
a solder joint. Figure 9.2(b) is a display of current density distribution in
the joint, where the cross section of the joint is plotted on the x–y plane and

Fig. 9.2. (a) Two-dimensional simulation of current distribution in a solder joint. (b)
Current density distribution in the joint, where the cross section of the joint is plotted
on the x–y plane and the current density is plotted along the z-axis.
SVNY339-Tu April 5, 2007 17:31

Electromigration in Flip Chip Solder Joints 251

e-

contact window
e-

e-

Fig. 9.3. SEM images of the damage in a flip chip solder joint caused by electromi-
gration. Electrons of the applied current entered the bump from the upper right corner
of the joint. (a) Up to 37 hr at 125◦ C with a current density of 2.25 × 104 A/cm2 , no
damage was observed. (b) After 38 hr and (c) after 40 hr, voids are seen at the upper
right interface, and the voids have propagated along the interface from right to left.
(d) After 43 hr, the joint failed by having a large void across the entire interface.

the current density is plotted along the z-axis. It is the current crowding or the
high current density shown at the upper right corner in Fig. 9.2(a) and (b) that
leads to electromigration damage in the solder joint, not the average current
density in the bulk of the joint. Consequently, electromigration damage in a
flip chip solder joint occurs near the cathode contact on the chip side, i.e., the
contact between the interconnect and the bump. The damage begins near the
entrance point of the electric current. How it propagates across the contact
will be explained below.
Figure 9.3 displays a set of SEM images of the damage in a flip chip solder
joint caused by electromigration. The upper contact of the solder joint to
Si consisted of a thin-film UBM of Cu/Ni(V)/Al. The total thickness of the
thin-film UBM is about 1 μm and the thickness of Cu is about 0.4 μm, hence
the UBM is not resolved in the SEM image. The applied current of electrons
entered the bump from the upper right corner of the joint. Up to 37 hr at
125◦ C with a current density of 2.25 × 104 A/cm2 , no damage was observed as
shown in Fig. 9.3(a). Yet, after 38 and 40 hr, voids are seen at the upper right
corner of the interface, and the voids have propagated along the interface from
SVNY339-Tu April 5, 2007 17:31

252 Chapter 9

Fig. 9.4. Corresponding curve of potential change versus time for Fig. 9.3.

right to left, shown in Fig. 9.3(b) and (c), respectively. After 43 hr, the joint
failed by having a large pancake-type void across the entire interface, shown in
Fig. 9.3(d). The corresponding curve of potential change versus time is shown
in Fig. 9.4. The curve shows that the potential change is insensitive to the
void formation until the end, where it shows an abrupt jump when the void
has propagated across the entire interface. The arrows in Fig. 9.4 indicate the
corresponding “time” when the images in Fig. 9.3 were taken.
Why the potential change of the solder joint is insensitive to void forma-
tion and propagation can be explained by two findings. The first is shown in
Fig. 9.5, which depicts the cross section of a solder joint with pancake-type
void formation at the upper interface. The formation and propagation of the
void displaced the entrance of the current to the front of the void, so there
is very little change in resistance of the solder bump by the void formation
as long as the current can enter the solder bump. Finally, an abrupt change
occurs only when the void has extended across the entire joint or when the
contact becomes an open. The second finding is shown in Table 9.1, which
compares the electrical behavior between Al (or Cu) interconnect and solder
joint. The resistance of a cubic piece of solder of 100 μm × 100 μm × 100 μm
(the size of a solder joint) is about 1 milliohm. The resistivity of Sn and Pb is
11 and 22 μΩ-cm, respectively. The resistance of an Al or Cu line 100 μm long
with a cross section of 1μm × 0.2 μm is about 10 ohms. The solder joint is a
SVNY339-Tu April 5, 2007 17:31

Electromigration in Flip Chip Solder Joints 253

Si
e-

e- Void
propagation

e- Cu

Fig. 9.5. Schematic diagram depicting the cross section of a solder joint with pancake-
type void formation and propagation at the upper interface.

low-resistance conductor, but the interconnect is a high-resistance conductor

and becomes the source of joule heating.
The simple calculation in the above shows that the resistance of the in-
terconnect will be very sensitive to the design of its dimension and to a slight
microstructure change and damage. Yet the resistance of the solder bump is
not, e.g., not even by the inclusion of a large void within the bulk of the sol-
der bump. Often, the matrix of a solder bump may contain a few very large
spherical voids due to residue flux from the solder paste, especially the Pb-
free solder paste, yet the voids have little effect on the resistance of the solder
bump, except when they move to the contact interface and spread out with
the residue.
Figure 9.6 is a SEM image of cross section of a daisy chain of flip chip
solder joints between a Si chip on the top and a substrate at the bottom. The
direction of electron current is indicated by the string of arrows. The cathode
contacts on the Si where the electron current enters the solder bump are
indicated by small circles. The UBM on the Si chip side was Cu/Ni(V)/Al with
0.8 μm Cu, 0.32 μm Ni(V), and 1 μm Al. The bond pad on the substrate side
was Au/Ni(V)/Cu with 0.08 to 0.2 μm Au, 3.8 to 5 μm Ni(V), and 38 μm Cu.
The solder bump composition was either eutectic SnPb or 95.5Sn4Ag0.5Cu.

Fig. 9.6. SEM image of cross section of a daisy chain of flip chip solder joints between
a Si chip on top and a substrate on the bottom. The arrows indicate the electron flow
directions.
SVNY339-Tu April 5, 2007 17:31

254 Chapter 9

The tests were carried out at 50◦ C ambient with an applied current of 1.7 A
for the SnPb solder bumps and 1.8 A for the SnAgCu bumps, and the current
density was about 3.5 to 3.7 × 103 A/cm2 [21].
The chip surface temperature was monitored by the temperature coeffi-
cient of resistance of 450-ohm serpentine aluminum metal resistors residing on
the test chip surface. These resistors were used to determine package thermal
resistance characteristics and for in situ monitoring of I 2 R or joule heating.
The joule heating was found to increase as electromigration-induced voiding
in the bump worsened. The temperature was calculated by adding dT to the
ambient temperature of 50◦ C.

Tdie = 50 + dT = 50 + RI 2 θja (9.2)

where θja is defined as the package “junction-to-air” thermal resistance value,

which is obtained by measuring dT and by calculating I 2 R. The measured
value of θja is about 62 to 72 C/watt. The chip temperature was found to
be as high as 175◦ C, indicating an increase of temperature of about 125◦ C
due to joule heating. The time for resistance changes of 15% as a function of
chip surface temperature was found to have a very sensitive dependence on
temperature.
Figure 9.7 shows a pancake-type contact void. The failure mode is unique
due to the flip chip configuration that void formation occurs only at the cath-
ode contact on the Si chip side, where the electron current enters the solder
bump. Hence, void or failure occurs only in one of a pair of bumps, not every
bump.

Fig. 9.7. Cross-sectional SEM image of pancake-type void formation. The location of
the void formation corresponds to the entrance of electrons into the flip chip bumps.
SVNY339-Tu April 5, 2007 17:31

Electromigration in Flip Chip Solder Joints 255

9.4 Electromigration in Flip Chip Eutectic

Solder Joints
Two kinds of solder bumps were tested for comparison of electromigration.
Eutectic SnPb and SnAg3.8 Cu0.7 , between electroless Ni UBM on Si and elec-
troplated Cu bond pad on printed circuit board (PCB), were tested at 120◦ C
with 1.5 A on a hot plate in atmospheric ambient. The solder bumps were
formed by solder paste printing through a stainless mask and reflowed twice
in a belt furnace with a peak temperature of 240◦ C. The first reflow was car-
ried out after printing the solder bumps on the chip, and the second reflow
was carried out to assemble the chip and PCB. After the assembly, the gaps
between the chip and the board were filled with epoxy, i.e., underfill. While
the IMC on the Ni UBM experienced two reflows, the IMC on Cu bond pad
has undergone only one reflow.
To have an in situ observation of electromigration of the solder bump,
a pair of solder bumps were cross-sectioned and polished mechanically and
chemically before electromigration. The diameter of the contact opening is
100 μm. The average current density through a half contact opening can
be calculated to be about 3.8 × 104 A/cm2 when 1.5 A was applied. It was
calculated on the basis of half of the SiO2 contact opening, not half of the
solder bump diameter.
The polishing has left behind embedded SiC and diamond particles on
the solder surface. The size of these particles is roughly 1 μm, and they
were used as inert diffusion markers to calculate the atomic flux driven by
electromigration. The marker motion and surface topographic changes due
to electromigration were observed with scanning electron microscopy (SEM),
energy-dispersive x-ray spectroscopy (EDS), and optical microscopy (OM).
In order to perform the observation, electromigration was stopped time to
time for observation, i.e., the observation was discontinuous and repeated at
various periods. At the end of the test, the samples were cross-sectioned a
second time, in a direction perpendicular to the first cross section, for SEM
observation. Schematic diagrams depicting the first and second cross sections
are shown in Fig. 9.8 [4, 5].

9.4.1 Electromigration in Eutectic SnPb Flip Chip Solder Joints

On the first cross-sectioned surface of a eutectic SnPb bump after electromi-
gration of 20 to 40 hr, accumulation of Pb in the anode side and void formation
in the cathode side were observed. The flux of atomic motion in the solder
bump can be measured with marker motion. In Fig. 9.9(a), the marker posi-
tions are shown, and the Pb is the brighter phase in the SEM image. Each
marker is either a SiC or a diamond particle. Their displacements are shown
in Fig. 9.9(b). Except for marker numbers 1, 10, and 11, all the other mark-
ers are similar in magnitude of movement. The average movement, except for
SVNY339-Tu April 5, 2007 17:31

256 Chapter 9

1st Cross-Section
a)

2nd Cross-Section

b)
Si

Ni electron flow Underfill

FR4 Cu

Fig. 9.8. Schematic diagrams depicting the first and second cross sections of flip chip
solder joints for observation of electromigration.

Fig. 9.9. (a) Marker positions on the cross-sectional surface. Each marker is either
a SiC or a diamond particle. (b) Marker displacements. Except for marker numbers
1, 10, and 11, all the other markers are similar in magnitude of movement. (c) The
average movement, except for marker numbers 1, 10, and 11, as a function of time. A
linear displacement is seen in the period from 20 to 39.5 hr.
SVNY339-Tu April 5, 2007 17:31

Electromigration in Flip Chip Solder Joints 257

marker numbers 1, 10, and 11, as a function of time is shown in Fig. 9.9(c).
A linear displacement is seen in the period from 20 to 40 hr.
In analyzing the atomic flux in the solder joint, electron wind force and the
mechanical force were considered to affect the motion of Pb atoms at 120◦ C,

D dσΩ D ∗
Jem = −C +C Z eE, (9.3)
kT dx kT

where Jem is atomic flux in units of atoms/cm2 sec, C the concentration of

atoms per unit volume, D/kT the atomic mobility, σ the hydrostatic stress
in the metal and dσ/dx the stress gradient along the direction of electron
flux, Ω the atomic volume, Z ∗ the effective charge number of electromigra-
tion, e the electron charge, E the electric field and E = ρj, where ρ is the
electrical resistivity and j is the current density. The value of atomic flux of
electromigration, Jem , can be estimated by the motion of marker [3,4]:

VEM u
Jem = = , (9.4)
ΩAt Ωt

where VEM is the volume of solder moved by electromigration, which was

calculated as the marker displacement multiplied by the cross-sectional solder
area A · u is the marker displacement, and t is the electromigration time.
By ignoring the back stress and by measuring the atomic flux of electro-
migration, Jem , the product of diffusivity and effective charge number, DZ ∗ ,
can be calculated from the following equation [4, 5]:

VEM D
Jem = ≈C· · Z ∗ · e · E. (9.5)
Ω · (A · t) kT

From the calculated values of DZ ∗ , we can estimate Z ∗ by knowing D and

assuming no back stress.
To examine whether the measured value of DZ ∗ is reasonable or not, we
have attempted to unravel Z ∗ from D by using the activation energy of mean-
time-to-failure (MTTF) of 0.8 eV measured by Brandenburg and Yeh [1] and
the prefactor of 0.1 cm2 /sec. The effective charge number is calculated to be
about 102 at 120◦ C. If we assume there is joule heating and assume the actual
solder temperature to be 140◦ C instead of 120◦ C, the effective charge number
would become 34. This value is in better agreement with the effective charge
number of pure Pb in Pb, which is about 47, than that of pure Sn in Sn, which
is about 17, as discussed in Chapter 8.
Figure 9.10 shows the second cross section of the solder bump which has
had the first cross section. The voids and the indented solder surface at the
cathode side and a bulge at the anode side can be seen, indicating that it
is not a constant volume transformation. More volume has been transferred
from the cathode to the anode. The polished surface, i.e., the free surface,
SVNY339-Tu April 5, 2007 17:31

258 Chapter 9

(a) (b)
1st Cross-Section Cu

Underfill
Al

Si
Al
electron flow

2 nd Cross-Section
Cross -Section FR4

Cu

Figure Sample layout (a) Plane view (b) Side view

electron

Ni : 18.4at%
Ni : 15.9at%
Cu : 28.4at% Cu : 29.6at%
Sn: 50.8 at%
Sn:52at%
Ni : 11.8at%
Cu : 23.6at%
Sn: 46.0at%
Pb: 18.6at%

50 μm

Cu : 61.8at% Cu : 42.9at%
Sn: 31.7at% Sn: 52.3at%
Pb: 5.2 Pb: 3.6

Fig. 9.10. The second cross section of the solder bump which has had the first cross
section. The voids and the indented solder surface can be seen at the cathode side.

enabled the indentation and bulge to take place, and both Pb and Sn were
moved to the anode. Without a composition mapping, the relative fluxes of
Sn and Pb in the transformation are unclear. The marker motion gives the
net flux, so there could be some reverse flux of Sn diffusing to the cathode
since Pb is the dominant diffusing species. Therefore, the calculation of Z ∗ in
the above could be inaccurate and it can only serve as an indication of the
SVNY339-Tu April 5, 2007 17:31

Electromigration in Flip Chip Solder Joints 259

general behavior of electromigration in eutectic SnPb solder joints. We shall

address the constant volume process and phase segregation issue in Sections
9.7 and 12.3.
In addition, the Ni-Cu-Sn compounds are found in the matrix of the solder
bump. The farthest location of this compound from the electroless Ni UBM is
about 20 μm, indicating that Cu atoms has diffused such a distance to form
the ternary compound.

9.4.2 Electromigration in Eutectic SnAgCu Flip

Chip Solder Joints
Figure 9.11 shows the first cross-sectioned surface before and after current
stressing at 120◦ C with 1.5 A for 20, 110, and 200 hr of a Pb-free solder
bump. The direction of electron flow is from the Ni UBM to Cu bump pad.
The voids are formed at the cathode side after 200 hr [see Fig. 9.11(d)]. The
void formation is much slower than that in eutectic SnPb solder as discussed in
Section 9.4.1. But the IMCs were squeezed out in the anode side as hillocks.
In contrast, no compound was pushed out in the eutectic SnPb solder in
electromigration.
Figure 9.12 shows (a) the marker and (b) marker motion on a cross-
sectioned surface. The marker motion is much less than that in eutectic SnPb.
The markers moved more in the bottom region of solder (marker Nos. 4 and 5),
which are close to the squeezed out IMC. However, the markers (Nos. 1 to 3)
close to the electroless Ni UBM moved very little. Compared with eutectic
SnPb, the marker motion in SnAg3.8 Cu0.7 is much smaller, which indicates
that the latter has a slower electromigration than the former.

Fig. 9.11. The first cross-

sectioned surface before (a)
and after current stressing at
120◦ C with 1.5 A for (b) 20,
(c) 110, and (d) 200 hr of a Pb-
free solder bump.
SVNY339-Tu April 5, 2007 17:31

260 Chapter 9

(a)

Marker
e-
1
2 3

5 4

(b) 6 20 hr
Displacement(μm)

5 20+44 hr
200 hr
4
3
2
1
0
1 2 3 4 5
Marker

Fig. 9.12. (a) The marker and (b) marker motion on a cross-sectioned surface of the
Pb-free solder joint.

The sample was also second cross-sectioned, perpendicular to the first

cross section. The SEM micrograph is shown in Fig. 9.13; the void formation
and dissolution of Ni UBM on the cathode side can be seen. However, the
first cross-sectioned surface appeared rather flat, and no obvious dimple or
bulge. The Ni-Cu-Sn ternary compounds (have a darker image in the micro-
graph) were found in the matrix of the solder, similar to eutectic SnPb. The
compounds grew across the entire cross section of the solder bump during
electromigration. The distance of the farthest compound from the electroless
Ni UBM is about 90 μm, almost reaching the Cu anode.

9.4.3 Marker Motion Analysis Using Area Array

of Nano-indentations
An area array of nano-indentation markers can be made on a cross-sectioned
solder joint surface using an instrumented system with a Berkovich diamond
pyramid tip. Figure 9.14 shows a set of 5 × 6 nano-indentation markers created
SVNY339-Tu April 5, 2007 17:31

Electromigration in Flip Chip Solder Joints 261

P :17.3at%
P :18at% P :13at% Ni:85.8at%
Ni:82at% Ni:85.8at%
Ni:16.43at%
Cu:24.8at%
Ag : 0.45
Ni:0.47at%
Sn : 58.4
Cu:0.42at%
Ag : 77.17
Sn : 21.9
Ni:10.95at%
Cu:23.8at%
Ag : 0.53
Sn : 64.7

Ag3Sn
Ni:12.17at%
Cu:28.50at%
Ag : 0.34
Sn : 58.99

Ni:6.3at%
Cu:28.1at%
Ag : 0.99
Sn : 64.6

Ni:1.5at%
Cu:38.0at%
Ag : 0.22
Sn : 60.3

Fig. 9.13. The second cross section, perpendicular to the first cross section of the
Pb-free solder joint after electromigration. Void formation and dissolution of Ni UBM
on the cathode side can be seen.

on a cross-sectioned eutectic SnPb solder joint before and after electromigra-

tion of current stressing of 0.32 A at 125◦ C for 24 and 48 hr. The diameter of
the solder joint is about 100 μm. The size of the nano-markers with depth of
1000 nm used in the study was about 5 μm. The electrons entered the sample
at the lower left corner and exited at the upper right corner. In Fig. 9.14(b)
and (c), hillock formation at the upper right corner can be seen. On the sur-
face, a bulge near the anode and a dimple near the cathode can also be seen.
The majority of markers moved down, against atomic mass flow, except a few
near the anode side.
Figure 9.15 shows an area array of nano-indentation markers created on a
cross-sectioned eutectic SnAgCu solder joint before and after electromigration
[15]. The electromigration test was performed with an average current density
of 1 × 104 A/cm2 at 125◦ C for 15 days. The electrons flowed from the top to
the bottom. The size of the solder bump was 300 μm. Upon completion of
the test, the specimens were examined by SEM for comparison. The surface
SVNY339-Tu April 5, 2007 17:31

262 Chapter 9

Pb Mass protrusion and became bigger

e- e- e-

e- Mass depletion
0 hr 24 hrs 48 hrs

Fig. 9.14. A set of an area array of 5 × 6 nano-indentation markers created on a

cross-sectioned eutectic SnPb solder joint before and after electromigration of current
stressing of 0.32 A at 125◦ C for 24 and 48 hr. The electron flow entered the sample at
the lower left corner and exited at the upper right corner.

topology of eutectic SnAgCu appears to be relatively flat compared to that

of eutectic SnPb shown in Fig. 9.14. This is in good agreement with the
results shown in Figs. 9.10 and Fig. 9.13. The marker displacements driven
by electromigration at different locations were determined by measuring the
distance from the marker center to the reference line at the anode interface
using SEM images.
As indicated by the arrows in Fig 9.15, the experimental measurement
showed that markers in the top four rows moved upward in the figure, which
was as expected to be against the atomic flux driven by a downward electron
flow. However, the markers at the sixth row from the bottom were found to
move downward in the figure. Especially some markers on the sixth row had
moved into the anode and disappeared from the SEM image. This indicates
that the atomic flux was upward near the anode, which was against the atomic
flux of electromigration. For the fifth row, the marker displacements were
found to be much smaller than the others. The displacement of markers at
different rows versus marker locations is plotted in Fig. 9.16. It seems that
there exists a “neutral plane” where no net flux moves toward either the
cathode or the anode. The location of the neutral plane can be determined
by the intersection between the x-axis and displacement curve. As shown in
Fig. 9.16, the distance from the neutral plane to the anode interface is 55 μm.
A possible mechanism for the reverse marker motion near the anode could
be the back stress or Kirkendall shift that induced a reverse flow of Sn from
the anode to the cathode. According to the Blech–Herring model of back stress
in electromigration, the stress gradient will produce a vacancy concentration
gradient, in turn an atomic flux to counteract the electromigration flux. If
the stress gradient is large enough, there will be no net electromigration or
no electromigration-induced damage of hillock or void formation. This im-
plies that the neutral plane should be at the anode interface. On Kirkendall
SVNY339-Tu April 5, 2007 17:31

Electromigration in Flip Chip Solder Joints 263

Fig. 9.15. A set of an area array of

5 × 6 nano-indentation markers cre-
ated on a cross-sectioned eutectic
SnAgCu solder joint (a) before and
(b) after electromigration of current
stressing of 1 × 104 A/cm2 at 125◦ C
for 360 hr. The electron flow entered
the sample at the top and exited at
the bottom.

shift, following Darken’s analysis of marker motion in interdiffusion, if lattice

planes start to shift from the anode interface, all the markers should move
toward the cathode. Hence, to analyze the motion of the sixth row markers,
a different mechanism is needed and it should take into account the effect of
current crowding at the bottom of the joint where the current exited from the
solder joint and the fact that the markers were on the surface. The plastic
deformation in hillock growth near the anode may play a role too.

9.4.4 Mean-Time-to-Failure of Flip Chip Solder Joints

The electronic industry uses the mean-time-to-failure (MTTF) analysis to
predict the lifetime of a device. In 1969, Black provided the following equation
SVNY339-Tu April 5, 2007 17:31

264 Chapter 9

Fig. 9.16. Plot of the displacement of markers at different rows after 239 hr versus
marker locations. It seems that a “neutral plane” exists.

to analyze failure in Al interconnects caused by electromigration [16]:

 
1 Q
MTTF = A exp . (9.6)
jn kT

The derivation of the equation was based on an estimate of the rate of

mass transport resulting in the formation of a void across an Al interconnect.
The most interesting feature of the equation is the dependence of MTTF on
the square power of current density, i.e., n = 2.
In the subsequent studies of the MTTF equation, whether the exponent
n is 1, 2, or a larger number has been controversial, especially when the
effect of joule heating is taken into account. However, assuming that mass
flux divergence is required for failure and the nucleation and growth of a void
requires vacancy supersaturation, Shatzkes and Lloyd have proposed a model
by solving the time-dependence diffusion equation and obtained a solution
for MTTF in which the square power dependence on current density was
also obtained [17]. Nevertheless, whether Black’s equation can be applied to
MTTF in flip chip solder joints deserves a careful examination.
To determine the activation energy, accelerated tests at high temperatures
are performed. Attention must be paid to the temperature range in which lat-
tice diffusion might overlap grain boundary diffusion and also grain boundary
diffusion might overlap surface diffusion. For eutectic SnPb solder, it is more
complicated because of the change of dominant diffusion species between Pb
and Sn above and below 100◦ C.
The formation of a void requires nucleation and growth. In the case of a
flip chip solder joint, Fig. 9.5 shows that the bulk part of the time to failure is
controlled not by the growth of a void across the contact interface, but by the
incubation time of void nucleation. The latter takes about 90% of the time of
failure. The propagation of the void across the entire contact takes only about
SVNY339-Tu April 5, 2007 17:31

Electromigration in Flip Chip Solder Joints 265

Table 9.2. Mean-time-to-failure of eutectic SnPb flip chip solder joints

1.5 A 1.8 A 2.2 A

(1.9 × 104 A/cm2 ) (2.25 × 104 A/cm2 ) (2.75 × 104 A/cm2 )
Calculated Measured Calculated Measured Calculated Measured
(h) (h) (h) (h) (h) (h)

100◦ C ··· ··· 380 97 265 63

125◦ C 108 573a 79.6 43 55.5 3
140◦ C 46 121 34 32 24 1
a
Not failed.

10% of the time. Furthermore, as shown in the earlier sections in this chapter,
the effect of current crowding on failure is crucial and cannot be ignored in the
analysis of MTTF. Black did point out the importance of current gradient or
temperature gradient on interconnect failure, although he did not take them
into account in his equation explicitly [16]. On the basis of the unique failure
mode of a flip chip solder joint as shown in Figs. 9.3 to 9.7, the major effects
of current crowding are to increase greatly the current density at the entrance
of the solder joint and also to increase the local temperature due to joule
heating. Furthermore, solder joint has IMC formation at both the cathode
and the anode interfaces, electromigration affects IMC formation, and in turn
IMC formation affects failure time and mode. This was not considered in
Black’s original model of MTTF. Therefore, we cannot apply Black’s equation
to predict flip chip solder joint lifetime without modification.
Brandenburg and Yeh used Black’s equation with n = 1.8 and Q = 0.8
eV/atom, without taking into account the effect of current crowding. The
equation, with n = 1.8 and Q = 0.8 eV/atom, has been found to have greatly
overestimated MTTF of flip chip solder joints at high current densities.
Table 9.2 compares the calculated and measured MTTF of eutectic SnPb
flip chip solder joints at three current densities and three temperatures. At
the low current density of 1.9 × 104 A/cm2 , the measured MTTF is slightly
longer than the calculated, but at 2.25 × 104 A/cm2 and 2.75 × 104 A/cm2 ,
the measured MTTF is much shorter than the calculated. This is also true for
the eutectic SnAgCu flip chip solder joints. These findings show that MTTF
of flip chip solder joints is very sensitive to a small increase of current density;
the MTTF drops rapidly when the current density is about 3 × 104 A/cm2 .
Also, the Pb-free solder has a much longer MTTF than the SnPb solder. For
example, at 2.25 × 104 A/cm2 at 125◦ C, the MTTF is 580 hr for the Pb-free
versus 43 hr for the SnPb.
Black’s equation can be modified to include the effect of current crowding
and joule heating [9]:
 
1 Q
MTTF = A exp , (9.7)
(cj)n k(T + ΔT )
SVNY339-Tu April 5, 2007 17:31

266 Chapter 9

where c is due to current crowding and has a magnitude of 10 and ΔT is due

to joule heating and may be higher than 100◦ C. Both parameters c and ΔT
will reduce the MTTF from Black’s equation, i.e., make the solder joint fail
much faster. Since ΔT depends strongly on j, the modified equation is much
more sensitive to the change of current density than Black’s equation. We
recall that the value of ΔT will depend on the design of flip chip solder joint
and interconnect, because of heat generation and heat dissipation.

9.4.5 Comparison between Eutectic SnPb and SnAgCu

Flip Chip Solder Joints
The marker motion as discussed in Sections 9.4.1 and 9.4.2 shows that elec-
tromigration in SnAg3.8 Cu0.7 is much slower than that in eutectic SnPb. In
addition, the MTTF of the latter is also shorter than the former. Why is the
Pb-free solder better? On the basis of Eq. (9.3), the flux driven by electromi-
gration has the driving force term, Z ∗ eE, and the mobility term, D/kT. In
driving force, both Z ∗ and resistivity differ for the two solders, yet the differ-
ence is small. In mobility, the difference in diffusivity can be very large; the
diffusivity in eutectic SnPb may be one order of magnitude faster than that
in eutectic SnAgCu. This is because the melting temperature of SnAg3.8 Cu0.7
(about 220◦ C) is higher than that of eutectic SnPb (183◦ C). Therefore, at
the same stressing temperature, the homologous temperature of the Pb-free
is lower than that of eutectic SnPb. Also, the smaller grain size and the eutec-
tic lamellar interfaces in the SnPb solder may enhance the diffusivity. Thus,
electromigration in eutectic SnPb will be faster.
Furthermore, in Eq. (9.3), we note the back stress term. The effect of
back stress in resisting electromigration in the Pb-free is larger than that in
the SnPb. A distinct difference in electromigration behavior between eutectic
SnPb and SnAg3.8 Cu0.7 is the squeezing out of IMC at the anode side of the
latter (see Fig. 9.11). It seems that in eutectic SnPb, the compressive stress
at the anode can be relaxed by the bulge of the solder surface as shown in
Fig. 9.10, indicating that lattice sites can be created easily because of more
grain and interface boundaries. But in the SnAg3.8 Cu0.7 , the Sn matrix is
mechanically harder and the surface oxide is protective so the cross-sectioned
surface remains rather flat as shown in Fig. 9.13. The higher compressive
stress or back stress was relaxed by squeezing out the hillocks of IMC. If the
Pb-free solder bump is confined by underfill which resists surface relief, the
buildup of compressive stress at the anode may be even higher.
It is worth mentioning that there is a very large reverse flux of Sn in
SnPb flip chip solder joints in electromigration at high temperatures, to be
discussed in Sections 9.5 and 9.7. The reverse flux is much smaller in the
Pb-free solder although the concentration of Sn is higher in the Pb-free, as
discussed in Sections 9.4.3 and 9.4.4. The reverse flux has a very serious effect
on the stability of thick Cu UBM at the cathode contact, in turn the MTTF
of the solder joint, to be discussed in Section 9.5. In thin-film Cu UBM, the
SVNY339-Tu April 5, 2007 17:31

Electromigration in Flip Chip Solder Joints 267

Cu will be consumed during the reflow, but in thick Cu UBM, the Cu will also
be consumed by solid-state Cu-Sn reaction, which is enhanced by the reverse
flux of Sn. The MTTF depends not only on the rate of electromigration, but
also on the rate of Cu-Sn reaction in consuming the Cu.

9.4.6 Kinetic Analysis of Pancake-Type Void Growth along

the Contact Interface
Figure 9.17 is a schematic diagram depicting the growth of a pancake-type void
at a contact interface [11]. The bent solid arrow represent current crowding.
The vertical solid arrow depict the atomic flux driven by the current crowding
from the top of the solder bump to the bottom. Meanwhile, the reverse flux
of vacancies moves from the bulk of the solder to the interface indicated by
the dotted arrow. If the vacancy flux in the Cu6 Sn5 is ignored, the vacancy
fluxes in the solder can be written as
bulk
v CSn DSn ∗
JSn = ZSn eρSn j, (9.8)
kT

where D is the diffusivity, e is the charge of an electron, ρ is resistivity, and j

is current density. Z ∗ is the effective charge number of electromigration.
The solder/IMC interface provides the transport path for excess vacancies
and enables them to diffuse along the interface. The lateral flux along the

Cu6Sn5
Jv
h
d
J
void Jint
a
d
b'
J*
Solder Sn

Fig. 9.17. In the model, the bent solid arrows represent current crowding. The vertical
solid arrows depict the atomic flux driven by the current crowding from the top of the
solder bump to the bottom.
SVNY339-Tu April 5, 2007 17:31

268 Chapter 9

interface due to the divergence of vacancies can be written as

ΔC ΔC
v
Jint = −Dint ≈ Dint  , (9.9)
Δx b

where Dint is the diffusivity in the interface, b is the width of current crowding
region, and ΔC is the concentration difference between the concentration in
the higher current density and the equilibrium concentration at the tip or
growth front of the void. In terms of mass conservation law, we have

v
Jint v
aδ = JSn ab , (9.10)

where δ is the effective width of interface and a is a unit length.

It is assumed that the initial width of void is d, and Jvoid is the flux of
vacancy at the tip of the void. The condition of conservation of flux is applied
again

v
Jint aδ = Jvoid ad. (9.11)

Substituting Eq. (9.10) into Eq. (9.11), the flux to grow the void can be
written as

v b
Jvoid = (JSn ) . (9.12)
δ

The volume of matter transported by Jvoid along the interface can be given
as

ΔV = Jvoid AΔtΩ, (9.13)

where A = aδ, ΔV = adΔl, and Ω is the atomic volume.

Inserting Eq. (9.12) into Eq. (9.13), the growth velocity of void becomes


Δl v b
v= = (JSn ) Ω. (9.14)
Δt d

If it is assumed that Cvbulk Ω = 1, we obtain


ej ∗ b
v= (DSn ρSn ZSn ) . (9.15)
kT d

To verify the mechanism of void propagation, the two key parameters are
the width of current crowding region, b , and the width of the void, d. As
shown in Fig. 9.17, the Gibbs-Thomas effect may play an important role in
SVNY339-Tu April 5, 2007 17:31

Electromigration in Flip Chip Solder Joints 269

Current Crowding Region

Void
Conducting Line

Solder Mask 10-4

10-1

Solder Bump
102

Fig. 9.18. Two-dimensional simulation of pancake-type void growth.

forming the tip of the void,

 
2γ Ω
Cr = C0 exp , (9.16)
r kT

where γ is the surface energy per area.

By applying the linear approximation, we have the void width as

C0 4γΩ
d = 2r = . (9.17)
ΔC kT

Since the model is two-dimensional, the void width is assumed to remain

constant. On the other hand, from Eq. (9.9) and (9.10), the width of current
crowding is
 1/2
 ΔC kT Dgb δ
b = ∗ ρ . (9.18)
C0 ejDSn ZSn Sn

In the two-dimensional simulation shown in Fig. 9.18, the contact window

length of eutectic SnAgCu solder joint is taken to be 224 μm and the current
crowding region is taken about 15% of the whole length, so the current crowd-
ing region, b , is estimated to be about 33.6 μm. From Fig. 9.7 the void width,
d, is measured as 2.44 μm. The test temperature is 146◦ C, the electric current
density is about 3.67 × 103 A/cm2 , the void length is 33 μm, the duration
of void propagation is 6 hr, thus the void growth velocity is about 5 μm/hr.
In the other case of eutectic SnPb solder bump stressed at 2.25 × 104 A/cm2
and 125◦ C as discussed in Section 9.4.1, the window length is 140 μm and the
SVNY339-Tu April 5, 2007 17:31

270 Chapter 9

Table 9.3. Comparison of measured and calculated values of growth rate of pancake-
type void in electromigration

Theory Experiment

b 25.49–44.15 μm 37.5 μm
d 0.81–2.42 μm 2.44 μm
v 1.24–6.44 μm/h 4.4 μm/h

width of current crowding region is about 9 μm. The voids are formed at 38 hr
and failed at 43 hr, therefore, the void growth velocity is about 28 μm/hr.
The diffusivity is taken to be DSn = 1.3 × 10−10 cm2 /sec for Sn, and the
diffusivity of the interface is taken to be 4.2 × 10−5 cm2 /sec. The effective
∗
charge is ZSn = 17 for Sn. The resistivity of Sn is ρSn = 13.25 μΩcm. The sur-
face energy γ = 1015 eV/cm2 and Ω is taken as 2.0 × 10−23 cm3 . The effective
interfacial width is about 0.5 nm. The only unknown parameter is the ratio
of ΔC and C0. In order to obtain reasonable results, we choose the range of
ΔC/C0 from 1% to 3%.
Using these parameters and experimental conditions, the theoretical val-
ues of current crowding length b have been calculated from Eq. (9.18), void
width d from Eq. (9.17), and void growth velocity v from Eq. (9.15). The com-
parison between theoretical values and experimental results are in reasonable
agreement as listed in Table 9.3.

9.4.7 Time-Dependent Melting of Flip Chip Solder Joints

A rather common cause of failure of flip chip solder joint is melting. For certain
eutectic SnPb flip chip solder joints, when the applied current density is above
1.6 × 104 A/cm2 and the test temperature is around 100◦ C, melting occurs.
For certain eutectic SnAgCu flip chip solder joints, when the applied current
density is above 5 × 104 A/cm2 and the test temperature is around 100◦ C,
melting occurs. These current densities have become the upper limit that can
be applied to flip chip solder joints. In melting, the temperature should reach
the melting point of the solder alloy, for example, 183 or 220◦ C for eutectic
SnPb and eutectic SnAgCu, respectively. This indicates that joule heating
must have raised the temperature from 100◦ C to the melting point. However,
the question is where is the joule heating coming from? Whether it is from
the solder bump itself or from the interconnect above the bump is unclear.
Melting in principle is a time-independent event. Generally speaking, no
superheating is needed in melting, so melting should occur instantaneously
when the temperature reaches the melting point as in reflow. However, the
observed melting of flip chip solder joints induced by electromigration takes
time. Typically, under an applied current density, it takes a while, from several
hours to a couple of days, for the joule heating to induce melting.
In Section 9.3, we reported the use of aluminum resistors to measure
the chip temperature and a very large joule heating was found even with an
SVNY339-Tu April 5, 2007 17:31

Electromigration in Flip Chip Solder Joints 271

applied current density of about 3.5 × 103 A/cm2 . The chip temperature was
measured to be as high as 175◦ C, indicating an increase of temperature of
about 125◦ C due to joule heating. When a current density of 5 × 104 A/cm2
is applied to a flip chip eutectic SnAgCu solder joint, current crowding will
increase it to 5 × 105 A/cm2 . For a flip chip with thin-film UBM, the current
crowding occurs within the solder bump, thus joule heating in the solder
bump is high.
In mainframe computers, the Si chips will be cooled in order to maintain
a device working temperature of 100◦ C, thus the solder bumps will not melt.
However, for most consumer electronic products, there is no cooling or the
cooling design is ineffective, e.g., by using a fan, so the solder bumps can
become very hot and melt.
Why joule heating is so large in a flip chip solder bump and why melting
takes time require explanation. While the unique current crowding in a flip
chip solder bump will cause a large joule heating because of the I 2 R depen-
dence, we must also consider heat dissipation besides heat generation. Since
solder alloy itself is a poor electrical conductor, it is also a poor thermal con-
ductor on the basis of the Wiedemann–Franz law [18]. This law states that for
a metallic conductor, the ratio of thermal conductivity to electrical conduc-
tivity is proportional to temperature. Therefore, it is important to have the
joule heating in the solder bump conducted away. Typically, the heat can be
conducted away from the bump by the Si chip via the UBM. Heat conduction
by Si is important since the on-chip Al or Cu interconnect is the source of
joule heating. Hence, the design of the UBM and interconnect is important in
heat management. Nevertheless, the design is static and independent of time.
On the basis of the design, only when the temperature reaches the melting
point of the solder should the bump melt, otherwise it should not. There is
no factor of time. However, the melting in flip chip solder joints is a dynamic
phenomenon; it is time-dependent [19, 20].
The change of microstructure of the solder joint due to electromigration
or thermomigration is a time-dependent event, and it affects the joule heat-
ing. When a pancake-type void forms and grows across the cathode contact
between the Si chip and the solder bump, there are two significant effects
that can influence heat generation and dissipation. First, as the pancake-type
void grows, the conducting path of the Al interconnect above the void must
increase. This will increase joule heating. When the current density is high,
electromigration can induce damage in the Al as discussed in Section 9.2.4. It
will cause more joule heating. Second, the void becomes a good heat insulator
and prevents the heat from dissipating through the Si chip. Consequently,
both joule heating and heat insulation become serious as the void grows, es-
pecially when the void grows to eclipse most of the contact opening, which
may lead to bump melting. While the Cu trace on the substrate can also con-
duct heat away from the solder bump, it nevertheless may have produced a
temperature gradient in the solder bump and caused thermomigration in the
bump. Thermomigration will be covered in Chapter 12.
SVNY339-Tu April 5, 2007 17:31

272 Chapter 9

9.5 Electromigration in Flip Chip Composite

Solder Joints
9.5.1 Thin-Film Cu UBM in Composite Solder Joints
In Sections 1.3.3 and 4.2, we discussed the trend of application of flip chip sol-
der joints to low-cost consumer products where polymer substrates are used.
Because of the low glass transition temperature of polymer, the solder on the
polymer substrate must have a low melting temperature. Therefore, a com-
posite solder joint was developed, which combines the high-melting 97Pb3Sn
solder on the chip side and the low-melting eutectic 37Pb63Sn solder on the
polymer substrate side. The major advantage of a composite solder joint is
that it yields a joint which is compatible with polymer substrates so that
direct chip attachment to polymer substrates can be accomplished. However,
electromigration is a concern [12, 21].
In Fig. 9.19, a set of cross-sectional SEM images of a pair of compos-
ite solder joints with thin-film under-bump metallization (UBM) are shown.

Fig. 9.19. Cross-sectional SEM images of a pair of composite solder joints with thin-
film UBM subjected to electromigration with a current density of 1.57 × 104 A/cm2
at 150◦ C for (a) 30 min, (b) 1 hr, and (c) 2 hr.
SVNY339-Tu April 5, 2007 17:31

Electromigration in Flip Chip Solder Joints 273

The composite joint was composed of 97Pb3Sn on the chip side and eutectic
37Pb63Sn on the substrate side. The contact opening on the chip side had a
diameter of 90 μm and the bump height was 105 μm. The UBM of trilayer thin
films on the chip side were Al (∼0.3 μm)/Ni(V) (∼0.3 μm) /Cu (∼0.7 μm). On
the substrate side, the bond pad metal layers were Ni (5 μm)/Au (0.05 μm).
Electromigration was conducted with a current density of 1.57 × 104 A/cm2 at
150◦ C for 30 min, 1 hr, and 2 hr, and the corresponding cross-sectional SEM
images of a pair of joints are shown in Fig. 9.19(a), (b), and (c), respectively.
In the pair of joints, electrons came in from the bottom of the right-hand side
joint, went up the solder bump, exited the bump at the upper left corner and
moved to the left and entered the left-hand side joint from the upper right
corner, went down the solder bump, and exited at the bottom. In Fig. 9.19,
the darker region in the solder bump is the eutectic phase and the brighter
region is the high-Pb phase.
The effect of current crowding on phase redistribution is very clear in
Fig. 9.19(a) to (c). We recall that Pb is the dominant diffusing species in
SnPb solder at temperatures around 150◦ C. After 30 min of electromigration
at 150◦ C, the eutectic phase in the right-hand side joint has been displaced
to the lower right corner, and correspondingly, in the left-hand side bump the
eutectic phase displaced to the upper left corner. After 1 hr, the situation
was about the same. After 2 hr, as shown in Fig. 9.20 which is a higher-
magnification image, a pancake-type void was formed at the cathode contact
on the Si side in the left-hand side joint. Interestingly, in Fig. 9.19(c) as well
as in Fig. 9.20, the dark eutectic phase has been displaced sidewise to the left
with the growth of the pancake-type void. The sidewise displacement of the
eutectic phase can be explained by the fact that as the void grows sidewise,
the current crowding moves with the tip of the void, electromigration will

Pancake void

Sn-rich

Pb-rich

Fig. 9.20. Higher magnification of pancake-type void formation in the left-hand side
bump in Fig. 9.19(c).
SVNY339-Tu April 5, 2007 17:31

274 Chapter 9

drive Pb away and back filled by Sn so the eutectic phase moves with the tip
of the void.
Consequently, at the end of the pancake-type void growth, the low-melting-
point eutectic phase is displaced to the upper left corner of the joint. Increased
joule heating due to current crowding, longer conducting path in Al intercon-
nect, and improved heat insulation of the void may lead to partial melting
of the upper left corner of the solder bump at a low temperature. Indeed,
frequently such partial melting in the composite solder joints with thin-film
UBM has been observed.

9.5.2 Thick Cu UBM in Composite Solder Joints

In the composite solder joint to be discussed here, the Cu in the UBM is 5 μm
thick. Figure 4.4 showed the cross-sectional SEM images. The UBM on the
chip side consisted of TiW(0.2 μm)/Cu (0.4 μm)/electroplated Cu (5 μm), and
the bond pad on the substrate side was electroless Ni (5 μm)/Au (0.1 μm). The
97Pb3Sn solder was electroplated onto the sputtered TiW/Cu/electroplated
Cu UBM and followed by a reflow at 380◦ C.
On the polymer substrate, eutectic 37Pb63Sn solder paste was stencil
printed on the electroless Ni/Au finished bonding pads and followed by a
reflow at 220◦ C. The reflowed eutectic solder bumps were flattened by the
caking process prior to assembly, discussed in Chapter 4. The thickness of
the flattened eutectic solder bump is about 40 μm. To assemble the two
solders, a water-soluble flux was coated onto the flattened substrate sol-
der bumps and the reflow had a peak temperature of 220◦ C and a dwell
time of 90 sec in nitrogen atmosphere. After assembly and cleaning the flux
residue, the gap between the chip and the board was filled with epoxy, i.e.
underfill.
Electromigration was conducted with 0.5 A and the average current den-
sity at the 50-μm-diameter contact opening was 2.55 × 104 A/cm2 . Such cur-
rent density should not cause electromigration damage in the 5-μm-thick Cu.
However, the reverse flux of Sn moving from the anode side of eutectic SnPb
to the cathode side of high-Pb transforms the Cu into Cu3 Sn, and in turn
Cu3 Sn to Cu6 Sn5 . Finally, failure occurs when the Cu is consumed, as shown
in Fig. 4.4.
Electromigration of the same composite flip chip solder joints at room
temperature showed similar failure mode but was found to have a strong
dependence on current density [21]. The composite solder joints did not fail
after 1 month stressed at 4.07 × 104 A/cm2 , but failed after 10 hr stressed
at 4.58 × 104 A/cm2 . At just a slightly higher current stressing of 5.00 × 104
A/cm2 , they failed after only 0.6 hr by the melting of the composite solder
bumps. Due to the growth of Cu6 Sn5 at the cathode side, the Cu UBM was
quickly consumed and was followed by void formation at the contact area.
The void reduced the contact area and displaced the electrical path, affecting
the current crowding and joule heating inside the solder bump.
SVNY339-Tu April 5, 2007 17:31

Electromigration in Flip Chip Solder Joints 275

It is known that at room temperature, Sn is the dominant diffusing species

in eutectic SnPb solder, and Sn has been found to move to the anode in elec-
tromigration. Without Sn moving to the cathode, the Cu3 Sn at the cathode
contact should be stable and the device should not fail easily. The samples
tested at room temperature failed only in tests at very high current density,
which tends to suggest that it was due to joule heating and the tempera-
ture in the solder joint was actually above 100◦ C so that Sn moves to the
cathodes.

9.6 Effect of Thickness of Cu UBM on Current

Crowding and Failure Mode
If Cu is part of the UBM, the thickness of the Cu can affect current crowding
greatly. For a thick Cu UBM of 5 μm in the composite solder joint, as discussed
in Section 9.5, the highest current density due to current crowding will occur
within the Cu. More importantly, the thick Cu will enable a redistribution of
current laterally in the entire Cu UBM, so the current density in the solder
bump will be much closer to the average value, i.e., only a slight current
crowding occurs in the solder near the Cu/solder interface. By using three-
dimensional simulations, the effect of current redistribution in the thick Cu
and the solder can be made clear. In the following, we shall consider four
cases of different Cu thickness; from 0.4 μm, 5 μm, 10 μm, and to 50 μm for
comparison. The last case is a Cu column bump.
In the first case of thin Cu, for example, the Cu in the thin-film UBM of
Al/Ni(V)/Cu is about 0.4 μm, the current crowding will occur in the solder
and will affect strongly electromigration of the solder. In the thin UBM, all
of the 0.4 μm Cu was consumed to form IMC. Since the resistivities of the
IMC and Ni(V) are high, very little electric current will be conducted by
them, so very little current will be redistributed to the periphery of UBM
which is the IMC under the dielectric. Typically, electromigration induces
void formation and propagation along the contact interface. Importantly, the
void has extended to the entire periphery of the contact, as shown in Fig. 9.7.
We note that how the void can extend itself to the low current density region
in the periphery of the contact under the dielectric is an interesting question.
Also, the nucleation site of the void is unclear; whether it is at the highest
current density region in the current crowding or in the periphery region of
low current density is unknown.
In the second case of a thick Cu UBM of 5 μm, most current crowding
occurs within the Cu. While part of the current crowding has extended into
the solder but the degree of crowding in the solder is small, the current density
in most of the solder joint is quite uniform and close to the calculated average
value of current density in the bulk of the solder. Besides, the thick Cu in the
periphery region will not be consumed completely by solder reaction during
reflow, so the remaining Cu UBM will conduct electric current and thus a
SVNY339-Tu April 5, 2007 17:31

276 Chapter 9

fraction of the current will be redistributed around the periphery of the con-
tact. The void formation in this case, as shown in the composite solder joint
in Fig. 4.4, is located in the entrance of the current to the bump and the void
will grow bigger with time, but it does not extend to the edge of the periphery.
This mode of failure indicates the important role of conduction by the thick
Cu UBM. Nevertheless, with time, the interaction between electromigration
and chemical reaction will convert Cu into Cu3 Sn and Cu6 Sn5 , and then the
resistance of the contact will increase quickly and lead to failure.
In the third case of a thick Cu UBM of 10 μm on eutectic SnPb solder
bump, current crowding occurs completely within the Cu; there is no current
crowding in the solder, not even in the region near the Cu/solder interface.
The failure is more gradual than the case of 5-μm-thick Cu. First, electromi-
gration dissolves the Cu into the solder and reduces the thickness of the Cu.
When the Cu thickness is reduced to 5 μm, the failure mode repeats itself as
discussed in the above. Figure 9.21 is a set of cross-sectional SEM images of
the sequence of failure of the 10 μm Cu UBM. The applied current was 0.6 A
and the average current density at the 50-μm-diameter contact opening was
3.0 × 104 A/cm2 .
Figure 9.21(a) to (d) are SEM images of the cross-sectioned flip chip sol-
der joints after current stressing for 50, 75, 100, and 120 hr, respectively.
Electrons entered from the upper left-hand side, went down the bumps, and
exited to the lower right-hand side of solder joints as indicated by the white
arrow. After 50 hr of current stressing, as shown in Fig. 9.21(a), the size
of Cu6 Sn5 IMC increased at the whole interface under Cu UBM and the
layer-type dissolution of Cu UBM was observed. After 75 hr as shown in
Fig. 9.21(b), the continuous and uniform decrease of Cu UBM thickness at
the whole interface of Cu UBM/solder increase was clearly observable. After
100 hr as shown in Fig. 9.21(c), accompanying the consumption of Cu UBM
many large Cu6 Sn5 IMCs started to form at the left-hand side corner of the
contact window. After 120 hr as shown in Fig. 9.21(d), there is no more Cu
or Cu6 Sn5 at the left-hand side corner. The final failure was the consump-
tion of almost all Cu UBM at the cathode interface. The failure sequence
of Fig. 9.21(c) to (d) is similar to the 5-μm-thick Cu UBM as shown in
Fig. 4.4.
In the fourth case of a Cu column bump of 50 μm or more in height and
in diameter, the bulk part of the solder bump is replaced by the Cu bump, so
the remaining solder is about 20 μm thick. There is no current crowding at
all in the solder part of the joint.

9.6.1 Cu Column Bumps

Figure 9.22 shows SEM images of the cross-sectioned flip chip joints of Cu
column bumps with eutectic SnPb solder bumps after current stressing for
SVNY339-Tu April 5, 2007 17:31

Electromigration in Flip Chip Solder Joints 277

Fig. 9.21. SEM images of cross-sectioned flip chip eutectic SnPb solder joints having
10-μm-thick Cu UBM after current stressing for (a) 50 hr, (b) 75 hr, (c) 100 hr, and
(d) 120 hr.
SVNY339-Tu April 5, 2007 17:31

278 Chapter 9

(a)
e-

50 μm

(b)
e-

50 μm

(c)
e-

50 μm

Fig. 9.22. SEM images of the cross-sectioned flip chip joints of Cu column bumps
with eutectic SnPb solder bumps after current stressing for 1 month at 100◦ C with
current density of (a) 3.4 × 103 , (b) 4.7 × 103 , and (c) 1 × 104 A/cm2 .

1 month at 100◦ C with current densities of (a) 3.4 × 103 , (b) 4.7 × 103 , and
(c) 1 × 104 A/cm2 [22]. The arrows labeled e− indicate the direction of electron
flow. The flip chip joints did not fail after 1 month of current stressing at these
three current densities. In a simulation of current distribution in a joint at
the initial state of applied current density of 1 × 104 A/cm2 , current crowding
SVNY339-Tu April 5, 2007 17:31

Electromigration in Flip Chip Solder Joints 279

occurs in the upper left corner of Cu column bump and has spread about
5 μm wide and 10 μm deep into the Cu column; the current distribution in
the solder region was uniform.
The effect of current density on the rate of IMC formation is clearly seen
in Fig. 9.22(a) to (c). What is surprising to find in these figures is the ab-
sence of polarity effect of electromigration; the amount of IMC formation at
the Cu/solder interface is almost the same in the pairs of joints, shown in
Fig. 9.22(a) to (c). Electrons went from Cu to IMC in the left-hand side one
but went from IMC to Cu in the right-hand side one, yet the amount of
Cu3 Sn and Cu6 Sn5 formed is almost the same [see Fig. 9.22(c)]. The thick
Cu-Sn IMCs were made up of two different kinds of IMCs. EDX identified the
thin dark layer near the Cu as Cu3 Sn and the thick white layer near the solder
as Cu6 Sn5 . After 1 month with a current density of 1 × 104 A/cm2 at 100◦ C,
the Cu3 Sn grew much thicker and its thickness became comparable to that
of Cu6 Sn5 as shown in Fig. 9.22(c). In addition, an important finding here is
that a large number of voids appeared in the Cu3 Sn layer, with the major-
ity of them being closer to the Cu3 Sn/Cu interface. The thickness of Cu3 Sn
and the number of voids increased significantly with the increase of current
density.
In the present system of a limited amount of Sn combined with an infinite
amount of Cu, Cu3 Sn grew thicker at the expense of Cu6 Sn5 and is accompa-
nied by an extensive formation of Kirkendall voids, as shown in Fig. 9.22(c).
These voids will fail the joint mechanically. When one molecule of Cu6 Sn5 is
concerted into two Cu3 Sn, it releases three Sn atoms which will attract nine
Cu atoms to form Cu3 Sn. The vacancies that are needed for the Cu diffusion
may condense to form voids. Therefore, in the combination of a very thick
Cu column bump and a relatively thin solder bump, the Cu6 Sn5 transforms
to the Cu3 Sn and the latter can grow very thick, so the vacancy flux that
opposes the Cu flux will form Kirkendall voids.
The effect of these Kirkendall voids on electrical and thermal conductivity
of Cu3 Sn, in turn the effect on joule heating and heat dissipation of the solder
joint is, of great concern. It might affect the temperature gradient in the solder
joint too.
Above 100◦ C, it is known that electromigration in a SnPb solder joint
will drive Pb to the anode, so Sn is pushed back to the cathode. This will
enhance the IMC formation at the Cu interface in the left-hand side one, but
not the right-hand side one. Why a similar amount of IMC forms at the Cu
interface in the right-hand side joint requires an answer; a plausible one is
thermomigration, which will be discussed in Chapter 12. Thermomigration
drives Pb to the cold side and Sn to the hot side. A temperature difference of
only 2◦ C across a solder joint 20 μm thick produces a temperature gradient of
1000◦ C/cm, which is sufficient for thermomigration. Thus, electromigration
accompanied by thermomigration and the formation of Kirkendall voids could
replace current crowding as a serious reliability issue regarding use of Cu
column bumps.
SVNY339-Tu April 5, 2007 17:31

280 Chapter 9

9.6.2 The Design of a Near-Ideal Flip Chip Solder Joint

Assuming that we must use solder in chip joint, how can we design a flip chip
joint that will have the best resistance to electromigration and other reliability
concerns? To improve reliability against electromigration, we must reduce
current crowding. This can be achieved by improving the design of the flip chip
configuration and materials. Since the basic principle in current distribution is
that electric current will take the least resistive path, there are options in the
design of a flip chip to reduce current crowding. Using finite element analysis,
current distribution in a flip chip solder joint can be studied as a function
of geometry and resistance of all the conducting elements associated with a
solder joint, including the Al or Cu interconnect, the UBM, and the solder
bump itself. The factors that affect current distribution the most have been
found to be the thickness and resistance of under-bump metallization. The
factor that affects joule heating the most is the Al or Cu interconnect over
the solder bump. The advantage of Cu column bump on current distribution
was presented in the last section. However, the Cu column bump has caused
the growth of thick Cu3 Sn accompanied by Kirkendall void formation. We
need to prevent the growth of Cu3 Sn. This can be achieved by using high-Pb
solder with a limited amount of Sn. Thus, the combination of Cu column and
high-Pb solder is attractive.
Among all solder joints, the high-Pb or C-4 solder joint has the best elec-
tromigration resistance. Although the high-Pb forms Cu3 Sn at the Cu/solder
interface, Cu3 Sn will not grow or transform if there is no Sn in the solder and
thus no Kirkendall void formation. If the thickness of the high-Pb in the joint
is about 10 μm, it is about the critical length below which electromigration
can be balanced by back stress and no electromigration damage will occur.
Hence, if we design a flip chip joint, consisting of Cu column/Cu3 Sn/high-
Pb/Cu3 Sn/Cu trace and if the thickness of the middle layer of high-Pb, which
we note is almost pure Pb after its Sn content has been consumed by the Cu3 Sn
formation, is below the critical length of electromigration, it is most likely the
best solder joint from the point of view of resisting electromigration. In ad-
dition, while the high-Pb may climb up the side wall of the Cu column, the
reflow reaction to form the Cu3 Sn causes no harm or there is no Kirkendall
void formation. The challenge is that ceramic substrate must be used since
the high-Pb has a high reflow temperature.
Because of the high melting point of high-Pb solder, we cannot use it on
polymer-based substrates; we need to use a composite solder, consisting of a
high-Pb solder on the Cu column side and a eutectic SnPb (or eutectic Pb-
free) on the polymer substrate side. But as we have shown in Section 9.5,
electromigration will drive the Sn from the eutectic solder to the cathode
and convert the Cu3 Sn to Cu6 Sn5 , and eventually failure will occur even for
a 10-μm-thick Cu UBM. Nevertheless, for a Cu column the time needed to
consume all the Cu may be too long to be of concern or not all the Cu can be
consumed. Instead, the problem of the column/composite will be the growth
SVNY339-Tu April 5, 2007 17:31

Electromigration in Flip Chip Solder Joints 281

of Cu3 Sn and Kirkendall void formation, especially on the side wall of the
Cu column. Hence, we must prevent the reverse flux of Sn in the composite
solder joint or we need to prevent the transformation of Cu6 Sn5 to Cu3 Sn.
This can be accomplished by using a diffusion barrier between the high-Pb
and the eutectic solder. The diffusion barrier can be a layer of Cu or Ni or a
bilayer of Cu/Ni 5 to 10 μm thick and it can be electroplated on the high-Pb
solder. In using a bilayer of Cu/Ni, the Cu will be on the high-Pb side and the
Ni on the eutectic side. Both the high-Pb solder and eutectic Pb-free solder
react with Cu or Ni, so there is no problem in joining them to the diffusion
barrier. To prevent the transformation of Cu6 Sn5 to Cu3 Sn, it is known that
(Cu, Ni)6 Sn5 is stable and by adding Ni to Cu6 Sn5 has prevented or slowed
down much of the growth of Cu3 Sn.
In the design, we can make the thickness of both the high-Pb and the
eutectic solder below their critical length of electromigration. However, there
is a problem with the thin layer of eutectic Pb-free solder because it will react
with Cu and the reaction can transform the entire Pb-free eutectic solder into
Cu-Sn IMC. While electromigration will be slower in IMC, the mechanical
properties of an IMC joint should be studied. The advantage of Ni or Ni(P)
is that the rate of IMC formation is much slower than Cu.
Concerning the side wall, even if the eutectic solder climbs up and forms
a coating on the high-Pb side wall, there will be no mixing during aging in
the solid state since the current density in the side wall is low.
Finally, we can slow down atomic diffusivity in the solder bump by alloying,
so the Pb-free solder may have an electromigration resistance as good as the
high-Pb solder.

9.7 Electromigration-Induced Phase Separation

in Eutectic Two-Phase Solder Alloy
Below eutectic temperature, the microstructure of a eutectic alloy consists
of two primary phases according to its eutectic phase diagram. Typically,
the two phases form a lamellar microstructure. They are in equilibrium with
each other, hence there is no chemical potential difference between them, and
it is possible to induce redistribution of the two phases without counterac-
tion. For example, there is no well-defined lamellar thickness of each phase
in the lamellar microstructure except the principle of reduction of the total
area of lamellar interfaces. Therefore, it is possible to have a near-complete
phase separation into two parts, one of each of the two primary phases, under
an external driving force. Indeed, this has been shown to occur in electro-
migration in eutectic SnPb solder joints at 150◦ C by Brandenburg and Yeh
[1]. Figure 4.6(a) and (b) show SEM images of a eutectic SnPb solder joint
before and after electromigration. A near-complete phase separation is seen
after electromigration. Also, in electromigration in composite solder joint as
discussed in Section 9.5, electromigration has induced the redistribution or
SVNY339-Tu April 5, 2007 17:31

282 Chapter 9

segregation of Sn and Pb to the cathode and the anode, respectively. The

kinetic analysis of phase segregation in a eutectic two-phase mixture driven
by electromigration will be presented in this section.
In a eutectic two-phase mixture, the change of concentration under a con-
stant temperature does not mean any change of chemical potential since the
two phases are coexisting with each other in equilibrium. When there is seg-
regation in the mixture, it means a change of volume fraction of the two
phases, in other words, it leads to a gradient of volume fraction, but not a
gradient of chemical potential, so we do not have flux governed by Fick’s first
law. Gradient of volume fraction is not a driving force for atomic diffusion,
therefore a redistribution of volume fraction is not counteracted by a chem-
ical force as in uphill diffusion. Thus, the segregation in eutectic two-phase
mixtures can be enormous. What is also unique is that due to the lack of
counteracting force, the segregation is not smoothed by diffusional process. In
the diffusion equation, the time rate change of concentration, dC/dt, equals
the second derivative of concentration by space coordinate, d2 C/dx2 , times
diffusivity. Thus, the second derivative tends to smoothen concentration with
time. Without it, because we do not have Fick’s first law to describe the flux,
the tendency for a stochastic process occurs. We shall show in the following
that this occurs in electromigration of the eutectic mixture. We will not find a
smooth change of concentration, but rather will find random states or random
phase distribution in the two-phase structure.
In Section 9.5, we discussed that upon electromigration in a composite sol-
der joint, while the Pb is being driven to the anode, the Sn diffuses reversely
to the cathode. It is this back-diffusion of Sn that leads to failure at the cath-
ode. Since we should also expect Sn atoms to be driven by electromigration
from the cathode to the anode, the back-diffusion is puzzling. Below, we shall
analyze the kinetics of flux migration in a two-phase structure assuming the
constraint of constant volume. The constraint leads to the back-diffusion.
In the two-phase structure of a eutectic system, the alloy composition is not
restricted at the eutectic point. Rather it can be a two-phase mixture below
the eutectic temperature having any composition between the two primary
phases. The main assumptions are the following. (1) Conservation of volume
and shape of samples (equalizing of volume fluxes) at under-critical regimes,
meaning no void or hillock formation. There are two ways of equalizing volume
fluxes, either by back stress or by Kirkendall lattice shift. (2) The fluxes do
not contain concentration gradient terms as in Fick’s first law. Instead, a flux
in terms of drift velocity, which is a product of mobility and driving force,
is used. Stability issues for the concentration profiles will be analyzed, and
it will be shown that the concentration profile in eutectic structures under
electromigration should demonstrate the stochastic tendency.
Consider a two-phase mixture of almost pure components; therefore,
hereafter the indices 1 and 2 correspond to phases as well as to species.
Since we shall limit ourselves within the “under-critical” regime, where the
shape of the sample remains unchanged, we have the constraint of constant
SVNY339-Tu April 5, 2007 17:31

Electromigration in Flip Chip Solder Joints 283

volume in every part of the sample. This means that in the laboratory
reference frame the sum of volume fluxes of two species should be zero
everywhere:

Ω1 J1 + Ω2 J2 = 0, (9.19)

where J1 , J2 are fluxes of atoms per unit area, and Ω1 , Ω2 are atomic volumes.
For convenience, we introduce p1 , p2 , the local volume fractions of phases:

ΔVi /Ωi ΔVi

Ω i n i = Ωi = = pi , (9.20)
ΔV ΔV
p1 + p2 = 1,

where ni is the number of atoms of either species 1 or 2 per unit volume. In

the coarsened space scale, the unit volume ΔV includes at least a few grains.
The electromigration fluxes of each component are determined by standard
expressions:

n1 D1 n 1 D1
Ω1 J1EM = Ω1 Z1 eE = Ω1 Z1 eρj,
kT kT (9.21)
n2 D2 n 2 D2
Ω2 J2EM = Ω2 Z2 eE = Ω2 Z2 eρj.
kT kT

Once more, the constraint of constant volume in Eq. (9.19) means that
these fluxes should contain additional terms for a convective flow. To satisfy
the constraint of constant volume in Eq. (9.19), we assume that there are
two alternative ways: back stress and lattice shift, or maybe a combination of
these two ways.

9.7.1 Electromigration-Induced Back Stress

in Two-Phase Structure
We apply the back-stress story in electromigration in Al interconnects to elec-
tromigration in a mixture of two coexisting phases. Accumulation of atoms at
the anode and of vacancies at the cathode leads to a stress gradient, changing
the fluxes, so that the total change of flux volume becomes zero. Namely,

 
p1 D1 ∂σ
Ω1 J1 = Z1 eE + Ω1 ,
kT ∂x
  (9.22)
p2 D2 ∂σ
Ω2 J2 = Z2 eE + Ω2 .
kT ∂x
SVNY339-Tu April 5, 2007 17:31

284 Chapter 9

Substituting Eq. (9.22) into the constraint of Eq. (9.19), we obtain an expres-
sion for arising stress gradient:

∂σ p1 D1 Z1 + p2 D2 Z2
= −eE . (9.23)
∂x p1 D1 Ω1 + p2 D2 Ω2

Due to this back-stress influence, the larger flux becomes smaller, and
the smaller flux reverses its sign, so that now they compensate each other.
Substituting Eq. (9.23) into Eqs. (9.22), we obtain
 
eE p1 p2 D1 D2 Z1 Z2
Ω1 J1 = Ω 1 Ω2 · − = −Ω2 J2 (9.24)
kT p1 D1 Ω1 + p2 D2 Ω2 Ω1 Ω2

We note that the above derivation is independent of the finite length of the
sample.
On the other hand, for a given length, Δx, of the sample, we obtain the
critical stress in Eq. (9.23) where there is no electromigration-induced damage
since the two opposing fluxes are equal to each other. Below or above the
critical stress, the fluxes become unequal and reverse themselves. We can also
interpret the above equation by stating that at a given current density, there
exists a critical length at which the two fluxes are equal and opposite to each
other. Then for lengths longer or shorter than the critical length, the fluxes
become unequal and change direction. We note that when the two fluxes are
equal and opposite, there is always electromigration, but no electromigration-
induced damage until complete phase separation occurs.
Equation (9.24) demonstrates the electromigration-driven segregation un-
der back stress. It is seen that in the case of equalizing fluxes by back stress,
the rate of segregation is determined by the slow species: if, say, diffusivity of
species 2 is much less than that of species 1, and if the fraction p1 is not too
small, we have

D1 D2 D1 D2 D2
≈ = .
p1 D1 Ω1 + p2 D2 Ω2 p1 D1 Ω1 + 0 p1 Ω1

Moreover, the sign of segregation is determined not by the difference of diffu-

Ω1 − Ω2 ). For example, the
sivities, but instead by the difference of ratios of ( Z 1 Z2

depletion of Pb at the cathode and corresponding enrichment by Sn at the

cathode does not necessarily mean that Pb is a faster diffusant than Sn. It
means ( Z ZSn
ΩPb > ΩSn ).
Pb

The redistribution of volume fractions is determined by the continuity

equation,

∂p1 ∂n1 ∂(Ω1 J1 )

= Ω1 =− ,
∂t ∂t ∂x
SVNY339-Tu April 5, 2007 17:31

Electromigration in Flip Chip Solder Joints 285

so that
   
∂p1 eE Z1 Z2 ∂ D1 D2
=− Ω 1 Ω2 − · p1 (1 − p1 ) .
∂t kT Ω1 Ω2 ∂x p1 D1 Ω1 + (1 − p1 )D2 Ω2
(9.25)

We shall discuss the stochastic behavior of Eq. (9.25) and the experimentally
observed random states later.

9.7.2 Electromigration-Induced Kirkendall Shift

in Two-Phase Structure
In the alternative case, there is no back stress at all, meaning all possible
stresses are immediately relaxed by lattice shift. In a usual polycrystalline
body with large grains, it is described by dislocation climb and corresponding
construction of extra planes in the region of atom accumulation, and decon-
struction of extra planes in the region of vacancy accumulation. Vacancy is
at equilibrium everywhere. Yet, in a eutectic two-phase mixture, the picture
may not be so simple because of growth and shrinkage of grains of the two
phases. We assume that by some mechanism the Kirkendall lattice shift with
a velocity U is guaranteed, and it equalizes fluxes without any back stress.
Then, in the laboratory reference frame:

p1 D1
Ω1 J1 = Z1 eE + p1 U,
kT (9.26)
p2 D2
Ω2 J2 = Z2 eE + p2 U.
kT

Substituting Eqs. (9.26) into the constraint of Eq. (9.19), we obtain the ve-
locity of Kirkendall shift,

eE
U =− (p1 D1 Z1 + p2 D2 Z2 ) .
kT

Substituting the last equation into the constraint of Eq. (9.26), we obtain the
final equations for fluxes of both species in the laboratory reference frame:

eE
Ω1 J1 = (Z1 D1 − Z2 D2 ) p1 p2 = −Ω2 J2 .
kT

As we can see in this case, the rate of segregation will be determined mainly by
the fast species and the sign of segregation is now determined by the diffusivity
in (Z1 D1 − Z2 D2 ) since the difference between the effective charge numbers is
SVNY339-Tu April 5, 2007 17:31

286 Chapter 9

not large.

∂p1 eE ∂
=− [(Z1 D1 − Z2 D2 ) p1 (1 − p1 )] . (9.27)
∂t kT ∂x

The implicit assumption in Kirkendall lattice shift is that vacancies are in

equilibrium everywhere in the diffusion zone, so there will be no void formation
at the cathode and no hillock formation at the anode. The above equations
should provide a correct description of electromigration in solder prior to void
formation.
A comparison between the model of back stress and Kirkendall shift is
made here. Experimentally when we perform electromigration in V-groove
samples of eutectic SnPb solder at temperatures above 100◦ C, Pb is driven
predominantly to the anode, but near room temperature Sn is driven pre-
dominantly to the anode. At temperatures above 100◦ C, solder has a very
high homologous temperature, therefore thermal activated process will relax
stress fast enough that back stress will not build up. Kirkendall shift model
applies. The selection will be controlled by (Z1 D1 − Z2 D2 ) and Pb has the
faster diffusivity and moves to the anode. However, at a very high current
density and after a long duration of electromigration, a lump of solder forms
at the anode and back stress exists. At room temperature, while diffusion in
solder is still quite fast, it is likely that both Kirkendall shift and back stress
operate together and Sn is being pushed to the anode.

9.7.3 Stochastic Tendency in Electromigration

in Two-Phase Structure
In the above two cases, back stress and Kirkendall shift, the equation for
redistribution of volume fractions is of the type

∂p1 ∂
=− [V (p1 )p1 (1 − p1 )] , (9.28)
∂t ∂x

but with different explicit expressions for the kinetic coefficient V (which has
the units of velocity). It is a first-order nonlinear equation, which is very
different from Fick’s second law or from Fick’s second law with a drift term.
The main effect is that it does not contain the second space derivatives, which
tend to smooth any local fluctuation of the composition profile by diffusion.
If V is a constant, Eq. (9.25) is reduced to the well-known Burger’s equation
with zero viscosity term, having peculiarities of solutions like shocks. This
means that even a smooth waviness of composition profile should evolve into
sharp breaks in the concentration profile.
Since noble and near-noble metals, such as Cu and Ni, are used as UBM
and bond pads in flip chip solder joints, their dissolution into the solder and
the formation of IMC are known to be enhanced by electromigration. More
SVNY339-Tu April 5, 2007 17:31

Electromigration in Flip Chip Solder Joints 287

importantly, the IMC of Cu6 Sn5 and Sn form a eutectic two-phase mixture.
Therefore, electromigration can induce the segregation of a large amount of
the IMC in Sn or in Pb-free solder. Figure 1.16 showed the growth of randomly
distributed Cu6 Sn5 in flip chip solder joint in electromigration. Electromigra-
tion has enhanced the dissolution of Cu from the thick Cu UBM into the
solder to form a large amount of IMC.

References

1. S. Brandenburg and S. Yeh, Proceedings of Surface Mount International

Conference and Exhibition, SMI98, San Jose, CA, Aug. 1998, p. 337–
344.
2. C. Y. Liu, C. Chen, C. N. Liao, and K. N. Tu, “Microstructure–
electromigration correlation in a thin stripe of eutectic SnPb solder
stressed between Cu electrodes,” Appl. Phys. Lett., 75, 58–60 (1999).
3. C. Y. Liu, C. Chen, and K. N. Tu, “Electromigraiton of thin strips of
SnPb solder as a function of composition,” J. Appl. Phys., 88, 5703–5709
(2000).
4. T. Y. Lee, K. N. Tu, S. M. Kuo, and D. R. Frear, “Electromigration
of eutectic SnPb solder interconnects for flip chip technology,” J. Appl.
Phys., 89, 3189–3194 (2001).
5. T. Y. Lee, K. N. Tu, and D. R. Frear, “Electromigration of eutectic SnPb
and SnAgCu flip chip solder bumps and under-bump-metallization,” J.
Appl. Phys., 90, 4502–4508 (2001).
6. K. N. Tu, “Recent advances on electromigration in very-large-scale-
integration of interconnects,” J. Appl. Phys., 94, 5451–5473 (2003).
7. W. D. Callister, Jr., “Materials Science and Engineering: An Introduc-
tion,” 5th ed., Wiley, New York (2000). Chapter 6 (Table 6.1) and Ap-
pendix B (Table B.2)
8. E. C.C. Yeh, W.J. Choi, K.N. Tu, P. Elenius, and H. Balkan, “Current-
crowding-induced electromigration failure in flip chip solder joints,” Appl.
Phys. Lett., 80, 580–582 (2002).
9. W. J. Choi, E. C. C. Yeh, and K. N. Tu, “Mean-time-to-failure study of flip
chip solder joints on Cu/Ni(V)/Al thin film under-bump metallization,”
J. Appl. Phys., 94, 5665–5671 (2003).
10. H. Gan, W. J. Choi, G. Xu, and K. N. Tu, “Electromigration in flip chip
solder joints and solder lines,” JOM, 6, 34–37 (2002).
11. L. Zhang, S. Ou, J. Huang, K. N. Tu, S. Gee, and L. Nguyen, “Effect of
current crowding on void propagation at the interface between intermetal-
lic compound and solder in flip chip solder joints, “ Appl. Phys. Lett., 88,
012106 (2006).
12. J. W. Nah, K. W. Paik, J. O. Suh, and K. N. Tu, “Mechanism of elec-
tromigration induced failure in the 97Pb-3Sn and 37Pb-63Sn composite
solder joints,”J. Appl. Phys., 94, 7560–7566 (2003).
SVNY339-Tu April 5, 2007 17:31

288 Chapter 9

13. Y. C. Hu, Y. L. Lin, C. R. Kao, and K. N. Tu, “Electromigration failure

in flip chip solder joints due to rapid dissolution of Cu,” J. Mater. Res.,
18, 2544–2548 (2003).
14. Y. H. Lin, C. M. Tsai, Y. C. Hu, Y. L. Lin, and C. R. Kao, “Electromi-
gration induced failure in flip chip solder joints,” J. Electron. Mater., 34,
27–33 (2005). (Dissolution of thick Cu UBM)
15. L. Xu, J. Pang, and K. N. Tu, Appl. Phys. Lett., to be published.
16. J. R. Black, Proc. IEEE, 57, 1587 (1969).
17. M. Shatzkes and J. R. Lloyd, J. Appl. Phys., 59, 3890 (1986).
18. N. F. Mott and H. Jones, “The Theory of the Properties of Metals and
Alloys,” Dover, New York, p. 242 (1958). (Wiedemann-Franz law)
19. A. T. Huang, Ph.D. dissertation, UCLA (2006).
20. F. Y. Ouyang, Personal communication.
21. J. W. Nah, J. O. Suh, and K. N. Tu, “Effect of current crowding and joule
heating on electromigration induced failure in flip chip composite solder
joints tested at room temperature,” J. Appl. Phys., 98, 013715 (2005).
22. J. W. Nah, J. O. Suh, K. N. Tu, S. W. Yoon, V. S. Rao, K. Vaidyanathan,
and F. Hua, “Electromigration in flip chip solder joints having a thick Cu
column bump and a shallow solder interconnect,” J. Appl. Phys., 100,
123513 (2006).
SVNY339-Tu April 5, 2007 17:34

10
Polarity Effect of Electromigration
on Solder Reactions

10.1 Introduction
In flip chip solder joints, electromigration-induced failure is dominated by cur-
rent crowding at the cathode area of contact. The high current density in the
current crowding region enhances the dissolution of UBM at the cathode and
transforms Cu to Cu3 Sn and to Cu6 Sn5 and finally leads to void formation and
failure; the interaction between electrical force and chemical force is a key fac-
tor in the failure. To understand the interaction and to avoid the complication
due to current crowding, straight V-groove samples having Cu wires as elec-
trodes were introduced. The V-groove samples have the following advantages.
First, their cross section is of similar dimension as that of a flip chip solder
joint, so they can carry the working current density. Second, they are highly
reproducible. Third, the effect of electromigration-induced mass transport or
damage can be observed directly, without cross-sectioning as in flip chip sam-
ples. Fourth and perhaps most important, there is no current crowding and
the interaction between electrical and chemical forces can thus be studied un-
der a uniform current distribution. The polarity effect of electromigration on
chemical reactions at the cathode and the anode can be analyzed [1] as well
as the effect of electromigration on phase separation in the two-phase eutectic
structure.

10.2 Preparation of V-Groove Samples

V-grooves were etched on (001) silicon wafers along {110} directions by us-
ing lithographic technique and anisotropic etching process. The width of the
V-grooves can be 100 μm and the length can be 1 cm. Then a thin Si diox-
ide of 100 nm was grown by wet oxidation at 1000◦ C, and was followed by a
trilayer of Ti/Cu/Au thin-film deposition of thickness 50 nm, 1 μm, and 50
nm, respectively. The silicon wafer was diced into rectangular pieces about
0.25 cm wide and 1 cm long; each contained a V-groove in the middle. Two
SVNY339-Tu April 5, 2007 17:34

290 Chapter 10

Si V-groove

Si (001)

(a)
Cu wire Cu wire

<110>

(b)
Cu wire
Si Wafer

Solder line

(c) 100 μm

Solder
69.4 μm
SiO2 (0.1 μm) 69.4 ∝m

Cr or Ti (0.05 μm)

Cu (1Cuμ(1
m) ∝

Au (0.05 μm)

Fig. 10.1. (a) V-groove sample with two Cu wires before joining. (b) Top view, and
(c) cross-sectional view of a V-groove sample.

Cu wires were placed at the ends of a V-groove, and the assembly was kept
in resin mild active (RMA) flux and placed on a hot plate at 230◦ C and a
solder bead was placed on the V-groove. The bead melted and ran into the
V-groove and joined the Cu wires, which will serve as electrodes. After the
sample was removed from the hot plate, a V-shaped solder line approximately
200 to 800 μm long, 100 μm wide, and around 69 μm deep was obtained.
Figure 10.1(a) illustrates a V-groove sample with two Cu wires as electrodes,
and Fig. 10.1(b) and (c) depict the top view and the cross-sectional view of
a V-groove sample. The length of the solder line can be controlled by the
SVNY339-Tu April 5, 2007 17:34

Polarity Effect of Electromigration on Solder Reactions 291

Solder V-groove sample after reflow

(a)
Si wafer

Cu Solder line Cu

Polished sample ready for EM test

(b) Si wafer
e-
Cu Solder line Cu
Cu Cu

(c)
IMC

Cu Cu

Solder

e- e-
Cathode Anode

Fig. 10.2. SEM images of sample (a) before and (b) after polishing for current stress-
ing. (c) Interfacial intermetallic compounds (IMC) can be seen to form at Cu/solder
interfaces.

spacing between the two Cu electrodes or by using several solder beads of

known diameter as spacers between the Cu wires before reflow. To make the
solder line the only conductive path between the two Cu electrodes, the sur-
face of the soldered V-groove sample was polished until the metal films on
the silicon (001) flat surface disappeared. Figure 10.2(a) and (b) show SEM
images of sample before and after polishing for current stressing. Interfacial
intermetallic compounds (IMC) can be seen to form at Cu/solder interfaces
in Fig. 10.2(c) [1–4].

10.2.1 Electromigration of Eutectic SnPb as a

Function of Temperature
Electromigration in the eutectic SnPb alloy depends on temperature. At
150◦ C, electromigration of eutectic SnPb V-groove samples was stressed at
SVNY339-Tu April 5, 2007 17:34

292 Chapter 10

Fig. 10.3. (a) Electromigration of eutectic SnPb at 150◦ C and 2.8 × 104 A/cm2 . The
growth of a lump at the anode and void at the cathode can be seen. (b) The void
becomes very clear after polishing away a top layer of the sample. (c) Electromigration
of eutectic SnPb at room temperature stressed at current density of 5.7 × 104 A/cm2
for 12 days. SEM image of the growth of hillock at the anode and void at the cathode
is shown.

2.8 × 104 A/cm2 for 8 days, and led to the growth of a hillock at the anode
and a void at the cathode, as shown in Fig. 10.3(a). The void becomes very
clear after polishing away a top layer of the sample, as shown in Fig. 10.3(b).
To analyze compositional change induced by electromigration along the
solder line, a series of EDX spots along the line was measured on the surface.
A very large compositional redistribution or phase segregation was observed
(see Fig. 10. 4). Electromigration has led to the accumulation of Pb at the
anode.
At room temperature, the V-groove solder lines of eutectic SnPb were
stressed at a current density of 5.7 × 104 A/cm2 for 12 days. Figure 10.3(c) is
a SEM image of the growth of hillock at the anode and void at the cathode.
Composition analysis by EDX has shown that the amount of Sn at the anode
side is consistently higher than that at the cathode side and the increase
SVNY339-Tu April 5, 2007 17:34

Polarity Effect of Electromigration on Solder Reactions 293

Fig. 10.4. Electromigration

of eutectic SnPb at 150◦ C and
2.8 × 104 A/cm2 has led to the
accumulation of Pb, instead
of Sn, at the anode. A very
large compositional redistribu-
tion or phase segregation was
observed.

of Sn has occurred along the entire length of the solder line. The average Sn
concentration on the surface appears higher than that in the bulk of the solder
line.
The accumulation of Sn at the anode reveals that at room temperature
Sn rather than Pb is the dominant diffusing species along the direction of
electron flow. The temperature dependence is in agreement with the findings
using tracer diffusion of radioactive Sn and Pb in eutectic SnPb alloy. Above
100◦ C, Pb diffuses faster than Sn, but Sn diffuses faster than Pb near room
temperature [3].
While Pb and Sn have been observed to be the dominant diffusing species
in the eutectic SnPb solder driven by electromigration at 150◦ C and room
temperature, respectively, it is unknown which one will diffuse faster at 100◦ C,
the device working temperature. Using V-groove samples of eutectic SnPb, it
was found that Pb accumulated at the anode, so Pb is the dominant diffusing
species at 100◦ C. Thus, we can perform accelerated tests for eutectic SnPb
solder above 100◦ C since the dominant diffusing species is the same [5].

10.3 Polarity Effect on IMC Growth at the Anode

Solder V-groove samples with a width of 100 μm and a cross section of 3.5 ×
10−5 cm2 were used to investigate the polarity effect of electromigration on
IMC formation. The two Cu wires at the two ends of the V-groove serve as
electrodes. The lead-free solder of SnAg3.8Cu0.7 (in wt%) was reflowed into
the V-groove between the two Cu wires with flux. The reflow temperature is
260◦ C and duration is 1 min. While interfacial IMCs form at the two Cu/solder
interfaces during reflow, we shall focus on the growth of these interfacial IMCs
under the influence of electric current density.
The V-groove samples were placed in a furnace with temperature settings
of 120, 150, and 180◦ C, and stressed with current density in the range of 103
to104 A/cm2 . The sample was polished slightly after a given time of stressing
in order to observe and identify the change of IMC by SEM and EDX. The
thickness of IMC is determined by dividing the area of IMC by the length
SVNY339-Tu April 5, 2007 17:34

294 Chapter 10

of interface. The measurements of area and length are obtained from digital
SEM pictures and imaging processing software. The area is measured in the
number of pixels in the IMC image, and the thickness and length are also in
units of pixels and then converted to micrometers. A standard having a length
of 30 μm as a reference was used to calibrate measured data.

10.3.1 IMC Growth without Electric Current (see Section 2.8)

As shown in Chapter 2 and Fig. 10.2(c), Cu6 Sn5 scallops formed after reflow,
but they grew into layer-type on aging. A very thin layer of Cu3 Sn may also
form between the Cu6 Sn5 scallops and Cu electrode at the initial stage, yet
it is too thin to be recognized in the SEM image. Planar Cu6 Sn5 along with
Cu3 Sn kept growing during solid-state annealing at 180◦ C. After 120 hr, the
thickness of Cu3 Sn layer is about half that of Cu6 Sn5 . At the Cu3 Sn/Cu
interface, Kirkendall voids (the dark dots) formed.

10.3.2 Growth of IMC at Anode and Cathode

with Electric Current
Figure 10.5 shows the thickness change of IMC in four pairs of SEM images at
both the anode and the cathode, after current stressing with current density
of 3.2 × 104 A/cm2 at 180◦ C for 0, 10, 21, and 87 hr. For easy comparison,
the SEM images of the anode and the cathode were placed side by side, the
anode at the left and the cathode at the right, using arrows beside the SEM
images to indicate the thickness of the IMC.
The major difference between the anode and the cathode is the polarity
effect or the direction of electron flow which is away from the cathode but
toward the anode, as indicated in Fig. 10.5. The long arrows outside the images
indicate the direction of electron flow. The same kind of IMC of Cu3 Sn and
Cu6 Sn5 formed at the anode and the cathode with or without application of
electric current. As can be seen in Fig. 10.5(a) and (b), the same scallop-type
Cu6 Sn5 compounds formed at solder/Cu interfaces in the initial stage, but
they transformed into layer-type after current stressing of 10 hr, as shown in
Fig. 10.5(c) and (d).
At the anode, as shown in the left column in Fig.10.5, both Cu3 Sn and
Cu6 Sn5 layers kept growing with current stressing time and the total thickness
approached 10 μm after 87 hr, comparable to that of thermal aging for 200 hr
at the same temperature without applying current. The Cu3 Sn phase between
the Cu6 Sn5 and Cu had darker color and became apparent in the SEM image
in Fig. 10.5(g). The total IMC at the anode was always much thicker than that
at the cathode, as indicated by the short arrows beside the images. Compared
to the case of thermal aging without current, there were fewer Kirkendall voids
formed at the Cu3 Sn/Cu interface at the anode.
At the cathode, it is hard to tell directly how the thickness of IMC changes
from SEM images in the right column in Fig.10.5. The IMC grew much slower
SVNY339-Tu April 5, 2007 17:34

Polarity Effect of Electromigration on Solder Reactions 295

Fig. 10.5. Thickness change of IMC at the anode and cathode, after current stressing
with current density of 3.2 × 104 A/cm2 at 180◦ C for 0, 10, 21, and 87 hr.

than that at the anode. Voids started appearing in the solder part just in front
of the solder/IMC cathode interface after 21 hr and grew bigger after 87 hr,
marked by arrows in Fig. 10.5(f) and (h). After the development of big voids,
IMC seems thicker in Fig. 10.5(h) due to the transformation of Cu6 Sn5 into
Cu3 Sn (darker phase). Void formation brings complexity into the analysis of
change of IMC thickness at the cathode.

10.3.3 IMC Thickness Change with Current

Density and Temperature
The thickness changes of IMC at 180, 150, and 120◦ C, with different cur-
rent densities, were measured and are shown in Fig. 10.6(a), (b), and (c),
SVNY339-Tu April 5, 2007 17:34

296 Chapter 10

(a) 4
(b)
Anode(4x10 A/cm )
2
Anode(4x104A/cm2)
4 2 4 2
Cathode(4x10A/cm ) Cathode(4x10 A/cm )
4 2
Anode(3x10 A/cm ) Anode(3x104A/cm2)
300 4 2 60
Cathode(3x10 A/cm ) Cathode(3x10 A/cm2)
4
4 2
Anode(2x10 A/cm ) Anode(2x104A/cm2)
4 2
Cathode(2x104A/cm2)
(Δ X)2 (μ m2)

(Δ X)2 (μm2)
250 Cathode(2x10 A/cm ) 50
3 2
Anode(4x10 A/cm ) Anode(4x103A/cm2)
3
200
Cathode(4x10 A/cm2) 40 Cathode(4x103A/cm2)
Without Current Without Current
Linear Fit Linear Fit
150 30

100 20

50 10

0 0
0 10 20 30 40 50 60 70 80 90 0 10 20 30 40 50 60 70 80 90 100
Time (hour) Time (hour)

(c)
Anode(4x104A/cm2)
Cathode(4x104A/cm2)
22 Anode(3x104A/cm2)
20 Cathode(3x104A/cm2)
Anode(2x104A/cm2)
(Δ X)2 (μ m2)

18
Cathode(2x104A/cm2)
16 Without current
Linear Fit
14

12
10
8
6
4
2

0 25 50 75 100 125 150

Time (hour)

Fig. 10.6. Measured thickness changes of IMC at (a) 180◦ C, (b) 150◦ C, and (c) 120◦ C,
with different current densities.

respectively. The measured thickness is the total thickness of Cu6 Sn5 and
Cu3 Sn. The figures contain the thickness data for both the anode and the cath-
ode for current stressing with current densities of 4 × 103 , 2 × 104 , 3 × 104 ,
and 4 × 104 A/cm2 . The square of thickness (Δx)2 of IMC is plotted as a
function of time, and the data for the no-current case is included as reference.
The solid symbols designate the data of the anode side, while the hollow sym-
bols of the same shapes represent the corresponding data of the cathode side.
The solid star is for the no-current case, or the case of thermal aging.
There are several common phenomena for IMC growth for all three test
temperatures, as shown in Fig. 10.6. First, the growth of IMC at the anode
has a parabolic dependence on time since the square of thickness (Δx)2 of
IMC increases linearly with time. Second, the thickness data of the anode are
above the no-current curve, while those of the cathode fall below it. Therefore,
IMC grows faster at the anode and slower at the cathode, compared with the
SVNY339-Tu April 5, 2007 17:34

Polarity Effect of Electromigration on Solder Reactions 297

no-current case. In other words, electric current enhances the growth of IMC at
the anode and retards it at the cathode due to the polarity effect. Third, IMC
grows much faster with higher current density at the anode than in the no-
current case. The enhancement of growth at the anode increases with current
density. After the same stressing time, the sample with the largest current
density (4 x 104 A/cm2 ) has the thickest IMC at the anode, and the one with
the lowest current density (4 x 103 A/cm2 ) has the thinnest, just slightly larger
than the no-current case. Finally, the thickness of IMC at the cathode does
not change as much, remaining around the initial thickness (around 2.5 μm)
or below. We shall discuss this behavior under dynamic equilibrium in a later
section. The situation at the cathode side also becomes more complicated due
to void formation in long time stressing.
The growth of IMC at the anode was studied at various temperatures when
current density was held constant at 4 × 104 , 3 × 104 , 2 × 104 , and 0.4 × 104
A/cm2 [4]. The growth at 180◦ C is above that of 150◦ C, while the growth
at 120◦ C is always the lowest, indicating that IMC grows faster at higher
temperature. The activation energy of the growth of total IMC is 1.03 eV,
obtained from the slope of ln(D) versus 1/kT . This is close to the reported
value of 0.94 eV [6].

10.3.4 Comparison among Electrodes of Cu, Ni, and Pd in

V-Groove Samples
In Chapter 7, interfacial reaction between solder and Cu was shown to be faster
than that between solder and Ni. However, the reaction between solder and
Pd is the fastest among the three. When electromigration in V-groove samples
is conducted using Cu, Ni, or Pd electrodes, it is expected that the chemical
force will dominant over the electrical force in the case of Pd electrodes, if the
applied current density is not too high. On the other hand, the electrical force
may dominant over the chemical force in the case of Ni electrodes. Indeed, the
IMC growth at the cathode and at the anode of Pd are fast and parabolic,
and they are much less affected by electromigration. In the case of Ni, a
linear growth was observed at the anode, indicating the dominant effect of
electromigration.

10.4 Polarity Effect on IMC Growth at the Cathode

The polarity effect slows down IMC growth at the cathode. The electric cur-
rent may enhance the diffusion of Cu in IMC since Cu is the dominant diffusing
species in IMC growth. The Cu diffusion is in the same direction as the elec-
tron current, and therefore may enhance the growth of Cu3 Sn and Cu6 Sn5 .
However, the electric current may also enhance the dissolution of Cu6 Sn5 into
the solder. To monitor the dissolution of the IMC at the cathode side, the
samples was aged at 150◦ C for 200 hr to form a thick layer of Cu6 Sn5 before
SVNY339-Tu April 5, 2007 17:34

298 Chapter 10

Previous: Reflow+EM
Cu6Sn5 Cu6Sn5

Cu Sn3.8Ag0.7Cu Cu

Cathode Anode
e-
Present:
Reflow+Aging+EM
Cu3Sn Cu6Sn5 Cu6Sn5 Cu3Sn

Cu Sn3.8Ag0.7Cu Cu
Reaction Dissolution
Cathode Anode
e-

Fig. 10.7. Schematic diagram of a V-groove sample before and after solid-state aging
for the purpose of monitoring the dissolution of IMC at the cathode side.

current stressing. This is shown in the schematic diagram of Fig. 10.7. In ag-
ing, a thin layer of Cu3 Sn compound is also formed between Cu and Cu6 Sn5 .
Since the change of Cu3 Sn layer after electromigration was not as obvious as
the Cu6 Sn5 layer, especially at low current density, we shall focus only on the
change of Cu6 Sn5 layer for a simple analysis.
At 150◦ C, current densities of 2 × 104 , 1 × 104 , and 5 × 103 A/cm2 were
applied to the aged V-groove sample to study the effect of current density
on the dissolution rate of Cu6 Sn5 compound. All the V-groove samples have
the same solder length of 550 μm. As shown in Fig. 10.8 by the triangular
symbols, at the current density of 2 × 104 A/cm2 , the thick layer of Cu6 Sn5
compound becomes very thin in 10 hr. This rapid dissolution corrupts the
cathode side and voids start to form.
When the current density was lowered to 1 × 104 A/cm2 , a uniform dis-
solution of Cu6 Sn5 compound at the cathode was observed. After 158 hr of
electromigration at this current density, a layer of Cu6 Sn5 compound of about
1 μm can still exist at the cathode. No void formation has been found at this
stage.
Further reducing the current density to 5 × 103 A/cm2 , the dissolution
behavior of the IMC became different from that under the higher current
densities; there was a fast dissolution of Cu6 Sn5 in the first 50 hr, but after
that, in some places it started to grow. This growth of the IMC was more
localized and much slower than the dissolution of the IMC. Due to this lo-
calized growth, somehow, the thickness of the IMC stabilized at the cathode
after 240 hr of electromigration. Figure 10.8 is a plot of the thickness change
of the Cu6 Sn5 as a function of time at different current densities.
SVNY339-Tu April 5, 2007 17:34

Polarity Effect of Electromigration on Solder Reactions 299

3
μm)
Δ (μ
ΔX

2
5×103 A/cm2
1×104 A/cm2
1 2×104 A/cm2

0
0 50 100 150 200 250 300
Time (hour)

Fig. 10.8. Plot of thickness of the anode of aged V-groove samples stressed at current
density of 2 × 104 A/cm2 versus time. The thick layer of Cu6 Sn5 compound becomes
very thin in 10 hr. This rapid dissolution corrupts the cathode side. Voids started to
form in some locations. When the current density was lowered to 1 × 104 A/cm2 , a
uniform dissolution of Cu6 Sn5 compound at the cathode was observed. After 158 hr
of electromigration at this current density, a layer of Cu6 Sn5 compound of about 1μm
can still be seen clearly at the cathode. Further reducing the current density to 5 × 103
A/cm2 , there was a fast dissolution of Cu6 Sn5 in the first 50 hr, but after that, in some
places where the IMC layer was very thin as identified by the arrow, it started to grow.

10.4.1 Dynamic Equilibrium

The most interesting curve in Fig. 10.8 occurs for the current density of
5 × 103 A/cm2 , using the square symbols. After 85 hr of electromigration,
a steady state of the IMC thickness is observed. As we have discussed, there
are two driving forces contributing to the change of thickness of the IMC;
they are the chemical force and the electrical force. The flux equation for the
mass transportation can be given as
 
D ∂μ D ∗
J = Jchem + Jem = C − +C Z ejρ, (10.1)
kT ∂x kT

where C is concentration, D is diffusivity, kT has the usual meaning, ∂μ/∂x

is chemical potential gradient, Z ∗ is effective charge number, e is electron
charge, ρ is resistivity, and j is current density.
At the cathode side, the chemical force grows the IMC, whereas,the elec-
trical force dissolves the IMC. These two forces compete during the entire
electromigration process. The solid-state aging study has illustrated that the
SVNY339-Tu April 5, 2007 17:34

300 Chapter 10

layer-type IMC growth driven by chemical potential gradient force alone is

diffusion-controlled. The growth rate satisfies dx/dt ∝ A/x. This means that
when a layer thickness approaches zero (x → 0), the growth velocity dx/dt
goes to infinity. Thus, the layer cannot disappear in the diffusion-controlled
mode. In other words, the chemical potential gradient becomes infinite as
x → 0. When the electrical force is weak, the chemical force may exceed the
electrical force, and the growth of IMC can occur at the cathode side. The
growth of IMC was observed at the cathode when the current density was
only 4 × 103 A/cm2 . When the electrical force is comparable to the chemical
force, a plateau in the IMC thickness curve is found, which indicates a dy-
namic equilibrium between growth and dissolution. We define this state as a
dynamic equilibrium state between chemical and electrical forces [7]. Why is
dynamic equilibrium intersting?
First, a critical product can be obtained from this equilibrium, similar to
that in the electromigration study in Al short strips, where a static equilibrium
can be achieved between the electrical force and back stress. We have the net
J = 0 in Eq. (10.1), so

Δμ ∗
= ZCu6 Sn5
ejρCu6 Sn5 . (10.2)
Δx

By rearranging the equation, we obtain a critical product as

Δμ
(jΔx)critical = ∗ . (10.3)
ZCu6 Sn5
eρCu6 Sn5

Knowing the critical product, we can calculate at a given current density

the thickness of IMC that can reach a dynamic equilibrium. As long as the
IMC thickness is equal to or larger than the critical thickness at the given
current density, there is no void formation at the cathode. Experimentally,
the thickness of the IMC at the equilibrium has been found to be about 2.9
μm at 5 × 103 A/cm2 .
Next, the dissolution rate of the Cu electrode can be obtained, from which
the lifetime of a Cu UBM of a given thickness can be calculated. At dynamic
equilibrium, the thickness of the IMC will not change with time, and the
moving velocities of the Cu/IMC interface and the IMC/Sn interface are equal
to each other. The moving velocity is related to the flux by the equation
J = Cv, and based on the experimental data, the measured dissolution rate
of Cu with current density 5 × 103 A/cm2 at 150◦ C is about 0.076 μm/hr.
In the last two sections, we discussed the polarity effect of electromigration
on IMC formation at the cathode and the anode. What is the effect of electro-
migration on the bulk of the solder in between the two electrodes? We recall
from Section 9.7 that the eutectic two-phase microstructure in the bulk of the
eutectic solder becomes intrinsically unstable driven by electromigration.
SVNY339-Tu April 5, 2007 17:34

Polarity Effect of Electromigration on Solder Reactions 301

10.5 Effect of Electromigration on the Competing

Growth of IMC
Greer and his co-worker published a paper recently about the effect of elec-
tromigration on the competing growth between two IMCs [8]. It is similar to
the competing growth between two-layered IMCs presented in Section 3.2.4.
However, in electromigration, the interaction across a solder joint between the
IMC at the cathode and the IMC at the anode should be considered. As shown
in Sections 10.3 and 10.4, the formation of IMC at the anode is accelerated
and that at the cathode is retarded, due to the interaction. The interaction
occurs because in conducting electromigration in a solder joint, there must be
a cathode and an anode. The solid-state diffusion of Cu and Ni in the solder

(a) (b)

Scan line

(c)

Fig. 10.9. (a) SEM image of a V-groove joint of 10-μm-thick eutectic SnAgCu solder
between two Cu electrodes. (b) The sample was current stressed with a current density
of 2 × 104 A/cm2 at 150◦ C for 144 hr. SEM image shows a trilayer of IMC formation
between the two Cu electrodes as in Cu/Cu3 Sn/Cu6 Sn5 /Cu3 Sn/Cu. (c) The composi-
tion profile shows that the Cu3 Sn at the cathode side grows faster than the Cu3 Sn at
the anode side. (Courtesy of Minyu Yan, UCLA).
SVNY339-Tu April 5, 2007 17:34

302 Chapter 10

is extremely fast, and generally speaking, electromigration in solder occurs at

a high homologous temperature. To address the effect of electromigration on
competing growth of IMC accompanied by the interaction, the dissolution of
IMC into the solder at the cathode side cannot be ignored and the precipi-
tation of IMC at the anode side must be considered. Even if an infinite sink
of solder is used as the anode, while there is no precipitation, the dissolution
will be infinite too.
However, owing to the trend of miniaturization, the size of solder joint
will approach 10 μm in diameter or in thickness, and the entire joint can
become IMC. Therefore, we may have a joint consisting of a trilayer of IMC
between two Cu electrodes as in Cu/Cu3 Sn/Cu6 Sn5 /Cu3 Sn/Cu. Such struc-
ture has been formed in a V-groove sample as shown in Fig. 10.9. A 10-μm-
thick eutectic SnAgCu solder was joined between two Cu wires in a V-groove
and followed by electromigration with a current density of 2 × 104 A/cm2
at 150◦ C for 144 hr. Interfacial reaction between the solder and the two Cu
electrodes has transformed the entire solder into a trilayer of IMC. The SEM
images before and after reaction are shown respectively in Fig. 10.9(a) and
(b), and composition profile after the reaction measured by electron probe
is shown in Fig. 10.9(c). The white line in Fig. 10.9(b) indicates the com-
position scanned by electron probe. Using such kinds of samples, it is pos-
sible to investigate the effect of electromigration on the competing growth
among the trilayer of IMC. The preliminary result shows that the Cu3 Sn at
the cathode side grows faster than the Cu3 Sn at the anode side as shown in
Fig. 10.9(c).

References
1. S.-W. Chen, C.-M. Chen, and W.-C. Liu, “Electric current effects upon the
Sn/Cu and Sn/Ni interfacial reactions,” J. Electron. Mater., 27, 1193–1197
(1998).
2. Q. T. Huynh, C. Y. Liu, C. Chen, and K. N. Tu, “Electromigration in
eutectic PbSn solder lines,” J. Appl. Phys., 89, 4332–4335 (2001).
3. H. Gan, W. J. Choi, G. Xu, and K. N. Tu, “Electromigration in flip chip
solder joints and solder lines,” JOM, 6, 34–37 (2002).
4. H. Gan and K. N. Tu, “Polarity effect of electromigration on kinetics of
intermetallic compound formation in Pb-free solder V-groove samples,” J.
Appl. Phys., 97, 063514–1 to –10 (2005).
5. R. Aquawal, S. Ou, and K. N. Tu, “Electromigration and critical prod-
uct in eutectic SnPb solder lines at 100◦ C,” J. Appl. Phys., 100, 024909
(2006).
6. T. Y. Lee, W. J. Choi, K. N. Tu, J. W. Jang, S. M. Kuo, J. K.
Lin, D. R. Frear, K. Zeng, and J. K. Kivilahti, “Morphology, kinetics,
and thermodynamics of solid state aging of eutectic SnPb and Pb-free
SVNY339-Tu April 5, 2007 17:34

Polarity Effect of Electromigration on Solder Reactions 303

solders (SnAg, SnAgCu, and SnCu) on Cu,” J. Mater. Res., 17, 291–301
(2002).
7. S. Ou, Ph.D. dissertation, UCLA, (2004).
8. H. T. Orchard and A. L. Greer, “Electromigration effects on compound
growth at interfaces,” Appl. Phys. Lett., 86, 231906 (2005).
SVNY339-Tu March 23, 2007 16:20

11
Ductile–to-Brittle Transition of Solder
Joints Affected by Copper–Tin Reaction
and Electromigration

11.1 Introduction
Concerning the mechanical properties of solder joints, there are two unique
characteristics worth mentioning again and again. The first is that solder alloys
have a very high homologous temperature at the device working temperature
or at room temperature. Take the melting point of eutectic SnAgCu at 217◦ C
as an example. Room temperature is 0.6 of the melting point on the abso-
lute temperature scale. Therefore, thermally activated atomic process must
be taken into account when the mechanical properties of a solder joint are
considered, especially at a slow strain rate. The second is that a solder joint
has two interfaces. Its mechanical properties can be very different from those
of a dog-bone-type bulk sample of solder without interfaces. Its mechanical
failure tends to occur near the interfaces because the interfaces can become
more and more brittle with time due to IMC formation and vacancy accumu-
lation from Cu-Sn reaction as well as from electromigration. At a high strain
rate impact, a brittle fracture occurs at the interface.
In this chapter, the effect of Cu-Sn reaction and electromigration on me-
chanical properties of flip chip solder joints will be emphasized. We recall that
chemical reaction can affect stress generation and stress relaxation in spon-
taneous Sn whisker growth as discussed in Chapter 6. Also, electromigration
can induce back stress and the back stress can influence electromigration as
discussed in Chapter 9. These are internal stresses. Here, we discuss the ef-
fect of Cu-Sn reaction and electromigration on mechanical properties of solder
joints when an external force is applied to the joint, especially the force of
impact in dropping by free fall. For portable and hand-held devices, the most
frequent failure is caused by an accidental drop to the ground.
Mechanical properties of solder alloys have been covered in many books.
What is covered in this chapter is the effect of Cu-Sn reaction and electromi-
gration on the mechanical properties of a flip chip solder joint, emphasizing
SVNY339-Tu March 23, 2007 16:20

306 Chapter 11

especially the interfaces of the solder joint. Due to interfacial reaction and
polarity effect of electromigration on the interfaces, a ductile solder joint can
become a brittle solder joint. The ductile-to-brittle transition is very sensitive
to a high-speed shear stress applied to the joint by an impact.
Because solder alloys are ductile by nature, it is of interest to understand
how electromigration and Cu-Sn reaction can affect the mechanical properties
of a solder joint interface and change its ductile nature. We shall examine the
mechanisms in the next section. Owing to the polarity effect of electromigra-
tion, vacancies will accumulate to form voids at the cathode interface of the
joint. Besides electromigration, the ductile-to-brittle transition in solder joints
can also be caused by aging because there will be much more intermetallic
compound (IMC) formation at the joint interfaces due to Cu-Sn reaction,
especially when Kirkendall voids accompany IMC formation. When a brittle
interface encounters a high-speed impact in dropping, fracture failure occurs
easily. Actually, impact stress is catching up with thermal stress to become
another important issue from the point of view of mechanical reliability of
devices because more and more portable consumer electronic products are
expected to fail by dropping.

11.2 Tensile Test Affected by Electromigration

To examine the effect of electromigration on mechanical properties of sol-
der joints, samples having the configuration of Cu (wire)–solder ball (Sn95.5
Ag3.8Cu0.7)–Cu (wire) were designed and prepared, as shown in Fig.11.1.

Si V-groove Si (001) 300 μm

Cu Cu
SiO 2

<110> Si

270 m 300 m

Cu wire with polymer coating Solder ball: 95.5Sn-4Ag-0.5Cu

Fig. 11.1. A Cu(wire)–solder ball(Sn95.5Ag3.8Cu0.7)–Cu(wire) sample was prepared

by etching a V-groove on a silicon chip with a width of 300 μm, and then two Cu
wires as electrodes and two identical solder balls 300 μm in diameter were aligned in
the V-groove.
SVNY339-Tu March 23, 2007 16:20

Ductile–to-Brittle Transition of Solder Joints 307

Fig. 11.2. (a) Optical micro-

graph of the sample. (b) Cross- a b
sectional image after polishing.

~440 μm Cross-section
300 μm

First, a V-groove on a silicon chip with a width of 300 μm was etched. Af-
ter the etching, the Si was oxidized to form an oxide layer on the wall of
the V-groove without depositing the trilayer of metal films as discussed in
Chapter 10; molten solder will therefore not wet the oxidized V-groove. Next,
two Cu wires 300 μm in diameter with polymer coating were placed in the
V-groove, and two identical solder balls 300 μm in diameter were aligned
in between the Cu wires in the V-groove. Then the assembly was heated
to 250◦ C for a few minutes, the solder balls melted, and joined to the two
Cu electrodes by interfacial IMC formation. The polymer coating on the Cu
wire will limit the molten solder to wet only the cross-sectional surface of
the wires. After cooling, the one-dimensional wire sample was removed from
the V-groove by ultrasonic vibration. Figure 11.2(a) is an optical micrograph
of the sample, and Fig. 11.2 (b) is its longitudinal cross-section image after
polishing [1].
The advantage of such one-dimensional samples is that tensile stress can be
applied to study solder joints. Furthermore, using the Cu wires as electrodes,
electric current and tensile stress can be applied serially or simultaneously.
Unlike dog-bone-type bulk test samples, these samples have two interfaces
with IMC formation, which makes them closer to real solder joints in a device.
The one-dimensional Cu–solder–Cu samples were divided into two groups.
The first group was subjected to the tensile test without applying current. The
strain rate was 10−2 /sec. The second group was subjected to electromigration
before the tensile test. The current density was 1.68 × 103 and 5.03 × 103
A/cm2 at 145◦ C for 24 and 48 hr. Then the tensile test was performed with
a strain rate of 10−2 /sec.
Figure 11.3(a) and (b) show stress–strain curves of the tensile test. Fig-
ure 11.3(a) demonstrates the current density effect of electromigration on
tensile strength. The top curve is a pure tensile test. The middle curve is after
electromigration with 1.68 × 103 A/cm2 at 145◦ C for 48 hr. The bottom curve
SVNY339-Tu March 23, 2007 16:20

308 Chapter 11

50
(a) No electromigration
40
Stress MPa 48hours, 145°C,
30
1.68×103 A/cm2
20
48hours, 145°C
10
5×103 A/cm2
0
0.0 0.1 0.2 0.3 0.4 0.5
Strain

50 (b)
No electromigration
40
Stress MPa

24 hours, 145°C,
30
5×103 A/cm2
20
48hours, 145°C
10
5×103 A/cm2
0
0.0 0.1 0.2 0.3 0.4 0.5
Strain

Fig. 11.3. Stress–strain curves before and after electromigration. (a) The upper curve
is pure tensile test. The middle curve is after electromigration with 1.68 × 103 A/cm2
at 145◦ C for 48 hr. The lower curve is after electromigration with 5.03 × 103 A/cm2 at
145◦ C for 48 hr. (b) The upper curve is the tensile test result without current stressing,
and the middle and lower curves are samples stressed with 5 × 103 A/cm2 at 145◦ C
for 24 and 48 hr, respectively.

is after electromigration with 5.03 × 103 A/cm2 at 145◦ C for 48 hr. It illus-
trates the change of tensile strength with electromigration. In Fig. 11.3(b), the
top curve is the tensile test result without current stressing, and the middle
and bottom curves are samples stressed by 5 × 103 A/cm2 at 145◦ C for 24
and 48 hr, respectively.
A longer time or a higher current density of electromigration will cause
more vacancies to move from the anode to the cathode, thus weakening the
interfacial mechanical strength of the cathode by vacancy condensation. To
analyze whether the weaker cathode interface is the reason for decreasing
tensile strength due to electromigration, fracture images were taken after ten-
sile test with and without electromigration, as shown in Fig. 11.4. Without
current stressing, the sample was broken in the bulk of the solder because
the Pb-free solder is much softer than the copper wire, as shown in Fig.
11.4(a). After 96 hr of electromigration at 145◦ C, the sample was broken
near the interface of the cathode even though there was some plastic defor-
mation observed in the solder joint, as shown in Fig. 11.4 (b). After current
stressing for 144 hr with the same current density, the sample was broken
abruptly at the interface of the cathode side while the bulk of the solder joint
SVNY339-Tu March 23, 2007 16:20

Ductile–to-Brittle Transition of Solder Joints 309

(a) (b) (c)

e- e-

No 96 hours 144 hours

electromigration electromigration electromigration
5×103 A/cm2, 145°C 5×103 A/cm2, 145°C

Fig. 11.4. Fracture images taken after tensile test with and without electromigra-
tion. (a) Without electromigration. (b, c) After electromigration for 96 and 144 hr,
respectively.

maintained the original shape, which indicates a brittle fracture, as shown in

Fig. 11.4(c).

11.3 Shear Test Affected by Electromigration

To examine the effect of electromigration on shear behavior of solder joints,

Fig. 11.5 is an optical image of a flip chip bonded to an organic board. The
large white arrow indicates the shear force which pushes the chip and produces
shear to the solder joints between the chip and the board. In the flip chip

No underfil
l
Shear
direction

Fig. 11.5. Optical image of a flip chip sample in

shear test. In the flip chip sample, a daisy chain of A
composite solder balls formed between the silicon B
+ -
chip and organic board.
SVNY339-Tu March 23, 2007 16:20

310 Chapter 11

A B
a

A B
d

Fig. 11.6. SEM fracture images of the second group of sample with electromigration.
(a) Chip side; (c) substrate side. (b, d) Magnified pictures of two pairs of powered
solder joints showing alternate fracture in (a) and (c), respectively.

sample, a daisy chain of composite solder joints formed between the silicon
chip and organic board. The composite consists of a high-Pb solder on the
chip side and eutectic SnPb solder on the substrate side. In shear tests, the
flip chip samples were divided into two groups. The first group was sheared
without electromigration. The strain rate was 0.2 μm/sec. The second group
underwent electromigration with 2.55 × 104 A/cm2 at 155◦ C for 10 hr before
they were sheared with the same strain rate [2].
Figure 11.6 shows top views of SEM fracture images of the second group
of sample. Figure 11.6(a) is from the chip side and Fig. 11.6(c) is from the
substrate side. The letters A and B in these two figures indicate two pairs
of solder joints in the powered daisy chain. Figure 11.6(b) and (d) are mag-
nified images of these two pairs, respectively. They fractured alternately. On
the other hand, the rest of the unpowered joints shown in Figure 11.7(a) and
(c) fractured at the high-Pb side. Figure 11.7 shows side views of SEM frac-
ture images of the second group which was sheared after electromigration.
Figure 11.7(a) is the chip side image and (c) is the substrate side image.
Figure 11.7(b) and (d) are the magnified images of (a) and (c), respectively.
SVNY339-Tu March 23, 2007 16:20

Ductile–to-Brittle Transition of Solder Joints 311

Fig. 11.7. SEM fracture images of the side view of second group of samples which
were sheared after electromigration. (a) Chip side image; (c) substrate side image. (b,
d) Magnified pictures of (a) and (c), respectively. The arrow shows the electron flow
direction, and it alternates downward and upward in the daisy chain of solder joints.

The arrow shows the electron flow direction, and it alternates downward and
upward in the daisy chain of solder joints.
To explain the fracture mode, we recall that a composite solder joint has
the high-Pb region and the eutectic SnPb region. It was found that without
electromigration, solder joints fail in the high-Pb region because Pb is softer
than eutectic SnPb. However, after electromigrtion, the fracture always occurs
at the cathode interfaces in the daisy chain, regardless of whether the cathode
was the high-Pb region or the eutectic SnPb region. The phenomenon of
alternately failure in a daisy chain in shear test, as shown in Figs. 11.6 and
11.7, shows that electromigration weakens the cathodes by void formation at
the cathode interfaces, which is similar to the results of tensile tests.

11.4 Impact Test

11.4.1 Charpy Test
Wireless, hand-held, and movable consumer electronic products are ubiqui-
tous. A frequent cause of failure of these devices is an accidental drop to the
SVNY339-Tu March 23, 2007 16:20

312 Chapter 11

ground. The impact tends to cause interfacial fracture of those wire-bonds

or solder joints between a Si chip and its packaging module, and especially
those ball-grid-array (BGA) solder joints without underfill. While the molding
compound on wire-bonding and the epoxy underfill in flip chip solder joints
are effective in preventing physical separation of the chip from its module
on impact, an interfacial crack induced by the impact is enough to cause an
electrical open failure. Hence, the reliability of solder joints concerning the
combined effect of electromigration-induced damage at the cathode interface
and impact-induced high-speed shear stress in dropping cannot be ignored.
The drop-induced fracture of BGA solder joints is of major concern. This is
because BGA solder balls are much heavier than flip chip solder balls. Without
the protection of underfill, they have a much higher probability of interfacial
fracture failure on dropping when a torque is involved.
At present, the microelectronics industry has the JEDEC specifications on
a free fall drop test, in which a board about 13 cm by 8 cm having an array
of 3 × 5 chip-size packaging substrates on the board is dropped with a drop
table. The board is attached to the drop table with four corners of the board
fixed on standoffs, so the board can have flexional vibration. In the drop test,
the board is positioned horizontally. The impact of the free fall drop will cause
bending and vibration of the board, which will induce fracture at the solder
joint interfaces with the substrate. However, the size of the test board is too
big to be meaningful for hand-held devices, especially the effect of impact
on small packaging boards. No standard drop machine and specifications are
available to test small packaging samples, e.g., a chip-size packaging in which
a piece of 1 cm2 flip chip is packaged on a board of similar dimensions. Further
discussion of the drop test will be given in Section 11.5. Next we discuss the
classic Charpy impact test.
The classic Charpy impact test is a standard test for fracture toughness of
bulk samples of steel, and the typical test sample is a rectangular bar of size
about 1 cm × 1 cm × 5 cm. The test measures the impact toughness of the
sample by measuring the potential energy loss of a pendulum before and after
fracturing the sample. The pendulum of the machine hits the backside of the
sample which has a notch on the front side. Impact toughness is measured by
the energy spent to create the two fracture surfaces in the test sample. In most
body-centered-cubic metals, including steels, a ductile-to-brittle transition
temperature (DBTT) exists. The metal is ductile above DBTT, yet it becomes
brittle below DBTT. The “Titanic” ship may have sunk in cold water after
hitting an iceberg which caused brittle fracture in its hull. The Charpy impact
test was invented to characterize the DBTT in a metal, so that the lower limit
of its application temperature which should be above DBTT is known [3].
The energy spent during the impact is measured by the gravitational
potential energy change of the pendulum before and after the impact. The
potential energy change is measured from the difference between the initial
pendulum height, h1 , and the maximum height achieved during the follow-
through, h2 , or the difference in height before and after the impact, as shown
SVNY339-Tu March 23, 2007 16:20

Ductile–to-Brittle Transition of Solder Joints 313

Fig. 11.8. Schematic diagram of geometry of Charpy impact test. The heights and
angles of the hammer before and after impact are shown.

in Fig. 11.8.

PE = mgh1 − mgh2 = mgΔh, (11.1)

where m is the mass of the pendulum (in grams), g is gravity constant or ac-
celeration of gravity (980 cm/sec2 ), Δh = h1 − h2 = L (cos θ1 − cos θ2 ), where
L is the length of the swing pendulum in centimeters, and θ1 and θ2 are the
angles of the pendulum before and after impact, respectively. The potential
energy has units of Newton-cm. In the Charpy test, the heights or angles are
measured from a readable angle-indicator (a needle) on a scale before and af-
ter the impact. The accuracy of the reading of angles is about 0.5 to 1 degree.
For large bulk samples and very tough materials such as steel bars, the reading
from the scale is good enough.
What we have learned from the Charpy test is that there are three key
factors which contribute to a brittle fracture: (1) low temperature, (2) high-
speed shear, and (3) a geometrical notch. When a structure of ductile material
has all three factors, it tends to show a brittle behavior. If these factors do
not exist, it remains ductile, except materials which are brittle by nature such
as glasses. Most materials become brittle at low temperatures, including pure
Sn which is known to have a sluggish phase change from beta-Sn (metal-
lic, with a body-centered tetragonal lattice) to alpha-Sn (semiconducting,
with a diamond lattice) at 13◦ C. Due to an extremely large molar volume
change, the phase change in pure Sn, known as tin pest, has caused struc-
tural fracture. Nevertheless, for most applications of eutectic solder joints
SVNY339-Tu March 23, 2007 16:20

314 Chapter 11

near room temperature, owing to the intrinsic nature of high homologous

temperature of these alloys, the first factor of low temperature is not a con-
cern. But when the application takes place in very cold weather, it will be
a concern. On the second factor of high-speed shear, this is serious because
the failure due to dropping involves a high-speed shear. Therefore, a standard
shear test by push which is a low-speed shear will not be able to character-
ize the brittle behavior of a solder joint in dropping. To do so, an impact
shear which is as fast as that in a drop is required. The third factor of a
geometrical notch suggests that in a solder joint, if a sharp geometrical cor-
ner exists between the solder bump and its substrate, it serves as a notch
and may contribute to the ductile-to-brittle transition. However, for solder
joints, we must add the fourth factor of IMC and Kirkendall void formation
at the joint interfaces. Void formation can be caused by interdiffusion and
electromigration.

11.4.2 Mini Charpy Machine to Test Solder Joints

To apply the Charpy impact test to small samples, e.g., a single BGA or a
single flip chip solder ball joint in electronic packaging, where the diameter of
a solder ball is from about 760 μm to 100 μm, a mini impact testing machine,
as shown in Fig. 11.9, was built in order to test the bonding nature of one of
these solder balls to their substrate [4–7].
Basically it is a portable mini Charpy testing machine. An electromagnet
was used to release the hammer having an arm of 1 foot. The hammer velocity
released from the height of 1 foot is about 2.44 m/sec at the lowest position.
The velocity is about three orders of magnitude faster than the shear speed in
a typical push test which is about 1 mm/sec. The solder ball sample is placed
on an XYZ positioning stage at the lowest position. The initial position of the
hammer and its final position after impact are recorded by a needle pointer
in an angle recorder on a hemispherical dial as shown in the upper left corner
in Fig. 11.9. On the dial surface, the angle can be read to an accuracy of
0.5 degree. Since the measured angle difference in a typical impact test is
about 10 degrees, the resolution is about 10 to 5% of the measured energy
change.
The mini Charpy impact test machine has been used to study the impact
toughness of solder balls bonded to BGA substrates. The arrangement of sol-
der balls on the BGA substrate is such that there is only one bump in the
path of swing of the hammer. Figure 11.10 shows SEM images of the fractured
surface of the solder joints. Figure 11.10(a) shows a fracture within the bulk
of the solder bump. The fractured surface has the marking of ductile defor-
mation by shear. Figure 11.10(b) shows a brittle fracture along the interface
between the solder bump and the IMC phase. The fractured surface appears
rather smooth. A change from ductile fracture to brittle fracture in a solder
joint as shown in Fig. 11.10 is undesirable. Qualitatively, the cause of the
SVNY339-Tu March 23, 2007 16:20

Ductile–to-Brittle Transition of Solder Joints 315

Electromagnet

Sample
Hammer
Angle Recorder

XYZ - positioning Stage

1 1inch
inch

Fig. 11.9. Photograph of a micro-impact testing machine which was constructed on

the principle of Charpy test to test the bonding of solder balls to their substrate.

ductile-to-brittle transition is due to heat treatment of the solder joint, in

turn the formation of a large amount of IMC as well as voids at the interface
between the solder and the metallization on the BGA board. Quantitatively,
to characterize the transition, we need to use the mini Charpy machine to
perform a systematic study of a large number of solder joints as a function of
temperature, time, solder composition, under-bump metallization, and loca-
tion of the bump on the BGA substrate.
We note that the ductile-to-brittle transition in solder joints is not due
to temperature change as in body-centered cubic metals, rather it is due to
interfacial IMC formation in aging or interfacial void formation in electro-
migration. When a thick IMC formation is accompanied by a large number
of Kirkendall void formation, a brittle behavior is expected. When electro-
migration has led to the accumulation of a large number of vacancies at the
SVNY339-Tu March 23, 2007 16:20

316 Chapter 11

(a) (b)

Fig. 11.10. SEM images of the fractured surface of the solder joints. (a) A fracture
within the solder ball. The fractured surface has the marking of ductile deformation by
shear. (b) A fracture along the interface between the solder ball and the IMC phase.
The fractured surface appears rather smooth.

cathode interface of a joint, it has produced a brittle interface at the cathode

too.
During the high-speed shear in the impact test, the impact energy should
have been distributed between the formation of the two fracture surfaces and
the deformation of the bulk of the solder ball. The softer the solder ball
is, the larger the deformation. The more brittle the joint interface is, the
smaller the deformation. The deformed balls can be collected after they were
knocked off the substrate and examined by SEM for the amount of plastic
deformation in them. For comparison, the energy needed to obtain the same
amount of plastic deformation in a set of free solder balls can be measured
in an Instron machine, and it was found that the average deformation en-
ergy is about 10% the total impact energy measured from the mini Charpy
machine. Thus, most of the impact energy is spent to create the fracture
surfaces.

11.5 Drop Test

11.5.1 JEDEC-JESD22-B111 Standard of Drop Test
We review briefly here the specification of a widely adopted industry standard
of drop test of JEDEC (Joint Electronic Device Engineering Council) standard
JESD22-B111 (board level drop test method of components for hand-held
electronic products), issued in July 2003. A schematic diagram of the test
machine and test sample is shown in Fig. 11.11. The test sample is a printed
circuit (PC) board measuring about 13 cm × 8 cm, which are placed upside
down on four standoffs on a drop table, as depicted in Fig. 11.11. On the board,
there are typically 5 × 3 arrays of 15 components or packaging modules. The
four corners of the board are fixed tightly on the standoffs by screws. The free
fall of the drop table is guided by two guiding rods from a height of 0.82 m to
SVNY339-Tu March 23, 2007 16:20

Ductile–to-Brittle Transition of Solder Joints 317

Guide Rods

PCB Assembly
Stand offs

Base Plate Accelerometer

Drop Table

Fig. 11.11. Schematic diagram of a drop test arrangement.

hit a shock pad, which is typically a stone or a cement block but its surface
can be modified for shock absorption.
When the table hits the shock pad, the PC board will bend and vibrate. An
accelerometer is attached to the table and another one to the board to record
the “input acceleration” and “output acceleration,” respectively, during the
impact when the drop table hits the shock pad. A typical dumping curve of
output acceleration is shown in Fig. 11.12. It shows the drop has reached an
acceleration of 1500g upon hitting the pad, in a period of half of the half-sine
pulse of 0.5 ms when the first dumping of deceleration has occurred. This is
the key specification in the drop test—it must achieve an acceleration of 1500g
in a period of half of the half sine pulse of 0.5 ms. Since force is measured by
the change of momentum in a very short time, 0.5 ms is the period when the
momentum change occurs in the impact, so it is the key parameter in a drop
test and is used to define the force acting on the solder joints [8–10].
In a free fall from a height of h, if we assume the velocity upon reaching
the ground or the shock pad is v, we have from the conservation of energy
that
1 2
mgh = mv ,
2
(11.2)
1
v = (2gh) /2 ,

2
where the gravity factor g = 9.8 m/sec . In a free fall, the velocity v is inde-
pendent of mass. The velocity will be 4 m/sec when the free fall height is
0.816 m.
SVNY339-Tu March 23, 2007 16:20

318 Chapter 11

Fig. 11.12. Dumping of acceleration in dropping test.

Upon hitting the ground, the velocity of the drop table as well as the PC
board will change from v to zero. Assuming the impact is elastic, especially
the impact of the board, it will have a reverse change of velocity from zero
to −v. Let Δt be the transition time of the change of velocity from v to −v,
and −a is deceleration in the event of impact. If we assume that Δt is 0.5 ms
as shown in Fig. 11.11, then we have

Δv 8 m/ sec
−a = = = 16, 000 m/ sec2 ∼
= 1600g, (11.3)
Δt 0.5 × 10−3 sec

where g = 9.8 m/sec2 . Thus, we note that −a is unrelated to the gravity

acceleration but is related to the rate of change of velocity in the impact, and
it can reach a value of 1600g in the drop test if we take into account the reverse
velocity in the change. Experimentally, ΔV can be measured from the area
under the first peak shown in Fig. 11.12, provided that there is no interference
from higher order harmonic vibrations.
Where is the 0.5 ms coming from? It is from the frequency of vibration
of the board attached to the drop table [11–13] The momentum change in
impact will be Δ(mv), where m is the mass of the table. Because of the
coupling between the table and the board, the momentum change will induce
a vibration or bending of the board that carries the packages. Assuming the
fundamental mode of vibration of the rectangular board has a frequency of
f = 1 kHz or 1000 sec−1 , the period of one cycle is 1 ms and a half-sine pulse
will be 0.5 ms. This is what is shown in Fig. 11.12. As we have mentioned, force
SVNY339-Tu March 23, 2007 16:20

Ductile–to-Brittle Transition of Solder Joints 319

is measured by the change of momentum in a very short time, so the time or

the frequency of vibration is the key parameter in a drop test of solder joints.
A low-frequency test, for example, a four-point bending applied mechanically
to a board, will not produce the force encountered in a drop.
We can calculate the frequency by dividing the acoustic velocity in the
board by the length of the board, using Eq. (11.4). The acoustic velocity in
the board is proportional to (Y /ρ)1/2 where Y and ρ are Young’s modulus
and density of the board, respectively, and l is the length.

1 Y
f= . (11.4)
2l ρ

For example, if we take the Young’s modulus and density of polyethylene

2
to be 0.011 × 1011 dyn/cm and 1 g/cm3 , respectively, and take the length of
a beam to be 0.41 m, we find that the frequency is about 1 kHz [14]. For a
quick estimate, we know that the velocity of sound in air is 5 × 104 cm/sec.
The time for sound to travel a distance of 0.5 m in air is 1 × 10−3 sec, so the
frequency, which is the inverse of time, is 1 kHz. Since sound travels faster in
solids than in air, the estimate shows that the discussion and calculation in
the above are correct.
Besides the vibration frequency, the amplitude of the vibration or bending
of the board is crucial in deforming the solder joints. The amplitude is de-
termined by the mechanical coupling between the drop table and the board.
Assuming a beam model or a plate model, the bending can be simulated
[9, 10].
In the drop test, the shock pad is typically a 100-kg stone and its surface
can be modified by covering with cloth to absorb the impact as a soft ground.
The oscillation of the board upon impact is measured by an accelerometer
of model No. 8704B5000, Kistler. If an acceleration sensor is placed on the
drop table, the input acceleration can be measured. This input acceleration
has a strong dependence on the hardness of the surface of the shock pad.
If we place a cloth over it, the input acceleration will be greatly reduced.
The output acceleration, measured on the board, can reach 2000g when the
drop height is 1.3 m (about 4 ft), which is the height of the pocket on a
shirt, and the velocity upon hitting the ground will be about 5 m/sec. The
output acceleration reduces to 1500g when the drop height is 0.82 m (close
to 3 ft) which is the height of a table. The vibration-induced failure of the
solder joint can be measured by a specific resistance change of the solder joint
interconnection beyond a minimum resistance value, for example, beyond 1000
ohms.
The PC board used in the standard drop test is much larger than most
modules or substrates used in hand-held-size electron devices, yet the large
board is needed for vibration. While the horizontal arrangement of the test
board enables us to measure the effect of vibration induced by impact, it
does not measure the effect of torque on solder joints when the board hits
SVNY339-Tu March 23, 2007 16:20

320 Chapter 11

Fig. 11.13. Schematic diagram of free fall

of a solder ball and its substrate board. The
board falls vertically so that its edge hits the
ground.

the pad vertically. We need to build a new drop test machine to do so, yet
the new machine must have the same characteristics of the standard drop test
discussed above, such as 0.5 ms and 1500g. Also, we should be able to measure
the torque on BGA solder joints by the drop test.

11.5.2 Dropping of a Packaging Board Vertically and

the Torque on Solder Balls
Figure 11.13 depicts the vertical free fall of a solder ball joined to a substrate
board. We assume the board falls vertically so that its edge will hit the ground
with a velocity of v. Upon hitting the ground, the velocity will change from
v to zero and to −v for both the board and the solder ball. The momentum
change of the solder ball will be Δ(mv), where m is the mass of the solder
ball. The change will induce a shear force F and a torque Q acting on the ball
and they tend to fracture the interface between the ball and the board. The
force can be calculated from the following relation,

F Δt = Δ(mv),
(11.5)
F = −m Δv
Δt = −ma,

where Δt is the transition time of the change of velocity from v to 0 and to −v

or the time of change of momentum. A short time will produce a large force
and a large torque. Since we can calculate (or measure) the free fall velocity
v and also know the mass m, the challenge in the vertical drop test is how to
measure the transition time Δt or the deceleration −a accurately because it
affects directly the magnitude of the force and the torque. Thus, in designing
a drop test, the interpretation of Δt is the most critical task.
SVNY339-Tu March 23, 2007 16:20

Ductile–to-Brittle Transition of Solder Joints 321

Fig. 11.14. In the vertical drop test, Substrate

the torque will cause a stress dis-
tribution at the interface between
the solder ball and the bond pad.
σmax
A schematic diagram of the stress
distribution is depicted, with tensile
stress at the top and compressive
stress at the bottom. Solder
Bump
r
F

Knowing the force, we obtain the torque Q = F × r, where r is the short-

est distance between the gravity center of the solder ball and the interface.
The force F tends to shear the interface between the ball and the board, and
the torque Q will exert a normal force on the interface. But the distribution
of the normal force due to the torque is such that it is tension at the up-
per end but compression at the bottom end of the interface, as depicted in
Fig. 11.14.
To analyze the stress distribution, we consider in Fig. 11.14 a schematic
diagram of the cross section of a solder ball attached to a vertical board. We
assume that the contact area between the solder ball and the substrate is
rectangular with a width of w and length of 2R. A linear variation of stress
distribution from tension in the upper half to compression in the lower half
of the contact is assumed. If the maximum stress at the two ends is taken to
be ±σmax , the following relationship is obtained from geometrical proportion.

z
σ= σmax , (11.6)
R

where σ is the normal stress at a distance z from the middle origin where the
normal stress is zero. The total moment produced by the stress distribution
should equal the torque. A very simple analysis gives the moment as

 R  R  R
z σmax 2
w zσdz = w z σmax dz = w z 2 dz = wσmax R2 . (11.7)
−R −R R R −R 3

In the last equation, σ(wdz ) is the force acting on a thin strip of (wdz) at a
SVNY339-Tu March 23, 2007 16:20

322 Chapter 11

distance of z from the origin, so zσ(wdz) is the moment. The total moment
should equal the torque, F × r, so we obtain

3 Fr
σmax = . (11.8)
2 wR2

The last equation enables us to calculate the σmax when the torque from
the impact has been measured. As discussed above, the σmax depends on the
torque, the torque depends on the force, and the force depends on the Δt.
From the magnitude of σmax , we may be able to determine whether the drop
will lead to crack formation at the interface or not. Besides the normal force
due to the torque, the interface also experiences a shear force. Therefore, at
the upper corner and lower corner of the joint, the normal and shear forces
must be combined together when considering crack initiation and propagation.

11.6 Converting a Mini Charpy Impact Machine to

Perform Drop Test
We shall use the mini Charpy impact machine to conduct the drop test and
measure the Δt for both horizontal and vertical drops. To do so, the major
change is to transfer the flexional vibration (frequency and amplitude) of the
large printed circuit board in the standard drop test to the flexional vibration
of a beam or the arm of the Charpy impact machine. The vibration of a hinged
beam will produce the acceleration required in both horizontal and vertical
drops. Therefore, the arm shall be made by using a material similar to the
polymer-based printed circuit board so that the Young’s modulus and density
are kept nearly the same, in turn the fundamental vibration mode is similar
to that in the standard test. A single flip chip and its packaging module can
be attached to the hammer at the end of the swing arm. One end of the arm is
hinged and the other end with the hammer and sample is free to swing. When
the hammer hits a fixed wall at the lowest position and the arm vibrates, it
produces the same effect on the attached sample as in the standard test. To
differentiate the horizontal drop and the vertical drop in order to measure the
torque, the chip-size package will be attached to the hammer in such a manner
that its surface is parallel to the swing direction of the hammer (see Section
11.6.2).
To achieve the free fall from a height of 0.82 m (close to 3 ft, which is the
typical height of a household table), we can build a table-top machine with
an arm of 0.41 m and swing the arm from a vertical position of 0.82 m height
as depicted in Fig. 11.15. At the end of the arm, we can add a hammer and
an acceleration sensor. A rigid wall is built at the lowest position. When the
hammer hits the wall, the arm vibrates. Also, we can design the mass of the
hammer to change the amplitude of the vibration of the arm.
An alternative way to conduct the test is to replace the rigid wall by
another arm that has an identical arrangement, i.e., the arm with a hammer
SVNY339-Tu March 23, 2007 16:20

Ductile–to-Brittle Transition of Solder Joints 323

Fig. 11.15. A table-top drop machine

with an arm of 0.41 m. The arm swings
from a vertical position of 0.82 m height. Start position
In the test, the test samples are attached
to the arm. At the end of the arm, there h = 2L
is a hammer to hit the fixed wall to in-
duce the vibration.
Rotation center

wall

M, hammer

and an acceleration sensor. In other words, there are two arms in the test; one
is the swing arm and the other is free hanging, and both of them are hinged.
We let the former hit the latter so that the momentum change is conserved in
the impact event. This is depicted in Fig. 11.16. Furthermore, the free hanging
arm can be hinged as the swing arm, but it can also be fixed at the upper end.
The advantage of the arm with one end fixed is that we can build a detective
circuit in the sample for in situ measurement of the failure induced by the
impact.
Thus, the impact test is the reverse process of a free fall drop. In the swing
arm, the velocity of the substrate and ball changes from v to 0, and in the free
hanging arm after the impact, the velocity of its substrate and balls changes
from 0 to v. While the velocity change is in the vertical direction in the free
fall, it becomes horizontal in the impact test. The conservation of momentum
ensures that the velocity change, Δv, is the same as in the free fall. To measure
the Δt, we can place acceleration sensors on the swing harmmer and the free
hanging harmmer.

11.6.1 Dropping of a Chip Size Package Horizontally

in Mini Charpy Machine
Figures 11.15 and 11.16 depict the arrangement of a flip chip and its sub-
strate glued near the end of an arm. The normal of the chip surface and its
substrate are parallel to the swing direction of the arm. At the end of the
arm, there is a weight or hammer. When the arm and the weight swing down
SVNY339-Tu March 23, 2007 16:20

324 Chapter 11

Free hang

Swing arm

Test
sample

Hammer

Fig. 11.16. To replace the fixed wall, the schematic diagram here depicts two arms
in the test; one is the swing arm and the other is free hanging, and both arms have
the hinged end. We let the former hit the latter so that the momentum change is
conserved in the impact event. In the test, the test samples are attached to the arms
in symmetrical positions.

together and hit the fixed wall, the arm bends and vibrates. No doubt the
weight will modify the frequency and amplitude of vibration of the arm. The
vibration will exert both normal and shear forces to the solder joints. This
sample arrangement is similar to the JESD22-B111 standard of drop test of a
horizontal board discussed in Section 11.5.1, because the direction of impact
is parallel to the normal of the chip and the substrate. Here we can test a
smaller sample with BGA solder joints on a chip-size packaging and also can
test flip chip solder joints between a single Si chip and its module. If we can
join the chip to the arm directly using solder bumps, the test will be more ef-
fective. Also, we can place the chip in the middle of the arm instead of near the
end of the arm. The amplitude of vibration in the middle of the arm is quite
large too.
Alternatively, we can remove the rigid wall and have two arms; one will
swing and the other one will be free hanging. Both of them will carry the same
load; a sample, a hammer, and an acceleration sensor. The swing arm will hit
the free hanging one. We can obtain cyclic impact with them. Actually, we
can study failure in the two samples on the two arms in the impact.
The simple tests described above will enable us to analyze the materials
behavior of solder joints in dropping and to correlate the behavior to pro-
cessing conditions. The effect of Cu-Sn reaction and the polarity effect of
SVNY339-Tu March 23, 2007 16:20

Ductile–to-Brittle Transition of Solder Joints 325

electromigration on the fracture behavior of solder joints in dropping can

be studied quantitatively and systematically. On the other hand, in elec-
tronic consumer products, the packaging design, structure, and materials, e.g.,
systems-in-packaging, may also affect its impact behavior in dropping.

11.6.2 Dropping of a Chip Size Package Vertically

in Mini Charpy Machine
If we rotate the flip chip by 90◦ from the arrangement shown in Fig. 11.16,
and glue it to the beam so that the surface of the chip is parallel to the swing
direction of the hammer, we can study the effect of torque on the solder ball
when the hammer hits a rigid wall. For this purpose, we can also attach the
sample to the side surface or bottom surface of the hammer. Similar to the
discussion presented in the last section, we can remove the rigid wall and
employ a pair of the same arms and let them hit each other. Again, the free
hanging arm can have a hinged or fixed end. The weight of the hammers can
be changed to adjust the momentum change in the impact.
To test an advanced packaging in which a stack of chips are wire-bonded to
a substrate and buried within molding compound, and the substrate is joined
by an array of ball-grid-array of solder balls to a packaging board, the mass
of the chips, the wires, the substrate, and the molding compound will add
to the force and/or torque on the interfaces between the solder balls and the
packaging board in a vertical and/or horizontal drop test of the packaging.
It is a much more complicated engineering problem. But the principle and
measurement methods presented above can be used to analyze the impact of
the drop step by step.

11.7 Creep and Electromigration

Creep is a long time event of deformation or a time-dependent deformation.
The basic process is atomic diffusion driven by a stress gradient. An example
of creep is the sagging of lead (Pb) pipes by their own weight on the wall
of some very old houses in Europe. Room temperature is at 0.5 of the ho-
mologous temperature for Pb, which melts at 327◦ C, and atomic diffusion is
thus sufficiently fast for creep to occur, even though the stress gradient due
to gravity force is quite low. Since eutectic SnPb or eutectic Pb-free solder
melts at a much lower temperature than Pb, creep is expected to occur at
room temperature or the device working temperature of 100◦ C.
Electromigration is also a long time event of diffusion driven by electron
wind force. We expect these two long time events to interact with each other.
No doubt, the basic interaction between electrical force and mechanical force
on atomic diffusion is similar to what we have discussed in Section 8.5 between
electromigration and back stress. Here the stress is external. Since a solder
joint has two interfaces, the polarity effect of electromigration on creep will
SVNY339-Tu March 23, 2007 16:20

326 Chapter 11

be different between the cathode and the anode interfaces. When electromi-
gration drives vacancies to the cathode interface, the creep rate there will be
enhanced. On the other hand, because vacancies will be driven away from the
anode, the creep will be reduced there. In other words, the back stress induced
by electromigration will add tensile stress to enhance the creep at the cathode
side, but it will add compressive stress to reduce the creep at the anode side.

References

1. F. Ren, J. W. Nah, K. N. Tu, B. S. Xiong, L. H. Xu, and J. Pang, “Elec-

tromigration induced ductile-to-brittle transition in Pb-free solder joints,”
Appl. Phys. Lett., 89 141914 (2006).
2. J.-W. Nah, F. Ren, K.-W. Paik, and K. N. Tu, “Effect of electromigration
on mechanical shear behavior of flip chip solder joints,” J. Mater. Res.,
21, 698–702 (2006).
3. J. M. Holt (Ed.), “Charpy Impact Test: Factors and Variables,” ASTM
STP 1072, Philadelphia, PA (1990).
4. T. Shoji, K. Yamamoto, R. Kajiwara, T. Morita, K. Sato, and M. Date,
Proceedings of the 16th JIEP Annual Meeting, 2002, p. 97.
5. M. Date, T. Shoji, M. Fujiyoshi, K. Sato, and K. N. Tu, “Ductile-to-brittle
transition in Sn-Zn solder joints measured by impact test,” Scr. Mater.,
51, 641(2004).
6. M. Date, T. Shoji, M. Fujiyoshi, K. Sato, and K. N. Tu, “Impact reliability
of solder joints,” Proceedings of the 54th ECTC, Las Vegas, NV, June
2004, pp. 668–674.
7. S. Ou, Y. Xu, K. N. Tu, M. O. Alam, and Y. C. Chan, “A study of
impact reliability of lead-free BGA balls on Au/electrolytic Ni/Cu bond
pad,” MRS 2005 Spring Meeting Proceedings, Symposium B, 10.5.
8. M. Alajok, L. Nguyen, and J. Kivilakti, “Drop test reliability of wafer level
chip scale packages,” IEEE 2005, Electronic Components and Technology
Conference, pp. 637–643.
9. E. H. Wong, Y.-W. Mai, and S. K. W. Seah, “Board level drop impact—
fundamental and parametric analysis,” Trans. ASME, 127, 496–502
(2005).
10. E. H. Wong and Y.-W. Mai, “New insight into board level drop impact,”
Microelectron. Reliab., 46, 930–938 (2006).
11. M. Roseau, “Vibrations in Mechanical System,” pp. 111–136, Springer-
Verlag, Berlin 1987.
12. L. Meirovitch, “Fundamentals of Vibrations,” pp. 14–39, McGraw–Hill
Higher Education, New York (2001).
13. W. Goldsmith, “Impact: The Theory and Physical Behavior of Colliding
Solids,” Edward Arnold Publishers, London (1960).
14. J. J. Tuma, “Handbook of Physical Calculations,” 2nd Enlarged & Re-
vised Ed., pp. 311–312, McGraw–Hill, New York (1983).
SVNY339-Tu April 5, 2007 17:37

12
Thermomigration

12.1 Introduction

When an inhomogeneous binary solid solution or alloy is annealed at con-

stant temperature and constant pressure, it will become homogeneous. On
the contrary, when a homogeneous binary alloy is annealed at atmospheric
pressure but under a temperature gradient, i.e., one end of it is hotter than
the other end, the opposite will happen and the alloy will become inho-
mogeneous. This de-alloying phenomenon is called the Soret effect [1, 2].
It is due to thermomigration or mass migration driven by a temperature
gradient. Since the inhomogeneous alloy has higher free energy than the
homogeneous alloy, thermomigration is an energetic process which trans-
forms a phase from a low-energy state to a high-energy state. It is unlike
a conventional phase transformation which occurs by lowering Gibbs free
energy.
In thermodynamics, under homogeneous external conditions defined by
a constant temperature and constant pressure (for example, if T is fixed
at 100◦ C and p is fixed at atmospheric pressure), a thermodynamic system
will minimize its Gibbs free energy, and it will move toward the equilibrium
condition at the given T and p. Both enthalpy and entropy are state func-
tions, so the equilibrium state is defined when T and p are given. On the
other hand, if the external conditions are inhomogeneous, for example, dif-
ferent temperatures at the two ends of a sample in thermomigration, the
equilibrium state of minimum Gibbs free energy is unattainable. Instead, if
the deviation from homogeneity is small, the minimum entropy production
should be reached, and the system will move toward a steady state, instead
of equilibrium. This is the Prigogine principle of irreversible thermodynam-
ics. It means that the entropy production inside the inhomogeneous system is
caused by the flux of heat due to the external inhomogeneity. Consider two
chambers 1 and 2, at different temperatures (T1 > T2 ) connected by a con-
ductive partition. If a quantity of heat δQ flows from 1 to 2, the net change in
SVNY339-Tu April 5, 2007 17:37

328 Chapter 12

entropy is
   
1 1 T1 − T2
dSnet = dS1 + dS2 = δQ − + = δQ . (12.1)
T1 T2 T1 T2

This is the entropy generated by the process of heat flow.

In thermomigration, a steady state of a linear concentration gradient is
obtained by the Soret effect in Fe-C system, to be discussed in Section 12.2.3.
Besides inhomogeneous temperature, we can also have inhomogeneous pres-
sure as in creep. Creep occurs under a stress gradient, hence it is an irreversible
process too.
Thermomigration should occur in a pure metal, as in electromigration
in Al. One would expect that an Al kitchen utensil, such as a cooking pot,
should expand in size after years of use. This is because if the outside of
the pot is hotter than the inside in cooking, Al atoms would have diffused
from the outside to the inside and the latter should have expanded. Yet, this
does not seem to happen. One reason is that the lattice diffusion in Al occurs
by vacancy mechanism. The outside of the pot which is hotter will have a
higher concentration of vacancies than the inside. The vacancy concentration
gradient induces a counter atomic flux which might have compensated nearly
all of the flux of Al atoms driven by the temperature gradient. The net change
may be too small to be noticed. Another reason is due to back stress. The
temperature inside the pod is 100◦ C, which is too low for creep to take place.
As thermomigration drives more and more Al atoms into the cold side and
builds up a high compressive stress there, the stress gradient will produce
an atomic flux of Al against the thermomigration. The equilibrium vacancy
concentration is affected by the stress.
Solder is typically a binary system, so the Soret effect can be found. Ac-
tually, the Soret effect has been reported to occur in PbIn alloy which forms
a solid solution over a wide range of concentration [3, 4]. On the other hand,
eutectic solder has a two-phase microstructure, and the effect of thermomigra-
tion in a eutectic two-phase structure is different from that in a solid solution,
to be discussed in later sections.
Thermomigration in solder joints has been a harder subject to study than
electromigration for two reasons. First, it is difficult to apply a tempera-
ture gradient across a small flip chip solder joint. For a solder joint 100 μm
in diameter, if we can apply a temperature difference of 10◦ C across it, we
have a temperature gradient of 1000◦ C/cm, which is sufficient to induce ther-
momigration in the solder, to be discussed in Section 12.3 [5–7]. Therefore, a
temperature difference of 10◦ C or even a few degrees centigrade in a solder
joint is of our concern. Second, the heat dissipation is hard to control because
of the two interfaces in a joint. Therefore, it is difficult to simulate temper-
ature distribution or temperature gradient in a solder joint because of the
complicated boundary conditions of UBM and bond-pad structure. We have
to simplify the test structure of thermomigration. On the other hand, solder
SVNY339-Tu April 5, 2007 17:37

Thermomigration 329

has a low melting point, so we can use the melting of the solder as an internal
calibration. The condition of heat generation and dissipation in the melting
experiment can be used to check the simulation.
Due to joule heating, electromigration may have caused a nonuniform tem-
perature distribution in a flip chip solder joint, and thus there may be a compo-
nent of thermomigration in any electromigration experiment. In other words,
electromigration in a flip chip solder joint is accompanied by thermomigration
when a large current density is applied and when the current distribution is
nonuniform due to current crowding. This is an advantage that one can com-
bine electromigration and thermomigration in a study. Nevertheless, one has
to design experiments in order to study thermomigration with and without
electromigration.
In Section 12.2, we shall discuss the design of a test structure of flip chip
solder joints which will enable us to conduct thermomigration with and with-
out electromigration. It will be shown that if a composite solder joint of high
Pb and eutectic SnPb is used, the redistribution of Sn and Pb in thermomigra-
tion can be recognized easily by optical microscopy, even though the original
compositional distribution is not homogeneous in the composite sample. Ob-
servation of thermomigration in eutectic SnPb flip chip solder joints will also
be covered.
In Section 12.3, the fundamentals of thermomigration will be presented and
the driving force of thermomigration and heat of transport will be discussed. In
Sections 12.4 and 12.5, thermomigration under DC and AC electromigration
will be covered. In Section 12.6, we will discuss the use of V-groove solder
line samples for studying thermomigration and its interaction with interfacial
chemical reactions. In Section 12.7, the interaction between thermomigration
and stress migration will be discussed.

12.2 Thermomigration in Flip Chip Solder

Joints of SnPb
12.2.1 Thermomigration in Unpowered Composite Solder Joints
Figure 12.1 shows (a) the schematic diagram of a flip chip on a substrate,
(b) the cross section of a composite flip chip solder joint, and (c) SEM image.
In Fig. 12.1(a), the small squares on the substrate are the electrical contact
pads. The composite joint was composed of 97Pb3Sn on the chip side and
eutectic 37Pb63Sn on the substrate side. It had a contact opening diameter of
90 μm on the chip side and a bump height of 105 μm. The trilayer thin films of
under-bump metallization (UBM) on the chip side were Al (∼0.3 μm)/Ni(V)
(∼0.3 μm) /Cu (∼0.7 μm). On the substrate side, the bond-pad metal layers
were Ni (5 μm)/Au (0.05 μm).
As a control experiment, uniform heating on flip chip samples was per-
formed in an oven at a constant temperature of 150◦ C and atmospheric
SVNY339-Tu April 5, 2007 17:37

330 Chapter 12

Fig. 12.1. Schematic diagram of a

flip chip on a substrate. The small
squares on the substrate are the elec-
trical contact pads. (b) Cross section
of a composite flip chip solder joint.
(c) SEM image of the cross section.
The darker image at the bottom is
the eutectic phase. The brighter im-
age is the high-Pb phase.

pressure for a period of 1 month. For cross-sectional examination, the sam-

ple was polished with SiC paper, then with Al2 O3 powder to the center of
the solder joints. The microstructures of the cross section were examined by
optical microscope (OM) and scanning electron microscope (SEM). Compo-
sition was analyzed using energy-dispersive x-ray analysis (EDX) and elec-
tron probe microanalysis (EPMA). No mixing between the high-Pb and eu-
tectic solders was observed and the image was the same as that shown in
Fig. 12.1(c).
To conduct thermomigration by using the temperature gradient induced
from joule heating, two sets of flip chip samples were prepared. In the first
set, as depicted in Fig. 12.1(a), the sample as received was used. There were
24 bumps on the periphery of the Si chip, and Fig. 12.2(a) depicts a total of
24 solder bumps from right to left at the periphery of a chip and each bump
has the original microstructure shown in Fig. 12.1(c) before electromigration
stressing. We recall that the darker region in the bottom area of each bump
is the eutectic SnPb and the brighter region in the top part is 97Pb3Sn.
Electromigration was conducted through only four pairs of bumps on the
periphery of the chip. They were the pairs of 6/7, 10/11, 14/15, and 18/19 as
numbered in Fig. 12.2(b). The arrows indicate the electron path. The electron
current went from one of the contact pads to the bottom of one of the bumps,
up the bump to the Al thin-film interconnect on the Si strip, then to the top of
the next bump, down the bump, and to the other contact pad on the substrate.
It is worth noting that one can pass current through just a pair of bumps or
several pairs in the row to conduct electromigration. The joule heating from
SVNY339-Tu April 5, 2007 17:37

Thermomigration 331

(a)
Silicon chip side
24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1

FR4 substrate side

(b)

Fig. 12.2. (a) Schematic diagram depicting 24 bumps on the periphery of the Si chip.
Each bump has the original microstructure shown in Fig. 12.1(c) before electromi-
gration stressing. Electromigration was conducted through only four pairs of bumps
on the periphery of the chip. They are the pairs of 6/7, 10/11, 14/15, and 18/19 as
numbered. (b) The effect of thermomigration is clearly visible across the entire row of
the unpowered solder joints—the darker eutectic phase has moved to the Si side, the
hot end.

the Al thin-film line on the strip is the heat source. Due to the excellent
thermal conduction of Si, the neighboring unpowered solder joints would have
experienced a thermal gradient similar to the pairs under DC or AC current.
Since both DC and AC were used, the arrows indicate both directions.
Cross sectioning was performed after the bump pair 10 and 11 failed after
5 hr of current stressing at 1.6 × 104 A/cm2 at 150◦ C. For studying ther-
momigration, the unpowered neighboring bumps were examined. The effect
of thermomigration is clearly visible across the entire row of the unpowered
solder joints, as shown in Fig. 12.2(b), because in all of them Sn has migrated
to the Si side, the hot end, and Pb has migrated to the substrate side, the
cold end. The redistribution of Sn and Pb was caused by a temperature gra-
dient across the solder joints since no current was applied to them. For the
unpowered bumps which were the nearest neighbors of the powered bumps,
the Sn redistribution is also tilted toward the powered bumps. For example,
the powered bump 10 is to the left of the unpowered bump 9, the Sn-rich re-
gion in bump 9 is tilted to the left, and a void is observed. Then the powered
bump 15 is to the right of the unpowered bump 16, and the Sn-rich region
in bump 16 is tilted to the right. In those bumps farther away from the pow-
ered bumps, for example, from bump 1 to bump 4 and bump 21 to 23, Sn
accumulated rather uniformly on the Si side.
SVNY339-Tu April 5, 2007 17:37

332 Chapter 12

(a)

e-

(b)

20μm

Fig. 12.3. (a) Schematic diagram of the Si strip and a row of four of the cross-
sectioned bumps. The chip was cut and only a thin strip of Si was kept. The strip has
one row of solder bumps connecting it to the substrate. The bumps were cut to the
middle so that the cross section of the bump was exposed for in situ observation during
electromigration. (b) The pair of joints on the left was powered at 2 × 104 A/cm2 for
20 hr at 150◦ C. The pair of joints on the right had no electric current at all. Yet all of
them showed composition redistribution and damage.

In powering the four pairs of bumps in the first set of samples, part of
the circuit consists of the thick Cu traces in the substrate, there will be joule
heating too. Nevertheless, the Al lines on the Si were the dominant heaters.

12.2.2 InSitu Observation of Thermomigration

In the second set, as depicted in Fig. 12.3(a), the chip was cut and only a
thin strip of Si was kept. The strip has one row of solder bumps connecting
it to the substrate. The bumps were cut and polished to the middle so that
the cross section of the bump was exposed for in situ observation during
electromigration. Figure 12.3(a) is a schematic diagram of the Si strip and
a row of four of the cross-sectioned bumps. Due to the excellent thermal
conduction of Si and the small strip used, when one pair of the bumps is under
DC or AC current, the other pair of solder joints would experience almost the
same thermal gradient as the powered pair. This set of samples can be used
to conduct in situ experiments by observing changes on the cross-sectioned
surface directly during electromigration. The major difference between the
first set and this set of samples is that the latter has a free surface during the
test. Thus, surface bulging can occur if a large amount of materials is driven
to the cold end by thermomigration, and the bulge can be observed easily.
The pair of joints on the left shown in Fig. 12.3(a) was powered at 2 × 104
A/cm2 for 20 hr at 150◦ C. The pair of joints on the right had no electric
SVNY339-Tu April 5, 2007 17:37

Thermomigration 333

current at all. Yet all of them showed composition redistribution and damage.
The pair on the right (unpowered) showed a uniform void formation at the
interface on the top side, i.e., the Si side which is also the hot side. In the bulk
of the joint, some phase redistribution can be recognized. The redistribution
of elements of Sn, Pb, and Cu can be measured by electron microprobe from
the cross sections of this unpowered pair on the right. The Sn has migrated
to the hot end and there is more Cu in the hot end too and Pb has moved to
the cold end.
If we assume there were no temperature gradient in the pair of bumps
on the right in Fig. 12.3, in other words, the temperatures were uniform
in these bumps, then its thermal history is similar to isothermal annealing,
no phase redistribution or void formation should be found since isothermal
annealing has no effect on phase mixing or unmixing, as shown in Fig. 12.1(c).
However, we may ask if there are other kinds of driving force that can lead
to phase change as observed. Besides electrical and thermal force, we could
have mechanical force. Yet, the mechanical force should have been present
in the isothermal annealing. The annealing does cause interfacial chemical
reactions between solder and UBM on the chip side and between solder and
bond-pad metal on the substrate side. The growth of intermetallic compound
may generate stress owing to molar volume change. However, this effect should
have occurred in the sample which was isothermally annealed at 150◦ C for
4 weeks, yet no noticeable change was detected. Furthermore, solder has a very
high homologous temperature at 150◦ C, so it is unlikely that stress would not
be relaxed in 4 weeks.
Thus, we conclude that the composition redistribution and the damage
(void formation) in the unpowered bumps are due to thermomigration. Then,
which is the dominant diffusing species in thermomigration or which species
diffuses with the temperature gradient is of interest. During electromigration
at 150◦ C, Pb has been found to be the dominant diffusing species. In ther-
momigration of the composite solder joint, the temperature gradient drives
Pb from the hot side to the cold side and Sn atoms from the cold side to the
hot side. The void found on the hot side indicates that Pb is the dominant
diffusing species and the flux of Pb is greater than the flux of Sn. The high-Pb
forms Cu3 Sn after reflow, but the Cu3 Sn will transform to Cu6 Sn5 after the
diffusion of Sn to the hot side. The formation of void and Cu6 Sn5 is rather
uniform across the entire contact area to the Si side. Why Sn diffuses against a
temperature gradient is an interesting question. To answer it, we shall discuss
the driving force and the flux motion in a two-phase microstructure under the
constraint of constant volume.

12.2.3 Random States of Phase Separation in Eutectic

Two-Phase Structures
Figure 12.4 shows a set of cross-sectional images of unpowered composite
solder joints after thermomigration of (a) 30 min, (b) 2 hr, and (c) 12 hr.
SVNY339-Tu April 5, 2007 17:37

334 Chapter 12

Fig. 12.4. Set of cross-sectional images of unpowered solder joints after thermomi-
gration of (a) 30 min, (b) 2 hr, and (c) 12 hr. Before thermomigration, the image is
similar to that shown in Fig. 12.1(a). A random state of phase separation is observed
in (a), the eutectic is segregated toward the hot end in (b), and a near-complete phase
separation is achieved in (c).

Before thermomigration, the image is similar to that shown in Fig. 12.1(a).

In Fig. 12.4(a) a random state of phase separation is observed. In Fig. 12.4(b)
the eutectic is segregated toward the hot end. In Fig. 12.4(c) a near-complete
phase separation is achieved. Many images similar to Fig. 12.4(a) in both DC
and AC stressing were obtained, and four of them are shown in Fig. 12.5 to
illustrate the random state of phase separation in the microstructure before
achieving complete phase separation. It appears that a fluidlike motion occurs
in the solid-state phase separation.
When an electron microprobe was used to measure composition distribu-
tion across a polished cross section of the flip chip sample after thermomigra-
tion, a highly irregular or stochastic composition distribution was observed
as shown in Fig. 12.6; no smooth concentration profile was observed. If the

Fig. 12.5. Four images of random states of phase separation. The two-phase mi-
crostructure in both DC and AC stressing are shown.
SVNY339-Tu April 5, 2007 17:37

Thermomigration 335

Mass% Concentration of Sn in Flip Chip Solder Joint

Sn %mass concentration 90
left line
80
70 center line
60
50 right line
40 Before
30 thermomigration
20
10
0
0 10 20 30 40 50 60 70 80 90 100 110 120
Distance from the Copper line on Chip Side (μm)

Fig. 12.6. Electron microprobe measurement of composition redistribution across a

polished cross section of the flip chip sample after thermomigration. A highly irregular
or stochastic composition distribution was found; no smooth concentration profile was
observed.

electromigration experiment was extended to several days, a clear and nearly

complete phase separation of Sn and Pb in the unpowered joints was found.

12.2.4 Thermomigration in Unpowered Eutectic

SnPb Solder Joints
The eutectic 37Pb63Sn flip chip solder joints used for the thermomigration
test were arranged similarly to that shown in Fig. 12.2(a) with 11 bumps.
The UBM thin films on the chip side were Al (∼0.3 μm)/Ni(V) (∼0.3 μm)
/Cu (∼0.7 μm) deposited by sputtering. The bond-pad metal layers on the
substrate side were Ni (5 μm)/Au (0.05 μm) prepared by electroplating. The
bump height between the UBM and the bond pad is 90 μm. The contact
opening on the chip side has a diameter of 90 μm.
Only one pair of bumps was current stressed with DC current of 0.95 A
at 100◦ C for 27 hr. The average current density at the contact opening was
1.5 × 104 A/cm2 . The unpowered bumps neighboring the powered pair were
used to study thermomigration.
Figure 12.7 displays SEM images of the cross section of four of the unpow-
ered bumps after the electromigration test. The lighter color in the images
is the Pb-rich phase and the darker color is the Sn-rich phase. Compared to
the as-received sample, the results show that the Pb–rich phase has moved
to the substrate side (the cold side) in the unpowered neighboring bumps of
the powered bumps. Also, one of the unpowered neighboring bumps, shown
in Fig, 12.7(c), has some dendritic structure of crystallization of a liquid
phase, indicating that it was partially melted in the test. It is worth not-
ing that crystallization of a molten eutectic phase should show a eutectic
SVNY339-Tu April 5, 2007 17:37

336 Chapter 12

Fig. 12.7. SEM images of cross section of 4 of the unpowered bumps. The lighter
color in the images is the Pb-rich phase and the darker color is the Sn-rich phase. In
(c), some dendritic structures of crystalization of a liquid phase are shown.

microstructure. The dendritic structure indicates that phase separation has

occurred before melting.
Figure 12.8(a) shows an enlarged SEM image of an unpowered bump. The
redistribution of Sn and Pb is shown by the accumulation of a large amount
of Pb to the substrate side (the cold side), yet there is no accumulation of Sn
to the chip side (the hot side). A very surprising finding is not only that the
microstructure in the bulk of the bump is quite uniform (except the accumu-
lated Pb-rich phase), but also the lamellar structure is much finer, indicating
the existence of many more interfaces in the microstructure after the phase
separation, in turn a higher energy state. We recall that when a eutectic two-
phase microstructure is annealed at constant temperature, grain growth and
coarsening, instead of refinement, of the two-phase lamellar microstructure
should occur to reduce the surface energy, as shown in Fig. 2.23. Since the
interface is disordered, the formation of a finer lamellar structure creates a
more disordered state or more entropy production. However, the diffusion
along the interfaces is faster than the lattice, so kinetically the rate of entropy
production is faster with a finer lamellar structure.
The concentration profiles across the bump by EPMA are shown in
Fig. 12.8(b) and (c). Three profile lines across the bumps were scanned and
SVNY339-Tu April 5, 2007 17:37

Thermomigration 337

Fig. 12.8. (a) Enlarged SEM image of bump No. 11. The redistribution of Sn and Pb
can be seen. It shows the migration of a large amount of Pb to the substrate side (the
cold side), yet there is no accumulation of Sn to the chip side (the hot side). (b, c)
Concentration profiles across the bump by EPMA of Pb and Sn, respectively. Three
profile lines across the bumps were scanned and every line is the average of three sets
of data points. Each point was taken at each 5-μm step from the chip side to the
substrate side.

every line is the average of three sets of data points. Each point was taken at
every 5-μm step from the chip side to the substrate side. The results shows
that the concentration of Pb on the substrate side is about 73% and that of
Sn on the chip side is around from 70% to 80%. However, the concentration
distribution reveals that there is no linear concentration gradient. Near the
cold end, the accumulation of Pb and depletion of Sn occurred. However, away
from the cold end, the average distribution of Pb and Sn in the rest of the
sample is quite uniform, except for some minor local fluctuations due to the
two-phase microstructure. Clearly, Pb moves with the temperature gradient
and is the dominant diffusing species in the thermomigration because it ac-
cumulates at the cold end, but there is no concentration gradient of Pb for
the diffusion, unlike precipitation. If Sn were the dominant diffusing species,
the concentration profile of Pb would have increased uniformly in the bulk
of the sample, but not a pile-up at the cold side. The higher concentration
of Sn at the end of the substrate side is due to the formation of Cu6 Sn5
IMC.
SVNY339-Tu April 5, 2007 17:37

338 Chapter 12

12.3 Fundamentals of Thermomigration

12.3.1 Driving Force of Thermomigration
Similar to the thermoelectric effect whereby a temperature gradient can drive
electrons to move, it can also drive atoms. In essence, the electrons in the
high-temperature region have higher energy in scattering or interaction with
diffusing atoms, hence atoms move down the temperature gradient. On the
driving force of atomic diffusion, we recall that the atomic flux driven by
chemical potential can be given as (see Appendix A)
 
D ∂μ
J = Cv = CMF = C − , (12.2)
kT ∂x

where v is drift velocity, M = D/kT is mobility, and μ is chemical potential

energy. Considering temperature gradient as the driving force, we have
 
D Q∗ ∂T
J =C − , (12.3)
kT T ∂x

where Q∗ is defined as heat of transport. Comparing the last two equations,

we see that Q∗ has the same dimension as μ, so it is the heat energy per atom.
In other words, if we take entropy S = Q/T , we have SdT which is thermal
energy.
The definition of Q∗ is the difference between the heat carried by the
moving atom and the heat of the atom at the initial state (the hot end or the
cold end) [see Eq. (12.1)]. For the element which moves from the hot end to
the cold end, Q∗ is negative since it loses heat. For an element moving from
cold to hot, its Q∗ is positive.
The driving force of thermomigration is given as
 
Q∗ ∂T
F =− . (12.4)
T ∂x

To make a simple estimation, we take ΔT /Δx = 1000 K/cm, and consider the
temperature difference across an atomic jump and take the jump distance to
be a = 3 × 10−8 cm. We have a temperature change of 3 × 10−5 K across an
atomic spacing. Thus, the thermal energy change will be

3kΔT = 3 × 1.38 × 10−23 (joule/K) × 3 × 10−5 K ≈ 1.3 × 10−27 joule.

As a comparison, we shall consider the driving force, F , of electromigration

at a current density of 1 × 104 A/cm2 or 1 × 108 A/m2 which we know has
SVNY339-Tu April 5, 2007 17:37

Thermomigration 339

induced electromigration in solder alloys:

F = Z ∗ eE = Z ∗ eρj.

We shall take ρ = 10 × 10−8 Ω-m, Z ∗ of the order of 10, and e = 1.602 ×

10−19 coulomb, and we have F = 10 × 1.6 × 10−19 (coulomb) × 10 × 10−8
(ohm-meter) × 108 amp/meter2 = 1.6 × 10−17 coulomb-volt/meter = 1.6 ×
10−17 newton.
The work done by the force in a distance of atomic jump of 3 × 10−10 m will
be Δw = 4.8 × 10−27 newton-meter = 4.8 × 10−27 joule. This value is close to
the thermal energy change we have calculated for thermomigration. Thus, if a
current density of 104 A/cm2 can induce electromigration in a solder joint, a
temperature gradient of 1000◦ C/cm will induce thermomigration in a solder
joint.
On heat of transport, we note that Q∗ can be positive or negative. In the
Fe-C system, carbon was found to move to the hot end interstitially with a
positive heat of transport. In alloys of SnPb, when thermomigration drives Pb
to move from the hot zone to the cold zone, it moves down the temperature
gradient. But the thermomigration drives Sn to move in the opposite direction,
against the temperature gradient. The Q∗ for Pb is negative or the heat
decreases, but for Sn, it seems that the Q∗ is positive since it moves to the
hot end and gains heat. This is because we have one temperature gradient in
thermomigration for both species, unlike interdiffusion in a diffusion couple,
in which the concentration gradient of the two interdiffusing species is in the
opposite direction, so the chemical potential change can be positive for both
species.
To measure Q∗ , if we know the atomic flux, we can use the flux equation,
i.e., Eq. (12.3), to determine Q∗ when diffusivity, the average temperature,
and temperature gradient are known. The heat of transport of Pb in ther-
momigration discussed in Section 12.2.4 is estimated below by using the flux
equation, Eq. (12.3).
By measuring the accumulation width of Pb (12.5 μm) on the substrate
side from Fig. 12.8(a), the total volume of atomic transportation can be ob-
tained from the product of the width and the cross section of the solder joint.
Taking the density of 27Sn73Pb as 10.25g/cm3 and the molecular weight of
27Sn73Pb as 183.3 g/mole, the flux of JTM = 4.26 × 1014 atoms/cm2 sec is ob-
tained. Assuming a temperature gradient of 1000 K/cm, a temperature of
180◦ C, which is very close to the melting temperature of eutectic SnPb, and
a diffusivity of DPb = 4.41 × 10−13 cm2 /sec, the molar heat of transport Q∗Pb
is estimated to be −25 kJ/mole.
The accuracy of determination of Q∗ may be affected by the measurement
of flux in Eq. (12.3) since the concentration distribution is non-uniform. The
assumed temperature gradient may be incorrect. However, more serious is
the basic assumption in the analysis that both Pb and Sn move with the
temperature gradient. Actually if Pb is the dominant diffusing species and
SVNY339-Tu April 5, 2007 17:37

340 Chapter 12

moves from the hot side to the cold side, the Sn will be pushed back in
the opposite direction if a constant volume process is assumed. The effect
of reverse flux of Sn on the calculation of heat of transport in a two-phase
microstructure should be studied.

12.3.2 Entropy Production

Onsager defined the conjugated flux and force in irreversible processes so that
their product is equal to the product of temperature and entropy production
per unit volume. In thermomigration, the major entropy production is due to
heat propagation under a temperature gradient, so
  
T dS dT 1 dT
= −κ − , (12.5)
V dt dx T dx

where V is volume and S is entropy. If we take heat conductivity in solder as

κ∼
= 50 joule/m-sec-K, dT /dx = 1000 K/cm, and T = 400 K, we obtain [8]

T dS joule
= 1.2 × 109 3 .
V dt m sec

Other source of entropy production during thermomigration will be much

smaller. The entropy production by atomic migration, the cross-effect, can be
estimated as
   2
T dS D D dT
= C F F =C 3k , (12.6)
V dt kT kT dx

where we have assumed the driving force F = 3k(dT /dx) where k is

Boltzmann’s constant, and by taking dT /dx = 1000 K/cm, we obtain

T dS joule
= 3 × 102 3 ,
V dt m sec

which is much smaller than that due to heat propagation.

For comparison, in electromigration, the major entropy production is due
to joule heating:
 
T dS dϕ
= jem − 2
= ρjem , (12.7)
V dt dx

where jem is electric current density and ϕ is electric potential or voltage. If

we take jem = 104 A/cm2 and ρ = 10−5 Ω-cm (for solder), we obtain

T dS joule
= 109 3 ,
V dt m sec
SVNY339-Tu April 5, 2007 17:37

Thermomigration 341

which is of the same order of magnitude of entropy production in thermomi-

gration in a temperature gradient of 1000 K/cm near the temperature of
400 K. Similarly, all other sources of entropy production in electromigration
are much weaker. The entropy production by atomic migration (electromigra-
tion) under electron wind force is

 
T dS D D
(Z ∗ eρj) .
2
= C F F =C (12.8)
V dt kT kT

If we take C = 1029 /m3 , T = 400 K, D = 10−12 m2 /sec, Z ∗ e = 10−18 coulomb,

ρ = 10−7 Ω-m, and j = 108 A/m2 , we obtain

T dS joule
= 10 3 .
V dt m sec

The value is much smaller than that from joule heating.

Since entropy production by joule heating or heat propagation is many
orders of magnitude larger than that by atomic migration, it is conceivable
that entropy production in electromigration or thermomigration can affect
microstructure greatly.

12.3.3 Effect of Concentration Gradient on Thermomigration

Thermomigration can cause a uniform single phase solid solution to become
nonuniform. From the point of view of kinetics, the change from a uniform to
a nonuniform solid solution requires uphill diffusion of the components against
their composition gradient, so thermomigration is opposed by the concentra-
tion gradient. It is worth noting that in thermomigration, the diffusion of the
element to the hot end is against both the temperature gradient and the con-
centration gradient. Thus, in the flux equation of thermomigration we need
to add the opposing flux as

D Q∗ ∂T ∂C
J = −C +D . (12.9)
kT T ∂x ∂x

The sign of the last term is positive because it is uphill diffusion. Accordingly,
when the concentration gradient in uphill diffusion is large enough, there can
be no net effect of thermomigration, i.e., J = 0. Thus, a steady state is reached,
in which a constant concentration gradient will be maintained. In the steady
state,
 
1 ΔC d(ln C) 1 Q∗ ΔT
= = (12.10)
C Δx dx kT T Δx
SVNY339-Tu April 5, 2007 17:37

342 Chapter 12

This was achieved in a thermomigration experiment in the Fe-C system by

Shewmon, and carbon was found to diffuse to the hot end. The concentration
of C in Fe reached a constant linear gradient [1].
The concentration gradient that has balanced the thermomigration can be
measured, and the Q∗ can be calculated by knowing the temperature gradient
and temperature. While there is an uncertainty in choosing the temperature,
the uncertainty is negligible because ΔT /T is typically small.
However, if we examine Eq. (12.10), it is unclear if the equation can be
satisfied arbitrarily. If we keep C, T , and Q∗ as constants, there are variables
of ΔC, Δx, and ΔT . Among the three variables, we have to fix two of them in
order to solve the third one. For example, if the length and the temperature
difference between the two ends of the sample are given, the variable ΔC is
fixed by Eq. (12.6). But it is unclear if it can be satisfied arbitrarily, since the
initial homogeneous concentration is given, so the maximum ΔC is limited by
(2C − 0).
The steady-state concentration gradient depends on the length of the sam-
ple, Δx, as well as the initial homogeneous concentration in the sample. When
the length is short, we can have a larger concentration gradient, yet there may
be a back stress in the cold end and we need to assume equilibrium vacancy
distribution in order to remove the back stress. Also, a larger temperature
gradient exists in a shorter sample if the same temperature difference is main-
tained. When the length is long, the concentration gradient may not be large
enough to balance thermomigration since the gradient depends on the orig-
inal concentration of the homogeneous solid solution. Yet, the temperature
gradient will be lower too if the same temperature difference is maintained in
a longer sample.

12.3.4 The Critical Length below Which No Thermomigration

Occurs in a Pure Metal
In electromigration in a pure Al strip, there exists a critical length of sample
below which there will be no net effect of electromigration due to back stress.
The back stress induced a reverse flux to balance the flux of electromigration.
In thermomigration of pure Al or Sn, this should also happen when back stress
occurs. And we have
Q∗ ∂T ΔσΩ
=
T ∂x Δx
or
 
∂T ΔσΩ
Δx = T. (12.11)
∂x Q∗

In the above equation, Δx(dT /dx) can be taken to be a critical product of

the sample below which there will be no net effect of thermomigration. Or
SVNY339-Tu April 5, 2007 17:37

Thermomigration 343

for a given temperature gradient, below a critical length, Δx, there will be
no net effect of thermomigration. However, this case is more complicated
than that in electromigration because of temperature gradient. Equilibrium
vacancy concentration is a function of temperature.

12.3.5 Thermomigration in a Eutectic Two-Phase Alloy

Thermomigration in a eutectic alloy is different from that in a solid solu-
tion. A eutectic alloy below the eutectic temperature is a two-phase alloy. In
a two-phase mixture the change of composition does not mean any change
of chemical potential, instead it means just the change of local volume frac-
tions of the two phases. Composition of each primary phase is determined by
thermodynamic equilibrium between them and is known from the equilibrium
phase diagram. Thus, if some segregation in a eutectic solder is induced by
thermomigration, it means a change of gradient of volume fractions, not gra-
dient of chemical potentials. Thus, the segregation can be enormous due to
the lack of a counteracting force.
Strictly speaking, the equilibrium phase diagram is obtained under the
assumption of constant temperature and constant pressure. Thus, the concept
of constant chemical potential between the two eutectic phases cannot be
applied to thermomigration as discussed in the last paragraph because the
temperature is not constant. It is an approximation.
At the end of thermomigration in a eutectic two-phase structure, no steady
state of a linear concentration gradient is achieved, instead a near-complete
segregation of the two eutectic phases occurs. Furthermore, since a gradient
of volume fractions is not a driving force, the lack of counteracting force in
the form of ΔC/Δx will not produce a smooth segregation. Thus, a tendency
of stochastic behavior occurs in thermomigration of the eutectic mixture, as
shown experimentally in Fig. 12.5, Section 12.2.3. No smooth concentration
gradient exists as in the Soret effect of a solid solution. A description of
thermomigration in a eutectic structure can be given, which is similar to the
presentation in Section 9.7 for electromigration in a two-phase structure.
The fluxes generated by the temperature gradient are given as

 
n 1 D1 ∂ ln T p1 D1 ∂ ln T ∂(1/kT )
Ω1 J1TM = Ω1 · −Q1 =− · Q1 = p1 D1 Q1 ,
kT ∂x kT ∂x ∂x
 
n 2 D2 ∂ ln T p2 D2 ∂ ln T ∂(1/kT )
Ω2 J2TM = Ω2 · −Q2 =− · Q2 = p2 D2 Q2 .
kT ∂x kT ∂x ∂x
(12.12)

If they have the same direction (if Qi have the same sign), they will not satisfy
the constraint of constant volume as given in Eqs. (9.19) and (9.20).
SVNY339-Tu April 5, 2007 17:37

344 Chapter 12

12.4 Thermomigration and DC Electromigration

in Flip Chip Solder Joints
In DC electromigration, there is a polarity effect. When a daisy chain of
bumps is tested by DC electromigration, the void formation at the cathode
contact to the Si occurs on every alternative bump. Therefore, it is very easy
to recognize DC electromigration. However, we have to consider the contri-
bution of thermomigration to DC electromigration. Thermomigration may
accompany electromigration when the joule heating of the latter has induced
a temperature gradient of the magnitude of 1000◦ C/cm across the solder
joint.
If we assume the temperature is hotter on the Si chip side, i.e., the top
side in Fig. 12.3, thermomigration will drive the dominant diffusing species
down, and it is in the same direction as electromigration in the downward
flow electrons, so the effects of thermomigration and electromigration are
combined. If we consider DC electromigration in the pair of bumps on the
left in Fig. 12.3(a), we assume electrons flow down on the left-side bump.
Both thermomigration and electromigration will drive vacancies to go to the
Si contact and void formation will occur near the contact. However, in the
right-side bump of the pair, electromigration will drive atoms in the opposite
direction and counteract thermomigration, i.e., the two effects tend to cancel
each other. Since we can obtain different experimental results in a pair of
bumps, we should be able to decouple the contribution of thermomigration
and electromigration. If we use pure Sn flip chip samples, we have a simple
case of diffusion of one element and we can use marker to determine the net
effect of fluxes. With two elements in eutectic SnPb bumps or in solid solu-
tion PbIn bumps, the problem is more complicated. In these cases, besides
marker motion, the concentration change of the flux of Sn and Pb (or Pb and
In) should be determined. Due to stochastic behavior, the analysis of eutectic
SnPb is even more complicated than that of PbIn.

12.5 Thermomigration and AC Electromigration

in Flip Chip Solder Joints
There is no difference in joule heating whether we apply AC or DC to stress
a pair of flip chip solder bumps. This is because the current distribution
in the pair of powered bumps as shown in Fig. 12.3 is independent of the
current direction or polarity, therefore joule heating is the same whether we
apply AC or DC in electromigration, except there may be a difference at a
very high frequency AC. However, unlike DC, it is generally assumed that
AC does not induce mass flow. If the assumption is correct that there is no
electromigration-induced mass migration in a pair of bumps stressed by AC,
we should expect only thermomigration in the pair of bumps powered by AC,
if AC has generated a temperature gradient in the pair of bumps. This is the
SVNY339-Tu April 5, 2007 17:37

Thermomigration 345

advantage of using AC to study thermomigration in a flip chip solder joint;

AC serves just as a heating source to generate a temperature gradient across
the solder bump and makes no contribution to electromigration.
We should be able to verify the assumption that AC does not induce
mass migration by examining very carefully a pair of bumps that have been
stressed by AC together with a neighboring bump that was unpowered. That
arrangement was depicted in Fig. 12.3, in which the bump pair on the left-
hand side can be stressed by AC, but the neighboring pair on the right-hand
side are dummies and will carry no current. The joule heating generated by
the left pair stressed by AC will cause the same thermomigration in both
pairs. Marker displacement experiments should be conducted in both pairs to
determine whether the mass migration is the same or not. Direct comparison
should be made to DC experiments too.

12.6 Thermomigration and Chemical Reaction

in Solder Joints
When Sn is driven to the hot end in a flip chip solder joint by thermomigra-
tion, more Cu-Sn intermetallic compound will form at the hot end. Solder line
in V-grooves on (001) Si wafer can be used to study the interaction between
thermomigration and chemical reaction. For thermomigration, we can keep
one end of the Cu wire at 100◦ C (boiling water) and the other end at 180◦ C
(heating source); there is thus a very large temperature gradient between the
two ends of the solder line, if the length of the wire is short. For thermomi-
gration in solder alloys, the temperature at the hot end will be limited by the
melting of the solder. Also, it is not good to cool the cold end to a very low
temperature which will limit atomic diffusion.
Compared to flip chip samples, the advantages of V-groove samples are
that the temperature gradient can be measured more easily and in situ ob-
servation of damage development in the sample during thermomigration can
be performed.

12.7 Thermomigration and Creep in Solder Joints

The sample of Cu-wire/solder-ball/Cu-wire as discussed in section 11.2 can be
used to study the interaction between thermomigration and mechanical stress
in solder joints, same as the study of the interaction between electromigration
and mechanical stress. The diameter of the Cu wire can be varied from 1000
μm (1 mm) to 300 μm, and the size of the solder bead should be similar to
the diameter of the Cu wire. The larger diameter samples will be better for
sample preparation and measurement in thermomigration.
After thermomigration, tensile test of the sample can be conducted and
compared to a sample without thermomigration. If vacancies move to the hot
SVNY339-Tu April 5, 2007 17:37

346 Chapter 12

end in thermomigration, the interface at the hot end will become weaker.
For shear test, a solder joint between two parallel Cu plates can be prepared,
instead of Cu wires. We note that a solder joint between two parallel Cu plates
has been used widely to study shear properties of solder alloy at constant
temperature. Here the two plates should be kept at different temperatures
during shear. Back stress is expected in thermomigration in solder joints,
especially at the cold end. The interaction between back stress and applied
stress is of interest.
More interestingly, a weight can be attached to the cold end for the study
of thermomigration and creep together. The sample will experience both a
gradient of temperature and a gradient of stress at the same time, but the
gradients are along different directions. To analyze the combined effect, the
effect of creep alone on phase change in a solid solution and in a eutectic
two-phase structure should be studied first.

References
1. P. Shewmon, “Diffusion in Solids,” 2nd ed., Chapter 7 “Thermo- and
electro-transport in solids,” TMS, Warrendale, PA (1989).
2. D. V. Ragone, “Thermodynamics of Materials,” Volume II, Chapter 8
“Nonequilibrium thermodynamics,” Wiley, New York (1995).
3. W. Roush and J. Jaspal, “Thermomigration in Pb-In solder,” IEEE Proc.,
CH1781, pp. 342–345 (1982).
4. D. R. Campbell, K. N. Tu and R. E. Robinson, “Interdiffusion in a bulk
couple of Pb-PbIn alloy,” Acta Metall., 24, 609 (1976).
5. H. Ye, C. Basaran, and D. C. Hopkins, “Thermomigration in Pb-Sn solder
joints under joule heating during electric current stressing,” Appl. Phys.
Lett., 82, 1045–1047 (2003).
6. A. T. Huang, A. M. Gusak, and K. N. Tu, “Thermomigration in SnPb
composite flip chip solder joints,” Appl. Phys. Lett., 88, 141911 (2006).
7. Y. C. Chuang and C. Y. Liu, “Thermomigration in eutectic SnPb alloy,”
Appl. Phys. Lett., 88, 174105 (2006).
8. Fan-yi Ouyang, K. N. Tu, Yi-Shao Lai, and Andriy M. Gusak, “Effect of
entropy production on microstructure change in eutectic SnPb flip chip
solder joints by thermomigration,” Appl. Phys. Lett., 89, 221906 (2006).
SVNY339-Tu March 2, 2007 22:4

Appendix A: Diffusivity of Vacancy

Mechanism of Diffusion in Solids

To analyze atomic diffusivity, we shall consider the vacancy mechanism of

diffusion in a face-centered-cubic metal. We make the following assumptions
in order to develop the analytical model.
1. It is a thermally activated unimolecular process. Unimolecular process
means that we consider a single atom in the diffusion process and it is
a near-equilibrium process. This is unlike chemical reactions that are bi-
molecular processes, such as rock salt formation, in which the collision of
two atoms of Na and Cl is involved and the process is far from equilibrium.
2. It is a defect-mediated process. Here the defect is a vacancy.
3. The activated state obeys a Boltzmann’s equilibrium distribution from
transition state theory. Hence, the Boltzmann distribution function is used.
4. It is assumed that the probability of reverse jumps is large due to small
driving force, so we have to consider reverse processes. In other words, the
process is not far from equilibrium.
5. Statistically, atomic diffusion obeys the principle of random walk.
6. A long-range diffusion requires a driving force.

v
ΔGm

x
A B
λ
Distance

At equilibrium, in a one-dimensional configuration, atoms are attempting

to jump over the potential energy barrier with the attempt frequency, ν0 ,
to exchange position with a neighboring vacancy, as depicted in Fig. A.1.
SVNY339-Tu March 2, 2007 22:4

348 Appendix A

The successful or exchange jump frequency is given below on the basis of

Boltzmann’s distribution:
 
−ΔGm
ν = ν0 exp ,
kT

where

ν0 = attempt frequency,
ν = exchange frequency,
ΔGm = saddle point energy (activation energy of motion).

We note that there is a reverse jump at the same attempt frequency.

v+
v–
ΔGm

Now consider a driving force F (which equals the slope of the base line
in Fig. A.2). The meaning of F will be discussed later. The forward jump is
increased by
 
+ λF
v = v exp + ,
2kT

where λ is the jump distance. The reverse jump is decreased by

 
λF
v − = v exp − .
2kT

And the net frequency is

 
− λF
vn = v − v = 2v sinh
+
.
2kT

Now, we take the “condition of linearization,”

λF
 1.
kT
SVNY339-Tu March 2, 2007 22:4

Appendix A 349

Then the net frequency jump vn is “linearly” proportional to the driving force
F:
λF
vn = v .
kT
We can define a drift velocity

vλ2
v = λvn = F.
kT
Then, the atomic flux J, which has units of number of atoms per unit area
per unit time, is

Cvλ2
J = Cv = F,
kT

where M = vλ2 /kT is defined as the atomic mobility. The atomic flux J is
“linearly” proportional to the driving force F. The driving force is generally
defined as a potential gradient,

∂μ
F =− .
∂x
In atomic diffusion, here μ is the chemical potential of an atom and is defined
at constant temperature and pressure to be
 
∂G
μ= ,
∂C T,p

where G is Gibbs free energy and C is concentration. For an ideal dilute solid
solution,

μ = kT ln C,
∂μ ∂C kT ∂C
F =− =− ,
∂C ∂x C ∂x
     
Cvλ2 Cvλ2 kT ∂C ∂C ∂C
J= F = − = −vλ2 = −D .
kT kT C ∂x ∂x ∂x

Hence, we have obtained Fick’s first law of diffusion:

J
 ∂C  = D = vλ2 ,
− ∂x

where D is the diffusion coefficient (or diffusivity) in units of cm2 /sec. Then,
M = D/kT . In the above derivation, as depicted in Fig. A.1, we have assumed
SVNY339-Tu March 2, 2007 22:4

350 Appendix A

that the diffusing atom has a neighboring vacancy. For the majority of atoms
in the lattice, this is not true, and we must define the probability of an atom
having a neighboring vacancy in the solid as
 
nv ΔGf
= exp − .
n kT

nv is the total number of vacancies in the solid, n is the total number of lattice
sites in the solid, and ΔGf is the Gibbs free energy of formation of a vacancy.
Since in a face-centered-cubic metal, a lattice atom has 12 nearest neighbors,
the probability of a particular atom having a vacancy as a neighbor is
 
nv ΔGf
nc = nc exp − nc = 12.
n kT

Next, we have to consider the correlation factor in the face-centered-cubic

lattice. The physical meaning of the factor is the probability of reverse jump;
after the atom has exchanged position with a vacancy, it has a high proba-
bility of returning to its original position before the activated configuration is
relaxed. The factor has a range between zero and unity. When f = 0, it means
the probability of reverse jump is 100%, so the atom and the vacancy are ex-
changing position back and forth, which will not lead to any random walk
but instead a correlated walk. When f = 1, it means that after the jump, the
atom will not return to its original position, and it is a random walk because
the next jump will depend on the random probability of a vacancy coming
to the neighborhood of this atom. In fcc metals, f = 0.78, so about 80% of
jumps are random walk, and about 20% are correlated walk. Finally, we have
the diffusivity as
 
ΔGm + ΔGf
D = f nc a2 ν0 exp − ,
kT
     
ΔSm + ΔSf ΔHm + ΔHf ΔH
D = f nc λ2 ν0 exp exp − = D0 exp − .
k kT kT
SVNY339-Tu March 3, 2007 7:35

Appendix B: Growth and Ripening Equations

of Precipitates

The distribution function of a set of precipitates is obtained by solving the

continuity equation in size space:

∂f ∂f
=v ,
∂t ∂x

where f is the size distribution function of the precipitates and v is the

growth/dissolution velocity of the precipitates. To solve the continuity equa-
tion, the first step is to find the growth/dissolution velocity. In this appendix
we shall derive the velocity of a spherical precipitate. When the diameter
of the precipitates is in nanoscale, it is important to take into account the
Gibbs–Thomson potential of curvature as in the LSW theory of ripening.
In other words, the equilibrium concentration at the precipitate/matrix in-
terface is a function of radius. When the precipitate diameter is large, we
can assume the equilibrium concentration to be constant, independent of the
radius.

B.1 Kinetics of Precipitation

We consider the growth or dissolution of a spherical particle or precipitate.
Letting R be the variable, the diffusion equation in spherical coordination,
assuming a steady state, is

∂2C 2 ∂C
+ = 0.
∂R2 R ∂R

The solution is

b
C= + d. (B.1)
R
SVNY339-Tu March 3, 2007 7:35

352 Appendix B

The boundary conditions are

b
At R = r0 , C = C0 , we have C0 = + d. (B.2)
r0
b
At R = r, C = Cr , we have Cr = + d. (B.3)
r
Now, if we take the difference between the last two equations, we have
 
1 1 r0 − r ∼ b
Cr − C0 = b − =b = where r0  r. (B.4)
r r0 rr0 r

This is an important assumption. It means that precipitates are far apart.

Note that if we take the volume fraction, f , the ratio of volume of the precip-
itate particles to the volume of the diffusion field, or the total volume of the
precipitated phase to the total volume of the matrix, is
4π 3
r r3
f= 3 = 3 → 0.
4π 3 r0
r
3 0
It is a very small value: f → 0. (This is a very important assumption in the
LSW theory of ripening to be discussed later.)
We have b = r(Cr − C0 ). Substituting b into Eq. (B.3), we have

r(Cr − C0 )
Cr = + d. (B.5)
r

We have d = C0 , and Eq. (B.1) becomes

(Cr − C0 )r
C(R) = + C0 . (B.6)
R
dC (Cr − C0 )r
Therefore, =− .
dR R2

At the particle/matrix interface for a particle of radius r, or R = r, we

have
dC Cr − C0
=− . (B.7)
dR r
Then the flux of atoms arriving at the interface is

∂C D(C0 − Cr )
J = +D = at R = r. (B.8)
∂R r
Note that when Cr > C0 , J < 0, the net flux is toward the particle, and thus
SVNY339-Tu March 3, 2007 7:35

Appendix B 353

it grows. When Cr < C0 , J > 0, the flux leaves the particle so the particle
dissolves.

B.2 Growth Rate of a Spherical Particle Assuming

Cr Is Constant
If Ω is atomic volume, in time dt, a volume is added to the spherical particle,

ΩJAdt = ΩJ4πr2 dt = 4πr2 dr,

where the last term is the increment of a spherical shell due to the growth.
Hence,

dr ΩD(C0 − Cr )
= ΩJ = . (B.9)
dt r

B.2.1 Case 1: The Growth of a Precipitate

By integration and assuming when t = 0, r = 0,

r2 = 2ΩD(C0 − Cr )t. (B.10)

Note here that if we follow Ham’s approach and take Cr as a constant, it is

not a function of r as given by the Gibbs–Thomson equation. From the above
equation, we see that r ∼= t /2 and r3 ∼
1
= t3/2 . Or we have

r3 = [2ΩD(C0 − Cr )t]3/2 . (B.11)

B.2.2 Case 2: The Depletion of Concentration in the Matrix

(Mean-Field Consideration)
On the other hand, we consider the loss of average concentration in the matrix,
ΔC = C0 − C, due to the formation of the precipitate, where the average
concentration in the matrix is C, which can be regarded as the “mean-field”
concentration (the conception of mean-field theory). In the beginning, the
average concentration is C0 , and it changes to Cwhen the precipitate grows.
Let 1/Ω = Cp be the concentration in the solid precipitate. We have simply
by mass balance,

4π 3 4π 3 1 4π
r0 (C0 − C) = r = [2ΩD(C0 − Cr )t]3/2 , (B.12)
3 3 Ω 3Ω
 3/2  3/2
2D(C0 − Cr )Ω1/3 2Bt
C = C0 − t = C0 − , (B.13)
r02 3
SVNY339-Tu March 3, 2007 7:35

354 Appendix B

where

3D(C0 − Cr )
B≡ 1/3
.
Cp r02

We note that the above equation is the same as Eq. (1–36) in Chapter 1 in
Shewmon.

B.2.3 Case 3: Consider Growth of Precipitate and Depletion of

the Matrix Together
We can derive the last equation in a slightly different way. The growth of
the precipitate reduces the concentration in the matrix. The amount of solute
atoms diffusing to the precipitate in time Δt is

J(r)4πr2 Δt = number of atoms.

It should be equal to the reduction of the average concentration in the volume

of the sphere of diffusion of r0 . Hence, if take the average concentration in the
matrix to be C,

4πr03
ΔC = J(r)4πr2 Δt.
3

Or, we have

ΔC 3 3D
= 4πr2 J(r) = − 3 (C0 − Cr )r. (B.14)
Δt 4πr03 r0

The conservation of mass requires that

4π 3 4π 3
r (C0 − C) = r Cp , (B.15)
3 0 3

where Cp is the concentration of solute in the solid precipitate and Cp = 1/Ω.

Hence,

 1/3
C0 − C
r = r0 . (B.16)
Cp

By substituting r into the rate equation above, we have

ΔC 3D 1
= − 2 (C0 − Cr ) 1/3 (C0 − C)1/3 . (B.17)
Δt r0 Cp
SVNY339-Tu March 3, 2007 7:35

Appendix B 355

Let

3D(C0 − Cr )
B≡ 1/3
.
Cp r02

We have

dC
= −B(C0 − C)1/3 .
dt
By integration we obtain

3
− (C0 − C)2/3 = −Bt + β.
2
At t = 0, C0 = C, so β = 0.

Thus, we have the solution,

 3/2
2Bt
C = C0 − , (B.18)
3

which is the same as what we have obtained. Hence, we have

C0 − C ∼
= t3/2 for 3-dimensional growth.

⎡  3/2 ⎤   3/2   
3/2
2Bt t t
Let C = C0 ⎣1 − 2/3
⎦ = C0 1− = C0 exp −
3C0 τ τ

(B.19)

if we assume t  τ,

1/3 2/3  1/3

Cp r02 C0 ∼ r2 Cp
where τ= = 0 . (B.20)
2D(C0 − Cr ) 2D C0

Usually D, Cp , C0 are known, and we can design the experiment to control

the growth of the precipitate.

B.3 Gibbs–Thomson Potential: Effect of

Surface Curvature
Consider a sphere with radius r and surface energy per unit area γ. The
surface energy exerts a compressive pressure on the sphere because it tends
SVNY339-Tu March 3, 2007 7:35

356 Appendix B

to shrink to reduce the surface energy. The pressure is

dE d4πr2 γ
F − − 8πrγ 2γ
p= = dr = dr
2
=− 2
=− . (B.21)
A A 4πr 4πr r
If we multiply p by the atomic volume Ω, we have the chemical potential

2γΩ
μr = − . (B.22)
r
This is called the Gibbs–Thomson potential due to the curvature of surface.
We note that it is not just the potential of the surface atoms of the precipitate,
it is the potential energy of all the atoms in the precipitate. We see that for
a flat surface r = ∞, μ∞ = 0 so we have

2γΩ
ur − μ∞ = . (B.23)
r
In the following, we shall apply this potential to determine the effect of cur-
vature on solubility. We consider an alloy of α = A(B), where B is solute in
solvent A. At temperature T, B will precipitate out. We consider two pre-
cipitates of B, one larger than the other. The solubility of B surrounding the
large one is less than that surrounding the smaller one. If we take X to be
the solubility, we have

X∞ < Xlarge < Xsmall .

To relate the solubility to Gibbs–Thomson potential, we have the chemical

potential of B as a function of its radius as

2γΩ
μB,r − μB,∞ = , (B.24)
r
where γ is the interfacial energy between the precipitate and the matrix. If
we define the standard state of B as pure B with r = ∞, we have

μB,r = μB,∞ + RT ln aB , (B.25)

where aB is the activity. According to Henry’s law,

aB = kXB , r,

where XB,r is the solubility of B surrounding a precipitate of radius r. At

r = ∞,

μB,∞ = μB,∞ + RT ln aB .
SVNY339-Tu March 3, 2007 7:35

Appendix B 357

This implies that RT ln aB = 0, or aB = 1. So k = 1/XB,∞ . Therefore,

XB,r
μB,r = μB,∞ + RT ln . (B.26)
XB,∞

Hence,

XB,r μB,r − μB,∞ 2γΩ

ln = = .
XB,∞ RT rRT

Or if we consider kT instead of RT, we have

 
2γΩ
XB,r = XB,∞ exp . (B.27)
rkT

B.4 Effect of Curvature on Solubility (Ripening)

The solubility of B around a spherical particle of B of radius r is given by
 
2γΩ
XB,r = XB,∞ exp ,
rkT

where r = ∞, the exponential equals unity. Thus, XB,r goes up when r goes
down. Now we replace XB,r by Cr and XB,∞ by C∞ , which is the equilibrium
concentration on a flat surface. We have
 
2γΩ
Cr = C∞ exp . (B.28)
rkT

If 2γΩ  rkT , we have

 
2γΩ
Cr = C∞ 1 + ,
rkT
2γΩC∞ α
Cr − C∞ = = , (B.29)
rkT r

2γΩ
where α = kT C∞ ,

α
Cr = C∞ + . (B.30)
r
SVNY339-Tu March 3, 2007 7:35

358 Appendix B

Thus, Cr is not a constant, but a function of r. Now we substitute Cr into

the growth equation of

dr ΩD(C0 − Cr )
= ΩJ = .
dt r
We have
dr ΩD  α
= C0 − C∞ − . (B.31)
dt r r
Note that C0 − C∞ > 0 always. We can define a critical radius r∗ such that
α
C0 − C∞ = .
r∗
Then we have
 
dr αΩD 1 1
= − . (B.32)
dt r r∗ r

The parameter r∗ is defined such that

r > r∗ , dr
dt > 0 The particle is growing.
r< r∗ , dr
dt <0 The particle is dissolving.
r = r∗ , dr
dt = 0 The particle is in a state of metastable equilibrium. It
has a concentration C at the interface, or Cr∗ = C.
In ripening, the larger particles grow at the expense of the smaller ones. It
will approach a dynamic equilibrium distribution of size of the particles. The
distribution function can be obtained by solving the continuity equation in
size space. Knowing dr/dt, it is the beginning of the LSW theory of ripening.
SVNY339-Tu March 3, 2007 7:40

Appendix C: Derivation of Huntington’s

Electron Wind Force

In the following we present the assumptions and step-by-step derivation of

Huntington’s model of electron wind force.
(1) Considerations are hemiclassical. Each electron is treated as a group
of waves or Bloch waves with an average wave vector k and group velocity of
V = h̄1 ∂E(k)
∂k
, where the function E(k)should be found from the electron
2 2
band theory (dispersion law). For free electrons, E(k) = h̄2mk∗ , and for elec-
2 2
trons at the bottom of the conduction band, E(k) = Emin + h̄2mk0 , where
2
m∗ = h̄2 ( ∂∂kE2 )−1 is the effective electron mass. We note that ∂E ∂k
means
∂E ∂E ∂E
gradient in k-space, e.g., a vector with components of ∂kx , ∂ky , ∂kz . For
Bloch waves, according to the Bloch theorem, we recall that each quantum
state of independent electron in the periodic potential U (r + R) = U (r) and
R = n1 a1 + n2 a2 + n3 a3 can be described by the product of a planar wave and
periodic function Ψh̄k (r) = eikr Wh̄k (r), where Wh̄k (r + R) = Wh̄k (r) and n is
the band index.
(2) h̄1 ( ∂E − ∂E∂k
) = V  − V is the change of electron’s group velocity as a
∂k
result of scattering.

(3) − mh̄0 ( ∂k
∂E
 −
∂kx ) = −(px − px ) is momentum along the x-axis, transfer
∂E
x
to defect during mentioned individual scattering.
(4) f (k) is the probability that the quantum state k is occupied by some
electron. The quantum cell in k-space with a “k-volume” is given by Ω =
2μ 2μ 2μ 8π 3
Lx · Ly · Lz = V , where V is the real total volume. At equilibrium, we have
1
f0 = E−μ (Fermi–Dirac distribution).
e kT +1
(5) 1 − f (k  ) is a probability that the quantum state k  was free or unoc-
cupied before scattering, so that the Pauli exclusion principle does not forbid
the k → k  transition.
(6) Wd (k → k  ) is a probability of this transition per unit time. It means
that the product Wd dt is a probability of transition during dt, if dt  τd .
SVNY339-Tu March 3, 2007 7:40

360 Appendix C

(7) According to the Pauli principle, each quantum cell in k-space (with
3
Ω = 8πV ) may contain up to two electrons with opposite spins, so the k-volume
3
per electron is Ω2 = 4π
V .
(8) Now we consider unit volume V = 1 m3 .
(9) The number of possible electron states in the “elementary” k-
3 3
volume d3k = dkx dky dkz is dΩk = d4πk3 . The elementary k-volume is physically
2
small.
(10) The momentum, Mx along the x-axis, transferring from electrons to
the defects in the unit volume V = 1 m3 per unit time is given as

d3 k d3 k  
− (p − px )f (k)(1 − f (k  ))Wd (k, k  ).
4π 3 4π 3 x

Or
 2   
dMx 1 m0 ∂E ∂E  ))W (k, k  d3 k  d3 k.
=− − )f (k)(1 − f (k d
dt 4π 3 h̄ ∂kx ∂kx

(11) We shall represent the last equation by two integrals.

dMx
= I1 + I2 ,
dt

where
 2 
1 m0 ∂E
I1 = − f (k)(1 − f (k  ))Wd (k, k  )d3 k  d3 k,
4π 3 h̄ ∂kx
 2 
1 m0 ∂E
I2 = − f (k)(1 − f (k  ))Wd (k, k  )d3 k  d3 k.
4π 3 h̄ ∂kx

Since the integration is being made over all k and all k  , we can interchange
the variables in the first integral as
 2 
1 m0 ∂E
I1 = − f (k  )(1 − f (k))Wd (k  , k)d3 k  d3 k.
4π 3 h̄ ∂kx

∂E
Then in I1 and I2 , we have the same ∂kx , and thus we have

dMx
= (−I2 ) − (−I1 ) (C.1)
dt
 2
1 m0 ∂E  
= f (k )(1 − f (k))Wd (k, k) −f (k )(1 −f (k))Wd (k , k) d3 k d3 k.
4π 3 h̄ ∂kx
SVNY339-Tu March 3, 2007 7:40

Appendix C 361

(12) Huntington showed that to simplify the expression of the last equa-
tion, he used the concept of relaxation time τd . This notion was first in-
troduced for the analysis of the kinetic Boltzmann equation for gases. With
certain approximation, the rate of change of the distribution function can be
represented as

∂f (t, k) 1
= f (k)(1 − f (k  ))Wd (k, k  )
∂t 4π 3

f (t, k) − f (k)
−f (k  )(1 − f (k))Wd (k  , k) d3 k  −
τd

∂f
for the equilibrium distribution. For the stationary case, ∂t = 0, so that

1

f (k)(1 − f (k  ))Wd (k, k  ) − f (k  )(1 − f (k))Wd (k  , k) d3 k 

4π 3
f (t, k) − f (k)
= . (C.2)
τd

In the above equation, f (k)(1 − f (k  ))Wd (k, k  ) is the probability per unit
time of k → k  transition, provided that the state k before transition was
filled and the state k  was empty. The function f (k  )(1 − f (k))Wd (k  , k) is
the probability per unit time of the inverse transition.
(13) By substituting Eq. (C.2) into Eq. (C.1), we have


dMx 1 m0 ∂E(k) f (k) − f0 (k)
= d3 k .
dt 4π 3 h̄ ∂kx τd

(14) Let the relaxation time be independent of k and τd = constant. Then

 
dMx m0 1 ∂E(k) m0 1 ∂E(k)
= d3 k f (k) − d3 k f0 (k).
dt h̄τd 4π 3 ∂kx h̄τd 4π 3 ∂kx

(15) Evidently, the average vector velocity of electrons in equilibrium is

zero:

1 ∂E 1 ∂E 1 ∂E
Vx = |eq = |eq = |eq = 0.
h̄ ∂kx h̄ ∂ky h̄ ∂kz

Therefore,

∂E
f0 (k)d3 k = 0.
∂kx
SVNY339-Tu March 3, 2007 7:40

362 Appendix C

Thus,

dMx m0 1 ∂E
= f (k)d3 k. (C.3)
dt h̄τd 4π 3 ∂kx

(16) To relate the momentum change to force, we have the current density
given as

d3 k 1 ∂E(k)
jx = (−e)nVx = (−e) f (k) · , (C.4)
4π 3 h̄ ∂kx
3
where n = d4πk3 f (k) is the number of electrons per unit volume with k belonging
3
to d3 k. Indeed, d4πk3 is the number of “single electron cells” in the “volume” of
d3 k of k-space, and f (k) is the “inhabitance” of cell.
(17) Combining Eqs. (C.3) and (C.4), we obtain

dMx jx m0
=− . (C.5)
dt eτd

This is a momentum change along the x-direction, transferred to defects (the

diffusing atoms) per unit time per unit volume.
(18) Let Nd be the density of defects (number of defects per unit volume).
Then, according to Newton’s second law, the force at one defect, caused by
electron wind, is

1 dMx jx m0
Fx = =− . (C.6)
Nd dt eτd Nd

This force has a clear physical meaning assuming the condition that during
atomic jump the defect feels much more than one collision. Characteristic time
of one successful jump is of the order of Debye time, τDebye ∼ 10−13 sec. So
for Eq. (C.6) to be reasonable, it is necessary that the product of scattering
frequency, υscatter , and Debye time be much less than unity:

kT VF
νscatter ≈
εp l

where l is the mean free path length of electron around defect, VF /l is the
frequency of “possible” collisions, and kT/εp is the fraction of electrons which
are able to be scattered according to the Pauli principle.
l ≈ nσ
1
, where σ is the cross section and is about 10−19 m2 (according to
Huntington’s estimate).

kT kT h̄kF
n ∼ 1029 m−3 (nex ≈ n ≈ 1027 m−3 ), ≈ 10−2 , VF = ≈ 106 m/ sec .
εp εp m0
SVNY339-Tu March 3, 2007 7:40

Appendix C 363

Thus, νscatter ≈ 10−1 106 nσ ≈ 10−2 106 1029 10−19 ≈ 1014 sec−1 .
So νscatter τDebye ≈ 10  1.

(19) Let us now transform Eq. (C.6) in terms of electric field: jx = ερx ,
where ρ is an average resistance of metal. According to the Drude–Lorentz–
Sommerfeld model, the resistance ρ of a metal can be written as

|m∗ |
ρ= ,
ne2 τ

h̄2
where m∗ = ∂2 E
is the effective electron mass.
∂k2
Huntington used the same expression for the resistance of defects,

|m∗ | |m∗ |
ρd = , so that we have τ d = .
ne2 τd ne2 ρd

Thus, from Eq. (C.6), we obtain

 
εx m0 ne2 ρd m0 ZN ρd
Fx = − =− eεx , (C.7)
ρ eNd |m∗ | |m| Nd ρ

where N is the density of ions and Z is the valence number; n = ZN .

Thus, we have the effective charge

ρd
m0 ZN ρd m0
Q∗ = −Z ∗ e, where Z ∗ = =Z ∗ Nd
ρ .
|m∗ | Nd ρ |m | N

(20) Now, let us take into account the fact that τd , ρd , and Fx change from
position to position. Obviously, they shall reach a maximum at the saddle
point of diffusion.
Assume that F (y) = Fm sin2 ( πy
d ), where y is not the y-axis. Rather it is a
coordinate along the jumping path, which usually does not coincide with the
x-axis. Work or change of potential barrier is

 aj /2  a/2
πy aj Fm
Uj = F (y)dy = Fm cos θj sin2 dy = cos θj .
0 0 a 4

After averaging in all possible jump directions, we have

D 1
Jx = C Fm .
kT 2
SVNY339-Tu March 3, 2007 7:40

364 Appendix C

The factor of 1/2 is due to the integral of

 a/2
πy 1a
sin2 dy = .
0 a 22

(21) Thus, we have finally the effective charge number:

 ρmax

d
∗ 1 ∗ 1 m0 Nd
Zeff = Zmax −Z =Z ρ −1 .
2 2 |m∗ | N
SVNY339-Tu March 2, 2007 20:27
Subject Index
A cakine, 115
accelerated test of Sn whisker
growth, 175
accelerometer, 319
Al electromigration, 212, 215, 216,
222, 224, 225, 226, 229, 230
anisotropic conductor, 235, 238
anisotropic conducting polymer
tapes, 31
Au/Cu/Cr under-bump
metallization, 15, 73, 94
Au/Cu/Cu-Cr, 97, 100
Au Sn4 , 183, 198
AuSnPb ternary phase diagram, 199
B (C-4) solder joint, 14, 97
back stress, 222, 225
back stress build-up, 228
back stress measurement, 229
back stress of electromigration, 222,
226
ball-grid array (BGA) solder balls, 16
ball-limiting metallization (BLM), 15
Burger’s equation, 286
beta-Sn (β-Sn), 153, 235
ball-grid array (BGA) board, 315
Blech structure, 212, 228, 231
bond pad, 9, 23, 24
bronze, 1
bulk diffusion couple of SnPb, 123
C 74
C-4 flip chip technology, 12
Coble creep model, 160
Charpy impact test, 311, 314
channel between scallops, 128, 150
chemical potential, 90, 170, 224
chip-packaging interaction, 30, 124
chip-size packaging, 29
cold joint, 7, 183, 198
composite solder joint, 20, 113, 272,
310, 329
compressive stress gradient, 16, 153,
160, 163
constant volume constraint, 283, 343
consumption rate of Cu, 48, 69
continuity equation, 173
controlled-collapse-chip-connection
Coffin-Manson mode of low cycle
fatigue, 22
copper-tin binary system, 1
copper-tin reaction, 3, 37, 73, 111,
127, 154
creep, 16, 74, 154, 325, 345
critical length of electromigration,
223, 225
critical product of electromigration,
224, 247
(Cu, Ni)6 Sn5, 102, 104, 193
Cu/Ni(V)/Al under-bump
metallization, 73, 100, 251, 273,
275, 329, 335
Cu/Sn room temperature reaction,
Cu-solder-Cu samples for tensile
test, 306
SVNY339-Tu March 2, 2007 20:27
366 Subject Index
Cu column bump, 276
CuSnPb ternary phase diagram, 56
Cu3 Sn, crystal structure, 45, 75, 78
Cu6 Sn5 , crystal structure, 44, 75
Cu6 Sn5 , crystallographic
orientation, 43, 45
Cu6 Sn5 , growth kinetics, 75, 80
Cu6 Sn5 , sequential formation, 81
current crowding, 230, 247, 250
current density gradient force, 230
D 89, 350
daisy chain of flip chip solder joint,
253
Darken’s analysis of interdiffusion,
157, 227
deformation potential, 225
deviatoric strain, 166
dewetting, 96
diffusion barrier, 178, 280
diffusion controlled reaction
(or growth), 63, 84
dilatation strain, 166
direct chip attachment, 17
direct force of electromigration, 218
drift velocity, 212, 220
drop test, 316
dual damascene structure, 214
ductile-to-brittle transition in solder
joint, 25, 306
ductile-to-brittle transition
temperature (DBTT), 312
dynamic equilibrium, 299
E
effective charge number of
electromigration, 220, 257, 364
electroless Ni(P) UBM, 19, 188
electromigration, 25, 119, 176, 212,
245
electromigration in eutectic solder
joint, 249, 255
electron wind force of
electromigration, 217
eutectic effect of phase separation,
119
eutectic SnAgCu, 6, 67
eutectic SnAg, 6, 67
eutectic SnCu, 6, 67
eutectic two-phase structure, 119,
281
F
facetted scallop, 53, 143, 15
flexional vibration, 322
flux divergence, 216
flip chip solder joint, 111, 245
flip chip technology, 12
focused ion beam thinning, 172
free energy of formation of vacancy,
free energy of motion of vacancy, 89,
348
frequency of vibration, 319
G
glancing incidence x-ray diffraction,
74
gradient of chemical potential, 112,
119
gradient of current density, 234
gradient of volume fraction, 112, 119,
282
grain boundary precipitation, 156,
158, 179
H
halo formation, 42
high speed shear, 314, 316
hillock, 155, 292
homogenization, 120, 327
homologous temperature, 215, 228,
305
I
intermetallic compound (IMC), 17,
38, 60, 62, 69, 73, 128, 131, 158,
183, 194, 198, 293, 296, 298, 301
immersion Sn, 11
impact fracture, 25, 316
input/output (I/O) pads, 12
interconnect, 214
interdiffusion coefficient, 62, 87
interfacial reaction coefficient, 89
interfacial reaction-controlled
reaction, 62, 87, 89
interfacial tension of SnPb alloy, 51
interstitial diffusion, 74, 117
SVNY339-Tu March 2, 2007 20:27

Subject Index 367

J (Ni, Cu)3 Sn4 , 193

JEDEC-JESP22-B111 standard of
drop test, 316
JEDEC specification of drop test,
312, 316
joule heating, 26, 211, 254, 270
K
Kirkendall effect of interdiffusion,
159
Kirkendall shift, 159, 285
Kirkendall void, 59, 101, 192, 279,
295
L pin-through-hole technology, 10, 12
lamellar microstructure, 112, 119,
281
Laue pattern, 164
layer-type (layered) morphology, 61,
84, 88, 131, 133
leadframe, 9, 16, 153
line-to-bump geometry, 26, 246, 249
low cycle fatigue, 1, 22
LSW theory of conservative ripening,
137, 148
M ripening, 41, 64, 139
marker displacement, 80, 262, 263
matte Sn, 17, 153
mean-time-to-failure (MTTF), 245,
264
mini Charpy machine, 314
miniaturization, 28
Moire interferometry, 124
mono-size distribution, 135, 137
monochromator, 167
morphological stability of scallops,
128
multi-chip module, 14
multi-level metal-ceramic module, 14
N size distribution function, 141, 148
Nabarro-Herring creep model, 160
near-ideal flip-chip solder joint, 280
NEMI (National Electronics
Manufacturing Initiative), 178
NiSnPb ternary phase diagram, 185
Ni3 P, 188, 192
Ni3 Sn4 , 184, 188
non-conservative ripening, 128, 139
O
Onsager’s reciprocity relations, 170,
225
over-hang, 232
P
pancake-type void, 245, 253, 254, 267
Pb-free solder, 4, 38, 68
PdSnPb ternary phase diagram, 197
phased-in Cu-Cr UBM, 14, 97
phase separation, 121, 248, 281, 333
polarity effect, 289
polarity effect on IMC growth, 293,
297, 301
R
random phase distribution, 282, 286,
334
random state, 282, 334
reflow, 2, 15
reliability of solder joint technology,
16
RMA (mildly activated rosin flux),
38
Rutherford backscattering spectrum
(RBS), 77, 83
S
scallop, 40, 41, 43, 53, 64, 68, 93,
127, 131, 1
scallop-type morphology, 131
Seeman-Bohlin x-ray diffractometer,
74, 92
silicide, 82, 129
single-phase growth, 82, 129
SIP (system-in-packaging), 29
Sn whisker, 16, 153
SOP (system-on-packaging), 29
SOC (system-on-chip), 29
solder finish, 153
solder-less joint, 31
solder bumping, 114
solder cap, 39
SVNY339-Tu March 2, 2007 20:27
368 Subject Index
solder fountain, 10
Soret effect, 119, 327, 343
spalling, 17, 93, 100, 101, 104, 117
spontaneous Sn whisker growth,
153
surface mount technology, 12, 153
stochastic behavior, 343
stochastic tendency, 286
synchrotron radiation, x-ray
micro-diffraction, 43, 163
T W
temperature gradient, 327
theory of non-conservative ripening,
139
thermal mechanical stress, 22, 30,
124
thermal mismatch, 2
thermomigration, 25, 327
thermomigration in eutectic
two-phase alloy, 329, 343
thermomigration in SnPb solder,
335
thermomigration, fundamentals,
338
thin-film under-bump-metalligation,
111
time-dependent melting, 270
torque, 237, 320
two level packaging, 14
U
ultra-low k materials, 30
under-bump metallization (UBM),
14, 60, 7
underfill, 1, 23, 30
V
v-groove, 105, 289, 306
very-large-scale-integration (VLSI),
12, 211
wear-out failure of electromigration,
232
wetting angle, 38, 40, 49, 50
wetting reaction, 2, 3, 38
wetting reaction as a function of
SnPb composition, 48
wetting tip, 43, 105
Wiedemann-Franz law, 271
wire bonding, 9, 12
X
x-ray micro-diffraction, 229
Y
Young’s equation, 49
Z
zincated Al surface, 188
SVNY339-Tu March 3, 2007 8:18
Springer Series in
M AT E R I A L S S C I E N C E
Editors: R. Hull R.M. Osgood, Jr. J. Parisi H. Warlimont
30 Process Technology
for Semiconductor Lasers
Crystal Growth and Microprocesses
By K. Iga and S. Kinoshita
31 Nanostructures and Quantum Effects
By H. Sakaki and H. Noge
32 Nitride Semiconductors and Devices
By H. Morkoç
33 Supercarbon
Synthesis, Properties and Applications
Editors: S. Yoshimura and R. P. H. Chang
34 Computational Materials Design
Editor: T. Saito
35 Macromolecular Science and Engineering
New Aspects
Editor: Y. Tanabe
36 Ceramics
Mechanical Properties,
Failure Behaviour,
Materials Selection
By D. Munz and T. Fett
37 Technology and Applications
of Amorphous Silicon
Editor: R. A. Street
38 Fullerene Polymers and
Fullerene Polymer Composites
Editors: P. C. Eklund and A. M. Rao
39 Semiconducting Silicides
Editor: V. E. Borisenko
40 Reference Materials in
Analytical Chemistry
A Guide for Selection and Use
Editor: A. Zschunke
41 Organic Electronic Materials
Conjugated Polymers and Low
Molecular Weight Organic Solids
Editors: R. Farchioni and G. Grosso
42 Raman Scattering in Materials Science
Editors: W. H. Weber and R. Merlin
43 The Atomistic Nature of Crystal Growth
By B. Mutaftschiev
44 Thermodynamic Basis of Crystal Growth
P–T–X Phase Equilibrium
and Non-Stoichiometry
By J. Greenberg
45 Thermoelectrics
Basic Principles
and New Materials Developments
By G. S. Nolas, J. Sharp,
and H. J. Goldsmid
46 Fundamental Aspects of Silicon Oxidation
Editor: Y. J. Chabal
47 Disorder and Order
in Strongly Nonstoichiometric Compounds
Transition Metal Carbides,
Nitrides and Oxides
By A. I. Gusev, A. A. Rempel,
and A. J. Magerl
48 The Glass Transition
Relaxation Dynamics in Liquids
and Disordered Materials
By E. Donth
49 Alkali Halides
A Handbook of Physical Properties
By D. B. Sirdeshmukh, L. Sirdeshmukh,
and K. G. Subhadra
50 High-Resolution Imaging and Spectrometry
of Materials
Editors: F. Ernst and M. Rühle
51 Point Defects in Semiconductors
and Insulators
Determination of Atomic
and Electronic Structure
from Paramagnetic Hyperfine
Interactions
By J.-M. Spaeth and H. Overhof
52 Polymer Films with
Embedded Metal Nanoparticles
By A. Heilmann
SVNY339-Tu March 3, 2007 8:18
Springer Series in
M AT E R I A L S S C I E N C E
Editors: R. Hull R.M. Osgood, Jr. J. Parisi H. Warlimont
53 Nanocrystalline Ceramics
Synthesis and Structure
By M. Winterer
54 Electronic Structure and Magnetism
of Complex Materials
Editors: D.J. Singh and
D. A. Papaconstantopoulos
55 Quasicrystals
An Introduction to Structure,
Physical Properties and Applications
Editors: J.-B. Suck, M. Schreiber,
and P.Häussler
56 SiO2 in Si Microdevices
By M. Itsumi
57 Radiation Effects in Advanced Semiconductor
Materials and Devices
By C. Claeys and E. Simoen
58 Functional Thin Films
and Functional Materials
New Concepts and Technologies
Editor: D. Shi
59 Dielectric Properties of Porous Media
By S.O. Gladkov
60 Organic Photovoltaics
Concepts and Realization
Editors: C. Brabec, V. Dyakonov, J. Parisi
and N. Sariciftci
61 Fatigue in Ferroelectric Ceramics
and Related Issues
By D.C. Lupascu
62 Epitaxy
Physical Principles and Technical
Implementation
By M.A. Herman, W. Richter, and H. Sitter
63 Fundamentals of Ion-Irradiated Polymers
By D. Fink
64 Morphology Control of Materials
and Nanoparticles
Advanced Materials Processing
and Characterization
Editors: Y. Waseda and A. Muramatsu
65 Transport Processes in
Ion-Irradiated Polymers
By D. Fink
66 Multiphased Ceramic Materials
Processing and Potential
Editors: W.-H. Tuan and J.-K. Guo
67 Nondestructive Materials Characterization
With Applications to Aerospace Materials
Editors: N.G.H. Meyendorf, P.B. Nagy,
and S.I. Rokhlin
68 Diffraction Analysis
of the Microstructure of Materials
Editors: E.J. Mittemeijer and P. Scardi
69 Chemical–Mechanical Planarization
of Semiconductor Materials
Editor: M.R. Oliver
70 Applications of the Isotopic Effect in Solids
By V.G. Plekhanov
71 Dissipative Phenomena
in Condensed Matter
Some Applications
By S. Dattagupta and S. Puri
72 Predictive Simulation of Semiconductor
Processing
Status and Challenges
Editors: J. Dabrowski and E.R. Weber
73 SiC Power Materials
Devices and Applications
Editor: Z.C. Feng
74 Plastic Deformation in Nanocrystalline
Materials
By M.Yu. Gutkin and I.A. Ovid'ko
75 Wafer Bonding
Applications and Technology
Editors: M. Alexe and U. Gösele
76 Spirally Anisotropic Composites
By G.E. Freger, V.N. Kestelman,
and D.G. Freger
77 Impurities Confined in Quantum Structures
By P.O. Holtz and Q.X. Zhao