Professional Documents
Culture Documents
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.
King-Ning Tu
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
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
vi Contents
Contents vii
viii Contents
Contents ix
x Contents
Contents xi
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
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
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
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
Introduction 5
Ag3Sn
Cu6Sn5
100μm
6 Chapter 1
Fig. 1.2. SEM image of crack at the corner of SnAgCu solder joint.
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.
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.
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.
Introduction 11
Fig. 1.6. Cross-sectional image of through holes plated with Cu and immersion Sn.
12 Chapter 1
Introduction 13
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
Introduction 17
Whisker
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.
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
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.
Introduction 23
Solder ball
Si chip
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
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.
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.
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.
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)
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.
Introduction 29
30 Chapter 1
Introduction 31
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
Introduction 33
2
Copper–Tin Reactions in Bulk Samples
2.1 Introduction
38 Chapter 2
Fig. 2.1. A solder bead melted and spread out to form a cap on the Cu surface.
Cu6 Sn5
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
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
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
160
140
120
100
80
60
40
20
0
0 1 2 3 4 5 6 7
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.
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:
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
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
|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.
(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.
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.
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
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%
1 dE
= σ + Γ(θ), (2.8)
2πr dr
γγlvlv
510
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%
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.
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
54 Chapter 2
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
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
58 Chapter 2
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.
60 Chapter 2
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.
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!
64 Chapter 2
γΩ2 DC0 ρAΩυ(t)
r3 = + dt,
3NA LRT 4πmNP (t)
dr1 At to
r1
r2
dr2
Δ = Exposed area
Ti or Cr Si Substrate
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
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.
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
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
3
Copper–Tin Reactions in Thin-Film Samples
3.1 Introduction
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.
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).
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
(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
Table 3.2. Reflections of Cu3 Sn found in bimetallic Cu-Sn thin films annealed at
100◦ C
5
Cu Sn Cu Cu6 Sn5 Cu Cu3Sn
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.
80 Chapter 3
24
20
THICKNESS (× 10nm)
2000 Å Sn/ 5600 Å Cu TOTALLY
16 REACTED
12
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.
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.
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.
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.
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
J = Cv, (3.2)
SVNY339-Tu July 12, 2007 15:5
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 .
dxαβ ∂C ∂C
(Cαβ − Cβα ) = Jαβ − Jβα = −D |αβ + D |βα (3.3)
dt ∂x ∂x
∂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
86 Chapter 3
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
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
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
dxβ
= Gβ ΔCβ Kβeff . (3.15)
dt
88 Chapter 3
Aa B AbB Ag B Ad B
xb xg
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
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
ΔG = AΔn. (3.23)
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)
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.
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
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
(a) Flux
Sn ripening
Cu
Cu-Sn compound 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
Molten Solder
rs
rh
Soild Soild
Cr Cr
Hemispherical Cu6Sn5 Cu6Sn5 Spheroid
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. 3.20. 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.
(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.
102 Chapter 3
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.
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
(Cu,Ni)6Sn5
Ni Sn
(b)
(Cu,Ni)6Sn5
Sn
(c)
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.)
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
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
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.
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
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.
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.
118 Chapter 4
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.
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
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
Before aging
80
60
40
20 Before Aging
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.
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
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
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.)
132 Chapter 5
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
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
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
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
(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
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
138 Chapter 5
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
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
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
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.
∞
N (t) = f (t, R)dR, (5.9)
0
∞
m 1
<R >= Rm f (t, R)dR. (5.10)
N
0
∞
free δ
S = 2πRf (t, R)dR. (5.12)
2
0
∞
2 3
Vi = πR f (t, R)dR. (5.13)
3
0
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
∂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
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
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
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
142 Chapter 5
. ϕ (η )versus h
.
.
k
The parameter b = 9(<η2 >−<η> ∼ k ∼ k
2 ) = 0.5535 = <η 3 > .
144 Chapter 5
(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
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
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.
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
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
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.
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
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
(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.
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
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.
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
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).
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.
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
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.
164 Chapter 6
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
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
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
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.
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
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
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 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.
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
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.
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
∂ 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.
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
D Bσ0 D
J =C Xr = . (6.12)
kT kT a
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
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.
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. 6.16. 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.
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
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
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.
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
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.
188 Chapter 7
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).
190 Chapter 7
(b) Ni P (d)
Ni3P
Ni3Sn4
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.
1/2 1/4
12δDi 2(bC 0 − C34 )
y0 = t1/4 , (7.1)
b C34 Dc
SVNY339-Tu April 5, 2007 17:13
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.
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
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.
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.
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
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
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.
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.
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.
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
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
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
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
206 Chapter 7
208 Chapter 7
II
8
Fundamentals of Electromigration
8.1 Introduction
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
(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
Current Crowding
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].
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
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
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
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
where n is the electron density and Nd is the density of diffusing ions. The
electron current density can be written as
−→ 2me j
Fwd = −
eτcol Nd
ρd n
=− eE, (8.6)
Nd ρ
−
→
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
−−→
FEM = Z ∗ eE (8.9)
and
ρd N
Z ∗ = Zel∗ − Z . (8.10)
Nd ρ
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 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.
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,
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
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.
Fig. 8.5. The temperature dependence of Z* calculated for Au is found to agree well
with the measured values.
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
dσΩ
Fme = −∇μ = − , (8.18)
dx
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
ΔσΩ
jΔx = . (8.21)
Z ∗ eρ
dφ
N∗ = − at Je = 0, (8.22)
Ω
dσ
dφ Z ∗ Dρe
=− . (8.23)
dσ kT
−1
The dimensions of dφ/dσ and N * are cm3 /C and C , respectively.
−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
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
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
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.
230 Chapter 8
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]
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.
232 Chapter 8
Over-hang
e-
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
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
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,
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
(a) 0 h
(b) 500 h
236 Chapter 8
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
Therefore, the resulting electrical fields along these two axes are
By rearrangement, we have
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)
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.
δ e- d
c a c
jv More jv More V jv
Compressive Tensile h
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
be different:
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:
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
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
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
9
Electromigration in Flip Chip Solder Joints
9.1 Introduction
246 Chapter 9
Al or Cu
interconnects Solder bumps
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.
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).
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.
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
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
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.
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.
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
256 Chapter 9
1st Cross-Section
a)
2nd Cross-Section
b)
Si
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
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
VEM u
Jem = = , (9.4)
ΩAt Ωt
VEM D
Jem = ≈C· · Z ∗ · e · E. (9.5)
Ω · (A · t) kT
258 Chapter 9
(a) (b)
1st Cross-Section Cu
Underfill
Al
Si
Al
electron flow
2 nd Cross-Section
Cross -Section FR4
Cu
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
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.
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.
262 Chapter 9
e- Mass depletion
0 hr 24 hrs 48 hrs
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.
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
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.
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
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)
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
Δl v b
v= = (JSn ) Ω. (9.14)
Δt d
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
Void
Conducting Line
10-1
Solder Bump
102
C0 4γΩ
d = 2r = . (9.17)
ΔC kT
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.
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
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
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.
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.
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
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
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.
282 Chapter 9
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:
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.
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
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.
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
∂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
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
288 Chapter 9
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.
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
(a)
Si wafer
Cu Solder line Cu
(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.
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
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].
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.
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.
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
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
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].
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
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.
300 Chapter 10
Δμ ∗
= ZCu6 Sn5
ejρCu6 Sn5 . (10.2)
Δx
Δμ
(jΔx)critical = ∗ . (10.3)
ZCu6 Sn5
eρCu6 Sn5
(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
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
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.
Cu Cu
SiO 2
<110> Si
270 m 300 m
~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
e- e-
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.
No underfil
l
Shear
direction
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
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.
312 Chapter 11
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.
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
Electromagnet
Sample
Hammer
Angle Recorder
1 1inch
inch
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.
Guide Rods
PCB Assembly
Stand offs
Drop Table
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
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
320 Chapter 11
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.
F Δt = Δ(mv),
(11.5)
F = −m Δv
Δt = −ma,
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.
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.
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
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
12
Thermomigration
12.1 Introduction
328 Chapter 12
entropy is
1 1 T1 − T2
dSnet = dS1 + dS2 = δQ − + = δQ . (12.1)
T1 T2 T1 T2
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.
330 Chapter 12
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
(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.
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.
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).
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
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.
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
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
Thermomigration 339
F = Z ∗ eE = Z ∗ eρj.
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.
T dS joule
= 1.2 × 109 3 .
V dt m sec
T dS joule
= 3 × 102 3 ,
V dt m sec
T dS joule
= 109 3 ,
V dt m sec
SVNY339-Tu April 5, 2007 17:37
Thermomigration 341
T dS D D
(Z ∗ eρj) .
2
= C F F =C (12.8)
V dt kT kT
T dS joule
= 10 3 .
V dt m sec
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
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.
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
Thermomigration 345
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
v
ΔGm
x
A B
λ
Distance
348 Appendix A
where
ν0 = attempt frequency,
ν = exchange frequency,
ΔGm = saddle point energy (activation energy of motion).
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
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
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
∂f ∂f
=v ,
∂t ∂x
∂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
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
r(Cr − C0 )
Cr = + d. (B.5)
r
(Cr − C0 )r
C(R) = + C0 . (B.6)
R
dC (Cr − C0 )r
Therefore, =− .
dR R2
∂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.
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
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.
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
4π 3 4π 3
r (C0 − C) = r Cp , (B.15)
3 0 3
1/3
C0 − C
r = r0 . (B.16)
Cp
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.
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,
356 Appendix B
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
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
aB = kXB , r,
μB,∞ = μB,∞ + RT ln aB .
SVNY339-Tu March 3, 2007 7:35
Appendix B 357
XB,r
μB,r = μB,∞ + RT ln . (B.26)
XB,∞
Hence,
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
2γΩ
where α = kT C∞ ,
α
Cr = C∞ + . (B.30)
r
SVNY339-Tu March 3, 2007 7:35
358 Appendix B
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
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
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
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
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
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
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
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
D 1
Jx = C Fm .
kT 2
SVNY339-Tu March 3, 2007 7:40
364 Appendix C
Subject Index
A cakine, 115
accelerated test of Sn whisker Charpy impact test, 311, 314
growth, 175 channel between scallops, 128, 150
accelerometer, 319 chemical potential, 90, 170, 224
Al electromigration, 212, 215, 216, chip-packaging interaction, 30, 124
222, 224, 225, 226, 229, 230 chip-size packaging, 29
anisotropic conductor, 235, 238 cold joint, 7, 183, 198
anisotropic conducting polymer composite solder joint, 20, 113, 272,
tapes, 31 310, 329
Au/Cu/Cr under-bump compressive stress gradient, 16, 153,
metallization, 15, 73, 94 160, 163
Au/Cu/Cu-Cr, 97, 100 constant volume constraint, 283, 343
Au Sn4 , 183, 198 consumption rate of Cu, 48, 69
AuSnPb ternary phase diagram, 199 continuity equation, 173
controlled-collapse-chip-connection
B (C-4) solder joint, 14, 97
back stress, 222, 225 Coffin-Manson mode of low cycle
back stress build-up, 228 fatigue, 22
back stress measurement, 229 copper-tin binary system, 1
back stress of electromigration, 222, copper-tin reaction, 3, 37, 73, 111,
226 127, 154
ball-grid array (BGA) solder balls, 16 creep, 16, 74, 154, 325, 345
ball-limiting metallization (BLM), 15 critical length of electromigration,
Burger’s equation, 286 223, 225
beta-Sn (β-Sn), 153, 235 critical product of electromigration,
ball-grid array (BGA) board, 315 224, 247
Blech structure, 212, 228, 231 (Cu, Ni)6 Sn5, 102, 104, 193
bond pad, 9, 23, 24 Cu/Ni(V)/Al under-bump
bronze, 1 metallization, 73, 100, 251, 273,
bulk diffusion couple of SnPb, 123 275, 329, 335
Cu/Sn room temperature reaction,
C 74
C-4 flip chip technology, 12 Cu-solder-Cu samples for tensile
Coble creep model, 160 test, 306
SVNY339-Tu March 2, 2007 20:27
solder fountain, 10 U
Soret effect, 119, 327, 343 ultra-low k materials, 30
spalling, 17, 93, 100, 101, 104, 117 under-bump metallization (UBM),
spontaneous Sn whisker growth, 14, 60, 7
153 underfill, 1, 23, 30
surface mount technology, 12, 153
stochastic behavior, 343 V
stochastic tendency, 286 v-groove, 105, 289, 306
synchrotron radiation, x-ray very-large-scale-integration (VLSI),
micro-diffraction, 43, 163 12, 211
T W
temperature gradient, 327 wear-out failure of electromigration,
theory of non-conservative ripening, 232
139 wetting angle, 38, 40, 49, 50
thermal mechanical stress, 22, 30, wetting reaction, 2, 3, 38
124 wetting reaction as a function of
thermal mismatch, 2 SnPb composition, 48
thermomigration, 25, 327 wetting tip, 43, 105
thermomigration in eutectic Wiedemann-Franz law, 271
two-phase alloy, 329, 343 wire bonding, 9, 12
thermomigration in SnPb solder,
335 X
thermomigration, fundamentals, x-ray micro-diffraction, 229
338
thin-film under-bump-metalligation, Y
111 Young’s equation, 49
time-dependent melting, 270
torque, 237, 320 Z
two level packaging, 14 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
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