You are on page 1of 22

PERGAMON

Engineering Fracture Mechanics 61 (1998) 141162

Adhesion and debonding of multi-layer thin lm structures


R.H. Dauskardt a, *, M. Lane a, Q. Ma b, N. Krishna c
a

Department of Materials Science and Engineering, Stanford University, Stanford, CA 94305-2205, U.S.A.
b
INTEL Corporation, Santa Clara, CA 95054, U.S.A.
c
Applied Materials Corporation, Santa Clara, CA 95054, U.S.A.
Received 10 August 1997; accepted in nal form 21 April 1998

Abstract
A fracture mechanics technique to quantitatively measure the adhesion or interfacial fracture
resistance of interfaces in thin lm structures is described. Adhesion values obtained for the
technologically important SiO2/TiN interface in microelectronic interconnect structures are related to a
range of material, mechanical and design parameters which include interface morphology and adjacent
ductile layer thickness. In addition, the interface was shown to be susceptible to environmentally-assisted
subcritical debonding similar to stress corrosion cracking of SiO2 glass in moist air environments.
Subcritical debonding behavior was sensitive to a range of material and design parameters, and is
expected to have important implications for long term device reliability. # 1998 Elsevier Science Ltd.
All rights reserved.
Keywords: Adhesion; Thin lm; Debonding; Stress-corrosion; Interconnect

1. Introduction
The yield and reliability of microelectronic devices containing multi-layer thin lm structures
are strongly inuenced by the adhesion and resistance to subcritical debonding of the many bimaterial interfaces present. Debonding events are driven by residual stresses in the complex
device structure. These include intrinsic growth stresses produced during deposition together
with thermal expansion mismatch stresses. Thermal stresses develop during cooling from
processing temperatures and from thermal cycling during device operation. Experience has
shown that enhanced adhesion and integrity of the constituent interfaces are closely related to
composition and processing conditions. These determine key interfacial parameters, such as
* Author to whom correspondence should be addressed.
0013-7944/98/$ - see front matter # 1998 Elsevier Science Ltd. All rights reserved.
PII: S 0 0 1 3 - 7 9 4 4 ( 9 8 ) 0 0 0 5 2 - 6

142

R.H. Dauskardt et al. / Engineering Fracture Mechanics 61 (1998) 141162

interfacial impurity content, morphology and adjoining microstructures. However, few


techniques are currently available to accurately measure adhesion. In addition, very limited
time-dependent debonding data of a quantitative nature currently exists for thin lm structures.
The paucity of such data and their material and processing determinants currently represent a
serious limitation to the design of complex multi-layered structures with superior structural
integrity and long term reliability.
The objective of the present paper is to report on progress towards characterizing and
understanding the relationship between debond behavior and the underlying structure of thin
lm stacks. We begin by describing a fracture mechanics based technique to accurately
measure adhesion in thin lm structures. The technique is based on a well established interface
fracture mechanics sample geometry containing the interface of interest sandwiched between
two massive elastic substrates [1, 2]. The macroscopic or eective work of fracture is measured
in terms of the critical strain energy release rate, Gc. The technique has been shown to provide
a quantitative and reproducible measure of adhesion [1, 2]. In addition, the technique is ideally
suited to characterizing progressive debonding behavior associated with processes of
environmentally-assisted subcritical debonding [3].
A typical multi-layer metallized conduction line containing the SiO2 interlayer dielectric
(ILD) of interest and representative length scales is shown in Fig. 1(a). Pertinent length
scales are of the order of the various lm thicknesses (typically 02600 nm). Among other
secondary functions, the TiN layer is employed as a barrier material to inhibit diusion of
Al and Cu during high temperature processing as well as to promote adhesion of the ILD
layer. However, debonding of the SiO2/TiN interface leading to delamination of the ILD
layer can occur as revealed by the cross-sectional SEM of a delamination formed under a
nanoindentation site in Fig. 1(b). Such debonding can substantially reduce device yield
during processing or subsequent integrity during service. Despite the importance of this
interface, only limited quantitative adhesion data currently exists together with little
fundamental understanding of the role of salient parameters such as the chemistry and
structure of the interface, the elastic and plastic properties of the adjacent materials, and the
thickness of adjacent layers.

2. Thin lm debonding
2.1. Thin lm adhesion
The general concepts of interface fracture mechanics including details of stress eld
rotation [4, 5], crack path selection [6], and even plasticity eects [79] have been extensively
reported and only brief comments pertinent to the present study are provided here. In
addition, the experimental complexities associated with thin lm adhesion techniques will be
discussed. By adhesion energy, we are primarily concerned with the macroscopic, or eective,
work of fracture per unit area to separate the interface of interest. This may be quantied in
terms of the critical strain energy release rate, Gc, which is a function of both material
properties such as the interface chemistry, adjacent microstructures and elasticplastic
constitutive behavior, and mechanical parameters such as the loading mode mixity near to the

R.H. Dauskardt et al. / Engineering Fracture Mechanics 61 (1998) 141162

143

Fig. 1. A typical multi-layer thin lm interconnect structure showing: (a) a schematic illustration with pertinent
length scales; and (b) a cross-section SEM micrograph of a delamination of the SiO2 ILD lm from the interconnect
stack.

debond tip (ratio of shear to normal stresses). Other design properties including the interface
morphology (roughness) and thickness of adjacent thin lm layers have an important eect on
adhesion and are addressed in the present study.

144

R.H. Dauskardt et al. / Engineering Fracture Mechanics 61 (1998) 141162

Fig. 2. Energy dissipation processes that contribute to the measured interface fracture resistance including the neartip fracture mechanism and a wake dissipation zone.

The interface fracture resistance during debonding is determined by two dierent energy
absorbing processes: the near-tip work of fracture and energy dissipation in a zone surrounding
the debond as schematically shown in Fig. 2. In a region close to the debond crack tip, the
intrinsic interface fracture resistance, Go, provides a direct measure of the bonding energy and
fracture process at the interface. Contributions to Go may arise from actual chemical bonding
and processes associated with the fracture mechanism including debond tip plasticity.
Alternatively, an energy dissipation zone arises from processes which include plasticity of
adjacent ductile layers and interaction of the crack faces behind the debond tip. Interaction
mechanisms may involve frictional sliding of contacting asperities and even plastic stretching of
bridges across the fracture surfaces. In some cases where chemical bonding does not occur
across the interface, these processes provide the only contribution to the macroscopic interface
fracture resistance [10, 11]. Since these energy dissipation mechanisms typically act behind the
debond crack tip, their eect increases with initial debond extension until a steady-state
interface fracture resistance is achieved. While such resistance curve (R-curve) behavior is often
observed during interface failure [10, 12], the length scales associated with thin lm structures

R.H. Dauskardt et al. / Engineering Fracture Mechanics 61 (1998) 141162

145

(roughness, thin lm thickness, etc.) generally precludes the experimental detection of R-curve
eects.
The essential features of a number of specialized techniques that have been developed to
measure adhesion in thin lm structures are shown in Fig. 3. Indentation methods generally
rely on the formation of a dilated plastic zone in the lm to cause the lm to blister [13].
Values of Gc may be related to the indentation volume (or plastic zone size) and extent of
debonding. However, the relationship generally involves a calibration and assumptions about
the complex deformation elds which may extend into underlying layers. The technique and its
various derivatives (e.g. scratch test) typically leads to qualitative results. Alternatively,
techniques have been developed in which the lm is pulled from the substrate. The most well
known of these is the peel test [14], although more recent developments involve the application
of a highly stressed lm [15] or even electrostatic forces [16] to produce the required loading
for debonding. In these techniques, Gc can be accurately related to the forces and moments
transmitted to the thin lm. A liability of these techniques, however, is caused by plastic
bending in the thin lm during debonding which is dicult to separate from the measured
adhesion energy [17]. Finally, the blister test involves debonding the lm by creating a cavity in
the substrate below the lm and causing the lm to bulge by pressurizing the cavity [18]. The
technique has proved successful for some thin lms systems, but is often compromised by the
inherently compliant loading system, chemical interactions between the debond and the
pressurized environment (stress-corrosion cracking), and the etching or machining procedures
needed to produce the cavity.
Perhaps the greatest limitation of all the above techniques, however, is that during
debonding, residual stresses in the thin lm relax and contribute to the debond driving force.
The eects of such relaxation on the measured adhesion values can be large [19], and while
they can, in some cases, be included in the analysis, it is often dicult to accurately measure

Fig. 3. Typical techniques for adhesion testing of thin lms.

146

R.H. Dauskardt et al. / Engineering Fracture Mechanics 61 (1998) 141162

Fig. 4. Schematic illustration of the diusion bonding procedure used to sandwich the multi-layer thin lm stack
between the two silicon substrates.

the residual lm stress, particularly with rigid substrates where substrate curvature techniques
cannot be applied.
Accordingly, the intent of the present study was to develop a fracture mechanics based
adhesion testing technique that would overcome some of the limitations associated with
existing thin lm tests. An essential feature of the technique developed was to sandwich the
thin lm stack between two massive elastic substrates using a diusion bonding procedure. In
the case of the interconnect stack, this simply involves bonding a similar silicon wafer to the
top of the thin lm structure (Fig. 4). Interface fracture mechanics samples may then be
employed to propagate a crack into the thin lm structure. An advantage of the technique is
that during debonding, stress relaxation of the lm is essentially constrained and so does not
contribute to the debond driving force. A small contribution to the measured value of G will
arise from elastic curvature of one of the wafers containing the thin lm after debonding.
However, by considering the change in the strain energy per unit area of the debonded region
for the sandwiched and fully debonded samples, it can be shown that the contribution to the
debond driving energy is [20]:
Gcurvature

3:5 h2f s2f


Es h s

where sf is the lm stress, Es the substrate elastic modulus, and hf and hs the lm and
substrate thickness, respectively. Using typical values of these parameters for the present lm
and substrate system, Gcurvature01.4 mJ/m2, which is generally three orders of magnitude less
than the measured adhesion energy. Another advantage of the sandwiched sample
conguration is that fracture mechanics based tests to characterize subcritical debond-growth
rate behavior associated with environmentally-assisted or fatigue processes are easily facilitated.
These are described in the following section.
Two potential limitations of the sandwiched sample geometry include microstructural and
compositional changes of the thin lm stack associated with the elevated temperature diusion

R.H. Dauskardt et al. / Engineering Fracture Mechanics 61 (1998) 141162

147

bonding process and plastic deformation of the bond layer. For the present thin lm materials
and copper diusion bonding procedure, compositional changes associated with diusional
processes are thought to be minimal since the diusion bonding temperature is typically below
those for a number of the deposition processes. However, the copper bond layer is relatively
weak and energy dissipation by plastic deformation of the bond layer must be considered
during adhesion testing. Estimates of this contribution to the measured adhesion value are
demonstrated to be quite small in subsequent sections.
2.2. Subcritical debondgrowth rate behavior
The previous section was concerned with adhesion values associated with processes of
interface fracture under critical loading conditions. However, thin lm structures may fail by
progressive or time-dependent debonding at stresses far below those required for catastrophic
failure. In these systems, such debonding may be associated with environmentally-assisted
crack growth (stress-corrosion cracking) or even the classic mechanisms of fatigue crack
growth. Debonding adjacent to the SiO2 layer in the present interconnect stacks strongly
suggests the possibility of environmentally-assisted subcritical crack growth. These processes
are well known in a wide range of glass compositions in moist air environments where they are
referred to as slow crack growth (see e.g. Refs [21, 22]). The crack-growth velocity, da/dt, is
typically characterized in terms of the applied stress intensity factor, K, or strain energy release
rate, G, and found to exhibit three distinct regimes of behavior. The controlling mechanisms
for these regions include stress-dependent chemical reaction rates, transport of corrosive
species, and steric hindrance eects, and have been extensively reported (e.g. Refs [2123]),
albeit with no complete description. The classic data of Wiederhorn for subcritical cracking of
SiO2 glass in various moist air environments is shown in Fig. 5.
The stress-corrosion cracking phenomenon described above for bulk glasses provides useful
guidance for the expected debonding behavior along the TiN/SiO2 interface. It should be
noted, however, that for interface cracks, a stress eld rotation is introduced by elastic
discontinuities across the interface [4] and this eect on the mechanisms of subcritical crack
growth remains unexplored. In addition, debond tip blunting associated with plasticity in
adjacent layers may also inuence expected behavior particularly at near threshold growth
rates where steric hindrance of the environmental species limits extension rates. The intent of
the present study was to quantitatively assess the extent of environmentally enhanced stresscorrosion cracking for a typical TiN/SiO2 interface. The eect of plasticity in adjacent ductile
metal layers, which has a marked eect on critical fracture energies, is also reported.

3. Experimental procedures
3.1. Sample preparation
The thin lms stacks shown in Fig. 1 were fabricated on silicon wafers using standard
sputtering (for metal and TiN) and PECVD (for SiO2) deposition processes. AlCu lm
thicknesses were nominally 650 nm, although selected samples were fabricated with thickness in

148

R.H. Dauskardt et al. / Engineering Fracture Mechanics 61 (1998) 141162

Fig. 5. Subcritical crack-growth curves reported for soda-lime glass for a wide range of humid air environments [21].

the range of 0.14 mm. A second silicon wafer was then diusion bonded onto the top of the
thin lm structure as shown schematically in Fig. 4. Copper was used for bonding because it
can be bonded at relatively low temperatures. A thin layer of chromium, to enhance adhesion,
followed by copper was initially evaporated onto the thin lm stack and top wafer. After
bonding, the combined copper layers produced a nominal bond layer thickness of 400 nm,
although other thicknesses were employed to assess the eect of bond layer thickness on
measured adhesion values. The copper surfaces were diusion bonded by applying a pressure
of 12 MPa at 4008C for 4 h in a vacuum press. For the 4-point exure samples described
below, a thin carbon layer was embedded in the copper layer at the center of the beam to serve
as a ``weak interface'' to ensure the deection of the vertical pre-crack to the direction parallel
to the interfaces. At the end of the carbon layer, the crack seeks out a new weak path for
further growth. The mode mixity dictates that the crack kinks downward to the multi-layer
containing the interface of interest [6].
3.2. Critical adhesion tests
In the present study, 4-point exure samples with nominal dimensions of 3 mm width,
1.4 mm thickness and 30 mm length were sectioned from the sandwiched silicon wafers (Fig. 6).
The top wafer was carefully notched to within a few microns of the sandwiched thin lm
structure with a diamond wafering blade. A high-stiness micromechanical test system

R.H. Dauskardt et al. / Engineering Fracture Mechanics 61 (1998) 141162

149

employing a piezo-electric actuator (displacement resolution 020 nm) was used to load the
samples under displacement control with a ramp rate of 0.05 mm/s. Loads were measured to a
resolution of 0.01 N and recorded as a function of the loading pin displacement. The beams
were placed in fully articulating four point bending xtures with adjustable inner and outer
symmetrical loading pin positions. The xture was placed under a high resolution optical
microscope to observe crack initiation and propagation events. Tests were conducted in a
laboratory air environment (R.H. 045%) at 258C.
Similar to conventional fracture mechanics, if plasticity near the debond tip is limited (smallscale yielding), then the ``debond driving force'' can be expressed in terms of the strain energy
release rate, G [4]. For the present samples, plasticity is constrained to the sandwiched thin lm
structure and only the elastic energy stored in the silicon substrates need be considered to
determine G [24]. When the debond has extended suciently far from the vertical pre-crack
(a>2 h), G is independent of the debond length, characteristic of steady-state debond growth.
Applying beam theory, G is given by [25]:


21 1 n 2 M 2
2
G
4E b 2 h 3
where the bending moment M = PL/2, with P being the load and L the spacing between the

Fig. 6. Schematic illustration of: (a) the 4-pt exure sample; and (b) details of the multilayer stack including the
weak layer under the initial notch.

150

R.H. Dauskardt et al. / Engineering Fracture Mechanics 61 (1998) 141162

Fig. 7. A typical load-displacement curve for debonding along a SiO2/TiN interface adjacent to the thin lm
metallized interconnect stack. Features of the curve corresponding to cracking events are indicated.

inner and outer loading lines, b is the beam width, h is the half thickness, and E and n are the
elastic modulus and Poisson's ratio of the bulk substrate, respectively. The phase angle of
loading (ratio of shear to normal stresses) for these beams was 0438. Note that in the absence
of R-curve behavior, the fracture resistance of the interface, Gc, can be determined from the
horizontal plateau in the load-displacement curve as illustrated by the typical loaddisplacement record shown in Fig. 7. Features of the curve related to cracking events are
indicated and described in detail elsewhere [2].
A particular concern during initial development of the testing technique was to assess the
extent of energy dissipation by plastic deformation in the copper bond layer. As noted in
Section 2, plastic deformation of the ductile thin lm structure adjacent to the debond is one
of the principle energy dissipation processes during debonding of the SiO2/TiN interface.
However, a plastic zone in the ductile bond layer might be anticipated to dissipate further
energy due to the proximity of the bond layer to the high stress elds associated with the

Fig. 8. Plastic zones formed in the ductile thin lm metallized stack adjacent to the debond and in the Cu-bond
layer are shown in (a). FEM calculations for the corresponding geometry indicate regions in white that have
exceeded the von Mises yield criteria (b).

R.H. Dauskardt et al. / Engineering Fracture Mechanics 61 (1998) 141162

151

debond tip as shown schematically in Fig. 8(a). A nite element analysis to determine the
extent of yielding and magnitude of plastic strains in the bond layer was therefore conducted
using a commercial FEM package (Fig. (8b)). Results of the analysis are discussed in Section
4.1.
Note, nally, that recent FEM studies of the detailed interface stress elds in mixed-mode
delaminating beam structures with sandwiched layers indicate that the region of K-dominance
is small compared to other fracture mechanics samples [26]. Such K-dominant zones, which are
characterized by asymptotic linear elastic stress eld solutions, are required for interface
fracture toughness values that are independent of the sample conguration. For the present
interconnect structures, however, the very thin metallized layers and similar elastic properties
of the metal and silicon substrates suggest that the sample may be treated as a homogeneous
sample (as anticipated in Eq. (2)). In addition, while yielding may occur in the entire metallized
layer at the debond tip, the resulting interface toughness may be considered a constrained layer
SSY interface toughness, and is independent of specimen geometry [26].
3.3. Subcritical debonding tests
A load relaxation technique using constant displacements was employed to measure crack
growth rates as a function of the debond driving force, G. The basis for the technique is the
linear elastic relationship between the compliance of the four point exure geometry and the
debond length. The general method has been described previously [3, 27] and only a brief
description is presented here. Essentially, it involves loading a sample to a predetermined load,
xing the displacement, and measuring the load relaxation as a function of time (Fig. 9). The
debond extension rate can be related to the load relaxation through the sample compliance

Fig. 9. Schematic loadrelaxation plots showing the relationship between P vs t and debond length a vs t plots.

152

R.H. Dauskardt et al. / Engineering Fracture Mechanics 61 (1998) 141162

relationship as:
da
dP P1 a1

:
dt
dt P 2

The entire debond vs time curve can now be constructed from the load vs time curve (Fig. 9).
Note that this method can easily be extended to include subcritical debonding during initial
loading of the sample when the displacement rate is constant. Tests were conducted in
laboratory air environments (R.H. 0 45%) and at 258C.
4. Results and discussion
4.1. Critical adhesion values
The critical interface adhesion values derived from plateaus on the loaddisplacement
records of approximately 50 tests conducted on nominally identical thin lm interconnect
structures is shown in Fig. 10. The samples were prepared in dierent batches (AE) and tested
on dierent test systems to assess the reproducibility of the adhesion test method. A number of
samples were rejected due to premature failure. The remaining samples reached a load plateau
and subsequent analysis conrmed debonding of the SiO2/TiN interface. A more detailed
statistical analysis showed that within a 95% condence limit all the data formed a single
normal distribution with mean and standard deviation of 10.42 1.3 J/m2. This result indicated
that there was no signicant dierence detected between the two testing systems, and that there

Fig. 10. Critical debond fracture energy values shown for a large number of samples prepared in dierent batches
and tested on two dierent mechanical test systems (light and dark bars).

R.H. Dauskardt et al. / Engineering Fracture Mechanics 61 (1998) 141162

153

was no appreciable dierence between the dierent sample batches or the sequence (time) in
which they were tested.
Further studies, reported elsewhere, conrmed that, due to the presence of a stress corrosion
process, standardized displacement rates during testing should be applied, although the
variation of Gc with small changes in displacement rate or debond length were not large [2]. In
addition, the eect of R.H. in the range of 3070% was also not appreciable. Based on the
present results, it appears that the adhesion technique provides both accurate and reproducible
measures of the SiO2/TiN interface adhesion in the present interconnect thin lm structure.
An outstanding issue, however, relates to the extent of energy dissipation associated with
plastic deformation of the copper bond layer. The results of the FEM analysis of the thin lm
structure and copper bond layer clearly indicate that the stresses in the bond layer during
debonding exceed the von Mises equivalent yield stress (Fig. 8). For the present simulations,
the metallized interconnect structure was treated as a single ductile layer with properties of the
dominant AlCu lm. Estimated values for the elastic modulus and yield stress of the AlCu
interconnect stack and copper bond layer were obtained from nanoindentation studies as 74
GPa and 185 MPa, and 183 GPa and 185 MPa, respectively [28, 29]. Yield properties of thin
metallized lms are dicult to obtain and actual values are likely to be higher, representing
smaller plastic zones than those anticipated from the present analysis. The plastic zone is
predicted to extend a few layer thicknesses ahead of the debond tip in the thin lm stack under
critical loading conditions. However, the average plastic strain in the bond layer plastic zone
was found to be much lower than the corresponding strains in the thin lm stack. An estimate
of the plastic energy partitioned into the two plastic zones was obtained from the following
simple model for the total plastic energy, E pl
total:

pl
pl
pl
pl
4
E total E film E bond
so de AlCu dV
so de pl
bond dV
AlCu zone

e pl
AlCu

bond zone

e pl
bond

where so is the ow stress,


and
the plastic strains in the AlCu thin lm and
copper bond plastic zones, respectively, and dV an innitesimal volume element of the plastic
zones. The model considers only the initial plastic zones and does not include R-curve behavior
during initial debonding [9].
Based on the strains found in the FEM analysis, for a bond layer thickness of 400 nm, a
maximum of 8% of the total plastic energy is dissipated in the bond layer zone, with the
remaining 92% in the thin lm structure. If the bond layer is decreased to 40 nm, the bond
layer plastic energy value decreases to only 0.6% of the total plastic energy. Such calculated
plastic energy dissipation in the bond layer represents a conservative analysis since actual
yielding in the copper thin lm is likely to be signicantly less. In order to compare the
estimates with experimental values of adhesion energy, identical samples were prepared with
bond layer thicknesses of 400 and 100 nm, respectively. Results of the adhesion measurements
are shown in Fig. 11. The predicted trends included in the gure are obtained by assuming that
plastic energy dissipation accounts for all but Go of the measured adhesion value, where Go04
J/m2 (see Fig. 13). Over the bond layer thicknesses examined, it appears that the measured
adhesion values are consistent with values predicted by the FEM calculations. We conclude
that while a small contribution to the measured adhesion value is associated with plastic energy
dissipation in the Cu-bond layer, it is generally expected to be relatively small (<few %).

154

R.H. Dauskardt et al. / Engineering Fracture Mechanics 61 (1998) 141162

Fig. 11. Measured adhesion energy as a function of the Cu-bond layer thickness. Also included are predicted trends
in the adhesion energy based on the plastic energy dissipated in the interconnect stack and bond layer obtained
from FEM simulations.

4.2. Subcritical debondgrowth rate behavior


Extensive debonding of the TiN/SiO2 interface was typically found to extend more than
3 mm during the load relaxation test in 45% R.H. laboratory air at 258C [Fig. 12(a)]. The
sample contained an AlCu layer thickness of 0.65 mm. The corresponding debond-growth
rates were found to span approximately three orders of magnitude, from 0104 to 107 m/s
[Fig. 12(b)]. It is important to note that the debond extension occurs subcritically i.e. below the
critical fracture energy, Gc. At low growth rates (<107 m/s), the applied value of G is less
than 60% of the critical interface fracture energy, Gc=10.6 J/m2. The debond-growth data
could be tted to a power-law expression of the type used for bulk glasses, namely:
da
CG m
dt

where the constant C and the crack-growth exponent m depend on the material and
environmental couple. The value m was found to range from 11 to 18 for a number of samples
tested under nominally identical conditions [3]. Such values are typical of values reported for
subcritical crack-growth behavior of SiO2 glasses in moist air environments [21].
An important consideration for the present subcritical debonding phenomena is the existence
of a debond-growth rate threshold value, GTH, below which debond growth is not observed.
Such thresholds are typically observed during subcritical crack-growth processes in bulk glasses
at growth rates typically below 1010 m/s. The threshold is thought to arise from adsorbate
and corrosion product formation along the crack surfaces which block access of the
environmental species to the crack tip. In the case of glass/metal interface debonding, even
limited plastic deformation of the metal at the debond tip is sucient to expose the strained

R.H. Dauskardt et al. / Engineering Fracture Mechanics 61 (1998) 141162

155

Fig. 12. Typical load relaxation data showing: (a) debond extension; and (b) debond-growth rate, da/dt, as a
function of the applied debond driving force G in laboratory air (R.H. 0 45%) at 258C.

debond tip bonds to more of the environmental species [30]. For the present SiO2/TiN
interface, debond tip blunting associated with plasticity in the adjacent ductile AlCu metal
layers may produce similar eects. As a result, threshold behavior may be eliminated or moved

156

R.H. Dauskardt et al. / Engineering Fracture Mechanics 61 (1998) 141162

Fig. 13. Interface fracture energy as a function of the AlCu layer thickness. Increasing fracture energy with metal
layer thickness is believed to be associated with increasing plasticity in the thicker ductile metal layers.

to lower values of applied G, in both cases, with signicant implications for interconnect
lifetime.

5. Optimizing interfacial adhesion and progressive debonding


With a quantitative and reproducible measure of adhesion available as described in the
preceding section, the eect of a range of material and design parameters can be explored on
both critical interfacial fracture resistance as well as progressive debonding behavior. In the
present paper, the important eect of interface morphology and adjacent AlCu ductile layer
thickness is reviewed. Other parameters including interface chemistry have been reported
elsewhere [2, 3].
5.1. Plasticity in adjacent ductile layer
Energy dissipation by plastic deformation provides a signicant contribution to the
measured interfacial fracture energy; it acts as a potent toughening mechanism. FEM
simulations to determine the extent of the initial plastic zone that forms in the adjacent AlCu
ductile layer under critical loading conditions (i.e. G = Gc) were described in Section 4.1
(Fig. 8). The problem of relating such plastic work to the measured interface fracture energy
and associated R-curve behavior with initial debond extension has been addressed in a number
of recent FEM [8] and theoretical studies [9]. However, understanding the coupling that occurs
between the intrinsic fracture resistance, Go, dictated by the fracture mechanism and energy

R.H. Dauskardt et al. / Engineering Fracture Mechanics 61 (1998) 141162

157

dissipation that occurs in the surrounding process zone from eects like plasticity is an area of
current interest in the mechanics community [31].
The thin lm structures used in the present study oer unique opportunities to address such
issues experimentally. Accordingly, to access the eect of an adjacent ductile layer on interface
fracture resistance, seven thin lm systems processed under identical conditions, diering only
in the thickness of an AlCu layer (0.14.0 mm), were studied. The measured fracture energies
increase markedly with increasing AlCu thickness as shown in Fig. 13. For the thinnest Al
Cu layers, the interface fracture resistance approaches values as low as 4 J/m2, which as
described below, should be close to the intrinsic interface fracture resistance, Go. Note that
some changes in the crack path were observed for thicker AlCu layers. While cracks stayed at
the SiO2/TiN interface in samples with AlCu layers less than 01 mm, they propagated parallel
to the interface at a distance 02050 nm in the SiO2 layer for thicker AlCu samples. This
behavior may be related to changes in the local mode mixity due to plasticity in the ductile
layer which relaxes shear stress near the debond tip [4].
In addition to the FEM simulations for the plastic energy dissipation described in the
present paper, further insights into the possible contributions of plasticity to the measured
adhesion energy can be obtained from the work of Tvergaard and Hutchinson [9]. In their
work, a reference length scale is employed (comparable to the plastic zone size in the bulk
ductile material) which is given by:
Ro

1
EG

 2o
2
s ys
3p 1 n

where sys is the ductile layer yield stress. For the present AlCu layers, Ro024 mm, so that the
ratio hAlCu/Ro is typically much less than 1, where hAlCu is the ductile AlCu layer thickness.
Depending on the extent of work hardening and peak stresses developed in the ductile layer,
steeply rising R-curves are expected to exhibit plateau values of Gc/Go013 for such small
values of hAlCu/Ro. In the absence of more detailed information on the near tip fracture
process and stress elds, this provides at least some bounds on the plastic work contribution to
Gc. We note, in particular, that in the case of the thinnest ductile AlCu layers, yielding is
expected to become increasingly dicult and the peak stresses developed only a few times
higher than sys. As a result Gc/Go 41, and the plastic work contribution to Gc is minimal.
Other experiments reported elsewhere by the authors compared the eect of increasing
hardness of the ductile layer on interface fracture resistance [2]. In nominally identical thin lm
stacks with similar interface structures, increasing the hardness of the adjacent ductile AlCu
layer by incorporation of a TiAl3 reaction layer resulted in marked decreases in the measured
Gc values. The results are consistent with the above observations and were attributed to
reduced plastic energy dissipation in the harder TiAl3 containing AlCu layer.
Environmentally-assisted subcritical debond behavior was also found to be sensitive to the
extent of plastic energy dissipation in the adjacent ductile layer. Subcritical debond-growth
rates as a function of the applied debond driving force, G, are presented in Fig. 14(a). Samples
with thicker AlCu layers were clearly more resistant to subcritical debonding. The entire
debond-growth curves are shifted to higher driving forces for thicker AlCu layers. It was
found that all data could be collapsed onto a single curve when G was normalized by the

158

R.H. Dauskardt et al. / Engineering Fracture Mechanics 61 (1998) 141162

Fig. 14. Subcritical debond-growth rate data showing: (a) the eect of adjacent AlCu layer thickness; and (b) the
single universal curve obtained by normalizing G by the interface fracture resistance, Gc.

R.H. Dauskardt et al. / Engineering Fracture Mechanics 61 (1998) 141162

159

critical interface fracture energy, Gc [Fig. 14(b)]. This leads to a ``universal'' power-law
relationship similar to those found for subcritical crack-growth in toughened ceramics [32]. The
existence of the universal power-law indicates that the individual curves are shifted from one
another based on the extent of plastic dissipation and that the near tip subcritical mechanism
for debond extension are unaected by plasticity in the adjacent layer. This suggests that the
near tip subcritical debond-growth process is essentially ``shielded'' from the applied loads
through energy dissipation by plastic deformation in the AlCu layer. Changes in the
subcritical debond-growth mechanism are typically expected to give rise to a change in slope of
the debond-growth rate curve which was not observed in the present study.
5.2. Interfacial morphology
It is generally accepted that adhesion can be improved by roughening the interface. This was
conrmed in an earlier study for SiO2/metal interfaces [1]. Briey, two types of samples (A and
E) of nominally identical structure (SiO2/TiN/AlCu) were processed under slightly dierent
conditions so that the SiO2/TiN interface in the type-A samples was much rougher than that in
type-E samples (Fig. 15). Interfacial fracture energy measurements showed that the type-A
interface was 050% more fracture resistant than the type-E interface.
The eect of interface morphology has a marked eect on energy dissipation during
debonding, particularly under mixed mode loading conditions. For example, increasing
interface roughness leads to contact and frictional sliding of debond surface asperities behind
the debond tip. A simple model to account for such frictional interaction of debond surfaces
under mixed mode loading has been proposed [33]. Based on the model, increased adhesion of
rougher interfaces was shown to be primarily due to the extent of the frictional contact zone

Fig. 15. Cross-section TEM of the SiO2/TiN interfaces in the type-A and E systems. The interface in the type-A
system is much rougher than that in the type-E system.

160

R.H. Dauskardt et al. / Engineering Fracture Mechanics 61 (1998) 141162

behind the debond tip [34]. In the present interface system, additional contributions to the
fracture resistance of rougher interfaces are associated with the crack path selected during
debonding and the possible formation of bridges spanning the debond surfaces. Evidence for
SiO2 bridges was obtained from careful analytical analyses of the debond surfaces. For the
smooth interface (type-E), Auger spectroscopy revealed no Si on the TiN side of the fracture
surface and no Ti on the SiO2 side, indicating a perfect interfacial fracture. For the rough
interface (type-A), similar Si and Ti analysis suggested that a fraction of the surface area on
the TiN side was covered by SiO2 although the size of the SiO2 patches was too small to be
resolved spatially. However, Ti was still undetectable on the SiO2 side. These observations
suggest that the debond had fractured through regions of SiO2. Further AFM scans of the TiN
surface before SiO2 deposition and that of the TiN fracture surface after debonding suggested
that the fractured SiO2 regions were associated with cavities in the TiN that had been lled
with SiO2. While extensive ``pullout'' of these SiO2 regions has not been observed, the
possibility of crack bridging across the debond by SiO2 bridges remains an issue for future
investigations.

6. Lifetime predictions
Reliability and lifetimes of multilayer interconnect structures are strongly inuenced by
interfacial adhesion and resistance to subcritical debonding. Accordingly, a fracture mechanics
based life prediction strategy may be implemented to estimate the lifetimes of interconnect
structures based on the above observations. The lifetimes of these devices will be sensitive to
residual stress in the thin lms, thermomechanical loads imposed on the structure during
processing and operation, and the initial defect size present in the structure. By tting a powerlaw relationship (Eq. (5)) to the subcritical debonding data and integrating the resulting
equation from some initial defect size to a critical debond length, lifetime predictions can be
obtained. The time to failure for a structure can be expressed as:
!
Em
1
1
m1
7
tf
ac
m 1C Q m p m s 2m a m1
o
where E is the elastic modulus, Q is a geometrical factor, s is the applied stress, ao is the initial
debond length, and ac is the critical debond length for failure. Such lifetime predictions allow
dierent structures to be compared directly and judgments made on which structure is the
most durable and dependable.

7. Conclusions
Techniques have been developed to quantitatively and reproducibly measure the adhesion or
interfacial fracture resistance for weak interfaces in thin lm structures. Macroscopic adhesion
values were obtained for the technologically important SiO2/TiN interface and related to a
range of material, mechanical and design parameters which include interface chemistry,

R.H. Dauskardt et al. / Engineering Fracture Mechanics 61 (1998) 141162

161

morphology and adjacent layer thickness. In addition, the interface was shown to be
susceptible to environmentally-assisted subcritical debonding. Subcritical debond-growth rates
were observed at applied G values less than 60% of the critical fracture resistance, Gc. The
debond-growth rate behavior was found to be similar to stress corrosion cracking of SiO2 glass
systems in moist air environments. Subcritical debond-growth rate behavior was also found to
be sensitive to a range of materials and design parameters. Such subcritical debond-growth
phenomena is expected to have important implications for long term device reliability. The
broader implications of the work, however, are that processing solutions and guidelines for
interface engineering strategies for promoting interfacial adhesion and resistance to subcritical
debonding are emerging.

Acknowledgements
The authors thank Dr Jin Lee of INTEL Corporation for assistance with the FEM analysis.
A Research Incentive Award, Oce of Technology Licensing, Stanford University, and the
INTEL and Applied Materials Corporations provided nancial support for the work.

References
[1] Ma Q, Fujimoto H, Flinn P, Jain V, Adibi-Rizi F, Dauskardt RH, 1995. Quantitative measurement of interface
fracture energy in multi-layer thin lm structures. In: Proc MRS Annual Meeting. San Francisco, CA.
Materials Reliability in Microelectronics V. p. 9196.
[2] Ma Q, Bumgarner J, Fujimoto H, Lane M, Dauskardt RH, 1997. Adhesion measurement of interfaces in multilayer interconnect structures. In: Proc MRS Annual Meeting. San Francisco, CA. Materials Reliability in
Microelectronics VII. p. 314.
[3] Lane M, Dauskardt RH, Ware R, Ma Q, Fujimoto H, 1997. Progressive debonding of multi-layer interconnect
structures. In: Proc MRS Annual Meeting. San Francisco, CA. Materials Reliability in Microelectronics VII.
p. 2126.
[4] Hutchinson JW, Suo Z, 1991. Mixed mode cracking in layered materials. In: Hutchinson JW, Yu TY,
Advances in Applied Mechanics. editors, New York, Academic Press. 63191.
[5] Evans AG, Ruhle M, Dalgleish BJ, Charalambides PG. The fracture energy of bimaterial interfaces. Met Trans
1990;21A:241929.
[6] Fleck NA, Hutchinson JW, Suo Z. Crack path selection in a brittle adhesive layer. Int J Sol Struct
1991;27(13):1683703.
[7] Shih CF. Cracks on bimaterial interfaces: elasticity and plasticity aspects. Mat Sci Eng 1991;A143:7790.
[8] Ozdil F, Carlsson LA. Plastic zone estimates in mode I interlaminar fracture of interleaved composites. Eng
Fract Mech 1992;41(5):64558.
[9] Tvergaard V, Hutchinson JW. Toughness of an interface along a thin ductile layer joining elastic solids. Phil
Mag A 1994;70:64156.
[10] Dauskardt RH, Kook S-Y, Kirtikar A, Ohashi KL, 1997. Adhesion and progressive delamination of polymer/
metal interfaces. In: Srivatsan TS, Soboyejo WO, High Cycle Fatigue of Structural Materials, editors. TMSASM Publication. .
[11] Ohashi KL, Romero AC, McGrowan PW, Maloney WJ, Dauskardt RH. Quantitative assessment of interface
fracture at bone cement/metal prosthetic interfaces in total hip arthroplasties. Trans Orthopedic Res Soc.
1997;23:(in print).

162

R.H. Dauskardt et al. / Engineering Fracture Mechanics 61 (1998) 141162

[12] Kook S-Y, Kirtikar A, Dauskardt RH. Reliability of interfaces in microelectronic packages: adhesion and progressive debonding. J Mat Sci 1997;12:(in print).
[13] Drory MD, Hutchinson JW. Proc Royal Soc A 1996;452:231941.
[14] Thouless MD, Jensen HM. J Adhesion 1992;38:18597.
[15] Bagchi A, Lucas CE, Suo Z, Evans AG. A new procedure for measuring the decohesion energy for thin ductile
lms on substrates. J Mater Res 1994;9(7):173441.
[16] Yang HS, Brotzen FR, Callahan DL, Dunn CF. Rev Sci Inst 1997;68(6):25425.
[17] Jrgensen O, Horsewell A, Srensen BF, 1996. Comment on a new procedure for measuring the decohesion
energy for thin ductile lms on substrates. Bagchi A, et al., editors. J Mater Res. 11(8) 210911.
[18] Jensen HM. The blister test for interface toughness measurement. Eng Fract Mech 1991;40(3):47586.
[19] Jensen HM, Thouless MD. Eects of residual stresses in the blister test. Int J Solids Struct 1993;30(6):77995.
[20] Dauskardt RH, Nix WD, unpublished analysis.
[21] Wiederhorn SM. Inuence of water vapor on crack propagation in soda-lime glass. J Am Ceram Soc
1967;50:40715.
[22] Michalske TA, Freiman SW. A molecular mechanism for stress corrosion in vitreous silica. J Am Ceram Soc
1983;66:28488.
[23] Wiederhorn SM, Freiman SW, Fuller ER, Simmons CJ. Eects of water and other dielectrics on crack growth.
J Mat Sci 1982;17:346078.
[24] Suo Z, Hutchinson JW. On sandwiched test specimens for measuring interface crack toughness. Mat Sci Eng
1989;A107:13543.
[25] Charalambides PG, Lund J, Evans AG, McMeeking RM. A test specimen for determining the fracture resistance of bimaterial interfaces. J Appl Mech 1989;111:7782.
[26] Becker TL, McNaney JM, Cannon RM, Ritchie RO. Limitations on the use of the mixed-mode delamination
beam test specimen: eects of the size of the region of K-dominance. Mech Mat 1997;25:291308.
[27] Ma Q. A four-point bending technique for studying subcritical crack growth in thin lms and at interfaces. J
Mat Res 1997;12:8405.
[28] Vinci RP, Zielinski EM, Bravman JC. Thermal strain and stress in copper thin lms. Thin Solid Films
1995;262:14253.
[29] Nix WD, 1997. Stanford University. private communication.
[30] Card JC, Cannon RM, Dauskardt RH, Ritchie RO, 1993. Stress-corrosion cracking at ceramic-metal interfaces.
In: Carim AH, Schartz DS, Silberglitt RS, Loehman RE, Joining and Adhesion of Advanced Inorganic
Materials, MRS Symposium Proceedings, editors. MRS, Pittsburg, PA. 314. pp. 109116.
[31] Hutchinson JW, 1997. Linking scales in fracture mechanics. In: Karihaloo BL, Mai Y-W, Ripley MI, Ritchie
RO, Proc 9th Int Conf On Fract, Sydney, editors. Pergamon, Oxford. 1. 114.
[32] Dauskardt RH, Marshal DB, Ritchie RO. Cyclic fatigue-crack propagation in Mg-PSZ ceramics. J Am Ceram
Soc 1990;73(4):893903.
[33] Evans AG, Hutchinson JW. Eects of non-planarity on the mixed mode fracture resistance of bimaterial interfaces. Acta Metall 1989;37(3):90916.
[34] Lane M, Ni W, Dauskardt RH, Ma Q, Fujimoto H, Krishna N, 1997. Eects of interface nonplanarity on the
interface fracture energy on the TiN/SiO2 system. In: Thin Film Stresses and Mechanical Properties. Proc MRS
Annual Meeting. Boston, MA. 357362.

You might also like