Sei sulla pagina 1di 9

496 IEEE TRANSACTIONS ON COMPONENTS, PACKAGING, AND MANUFACTURING TECHNOLOGY—PART A, VOL. 20, NO.

4, DECEMBER 1997

Fatigue Crack Propagation Along Polymer-Metal


Interfaces in Microelectronic Packages
John Guzek, Hamid Azimi, and Subra Suresh

Abstract—In this study, a fracture mechanics-based technique


was used for characterizing fatigue crack propagation (FCP)
at polymer-metal interfaces. Sandwich double-cantilever beam
(DCB) specimens were fabricated using nickel and copper-coated
copper substrates bonded with a thin layer of silica-filled poly-
mer encapsulant. Under cyclic loading, crack propagation was
found to occur at the polymer-metal interface. The interfacial
failure mode was verified by scanning electron microscopy (SEM)
analysis of the fatigue fracture surfaces. The crack growth rate Fig. 1. Schematic drawing showing the components of a plastic pin grid
was found to have a power-law dependence on the strain energy array package (note: not drawn to scale).
release rate range, and exhibited a crack growth threshold,
much like the fatigue crack growth threshold stress intensity
factor range for monolithic bulk metals, polymers, and ceramics. a crack at an interface, and may originate from internal stresses
Interfacial FCP data for three candidate encapsulants predicted due to different thermal expansion coefficients between mate-
cracking resistances that were well correlated with package-level rials, moisture uptake, or defects at the interface. Delamination
reliability tests. By varying the surface roughness of the copper
and nickel plating, it was shown that interfacial FCP resistance can occur at any interface, but it is most commonly observed
increased with increasing roughness. The observed increases in at the polymer-metal interfaces.
FCP resistance were attributed to a reduction in the effective During the development of a new package, it is very
driving force for fatigue fracture along the rougher interfaces, costly and time-consuming to build actual packages, subject
and could be accounted for by a crack-deflection model.
them to environmental stresses, and then perform failure
Index Terms—Delamination, fatigue crack propagation, inter- analysis to determine the extent of delamination. Considering
faces, microelectronic packaging. the number of modifications that can be made to either
interface material, such full-loop experiments quickly become
I. INTRODUCTION impractical. Short-loop experiments that mimic the package
structure are much more feasible and can significantly aid the
A S the products of microelectronics technology move
toward greater levels of integration, increasing function-
ality, and enhanced performance, the demands on microelec-
development process. Therefore, characterization techniques
that quantify interfacial adhesion and correlate to the in-
tronic packaging grow in direct proportion. Packaging now service delamination characteristics of the materials are greatly
plays a fundamental role in allowing products to achieve the needed.
levels of performance and reliability that are necessary to take Although many studies, both experimental and numeri-
full advantage of today’s microelectronics technology. High- cal modeling, have addressed adhesion, delamination, and
performance microprocessor packaging is shifting from the cracking at polymer–metal interfaces found in microelectronic
established ceramic-based technology to newer, organic-based packages [1], [2], only very recently have studies of the
designs. This necessary transition brings with it an array of interfacial crack growth resistance under cyclic loading con-
materials-related challenges that are critical to the delivery of ditions been reported [3]. It should, however, be mentioned
the next generation of microelectronic products. that quantitative analysis of fatigue crack propagation (FCP)
To meet these challenges, modern packages have evolved resistance of adhesive joints for large structural components
into complex structures, composed of a wide variety of materi- have been done successfully within the last ten years, mostly
als. A schematic cross-sectional view of a typical cavity-down for composite materials [4]–[6] and large scale metal/adhesive
plastic pin-grid array package is shown in Fig. 1. One of the bonds [7]–[10] for automotive and aerospace applications.
most common types of failures observed in microelectronic Yet, none of these earlier studies focused on interfacial crack
packages after reliability testing is delamination at an interface. growth. Therefore, very little is known about mechanisms of
Delamination is the result of the formation and propagation of crack propagation at polymer–metal interfaces.
The intention of this study, therefore, is to investigate the
Manuscript received November 7, 1996; revised May 12, 1997. This paper fatigue crack growth processes at polymer/metal interfaces
was supported in part by Intel Corp. and the Office of Naval Research under
Grant N00014-93-1-1277. commonly found in microelectronic packaging. The objectives
J. Guzek and H. Azimi are with Intel Corp., Chandler, AZ 85226 USA. of this study are two-fold.
S. Suresh is with the Department of Materials Science and Engineering,
Massachesetts Institute of Technology, Cambridge, MA 02139 USA. 1) The first objective is to evaluate the feasibility of inter-
Publisher Item Identifier S 1070-9886(97)07002-9. facial fatigue testing as a tool for fatigue life estimation
1070–9886/97$10.00  1997 IEEE
GUZEK et al.: FATIGUE CRACK PROPAGATION ALONG POLYMER-METAL INTERFACES 497

of bi-material interfaces in microelectronic packages.


Interfacial FCP data on three candidate encapsulant
materials were collected, and the results were compared
with full-scale, package-level delamination experimental
results.
2) The second objective is to determine the relationship be-
tween FCP resistance and interfacial roughness. To this
end, specimens were fabricated using a single polymer
encapsulant bonded to copper or nickel substrates, which
were either in an “as-plated or “blast-treated” condition.
The blast treatment induced a rougher interface.
Fig. 2. Schematic drawing of the DCB geometry.
II. CHARACTERIZATION OF INTERFACIAL CRACK GROWTH
The case of a crack at an interface between two different of the adhesive layer. No assumption is made here regarding
materials has received much attention in the literature since the location of the crack (e.g., interfacial versus cohesive).
the work by Williams [11] in 1959. This intense activity However, Ozdil and Carlsson [16] performed a numerical
revolves around developing methods to describe the stress analysis on the DCB geometry comparing both cohesive
state at the crack tip. Due to the inhomogeneity at the bi- and interfacial cracks. They found that the location of the
material interface, the traditional stress intensity factor for crack did not noticeably alter the specimen compliance or the
homogeneous materials cannot be used [12]–[14]. Even when energy release rate as compared to the homogeneous case for
the joint is subjected to purely tensile loads applied normally thin adhesive layers. Therefore, data reduction techniques for
to the crack plane, the elastic mismatch at the interface will the homogenous case were found to be adequate for DCB
induce both tensile and shear stresses around the crack tip. interfacial fracture specimens, and this has been the accepted
Much progress has been made in modeling the interfacial practice.
crack tip stress state; however, there are still many difficulties Using the homogenous solution for interfacial specimens in-
and no universally accepted models exist. Therefore, many troduces minimal error for adhesive layers that are thin relative
workers have chosen to adopt the energy balance approach to the substrate height (i.e., 0.08) [16], but becomes
when studying crack growth at interfaces or in adhesive joints. invalid as the thickness of the adhesive layer is increased.
For an adhesive joint exhibiting bulk linear-elastic behavior, Recently, an exact analytical solution for the interfacial strain
i.e., away from the crack tip region, the strain energy release energy release rate of a tri-layer DCB specimen has become
rate is given by [15] available [17]. The solution is based upon the following
assumptions: linear elasticity (i.e., small deformations), a
(1) fully developed crack across the specimen width, plane stress
conditions, only one interface failing during crack propagation
where is the applied force and is the compliance (i.e., no crack-jumping between interfaces), and negligible
of the structure and is derived from the displacement/load strain energy associated with transverse shear. Strain energy
plot. As long as the adhesive layer is thin compared to the release rate is calculated by determining the potential energy
substrate, the compliance of the structure is not affected by function (stored strain energy minus the work of displacement)
the adhesive layer, i.e., it is equivalent to a homogeneous and differentiating with respect to crack area. For the tri-layer
sample made of the substrate material. Equation (1) forms DCB specimen (Fig. 2), the interfacial strain energy release
the basis for many calculations of . In practice, is found rate is given by
either analytically, experimentally, or numerically and is used
to determine for the specific test specimen geometry.
A common geometry for adhesion tests is the double can-
tilever beam (DCB), shown in Fig. 2. For thin adhesive layers (4)
[15]
where
(2)

where is the tensile modulus of the substrate and and


are the width and height of the substrate beam, respectively.
Combining (1) and (2), the strain energy release rate, as a
function of applied load and crack length, is given by
and
(3)

This expression is identical to the solution for in a ho- Note that and are all constants that depend upon
mogeneous material, since it does not consider the presence specimen geometry and material properties only. Equation (4)
498 IEEE TRANSACTIONS ON COMPONENTS, PACKAGING, AND MANUFACTURING TECHNOLOGY—PART A, VOL. 20, NO. 4, DECEMBER 1997

TABLE I
ENCAPSULANT CHEMISTRY, FILLER LOADING, ELASTIC
MODULUS, AND RECOMMENDED CURE SCHEDULES

solid pellet which was melted into a liquid prior to curing.


1
Fig. 3. Percent error between G calculations using the homogenous solu- All three were highly filled materials, containing 70–80 wt%
tion (3) and the exact solution (4) as a function of dimensionless crack length silica particles. Table I summarizes the basic chemistry, filler
(a=W ).
loading, elastic modulus and recommended cure schedule for
each encapsulant.
is an exact solution for the interfacial strain energy release
rate, and is valid for any adhesive layer thickness. C. Specimen Preparation
Fig. 3 shows the percent error between calculations
The DCB specimens used in this study were fabricated using
using the homogenous solution (3) and the exact solution (4)
the following general method. One substrate was sprayed with
(5) a dry lubricant mold release agent over a potion of one end
to create a precrack along the substrate-polymer interface.
Another substrate was selected, and the two were placed in
as a function of dimensionless crack length ( ). The
a specially designed mold, such that the bonding surfaces
homogeneous solution consistently overestimated the strain
were vertical and facing each other. Two 10 mil (254 m)
energy release rate range. Furthermore, the magnitude of the
shims were used to align the substrates and set the bondline
error was larger for short crack lengths and diminished as
thickness. The substrates were then clamped into place, and the
increased. For all of the data collected in this study, the error
shims were removed. The mold assembly was then transferred
was – %, so that the choice of using the homogeneous or
to a hot plate for adhesive dispensing. The mold was heated
exact solution was not critical. However, if the thickness of
to a temperature of approximately 80 C, as monitored by a
the adhesive layer were increased, one would expect the error
thermocouple. The encapsulant was dispensed from a syringe
to increase, which would necessitate using the exact solution.
along the bond line and flowed into the bond region. Finally,
the mold was transferred to an air-circulating oven, and the
III. MATERIALS AND EXPERIMENTAL METHODS
specimens were cured according to the appropriate cure profile
(see Table I). Once the cure was complete, the specimens
A. Substrates were removed from the mold. Any excess encapsulant was
Nickel- and copper-plated copper beam test specimens were removed with sandpaper. The specimens were then sent to
acquired from the same vendor who manufactures the package an independent contractor for attachment of the Krak-Gage .
heat slugs. The substrate base metal was OFHC copper ( The Krak-Gage is a thin metal foil bonded to the side of the
120 GPa, 260 MPa, / C). The nickel specimen and is used to measure crack length via the “indirect
and copper were electroplated over the copper substrate to a potential drop” method [18]. The Krak-Gage was mounted
nominal plating thickness of approximately three micrometers. such that the mouth of the gage was coincident with the end
To investigate the effect of surface roughness, one lot of of the precrack region.
each type of plated specimens was subjected to a “blast”
treatment after plating to induce a macroscopically rough D. Fatigue Crack Propagation Experiments
surface. Specimen beam dimensions were chosen such that The dry lubricant mold release agent was found to create an
yielding did not occur in the copper during fatigue crack area of weak adhesion, not a true precrack. Therefore, fatigue
growth testing. The double cantilever beam (DCB) specimens precracking was necessary to introduce a sharp, interfacial
were also made excessively thick to reduce beam deflection crack. To facilitate precracking the specimen was notched
to negligible levels. using a diamond-impregnated wire saw (203 m diameter
wire). All encapsulant ahead of the loading axis was removed.
B. Polymer Encapsulants Then, two cuts were made from either side of the bond
Three candidate encapsulants were used in the screening line to produce a Chevron “v-notch” type cross-sectional
portion of this study. The materials were procured through profile. Fatigue Crack Propagation (FCP) tests were conducted
Intel Corp., Chandler, AZ, and will be referred to as mate- on an MTS 858.02 servo-hydraulic test system equipped
rials A, B, and C. Encapsulants B and C were both single with an MTS 458 controller. A Krak-Gage /FRACTOMAT
component liquid systems. Whereas, material A came in a system (Hartrun Corporation, St. Augustine, FL) was used to
GUZEK et al.: FATIGUE CRACK PROPAGATION ALONG POLYMER-METAL INTERFACES 499

Fig. 4. Graph of crack length versus time showing that no crack growth
occurred during a 30 min hold at maximum load.

monitor crack length. The FRACTOMAT controller supplies 2


Fig. 5. SEM micrograph (150 ) showing the crack path along the top
a constant current to the Krak-Gage while monitoring the encapsulant-copper interface.
potential drop across the gage. As the crack propagates through
the specimen and gage, the change in potential drop is related IV. RESULTS
to crack length. The fatigue testing was automated through
the use of a personal computer. All tests were run under load
control in the 245 range of a 2450 load cell. The load A. Failure Mode
signal was monitored via a digital oscilloscope. Since subcritical crack growth can occur under static loading
Fatigue precracking was performed in constant load ampli- conditions, it was necessary to verify that the crack growth
tude ( ) mode under the following conditions: sinusoidal mechanism was associated with the cyclic loading, and not
waveform, frequency of 10 Hz, and ratio of minimum load the maximum stress. A test was interrupted during crack
to maximum load ( ) of 0.1. was gradually increased propagation in the Paris regime and held at maximum load.
periodically until 0.1 mm of crack growth was registered in After 30 min, cyclic loading was resumed. No crack growth
the gage. In an attempt to circumvent possible load interaction occurred during the hold period (see Fig. 4). It was, therefore,
effects, a precracking procedure was performed such that the inferred that subcritical crack growth was not time-dependent
final during precracking was less than the initial and that it was due to the cyclic loading.
of the decreasing portion of the test. Examination of the fractured surfaces via SEM proved
Once stable crack growth was achieved, the test was placed that fatigue crack growth progressed along the polymer-metal
in computer-controlled decreasing mode. The frequency interface. Fig. 5 shows the crack path for a specimen made
and ratio remained unchanged at 10 Hz and 0.1. The soft- from blast-treated copper substrates. Note that crack propa-
ware gradually reduced the load according to an exponential gation occurred along the encapsulant/copper interface. No
load-shedding scheme encapsulant could be detected optically or via SEM on any
of the metal fracture surfaces; therefore all of the failures
(6) were considered to occur at the interface. In summary, it was
verified that all samples failed interfacially due to the action
where instantaneous value of stress intensity factor of a cyclic fatigue mechanism.
range ( instantaneous value of crack
length, initial value of stress intensity range,
initial value of crack length, and normalized stress B. Effect of Polymer Encapsulant Chemistry
intensity gradient ( ), by setting to 0.1 per Interfacial FCP data were successfully collected on the three
mm of crack extension. The crack growth rate ( , where candidate encapsulant materials. The crack growth curves are
is cycles) was calculated according to an incremental shown in Fig. 6. All three material systems exhibited a well-
polynomial procedure. After the crack growth rate decreased defined Paris regime. A linear regression was used to calculate,
to mm/cycle, the test was continued under increasing , the slope of the FCP curve in the Paris regime. Near-
conditions until a crack growth rate of 10 mm/cycle threshold crack growth rates were reached for materials B
was reached, or until the crack had propagated through the and C. was not reached for material A, due to the
gage region. and load data were used to calculate via large amount of testing time in the Paris regime and crack
(4). The same values were calculated for overlapping advance needed to obtain the near-threshold data. However,
values during both the decreasing and increasing based on the trends of the FCP curves, it can be assumed that
portions of the test. This indicates that the nominal materials B and A should have similar thresholds, significantly
is a unique characterization parameter for fatigue crack growth above that of material C. For the purpose of comparison,
along polymer-metal interfaces in this work. was defined as the strain energy release rate range required to
500 IEEE TRANSACTIONS ON COMPONENTS, PACKAGING, AND MANUFACTURING TECHNOLOGY—PART A, VOL. 20, NO. 4, DECEMBER 1997

Fig. 6. Interfacial FCP data for three candidate encapsulants bonded to the Fig. 7. Interfacial FCP data for material B encapsulant bonded to as-plated
as-plated nickel. and blast-treated nickel.

TABLE II
SUMMARY OF KEY INTERFACIAL FCP PARAMETERS FOR
CANDIDATE ENCAPSULANTS ON AS-PLATED NICKEL

propagate a crack at a rate of 10 mm/cycle. All of these key


parameters ( ) are summarized in Table II.
Among the three materials tested, material C clearly exhib-
ited the least resistance to interfacial FCP in terms of both
threshold strain energy release rate range and crack growth
rate sensitivity to changes in driving force ( ). The brittleness
of this system was demonstrated by the relatively large value Fig. 8. Interfacial FCP data for material B encapsulant bonded to as-plated
of the slope in the Paris regime. The entire range of tested and blast-treated copper.
only extended from 2–5 J/m , compared to a range of
4–14 J/m in the case of materials B or A. Materials B and TABLE III
A exhibited similar near-threshold behavior, but material B SUMMARY OF KEY INTERFACIAL FCP PARAMETERS FOR
MATERIAL B ENCAPSULANT ON VARIOUS SUBSTRATES
displayed a much higher slope in the Paris regime. The crack
growth rate in the material A/nickel system was therefore
much less sensitive to changes in . If the linear trend
is extended into the regime of unstable crack growth (as
), material A would exhibit a fracture toughness
approximately twice that of material B.

C. Effect of Substrate Plating Metal and Interface Roughness


Interfacial FCP data were successfully collected on spec-
imens made from material B encapsulant on Ni as-plated, For each plating metal, roughening the surface increased the
Ni blast-treated, Cu as-plated, and Cu blast-treated substrates. overall interfacial fatigue crack propagation resistance, both at
The crack growth curves are shown in Figs. 7 and 8. Again, threshold levels, and at higher crack growth rates. For nickel,
all four material systems exhibited a Paris regime, and a inducing a higher roughness not only increased the ,
fatigue crack propagation threshold. The key FCP parameters but it led to a lower slope in the Paris regime. This would
defined previously were also calculated for these material indicate that the roughness-induced toughening mechanism
systems and are shown in Table III, along with a relative was not confined to the threshold region, but was active in all
roughness parameter, . is defined as the ratio of root mean FCP regimes. For the copper specimens, the rougher interface
square (RMS) surface peak height to average peak spacing as exhibited a much larger , but the difference in growth
determined via laser profilometry. rates at higher levels of became less. That is, the FCP
GUZEK et al.: FATIGUE CRACK PROPAGATION ALONG POLYMER-METAL INTERFACES 501

rates at high growth rates were similar. This indicated that


the roughness-induced toughening mechanism(s) was more
dominant at lower growth rates for the copper systems.

V. DISCUSSION

A. Implications for Service-Life


In the encapsulant comparison, the focus of the study was
to compare three candidate materials in order to identify
the one that would have the best in-service delamination
resistance. The FCP characteristics of these systems have many
implications for the service lives of the interfaces under cyclic
loading. Assuming that a crack-like flaw exists, the service
life of the component is determined by the time required
for the crack to grow to a critical length, at which point it 2
Fig. 9. SEM micrograph (500 ) showing encapsulant fatigue-fracture sur-
face for material A on as-plated nickel substrate.
propagates catastrophically. If the loading conditions at the
crack tip result in a stress state below the FCP threshold, then
the service life is essentially infinite; no failure will occur via B interface, the FCP data indicates that material C should
a fatigue mechanism. If the FCP threshold is exceeded, then have failed completely, which it did. Therefore, the interfacial
the fatigue-life estimate may be based on the integration of FCP data appeared to be a good predictor of delamination
performance of actual package structures. Had the temperature
(6) cycling been extended beyond 1000 cycles, it would have
been interesting to see if the trend predicted by the FCP
Thus, the number of cycles to failure, , is found by data continued for the material B and material A packages.
integration However, due to demands on the reliability testing facilities,
this was not possible.
(7)

where is the initial crack length and is the critical crack B. Shielding Mechanisms
length. , therefore, is highly dependent on , the slope of On the basis of the results obtained from FCP experi-
the FCP curve in the Paris regime. ments and subsequent fractography, one may postulate some
The low threshold of the material C system indicates that mechanisms associated with crack growth at an interface. The
a crack will begin to propagate under loading conditions that mechanisms postulated here include bridging by metal asper-
would not cause crack propagation in either the material B or ities and interfacial crack deflection and premature closure.
A systems. Furthermore, the steep slope in the Paris regime Microscopic analysis of the encapsulant fatigue-fracture
implies that the crack, once propagating, will grow to a critical surfaces was difficult for two reasons. First, the presence
length very rapidly. It was interesting to note that the strain of such a large volume fraction of filler particles on the
energy release rate range necessary to cause unstable crack fractured encapsulant surface tended to obscure any detail. No
growth along the interface between material C and nickel fatigue striations were observed on any of the fatigue-fracture
was on the order of the threshold in both materials A and surfaces. Second, due to the intimate contact of the encapsulant
B. Since the A and B systems had similar thresholds, crack with the metal, the encapsulant fracture surfaces were basically
growth is expected to occur under similar conditions for the “mirror images” of the metal substrate surfaces. Therefore,
two materials. However, since material A exhibits a smaller any detail on the encapsulant fatigue-fracture surface was also
, it should have a longer fatigue life. As seen in Fig. 6, at obscured by the topography of the metal plating.
the necessary for the onset of unstable crack growth in The only interesting feature observed was on the fatigue-
material B, the crack growth rate in material A was still stable fracture encapsulant surface from material A bonded to as-
and on the order of 10 mm/cycle. plated nickel substrate. Metal particles were found embedded
The interfacial FCP results were in very good agreement in the encapsulant surface (see Fig. 9). These embedded par-
with the package-level temperature cycling tests. After 1000 ticles were only observed in the material A specimen. This
cycles, material C was observed to completely delaminate suggested that unlike all the other specimens, where crack
from the nickel heat slug surface, material B exhibited slight propagation involved only debonding at the interface, crack
delamination, and material A showed no delamination. This propagation in this specimen involved some fracturing of
suggested that the loading conditions were severe enough metal asperities embedded in the polymer material. Thus, it is
to induce noticeable crack growth in material B, but not in possible that as the crack front approached and advanced past
material A. Since the FCP results predicted that cracks would these embedded asperities, the asperities could have bridged
grow more slowly at the material A interface, this was a the crack tip, much like fiber-bridging in fiber-reinforced
reasonable result. Furthermore, if the package experienced composites. This would improve the adhesion of this system,
conditions severe enough to cause crack growth at the material which is consistent with the FCP data shown in Fig. 6.
502 IEEE TRANSACTIONS ON COMPONENTS, PACKAGING, AND MANUFACTURING TECHNOLOGY—PART A, VOL. 20, NO. 4, DECEMBER 1997

where is the periodic deflection angle, is the deflected


distance, and is the undeflected distance. Thus, the shielding
effect is minimal when is small, but becomes more
pronounced when gets larger. Recognizing that the
parameter is similar to the relative roughness parameter
, the shielding effect may be seen to increase with .
Qualitatively, this trend was observed in both the Ni and Cu
systems. Suresh [20] also analyzed the effect of closure via
a parameter , which represents the ratio of in-plane sliding
displacement to normal opening displacement. Larger values
Fig. 10. Illustration of the proposed asperity bridging process. of imply greater levels of asperity contact and closure. The
effective stress intensity factor range, when deflection and
closure are present, is

(a)

(9)

Deflections in crack path also cause apparent changes in


crack propagation rates [20]. The crack growth rate of a linear
(b) crack, represented by , may be related to the average
Fig. 11. (a) Schematic drawing of a crack with periodic tilts and (b) propagation rate of a tilted crack, at equal values of by
schematic drawing of a deflected crack in fully opened condition at the peak
load of the fatigue cycle (on the left) and relative mismatch between the
fracture surfaces at the point of first contact during unloading (on the right). (10)

when crack length is measured along the Mode I direction.


Fig. 10 illustrates the hypothesized asperity-bridging mech- Thus, given FCP data for an ideally straight crack (
anism. In effect, the metal asperity could have bridged the
0, 0), (9) and (10) can be used to predict the
crack, remaining intact even after the crack tip had traversed
measured FCP response under various conditions of deflection
past. Eventually, the asperity would deform to the point of
and closure. Among all the material systems tested, the as-
fracture and break. The plastic deformation and subsequent
plated copper had the smoothest interface; hence, it will be
fracture of the metal would absorb much energy during the
considered a “straight” crack for the purpose of this analysis.
crack propagation process, acting as a toughening mechanism.
The predictions based on the data from as-plated copper
Fiber bridging in fiber-reinforced composites has been found
interfaces can be used to analyze the data from the rougher,
to shift the FCP curve to the right toward higher apparent
blast-treated copper specimens. Fig. 12 shows the FCP curves
driving forces, and decrease the slope of the curve in the Paris
for the as-plated and blast-treated copper specimens, along
regime [19]. A similar trend is seen when the FCP curve of
with predicted FCP curves for 0 and 0.05, assuming
material A (asperity bridging) is compared to material B (no
45 and 0.75. For high growth rates, the data
asperity bridging) in Fig. 6.
from the blast-treated specimen approach the prediction for
For both the copper and nickel systems, increasing the
a deflected crack in the absence of closure. At lower growth
interface roughness resulted in improved FCP resistance in
rates, the data approach the prediction for a deflected crack
all crack growth rate regimes. In bulk materials, the crack
with slight closure, given by 0.05. Since the crack-tip
may be deflected from a planar growth profile by impurities,
opening displacements are large at high growth rates, closure is
second phases, or other microstructural features. An interfacial
not expected. However, at near-threshold growth rates, crack-
crack is constrained to propagate along the interface; therefore,
tip opening displacements are very small, and closure is more
any nonplanarity or roughness in the interface will result in
likely. These trends are clearly seen when the experimental
crack-tip shielding.
data is compared to the model predictions. These trends are
Suresh [20] has modeled the effect of crack-tip deflection on
also similar to trends seen for near threshold fatigue crack
fatigue driving force, assuming the geometry shown in Fig. 11.
growth in bulk aluminum alloys [21].
The effective fatigue driving force ( ) can be related to
the applied stress intensity range ( ) through
C. Considerations for Further Research
The work reported in this study demonstrated the usefulness
(8) of a newly developed analytical technique. However, there
are many issues that still need to be explored. This study
GUZEK et al.: FATIGUE CRACK PROPAGATION ALONG POLYMER-METAL INTERFACES 503

3) The experimental technique was able to discriminate be-


tween three encapsulant materials bonded to nickel, and
the FCP resistance of each system correlated to levels
of interfacial delamination observed after package-level
reliability testing. Thus, interfacial FCP resistance was
shown to be a good indicator of material performance
in microelectronic packages which were subjected to
thermal cycling reliability tests.
4) By testing specimens made from substrates with differ-
ent plating metals and different levels of macroscopic
roughness, it was shown that interfacial FCP resis-
tance increased with substrate surface roughness, both
at threshold, and at higher crack growth rates.
Based on the collected fatigue crack propagation data and
subsequent fractography, three interfacial crack-tip shielding
Fig. 12. Predicted FCP rates for deflected cracks as functions of the closure
mechanisms were proposed.
= + =
factor  for  45 and D=D S 0.75 along with data for the as-plated 1) Crack deflection along the interface provided a reduc-
and blast-treated copper-encapsulant interfaces. tion in the effective driving force in all regimes of
crack growth. This was supported by the data showing
focused on external Mode I loading only; however, in actual increasing FCP resistance with interface roughness.
microelectronic packaging structures, Mode II loading may 2) Roughness-induced closure was another source of crack-
also be important. Testing over a range of mode-mixities could tip shielding, especially at low growth rates when crack
help elucidate crack propagation mechanisms and provide for opening displacements were small. This was supported
more accurate simulations of actual loading conditions. Also, by the results showing enhanced FCP resistance at low
all testing was performed at room temperature. The mechanical growth rates for the encapsulant/copper interfaces, where
integrity of polymer-metal interfaces (and microelectronic the difference in interface roughness is greatest between
packages) is very dependent on environmental conditions, the as-plated and blast-treated specimens.
such as temperature and humidity. FCP testing in aggressive 3) A crack-tip shielding mechanism involving crack-
environments would provide information on how temperature bridging by metal asperities was proposed to explain
and humidity affect crack growth, but for this study, it was the enhanced FCP resistance of material A on as-plated
assumed that crack growth will be the same if the local stresses nickel. This notion was supported by the presence of
at the crack tip are due to thermal or mechanical loading. metal particles embedded in the encapsulant fatigue-
Finally, the calculation of the strain energy release rate did fracture surface.
not consider residual stresses in the encapsulant. However, this
REFERENCES
should not affect the comparisons made in these experiments.
As long as the residual stresses are in the linear elastic regime, [1] P. D. Brandendburger, “Mixed-mode fracture of organic chip attachment
then the residual stress field can be superimposed on the adhesives,” M.S. thesis, Lehigh Univ., Bethlehem, PA, 1995.
[2] S. Liu, Y. Mei, and T. Y. Wu, IEEE Trans. Comp., Packag., Manufact.
mechanical loading. Technol., vol. 18, pp. 618–625, 1995.
[3] J. K. Shang, “Interface fatigue-crack growth in layered materials,” in
Proc. 6th Int. Fatigue Congr., Berlin, Germany, May 6–10, 1996, pp.
VI. CONCLUSION 43–54.
[4] S. Mall and K. T. Yun, J. Adhesion, vol. 23, pp. 215–231, 1987.
A test method was developed to study subcritical crack [5] S. Mall and W. S. Johnson, “Debonding characteristics of adhesively
propagation along polymer-metal interfaces using nominally bonded woven kevlar composites,” Adhesively Bonded Joints: Testing,
Mode I double cantilever beam sandwich test specimens. Analysis, and Design, ASTM STP 981. W. S. Johnson, Ed. Philadel-
phia, PA: ASTM, 1988, pp. 194–206.
The results of this experimental work led to the following [6] , “Characterization of mode I and mixed-mode failure of adhesive
conclusions. bonds between composite adherends,” Composite Materials: Testing
and Design (Seventh Conf.), ASTM STP 893. J. M. Whitney, Ed.
1) Cyclic crack growth rate versus strain energy release rate Philadelphia,PA: ASTM, 1986, pp. 322–334.
range ( versus ) results were obtained for fa- [7] D. A. Jablonski, J. Adhesion, vol. 11, pp. 124–133, 1980.
tigue crack propagation along polymer/metal interfaces. [8] R. Joseph et al., J. Adhesion, vol. 41, pp. 169–187, 1993.
[9] A. J. Kinloch and S. O. Osiyemi, J. Adhesion, vol. 43, pp. 79–90, 1993.
The crack growth rate was found to have a power-law [10] D. W. Schmuesser, J. Adhesion, vol. 36, pp. 1–23, 1991.
dependence on , in agreement with relationships [11] M. L. Williams, Bull. Seismol. Soc. America, vol. 49, pp. 199–204, 1959.
[12] A. H. England, J. Appl. Mech., vol. 44, pp. 400–402, 1965.
found for bulk fatigue crack propagation in metals, [13] F. Erdogan, J. Appl. Mech., vol. 44, pp. 403–410, 1965.
ceramics, and polymers. [14] J. R. Rice, J. Appl. Mech., vol. 55, pp. 98–103, 1988.
[15] Mostovoy et al., J. Mater., vol. 2, pp. 661–681, 1967.
2) Fatigue crack propagation was found to occur at levels [16] F. Ozdil and L. A. Carlsson, Engineering Fracture Mech., vol. 41, pp.
orders of magnitude below typical mode I quasistatic 475–485.
fracture energies for polymer/metal interfaces. The poly- [17] F. Gaudette, Ph.D. dissertation, Mass. Inst. Technol., Cambridge, 1996.
[18] H. R. Hartman and R. W. Churchill, “KRAK-GAGE, a new transducer
mer/metal interfaces were also found to exhibit a fatigue for crack growth measurement,” in Proc. SESA Fall Meeting, Keystone,
crack growth threshold ( ). CO, Oct. 1981.
504 IEEE TRANSACTIONS ON COMPONENTS, PACKAGING, AND MANUFACTURING TECHNOLOGY—PART A, VOL. 20, NO. 4, DECEMBER 1997

[19] H. Azimi, “Toughened epoxy polymers: Fatigue crack propagation Subra Suresh received the B.Tech. degree from the Indian Institute of
mechanisms,” Ph.D. dissertation, Lehigh Univ., Bethlehem, PA. Technology, Madras, India in 1977, the M.S. degree from Iowa State Uni-
[20] S. Suresh, Metall. Trans. A, vol. 16A, pp. 249–260, 1985. versity, Ames, IA, in 1979, and the Sc.D. from the Massachusetts Institute of
[21] , Fatigue of Materials. Cambridge, U.K.: Cambridge Univ. Technology (MIT), Cambridge, in 1981.
Press, 1991. He is the R. P. Simmons Professor in the Department of Materials Science
and Engineering and Professor of Mechanical Engineering, MIT. Between
1981 and 1983, he was a Lecturer and Assistant Research Engineer in the
Department of Materials Science and Mineral Engineering at the University
John Guzek received the B.S. and M.S. degrees in of California, Berkeley, and the Lawrence Berkeley Laboratory. He joined the
materials science and engineering from the Mass- faculty at Brown University, Providence, RI, in December 1983, as Assistant
achusetts Institute of Technology, Cambridge, in Professor of Engineering and was promoted to the rank of Associate Professor
1995 and 1996, respectively. of Engineering in 1986, and to Professor of Engineering in 1989. He joined
After graduation, he joined Intel Corporation as MIT in 1993. He is currently a Principal Editor of the international journals
a Package Integration Engineer in the Chandler Acta Metallurgica et Materialia and Scripta Metallurgica et Materialia, an
Assembly/Test facility, Chandler, AZ. He is cur- Associate Editor of Materials Science and Engineering, and a Series Editor
rently working on the integration of next-generation for the Cambridge University Press Solid State Science Series. His current
assembly technologies into a high-volume manufac- research interests focus on quantitative investigations of the microscopic and
turing environment. macroscopic aspects of mechanical behavior of metals, ceramics, thin films,
and composites.
Dr. Suresh is a member of the Executive Committee, Materials Division,
ASME.

Hamid Azimi received the B.S. degree in metallurgical engineering from


Sharif University, Tehran, Iran, and the M.S. and Ph.D. degrees, in materials
science and engineering, from Lehigh University, Bethlehem, PA.
After graduation, he worked as a Post Doctorate for six months and in
March 1995, he joined Intel Corporation, Chandler, AZ. He is currently a
Senior Packaging Engineer in the Assembly Technology Development Group,
Chandler, AZ. His areas of concentration are material characterization, adhe-
sion, fracture, and fatigue of polymer-metal interfaces, and failure analysis of
microelectronic packages. He is also working jointly with the polymer material
suppliers to develop new polymeric materials for packaging applications.