fig. 0 ••• use of modified accumulated damage model to ... · the accumulated damage at each...
TRANSCRIPT
1
Terasaki et al.: Use of Modified Accumulated Damage Model (1/11)
[Technical Paper]
Use of Modified Accumulated Damage Model to Predict Fatigue
Failure Lives of Sn-Ag-Cu-based Solder Joints in
Ball-Grid-Array-Type PackagesTakeshi Terasaki*, Hisashi Tanie*, Nobuhiko Chiwata**, Motoki Wakano***, and Masaru Fujiyoshi****
*Hitachi Research Laboratory, Hitachi, Ltd., 832-2, Horiguchi, Hitachinaka, Ibaraki 312-0034, Japan
**Corpolate Development Center, Hitachi Metals, Ltd., 1-2-1 Shibaura, Minato-ku, Tokyo 105-8614, Japan
***Technical Department, NEOMAX KAGOSHIMA Co., Ltd., 50-8 Midori, Izumi, Kagoshima 899-0201, Japan
****Metallurgical Research Laboratory, Hitachi Metals, Ltd., 2107-2 Yasugi, Yasugi, Shimane 692-8601, Japan
(Received December 27, 2011; accepted July 12, 2012)
Abstract
We have developed a modified accumulated damage model that can be used to predict fatigue failure lives of solder joints
in electronic devices. Our model calculates the fatigue failure life of solder on the basis of the damage that accumulates
during crack propagation by using a finite element method and corrects for the dependence of element size on the cal-
culated life by using the Hutchinson-Rice-Rosengren singularity theory. We predicted the fatigue lives of conventional and
copper-core solder bump joints in ball-grid-array packages in thermal cycling tests. The good agreement between these
predictions and experimental results indicates that our model can effectively predict fatigue failure life in solder joints.
Keywords: Solder Joint, Fatigue Life, Fatigue Crack Propagation, Modified Accumulated Damage Model, Finite
Element Method, HRR Singularity Theory, Sn-Ag-Cu, Ball-Grid-Array-Type Package
Fig. 0 •••
1. IntroductionThe trend in recent years toward greater compactness
and component mounting density in electronic equipment
has led to solder joints of electronic components that are
smaller and more diverse in shape. The thermal expansion
properties of the solder joints differ from those of the elec-
tronic components and the printed circuit board, so there
is repeated deformation during a change in temperature
owing to the difference in the thermal deformations of the
materials. This repeated deformation results in fatigue
cracking that originates on the surface of the solder and
propagates slowly within the solder until the joint fails.
The fatigue failure life of solder joints has conventionally
been investigated by using the correlation between the
fatigue failure life determined through durability testing
and fatigue-crack-initiation lives predicted by simulation.
[1, 2] However, that approach requires durability testing
for each solder joint size and shape, and the lifetime mea-
surements are time-consuming and expensive. To shorten
the development time and expedite reliability improve-
ments, there is a need for simulation techniques that
enable quantitative prediction of the fatigue failure lives of
various solder joint shapes.
Fatigue crack propagation in solder is measured at the
test-piece level on the basis of nonlinear fracture mechan-
ics,[3, 4] but there are few reports on applying simulation
methods based on nonlinear fracture mechanics to solder
joints because of the difficulty in creating the analysis
model.[5] In recent years, simulation methods for repro-
ducing the fatigue cracking path by evaluating the occur-
rence and development of fatigue cracking on the basis of
the accumulated damage at each location in the solder
joint have been proposed.[6–8]
Although the accumulated damage model that we pro-
posed[7] has the issue that the calculated fatigue crack
propagation life is dependent on the crack tip element size,
room-temperature fatigue crack propagation experiments
with solder center-cracked plate (CCP) test pieces[12, 13]
This paper was published in Journal of Japan Institute of Electronics Packaging, Vol. 14, No. 4 (2011), pp. 287–295 (in Japanese).
2
Transactions of The Japan Institute of Electronics Packaging Vol. 5, No. 1, 2012
show that this problem can be solved by applying a correc-
tion factor calculated according to the Hutchinson-Rice-
Rosengren (HRR) singularity theory.[9–11] We refer to
this as the modified accumulated damage model.
In the work reported here, we tested the validity of the
modified accumulated damage model by comparing the
results for solder joint temperature cycle testing per-
formed on two types of ball grid arrays (BGAs) with simu-
lation results for fatigue failure life, crack propagation
path, and dispersion in the fatigue failure life.
2. Temperature Cycle Testing Method[14]The dimensions and material properties of the BGA
package used in this research are shown in Fig. 1. The sol-
der bumps are arranged with a pitch of 0.5 mm. The pack-
age has a 9 × 9 array of bumps at the center and three rows
of bumps around the outside for a total of 333 bumps. The
BGA package is mounted on one side of an FR-4 board.
The lands on the BGA package have a solder mask defined
(SMD) structure and the lands on the FR-4 board have a
non-SMD (NSMD) structure. The package’s land surfaces
are covered with electrolytic Ni/Au. For the board’s lands,
the combination of Cu and organic solderability preserva-
tive (OSP) is generally used.
Two types of solder joints were evaluated: ordinary sol-
der bumps (conventional bumps) and solder bumps that
have a copper core and excellent resistance to thermal
fatigue (copper-core bumps). The solder composition was
Sn, 3.0 mass% Ag, and 0.5 mass% Cu (Sn-3Ag-0.5Cu) for
both types. The copper cores were coated with a 2-µm Ni
base layer.
The temperature cycle testing was performed in the
temperature range from –40°C to 125°C, with a 20-minute
cycle; 4000 cycles were performed for a set of 15 packages.
The number of cycles to fatigue failure lifetime was defined
as the number of cycles at which the electrical resistance
becomes twice the initial value under continuous monitor-
ing of the resistance.
3. Fatigue Failure Life Prediction Method3.1 BGA solder joint analysis model
To obtain a quantitative fatigue failure lifetime value for
BGA solder joints by using the modified accumulated dam-
age model, it is necessary to obtain the equivalent elastic
strain range in the BGA solder joints. Because the BGA
package used in the temperature cycle testing has 333 sol-
der bumps, a two-step analysis method[15] was used to
both reduce the analysis time and increase the analysis
accuracy. Specifically, we used a global model that has rel-
atively coarse elements for the solder bump to obtain the
solder bumps where the strain is maximum, and then
input the displacement calculated in the vicinity of those
locations to a sub-model for crack propagation analysis.
For the finite element analysis, we used general-purpose
software that was developed in house.
The global finite element method (FEM) model that we
used for conventional bumps is shown in Fig. 2. The con-
ventional bump sub-model is shown in Fig. 3 (a) and the
copper-core bump sub-model is shown in Fig. 3 (b). The
analysis region of the global model was restricted to 1/4 of
the entire model by utilizing symmetry, and 20-node hexa-
hedron elements were used for good accuracy. The sym-
Fig. 1 Cross-sectional diagrams of thermal cycle test pack-age.[14]
Fig. 2 FEM global model.
Fig. 3 FEM sub-model.
3
Terasaki et al.: Use of Modified Accumulated Damage Model (3/11)
metry boundary plane is constrained in the direction nor-
mal to the plane, and the nodes for which the plane of
symmetry and the bottom face of the board cross are con-
strained in the z direction. In the sub-models, the solder
and the copper core are approximated by sets of 10-µm
cubes divided into eight-node hexahedron elements. The
element division of cubes in the crack propagation region
has the major advantage of not requiring the time-consum-
ing redivision of elements as the crack shape changes, but
there is the disadvantage of errors in strain distribution
due to surface roughness. Here, our priority was on com-
putational time rather than accuracy, so we chose the divi-
sion of elements by cubes. The node counts of the sub-
models were 29,781 for the conventional bump and 31,529
for the copper-core bump.
To test the effect of correction for the dependence of
crack propagation life on the element size, crack propaga-
tion life was calculated using a sub-model with 5-µm cube
element division. The sub-model node counts were 205,461
for conventional bumps and 219,456 for copper-core
bumps, about seven times as many nodes as for sub-mod-
els using cubic elements with 10-µm sides.
3.2 Material constants and temperature load con-ditions
The material constants used in the analysis are listed in
Table 1. Materials other than the copper core and the sol-
der were assumed to have isotropic elasticity. The plastic
deformation behavior of the copper core was determined
by a reverse analysis so that the load-displacement rela-
tionship of ball compression testing can be reproduced.
The material constants of the solder used were the values
for Sn-3.5Ag-0.75Cu[2] because there are no measured
values for Sn-3Ag-0.5Cu stress-strain behavior.
Those material constants were obtained by approximat-
ing measurements from torsion testing of thin cylindrical
test pieces (cycling frequency: 0.01 Hz) using a multiple-
step increasing amplitude method[17] (Fig. 4) with an
elasto-plastic linear hardening model. The measured cyclic
stress-strain behavior does not exhibit remarkable work
hardening and softening over the entire temperature
range, so the kinematic hardening rule was used. The
approximate material constant, noted in Fig. 4, does not
have good accuracy for the strain range of 0.5% or less, but
from the constraints of the finite element analysis software
that we used and previous experience,[2, 7, 13, 15, 16] we
Fig. 4 Cyclic stress-strain curves for Sn-Ag-Cu at 0.01Hz.[2]
Table 1 Material properties.
ComponentTempe rature (°C)
Young’s modulus
(GPa)
Poisson’s ratio
Coefficient of thermal expansion (×10–6/°C)
Yield stress (MPa)
Strain hardening coefficient
(MPa)
Chip — 170 0.07 3 — —
DAF — 0.28 0.3 115 — —
Interposer — 31 0.2 13 — —
Resin — 29 0.3 8 — —
Pad — 127 0.3 16 — —
PCB — 25 0.2 14 — —
Solder
–50 41.3
0.35 23
48 1100
–15 40.8 36 900
26 38.2 30 700
75 35.5 22 450
125 33.8 17 250
Cu core — 127 0.3 16 60 900
4
Transactions of The Japan Institute of Electronics Packaging Vol. 5, No. 1, 2012
decided to use it in this research.
Because the solder exhibits strain rate dependence of
the stress-strain behavior and stress relaxation character-
istics, the elasto-plastic creep model[1] or the visco-plastic
model[18] describes the stress-strain behavior with higher
accuracy than the elasto-plastic model does, but the com-
putation time is longer. For that reason, the strain rate
dependence was approximated by using data from the low-
est possible cycling frequency (0.01 Hz) in the testing
range as data that is close to the strain rate that occurs
during the temperature cycle testing. The stress relaxation
characteristic was ignored.
The range of plastic strain that occurs in the solder
bump is obtained as half of the accumulated equivalent
plastic strain that arises in one cycle, given 1.5 cycles of
temperature loading and excluding the initial 0.5 cycles.[1]
To reduce the computation time in the work reported
here, however, we took the equivalent plastic strain that
develops in the solder bump when a change in tempera-
ture from 125°C to –40°C is applied uniformly to the entire
model as the temperature load to be the approximately
equivalent plastic strain range. That approximation pro-
duces error, but we put priority on the computation time.
3.3 Accumulated damage model[7]An overview of the accumulated damage model previ-
ously proposed[7] is presented in Fig. 5. The region of the
analysis object in which cracking is permitted to occur is
divided into elements of equal dimensions and each ele-
ment is assigned a number. Then, function fi, which repre-
sents the accumulated damage to each element, is initial-
ized to 0. This function increases as damage to each
element accumulates, and when it reaches a value of 1,
that element has reached the end of its life (is removed).
Here, the subscript i is the element number.
Next, an elasto-plastic stress analysis that models tem-
perature cycle testing or mechanical load testing is per-
formed, and the equivalent plastic strain that arises in each
element for one cycle of testing, Δεipeq, is computed. Next,
the amount of damage for one cycle, Δ fi, is calculated from
the plastic strain value for each element.
∆
∆f
Cipeqi
p
p
=
ε α
1
(1)
Here, Cp and αp are coefficients of the Coffin-Manson
rule,[19, 20] which shows the relation between the plastic
strain range Δεp and the low cycle fatigue life Nf.
∆ε α
p p fC N p= − (2)
From the obtained Δfi and the accumulated damage
already incurred by each element fi, the number of cycles
needed to reach the end of each element’s life, ΔNi, is cal-
culated using the following equation.
∆
∆N
ffi
i
i= −1
(3)
In the analysis, ΔN is the minimum value of ΔNi because
the finite element analysis should be performed whenever
the crack propagates and the object shape changes. Given
that, the additional damage to each element caused by
increasing the number of cycles by ΔN is given by the fol-
lowing formula.
f f N fi i i→ + ⋅∆ ∆ (4)
The function fi excludes elements that are equal to or
exceed (1-e), and the stress analysis is performed again for
the changed shape. The term e is a constant for eliminat-
ing elements whose accumulated damage value is close to
1 and is introduced to reduce the number of repetitions of
the stress analysis. In the work reported here, an e value of
0.1 was used.
Taking the calculations described above as one step and
repeating that step, we can evaluate the accumulated dam-
age for each element by the linear cumulative damage rule
of Eq. (4),[21] and we can remove elements in order as
they reach the ends of their lives. In this way, the crack
path is calculated automatically and the crack propagation
life for each the crack state is obtained at the same time.
By repeating this procedure from the initial shape up to
failure, we can obtain the failure life.
In this research, the values obtained as the least-squares Fig. 5 Steps in method for predicting solder fatigue life.[7]
5
Terasaki et al.: Use of Modified Accumulated Damage Model (5/11)
approximation from the low-cycle fatigue data for Sn-3.5Ag-
0.75Cu at the upper-limit temperature of the temperature
cycle testing (125°C)[2] (Fig. 6) were used as the coeffi-
cients of the Coffin-Manson relationship of Eq. (2), (Cp =
0.316 and ap = 0.5). The reason for using that approxima-
tion is that there is no low-cycle fatigue data for Sn-3Ag-
0.5Cu. The low-cycle fatigue life was measured by torsion
testing of a thin cylindrical test piece with a cycling fre-
quency of 0.1 Hz in the same way as for the stress-strain
behavior.
The least-squares approximated lifetime curve for 125°C
is shown as the solid line in Fig. 6, accompanied by dotted
lines that indicate twice and half the fatigue lifetimes in
each plastic strain range. The data points for –50°C, room
temperature, and 125°C lie mostly in the region between
the dotted lines, and no remarkable temperature depen-
dence is seen. The Coffin-Manson coefficients used in this
research were calculated from this dispersed measure-
ment data, so the lives predicted below exhibit the same
dispersion.
3.4 Crack propagation life correction method[12, 13]
What requires attention in the accumulated damage
model is that the calculated crack propagation life is
dependent on the element size. The reason for that depen-
dence is a strain singularity in which the strain becomes
abruptly large in the vicinity of the crack tip.
The strain singularity at the crack tip in solder and other
materials that undergo plastic deformation was clarified
theoretically by Hutchinson,[9, 10] Rice, and Rosengren[11]
in the HRR singularity model. That model expresses the
crack tip strain component εij for two-dimensional cracking
of a plane stress state in an infinite region by the following
equation.
ε αεασ ε
θij
nn
nn
ijJ
Ir E=
⋅ ⋅ ( )+ −+
00 0
11 (5)
Here, α, σ0, ε0, and n are material constants that approxi-
mate the nonlinear relationship of the equivalent plastic
strain εpeq and the equivalent stress σeq of the material by
the power law of Eq. (6), J is the J integral value, which is a
nonlinear fracture mechanics parameter, I is the integral
constant, r is the distance from the crack tip in a cylindri-
cal coordinate system, and Eij(θ) is a dimensionless func-
tion that depends on the angle θ in the cylindrical coordi-
nate system.
εε
ασσ
peq eqn
0 0=
(6)
Because the J integral is the intensity factor of the strain
distribution, we know from Eq. (5) that crack tip strain is
proportional to the -n/(n+1) power of the distance r from
the crack tip. Equation (5) holds for the vicinity of the two-
dimensional crack tip in an infinite region and is thus appli-
cable for asymptotic solutions of crack tips for general
crack shapes that include three-dimensional cracking.
Nevertheless, it does hold approximately for any crack
shape when the material deformation behavior is modeled
by Eq. (6).
Our proposed crack propagation life correction method
using the HRR singularity model[13] is shown in Fig. 7.
First, we determine the life NUL taking into account appro-
priate limits for the Coffin-Manson relationship[14, 15] of
Eq. (2) and take the plastic strain range as a suitable value
for the equivalent plastic strain range of the crack tip ele-
ments ε ELcorr (Fig. 7 (a)).
ε αcorrEL
p ULC N p= − (7)
The value of NUL is 25: the same as has been reported pre-
viously.[13]
Because we know from Eq. (5) that the mean equivalent
plastic strain range of the crack tip element ε EL is propor-
tional to the (-n/(n+1)) power of the element size LEL, the
appropriate element size LELcorr for calculating the quantita-
tive crack propagation life can be obtained by using Eq.
(8), as shown in Fig. 7 (b).
L LELcorr
ELcalc
ELcorr
ELcalc
nn
= ⋅
− +
ε
ε
1
(8)
Here, LELcalc is the element dimension used in the calcula-
tions, and ε ELcalc is the mean equivalent plastic strain range
of the crack tip element at the crack length taken to be the
crack life. On the other hand, if we substitute Eq. (5) into
Fig. 6 Fatigue life of Sn-3.5Ag-0.75Cu subjected to torsion testing.[2]
6
Transactions of The Japan Institute of Electronics Packaging Vol. 5, No. 1, 2012
Eq. (2), the crack propagation life N is proportional to the
µ-th power of the element size.[13] The term µ is
expressed by Eq. (9), where the terms µCS and µAD correct
for the effects of crack shape and accumulated damage,
respectively, and are given values of 0.05 and 0.1, respec-
tively.[12]
µα
µ µ=+
⋅ − + +nn p
CS AD11
1 (9)
Accordingly, the quantitative crack propagation life Ncorr
can be calculated from life Ncalc, obtained by calculation,
using Eq. (10) (Fig. 7(c)).
N NLL
corr calc
ELcorr
ELcalc
= ⋅
µ
(10)
In the crack propagation analysis of actual solder joints,
ε ELcalc changes in each step, so at step j of the repeated
damage calculations, the life increase ΔN jcalc is converted
to quantitative crack propagation life increase ΔN jcorr,
using Eq. (11), which summarizes Eqs. (7), (8), and (10).
∆ ∆N NC N
corrj
calcj p UL
ELcalc
nnp
=⋅
− − +
⋅α µ
ε
1
(11)
The quantitative crack propagation life Ncorr is obtained by
summing ΔN jcorr up to the number of steps m that reach
the crack length taken to be the life (Eq. (12)).
N Ncorr corrj
j
m
==∑∆
1 (12)
The material constant n of the power law of Eq. (6) is
obtained by approximating the stress-strain behavior[2] of
Sn-3.5Ag-0.75Cu for each temperature (Table 2). In the
work reported here, the mean of the n values for all tem-
peratures (8.6) was used to correct the calculated life.
Thus, the correction value µ is 0.942.
4. Results of Temperature Cycling Testing[14]The Weibull plot of the fatigue failure life is shown in
Fig. 8. There were 14 packages failures in the conventional
bump type between 1350 and 3600 cycles and 8 packages
failure in the copper core bump type between 2000 and
3300 cycles. The 50% failure probability life was 2460
cycles for the conventional bumps and 3300 cycles for the
copper-core bumps. The failed bumps were all located in
the corners of the BGA package.
5. Results of Fatigue Failure Life Prediction5.1 Crack propagation behavior
The equivalent plastic strain distribution of the solder
joints of each global model for conventional and copper-
Fig. 7 Procedure for fatigue life adjustment.[13]
(a) Decision on upper limit of plastic strain range
(b) Calculation of appropriate element size
(c) Fatigue life correction
Table 2 Values of power-law index n.
Temperature (°C)
–50 –15 26 75 125 Average
n 4.5 4.5 9.1 11.7 13.3 8.6
Fig. 8 Thermal cycle test results.[14]
7
Terasaki et al.: Use of Modified Accumulated Damage Model (7/11)
core bumps is shown in Fig. 9. For both, the maximum
equivalent plastic strain occurred in the solder bumps in
the BGA package corners, which is consistent with the
test results. Therefore, we input the corner solder bump
displacement to the sub-model and calculated the fatigue
crack propagation behavior. The crack propagation behav-
ior is shown in Fig. 10 (a) for the conventional bump and in
Fig. 10 (b) for the copper-core bump; the corrected crack
propagation life is also shown. The black areas in the fig-
ure are surfaces of equal accumulated damage of 0.5,
which is the equivalent of a fatigue crack. The number of
computation steps to failure was 77 for conventional bumps
and 121 for copper-core bumps.
The conventional bumps developed fatigue cracking
near the BGA-side lands that were coated with solder
resist around the margins. Almost no fatigue cracking
developed near the board-side lands, which were not
coated with the solder resist. The fatigue cracking surface
area near the BGA lands occupied about 2/3 of the joint
surface area at about 1300 cycles, and failure was pre-
dicted to occur at about 1600 cycles. The test results show
failures between 1350 and 3600 cycles, confirming that
fatigue failure life can be predicted roughly.
Fatigue cracking in copper-core bumps, on the other
hand, was seen near the BGA lands in addition to the main
fatigue cracking near the board-side lands. The fatigue
cracking surface area near the board lands covered about
half of the joint surface area at about 1300 cycles and
cracking in about 1/3 of the joint surface area was seen
near the BGA lands. After that, the cracking developed
mainly along the copper core interface, with failure pre-
dicted at 4000 cycles. In the testing, 8 out of 15 packages
failed before 4000 cycles, confirming that fatigue failure
life can be predicted roughly for the copper-core bumps as
well. Thus, fatigue cracking in the copper core solder
bumps was not concentrated at a particular joint interface,
but developed at multiple joint interfaces, so we think that
copper core bumps have superior thermal fatigue charac-
teristics.
A cross-section observation of a copper-core bump after
2200 cycles of temperature cycle testing[17] is presented
in Fig. 11 (a). Cracking has developed over most of the
bottom of the copper core (board side), but only half that
amount of cracking is seen at the top (BGA side). The
crack propagation prediction results shown in Fig. 11 (b)
are not perfectly consistent with the cross-section observa-
tion, but they do reproduce well the tendencies, such as
greater crack propagation at the bottom of the copper core
than at the top.
5.2 Effect of correcting for element size depen-dence
The crack propagation life and the crack shapes for each
element size and each bump shape before and after correc-
tion for element size dependence are shown in Fig. 12. The
cracks are represented by the surfaces of equal accumu-
lated damage of 0.5, as described in section 5.1 . The crack
Fig. 9 Equivalent plastic-strain distribution in solder joints.
(a) Conventional solder bump
(b) Cu-cored solder bump
Fig. 10 Crack propagation behavior examples.
(a) Conventional solder bump
(b) Cu-cored solder bump Fig. 11 Crack locations in Cu-cored solder bump after 2200 cycles.
(a) Thermal cycle test (b) Calculation
8
Transactions of The Japan Institute of Electronics Packaging Vol. 5, No. 1, 2012
propagation lives after correction were about 800 cycles
and 1300 cycles.
For both the conventional and copper-core bumps, the
crack propagation was about the same for any element size
if we look at the number of cycles after correction. For the
same crack propagation state, the crack propagation life
before correction differed by a factor of about 1.5, confirm-
ing that the proposed correction method can correctly
compensate for the dependence of the solder bump crack
propagation life on element size.
For the conventional bump, the model with an element
size of 5 µm produced somewhat more crack propagation
on the board side and less crack propagation on BGA side
by the same amount. The balance in crack propagation at
the top and bottom of the solder bump had a large effect
on the fatigue failure life, as shown in the next section, so
using a smaller element should lead to somewhat longer
fatigue failure life predictions.
The corrected element size LELcorr at each step for con-
ventional bumps and copper-core bumps is shown in Fig.
13. The value of LELcorr is close to 1 µm regardless of the
bump or element size. Furthermore, LELcorr is small in the
initial step for either type of bump. It increases as the steps
advance and becomes more or less constant except for the
10-µm conventional bump. As we see from Eq. (8), LELcorr
is determined by ε ELcalc, the mean equivalent plastic strain
range of the crack tip element. Thus, the change in LELcorr
described above means that ε ELcalc is large in the first step
and decreases as the steps advance and thereafter has a
roughly constant value.
5.3 Evaluation of the effects of factors affecting life-time
In crack propagation analysis, a single value is obtained
as the fatigue failure life, but in temperature cycle testing,
Fig. 12 Correction of calculated crack-propagation life.
Fig. 13 Proper element size at each step.
(a) Conventional bump around 800 cycles
(b) Conventional bump around 1350 cycles
(c) Cu-cored bump around 750 cycles
(d) Cu-cored bump around 1350 cycles
(a) Conventional solder Bump
(b) Cu-cored solder Bump
9
Terasaki et al.: Use of Modified Accumulated Damage Model (9/11)
there is much dispersion in the fatigue failure life values
(Fig. 8). The causes of that dispersion, in addition to the
dispersion in the fatigue life of the solder material shown
in Fig. 6, include voids in the solder bumps and dispersion
in the diameters of openings in the solder resist (SR open-
ing diameter) and land diameters, and variations in the vol-
ume of the solder bumps and the amount of soldering
paste supplied. The effects of voids on fatigue failure life
have already been investigated using crack propagation
analysis, and the results agree well with test results.[16]
The effects of SR opening diameter have been investi-
gated experimentally by Matsuzaki et al., who described
them as the main cause of dispersion in fatigue failure
life.[22] Although there are no quantitative measurements
of variations in SR opening diameter, it is empirically possi-
ble that the ordinary deviation from the design value of
φ0.25 mm is ±0.01 mm, or at most ±0.02 mm. Therefore,
we predicted the fatigue failure life using our crack propa-
gation analysis for the conditions described in section 3.2
for the case of ±0.02-mm variation from the design value
for the SR opening diameter of 0.25 mm. The sub-model
element size was 10 µm.
Our predictions for conventional and copper-core bump
fatigue failure lives for the SR opening diameter range
from 0.23 mm to 0.27 mm are presented in Fig. 14. The
error bars in the figure indicate ranges predicted on the
basis of the variation in the solder material fatigue life (the
number of cycles for half and double the fatigue life in
each plastic strain range). The range in which failures
occurred in the temperature cycle testing and the 50% fail-
ure probability life are also noted in the figure. It appears
that the dispersion in the temperature cycle testing results
can be mostly explained by the dispersion in the solder
material fatigue life, but confirmation requires a detailed
fatigue life distribution of the solder material, which is cur-
rently unavailable.
For the conventional bumps, an increase in the SR open-
ing diameter extends the fatigue failure life. In particular,
the fatigue failure life at φ0.27 mm is more than twice that
at φ0.26 mm (5000 cycles or more). The predicted fatigue
failure life obtained for conventional bumps roughly over-
laps the range of the temperature cycle testing results.
The opposite is true for the copper-core bumps. The
fatigue failure life increases as the SR opening diameter
decreases, reaching 5000 cycles or more at φ0.24 mm or
less. The fatigue failure life range obtained for the copper-
core bumps does not cover the interval between 2000 to
3000 cycles, which is often seen in the temperature cycle
testing. We think that this results from the actual bump
shape differing somewhat from the shape used in the anal-
ysis model (Fig. 11). The 50% failure probability life of con-
ventional bumps and copper-core bumps is roughly consis-
tent with the fatigue failure life for SR opening diameters
near φ0.26 mm. SR opening diameter deviations to about
φ0.26 mm were considered, but the effects of other shape
factors must also be studied and taken into account.
The crack shapes at the time of failure for conventional
bumps with diameters ranging from 0.24 mm to 0.26 mm
are shown in Fig. 15. As the SR opening diameter
increased, the amount of crack propagation on the board
NSMD pad side increased. That is the result of higher
bump rigidity on the BGA side for larger SR opening diam-
eters and distribution of the load on the NSMD side, where
rigidity was already high. That load distribution increased
fatigue failure life. The crack shapes at the time of failure
for copper-core bumps ranging from φ0.24 mm to φ0.26
mm are shown in Fig. 16. Copper-core bumps failed on the
board side, and as the SR opening diameter increased,
crack propagation decreased on the BGA side, and the
load concentrated on the board side.
The investigation described above shows that the SR
opening diameter has a large effect on fatigue failure life
and that the predicted fatigue failure life roughly repro-
duces the temperature cycle test results.
(a) Conventional solder bump
(b) Cu-cored solder bump
Fig. 14 Effect of diameters of solder resist opening on fatigue failure life.
10
Transactions of The Japan Institute of Electronics Packaging Vol. 5, No. 1, 2012
6. ConclusionTo reduce the great amount of work required to test sol-
der joint fatigue failure life, we developed a simulation
method (accumulated damage model) that can predict
quantitative fatigue failure life in a short time and verified
the prediction accuracy for two types of BGA solder joints
with temperature cycle testing. This method places greater
priority on a large reduction in computation time than on
prediction accuracy. It uses (a) cube-shaped elements in
the crack propagation region, (b) an elasto-plastic model
that ignores the strain rate dependence of the solder
stress–strain behavior and the stress relaxation character-
istics, and (c) an approximation of the equivalent plastic
strain range; moreover, it assumes that the HRR stress sin-
gularity holds for three-dimensional crack tips. In this
report, we obtained the following conclusions.
(1) Fatigue failure lives predicted by the developed
simulation method agree with BGA solder joint
temperature cycle testing results for both the con-
ventional solder bumps and the copper core solder
bumps.
(2) The proposed life correction method eliminates
the dependence of calculated number of cycles on
element size. Accordingly, a quantitative fatigue
failure life prediction can be made in a short time
with a small analysis model.
(3) Crack-propagation-analysis of the degree of effect
of SR opening diameter, which is considered to be
a cause of dispersion in fatigue failure life results
obtained from temperature cycle testing, showed
that an SR opening diameter variation of ±0.02 mm
can reproduce most of the dispersion in the tem-
perature cycle test results.
References [1] Q. Yu, M. Shiratori, S. Wang, S. Kaneko, and T.
Ishihara, “Analytical and Experimental Hybrid Study
on Thermal Fatigue Strength of Electronic Solder
Joints: 1st Report, Rationalization of Accelerated
Thermal Cyclic Test and Evaluation of Thermal
Fatigue,” Transactions of the Japan Society of
Mechanical Engineers Series A, Vol. 64, No. 619, pp.
550–557, 1998, Mar. (in Japanese).
[2] T. Terasaki, M. Nagano, H. Miura, and T. Nakatsuka,
“Evaluation of Thermal Fatigue Life of Pb-free Solder
Joints for use in Peripheral-Lead-Type LSI Packages”
Proceeding of 7th Symposium on “Microjoining and
Assembly Technology in Electronics,” MATE2001,
pp. 441–446, 2001, Feb. (in Japanese).
[3] Z. Guo and H. Conrad, “Fatigue Crack Growth Rate
in 63Sn37Pb Solder Joints,” Transactions of ASME,
Journal of Electronic Packaging, Vol. 115, pp. 159–
164, 1993.
[4] H. Nose, M. Sakame, M. Yamashita, and K.
Shiokawa, “Crack Propagation Behaviour of Four
Types of Lead and Lead-Free Solders in Push-Pull
Low Cycle Fatigue,” Transactions of the Japan Soci-
ety of Mechanical Engineers Series A, Vol. 68, No.
665, pp. 88–95, 2002, Jan. (in Japanese).
[5] M. Mukai, H. Takahashi, T. Kawakami, K. Takahashi,
K. Kishimoto, and T. Shibuya, “Evaluation of Crack
Propagation for Sn63-Pb37 Solder Bumps,” Transac-
tions of the Japan Society of Mechanical Engineers
Series A, Vol. 67, No. 655, pp. 483–489, 2001, Mar.
(in Japanese).
[6] A. Yasukawa, “Simple Analysis of Thermal Fatigue
Crack Propagation Behavior of Semiconductor-Chip-
Bonding Solder Layers,” Transactions of the Japan
Society of Mechanical Engineers Series A, Vol. 60,
No. 570, pp. 309–316, 1994, Feb. (in Japanese).
[7] H. Tanie and T. Terasaki, “Crack Propagation Model
for Reproduction of Crack Paths in Micro-Solder
Joints,” Transactions of the Japan Society of Mechan-
ical Engineers Series A, Vol. 72, No. 717, pp. 638–
645, 2006, May (in Japanese).
[8] M. Mukai, T. Monda, K. Hirohata, H. Takahashi, T.
Kawakami, and K. Takahashi, “Damage Path Simula-
tion of Solder Joints,” Transactions of the Japan Soci-
ety of Mechanical Engineers Series A, Vol. 72, No.
Fig. 16 Crack shapes in each Cu-cored solder bump at fatigue failure.
(a) 0.25 mm (b) 0.26 mm (c) 0.27 mm
(a) 0.24 mm (b) 0.25 mm (c) 0.26 mm
Fig. 15 Crack shapes in each conventional solder bump at fatigue failure.
11
Terasaki et al.: Use of Modified Accumulated Damage Model (11/11)
721, pp. 1364–1369, 2006, Sep. (in Japanese).
[9] J. W. Hutchinson, “Singular behaviour at the end of a
tensile crack in a hardening material,” Journal of the
Mechanics and Physics of Solids, Vol. 16, Issue 1,
pp. 13–31, 1968, Jan.
[10] J. W. Hutchinson, “Plastic stress and strain fields at a
crack tip,” Journal of the Mechanics and Physics of
Solids, Vol. 16, Issue 5, pp. 337–342, 1968, Sep.
[11] J. R. Rice and G. F. Rosengren, “Planar strain defor-
mation near a crack tip in a power-law hardening
material,” Journal of the Mechanics and Physics of
Solids, Vol. 16, Issue 1, pp. 1–12, 1968, Jan.
[12] T. Terasaki and H. Tanie, “A Method Based on a
Crack Propagation Model for Predicting Solder
Fatigue Life,” Transactions of the Japan Society of
Mechanical Engineers Series A, Vol. 74, No. 740, pp.
574–582, 2008, Apr. (in Japanese).
[13] T. Terasaki and H. Tanie, “Method of Estimating
Fatigue-Crack Propagation Properties Based on
Accumulated Damage Model,” Transactions of the
Japan Society of Mechanical Engineers Series A, Vol.
74, No. 740, pp. 583–591, 2008, Apr. (in Japanese).
[14] M. Wakano, T. Itabashi, N. Chiwata, M. Fujiyoshi,
and H. Tanie, “Solder Joint Reliability of Cu-cored
Solder Sphere with Low-Ag Sn-Ag-Cu Plating, 15th
Symposium on “Microjoining and Assembly Tech-
nology in Electronics,” pp. 47–52, 2009, Feb. (in Japa-
nese).
[15] T. Terasaki, N. Saito, and K. Nagano, “Development
of Fatigue Life Evaluation System for Solder-Joints in
Ball-Grid-Array-type LSI packages,” 10th Symposium
on “Microjoining and Assembly Technology in Elec-
tronics,” pp. 323–328, 2004, Feb. (in Japanese).
[16] T. Terasaki and H. Tanie, “Analysis of Fatigue Crack
Propagation in Micro-solder joints with Dispersed
Voids,” 12th Symposium on “Microjoining and
Assembly Technology in Electronics,” pp. 247–252,
2006, Feb. (in Japanese).
[17] Japan Society of Material Science, “Fatigue Design
Handbook,” pp. 127–129, Yokendo, 1995, Jan. (in Jap-
anese).
[18] M. Kobayashi, M. Mukai, H. Takahashi, T. Ishikawa,
T. Kawakami, and N. Ohno, “Implicit Integration and
Consistent Tangent Modulus of a Time-Dependent
Non-Unified Constitutive Model,” Transactions of the
Japan Society of Mechanical Engineers Series A, Vol.
69, No. 678, pp. 302–309, 2003, Feb. (in Japanese).
[19] L. F. Coffin, Jr, “A study of the effects of cyclic ther-
mal stresses on a ductile metal,” Transactions of
ASME, Vol. 76, pp. 931–950, 1954.
[20] S. S. Manson, “Behavior of Materials under Condi-
tions of Thermal Stress,” NACA Technical Note
2933, 1953.
[21] M. A. Miner, “Cumulative Damage in Fatigue,”
Trans. ASME, Journal of Applied. Mechanics, Vol.
12, No. 3, pp. A159–A164, 1945.
[22] T. Matsuzaki, Q. Yu, T. Shibutani, and T. Matsumoto,
“The Relation between the Thermal Shock Test Life-
time Weibull Distribution Tendency and the Destruc-
tion Mode of the Solder Bump Connection,” Journal
of the Japan Institute of Electronics Packaging, Vol.
12, No. 4, pp. 320–332, 2009, July (in Japanese).
Takeshi TerasakiHisashi Tanie Nobuhiko ChiwataMotoki Wakano Masaru Fujiyoshi