1997 IEMT/IMC Proceedings

THERMAI, FATIGUE LIFE PREDICTION OF SOLDER JOINTS USING STRESS ANALYSIS

Minoru MUKAl and Takashi KAWAKAMI
Research and Development Ccnter, Toshiba Corporation
4- 1 Ukishima-cho, Kawasaki-ku, Kawasaki 210, Japan
I

Kuniaki TAKAHASHI
Omc Works, Toshiba Corporation
2-9, Suehiro-eho, Om-city, Tokyo 108, Japan
Nobuo IWASE
Materials and Devices Devclopmcnt Centcr, Toshiba Corporation
2-4, Suchiro-cho,Tsurumi-ku, Yokohama 230, Japan

ABSTRACT
The fatigue life prediction of solder joints has become one of the most important research areas in
the development of reliable electronic components. In this study. elastic-creep stress analysis proposed in
previous papers was carried out for the solder joints of a recently developed LSI package (ball grid array
package). The results of fatigue life estimation agreed well with those of the actual thermal cycle tests (TCT).
The acceleration factor, which serves as the quantitative correlat,ion between the TCT condition and field
conditions, was determined from the analytical results using the Coffin-Manson equation. It was clear that
temperature hold-time under field conditions h a d a noticeable effect on the fatigue life. Thus, careful
consideration of field conditions i s essential in the reliability design. From the results of this study it w a s
concluded that the fatigue life estimation of solder joints based on the present method is satisfactory for
engineering purposes.
INTRODUCTION Recently, ball gpid array packages (BGA), as
shown in Fig.1, have been widely adopted. In the
LJsing surface mount technology, devices present study, this method using the elastic-creep
such as LSI packages a n d chip components are model was applied to the BGA's solder joints
soldered directly onto the surface of printed circuit (humps). A quantitative correlation between
boards. Solder joints connect the devices thermal cycle tests and the field conditions and
mechanically as well as electronically t o the printed effects of temperature hold-time on fatigue life are
circuit board. Devices mounted on the print,ed examined.
circuit board are subject t o temperature variations
arising as a result of the power switching and ELASTIC-CREEP MODEL OF 63SN-37PB
ambient temperature changes. Because of the
diffexnces in thermal expansion coefficients of the Constructing a suitable material model of
various materials on the printed circuit board, the G3Sn-37Pb solder is the most important issue in
thermal stresses arise under thermal cycling. Since the stress analysis. Solder materials are alloys with
thc solder material i s mechanically the most ductile l o w melting points, and their deformation behavior
part, l o w cycle fatigue phenomenon may occur in is nonlinear even at room temperat,ure. In
t,lir solder joints. Recent trends in LSI package paiticular, their creep behavior, which i s a time-
tlevelopinent indicate a shift toward more solder dependent inelasticity, is significant. The authors
joints and larger package sizes. and temperature have previously proposed a simple elastic-creep
changes under field conditions are also becoming model to express the deformation behavior of
greater. Consequently, fatigue life prediction of G3Sn-37Pb solder, and also used this model in this
solder joints has become one of the most important study. This model is shown in Fig. 2 .
research areas in the development of reliable
electronic components. There have been therefore
inaiiy studies of fatigue life estimation for solder
joints [l],[2]. In previous papers, the authors have c
proposed a fatigue life estimation method using the
clastic-creep model in solder material (63Sn-
3'iPb)[3]-[6]. In these papers. t,his method was
applied t o the solder joints of a quad flat package
(QFP) and a pili grid array packcge (PGA), and the
iwults i w c a l c d thc advantage of this method.

E, : Elastic strain
E, : Creep s t r a i n r a t e
A E, : Creep s t r a i n range
0 : Stress
Printed curcuit board E : Young's modulus
A. n : Material constants
Fig. 1 BGA
Fig. 2 Elastic-creep model

204
The creep behavior is assumed t o be expressed by The temperature histories considered are shown in
Norton's law; Fig. 4; in which the TCT condition and field
conditions are presented. In both conditions,
c,!=Aon (1) temperature-changing periods and hold-times
constituted the temperature histories. The hold-
where E is the equivalent creep strain rate. 0 times of the field conditions were assumed t o be 5
is the equivalent stress, and A and n are material minutes, 1 hour and 12 hours.
constants. The elastic and creep properties of
63Sn-3'iPb solder a t typical temperatures are listed
in Table 1 [3],[7]. Experiniental results of Sn63-
Pb37 solder materials have shown that an elastic-
creep model is adequate for characterizing the creep
behavior of inaterials in the low strain rate range.
In this study, this elastic-creep model was employed
for solder material in all numerical analyses as
follon~s.
Table 1 Material properties of SnG3Pb37 soldcr
'uoung's
modulus Poisson's
Ttierma I
expans ion
-
1.
-
1 Fig.3 Global model

(MPa'l r a t i o coefficient
i "C)
I U

Solder 1 2. 0Ox I O 4 I 0.39 j 2. 4 7 x 1 0 - 5 I

I I I
l'rlll~i~d
1. 8 0 x 104 0. 2 0 1. F O X 1 0 - 5

THERMAL FATIGUE LIFE O F SOLDER BLJI11PS

Two stress analyses were achieved in
sequence t o estimate the fatigue life of the solder
bumps in the BGA package. In the first stage,
preliminary global deformation analysrs wc?rr
carried out t o obtain load conditions for each solder
bump. In the second stage, the individual solder
bump, which is located in a region under a severe
stress condition, was analyzed using the elastic- (a) '['U''::ontl i t i o n
creep model of the solder material. l'hc coniputation
was carried out using the structural analysis
program ABAQUS (Version 5.4). The fatigue life
estimation of the solder bump was performed based
on the results of numerical analyses and t,he
Coffin-Manson equation.
Global deformation analysis
(h) I*I c: I d ori id^ t I o n s
Owing t o the large number of solder bumps,
it was difficult t o achieve analysis for the global Fig. 4 Trmpcrature histories
structure including the package, the P C U , and each
bump using a detailed analyt,ical model. From an
economic standpoint. analysis of individual solder The distribution of the Mises e q u i v a l a t
bumps is more reasonable for detailed analysis. s t r w s in the solder region under the TCT condition
This, however, requires knowledge of the relative is shown in Fig. 5 . Because of the stiffness of silicon
displacement between the package and the PCR for chill, the stress concentration area occurred in the
use as the boundary condition. For this reason, the edge of the chip, and this distribution trend
global deformation analysis was carried out t o corresponded well with the TCT results. The values
assess the boundary conditions at the u p p e r surface of relative displacement calculated at the bum11 A in
of the bump. Due t o the symmetry of this problem, Fig. 5 in consideration of warps of the BGA and the
the BGA, its bumps, and the PCB were modeled as PCB are l i s k d in Table X In the present, problem,
shown in Fig. 3 . The material properties of RGA t.hese values were eml~loyed in the following
and PCB used in this analysis are listed in Tablc 2. detailed analyses of the solder bump.

20 5
 Solder bumu analvses
Mises(MPa)
The detailed solder bump model of the BGA
is illustrated in Fig. 13.A half section of the solder
bump was considered due t o the symmetry of the
system. Quadrat,ic brick elements were used over
the whole region, with fine elements generated
around the corner of the bump t o obtain detailed
stress-strain singularity. The relative displacement
obtained b y the global analysis was applied t o the
upper surface of the bump in the direction indicated
by the arrow in Fig. 6.
Results and considerations
The distributions of the equivalent creep
st,rain range for the t,wo conditions are shown in Fig..
7.Although the strain ranges under TCT and field
conditions are different. the strain concentration
Fig.5 Mises equivalent stress distributions i n occurs in the same place (arrows in Fig. 7 ) . Figure 8
bumps of BGA shows a cross-sectional view of t,he solder bump
after the actual TCT. It is clear that the strain
Table 3 Relative displacement values concentration region under TCT conditions
coincides with the crack initiation site in the
Relative I vicinity of the interface. These results indicate that
displacement TCT can serve as an accelerated test in the
values A X (mm) reliability design of solder joints.
TCT condition
...-
.-...
I
2- - 1 x 1 0 3
I

I 5 rnin I 4.2x 10-4

Field Hold I '
5.3x 10
"L"-A"1 1 12 h 1 7. Z X

(a) TCT condition

Fig G Solder bump model of BGA

L..- (b) Field condition (Hold-time 12h)
Fig.7 Equivalent creep strain range distributions

20 6
 acceleration factor, at., may be expressed using the
Coffin-Manson equation as follows:

as = (A E F / A E ~)-1.94 (3)
where A E is the equivalent creep strain range
I?

under field conditions and A E T is the equivalent

creep strain range under TCT conditions. In the
case of the hold-time condition of 12 hours, the
acceleration factor of 24 is obtained from the
analytical results of A E F = 0.0036 and A E T =
0.01%. The knowledge about electric resistance is
not obtained in results of fatigue life prediction
based only on a stress analysis. However, the
Fig.8 Solder bump after TCT increase in electric resistance due t o the thermal
fatigue under field conditions can be estimated by
a multiplying the results of actual TCT by this
20.035 acceleration factor.
k The effects of hold-time a n d the frequency of
0.03 thermal cycle also pose ail interesting problem. The
.i

0.025 modified Coffin-Manson equation is widely used

c
.' in the reliability design a s an approximate method
v,
0. 02 t o consider the effect of the frequency of thermal
C
a, cycle. In this equation, it has been assumed that the
20.015 number of cycles t o fatigue is proportional t o the 113
0 power of the frequency o-F thermal cycle. However,
0.01 modulation term concerned with the frequency of
L
a thermal cycle is not required in our study, because
7 0.005 the creep deformation depending on time is
> positively considered in the elastic-creep analysis.
.ri
3 0' I Referring t o Fig. 10, it is shown that the
wIs 0 0. 05 0. 1 0. 15 0. 2 0. 25 effect of the hold-time on the equivalent creep
Distance from surface r (mm) strain range is significant, since the increase in
creep strain during the hold-time occurs due t o the
Fig. 9 Relationshil) bet,ween equivalent creep strain clastic follow-up and stress redistribution
range and the distancc from surface (TCT phenomena. The number of cycles t o the crack
condition) initiation under field conditions was obtained by
substituting values of the equivalent creep strain
Figure 9 shows the detailed distribution of range, which were obtained a t a distance 0.05 mm
the cquivalent creep st,rain range around t h r strain from the surface, into E q . (2). The relationship
concentration region under the TCT condition. between the frequency of thermal cycle (iday) and
Sincc the crack initiation site was located in the the number of cycles t o the crack initiation is shown
vicinity of the singularity point, values of the in Fig. 11. The number of cycles t o fatigue under the
equivalent creep strain occurring a t thc surface hold-time condition of 12 hours is about 113 times
were not reliable. In consideration of the grain size that under the hold-time (condition of 5 minutes. As
of solder, i t has been assumed that an estimation a consequence, the temperature hold-time has a
point of the equivalent creep strain range is located noticeable effect on the fatigue life.
at a distance 0.05 inin from the surface of thc bump.
lieferring to Fig. 9, a value of 0.0186 has been
obtained a t a distance 0.05 mm from the surface of 007

h
the bump. The following Coffin-Manson cquation 006 --
has previously been obtained by the authors: Hold time 12h

NI = 0.14G(A E ,,,) - I $I4 (2)

005 I................ Hold
- .- .- .- . Hold
~~
time l h
time Smin
004
where NI is the number. of cycles t o failure, and 003
A E ,,, is the inelastic strain range. From a value
of 0.0186, the number of cycles t o t,he cracli 002 .....
initiation was estimated t o be 330 cycles. Since
niac:roscoyic cracks with depth of about 0.04 niin
were observed in the vicinity of the interrace after
250 cycles of actual TCT, this estimation result is
satisfactory.
.ri
3
W
U
" " 0
00 1
I
0 0. 05 0. 1
.I.,..

0. 15 0. 2
" "
.....
.......3

"
-I'
..,.. .....

0. 25
Figuine 10 shows the detailed distribution of Distance from surface r (mm)
the equivalent creep strain range around the strain
concentration region under the field conditions. Of Fig. 10 Relationship between equivalent creep
particular interest in terms of reliability design is strain range and the distance from surface (Filed
an acceleration factor., which is a quantitative conditions)
correlation between TCT and field conditions. The

207
 50000 ACKNOWLEDGMENTS
40000
The authors would like t o thank Dr Osamu
Horigami, Dr Koichi Ohtoml and Mr Setuo
30000 Yamamoto of the Toshiba R&D Center for
Hold time 5min

/
supporting this study
20000 REFERENCES
rcl
2.
p
/
1d time l h Lau, J. H., Rice, D.W., and Avery, P. A., 1987,
"Elastoplastic Analysis of Surface-Mount Solder
Joints," IEEE Transactions on Components,
10000 Hybrids, and Manufacturing Technology, Vol.
9000 CHMT-10, No.3, pp.34G-357.
8000 [2] Kitano, M., Kumazawa, T., and Kawai, S., 1992,
7000 "A New Evaluation Method for Thermal Fatigue
6000 of Solder Joint," in Advances in Electronic
5000 Packaging, Proceedings of the 1992 Joint
1 10 100 ASMEiJSME Conference on Electronic
Packaging, EEP-Vol. 1-1,pp.301-308.
Cyclic f r e q u e n c y (/day) [3] Mukai, M., Kawakami, T., Endo, T., and
Takahashi, K., 199 1, "Elastic-Creep Behavior
Fig. 11 Effect of cyclic frequency on fatigue life and Fatigue Life in an IC Package Solder Joint,"
Proceedings of JSME The 4th Computational
Mechanics Conference, No.910-79, pp.223-224.
[4] Mukai, M., Kawakami. T., Endo, T., and
CONCLUSIONS Talrahashi, K., 1992, "The Demonst,ration on
Elastic-Creep Thermal Stress Analyses for IC's
The elastic-creep analysis presented by the GYSn-37Pb Solder Joint,s." Proceedings of JSME
authors was applied to the BGA's solder joints The 70th JSME Fall Annual Meeting, No. 920-78,
(bumps). The estimation results related t o the crack pp.3GO-3G2.
initiation of solder corresponded well with the [ 5 ] Hiruta, Y., Hirano, N., Yamaji, Y., Mukai, M..
results of TCT. The fatigue life estimation under Motoyama, Y., Homma, R., Ohno, CJ., and Sudo,
field conditions, which have hold-times of 5 minutes, T., 1993, "An 820 Pin PGA for Ultralarge-Scale
1 hour a n d 12 hours, was also carried out. Since BiCMOS Devices," IEEE Transactions on
close agreement between the TCT and the field Components, Hvbrids, and Manufacturint.
conditions was obtained for the location of the Techiiology, Vol. iG, No. 8, pp.893-901.
strain concentration, it was clear that the TCT [GI Mukai, M., Kawakami, T.. Endo, T.: Hiruta, Y..
could serve as an accelerated test in the reliability and Takahashi, K.. 1994, "Elastic-Creel,
design. The acceleration factor, which provides the Thermal Stress Analyses for the SMT-PGk
quantitative correlation between the two conditions, Package's Solder Joint." Proceedings of the
was determined from analytical results with the ASMEWAM, AMD-Vol.187. p p . 195-203.
Coffin-Manson equation. The results of the fatigue [7] Yebisuya, T., and Kawakubo, T., 1993, "Creep
life estimation using the elastic-creep analysis and Tensile Properties of Cast Bi-Sn. Bi-Pb and
reflect accurately the effects of temperature hold- Bi-Sn-Pb Solders," J.Japan Inst Metals, Vol. 57,
time without using the modified Coffin-Manson No. 4, pp.455-462
equation. As a result of comparison of different
hold-time, i t was shown that the temperature hold-
time, which is namely the frequency of thermal
cycle, has a noticeable effect on the fatigue life. This
is attributable to the increase in creep strain
occurring during the hold-time due t o a n elastic
follow-up phenomenon. Therefore, proper
assessment of field conditions is the important
issue i n t h e reliability design. From the results of
this study it was concluded that the fatigue life
estimation of solder joints based on the present
method is satisfactory for engineering purposes,
and the advantage of this method is especially
evident when the fatigue life under field conditions
is estimated accurately.

208