Effect of Simulation Methodology on Solder Joint Crack Growth Correlation
Robert Darveaux
1900 South Price Road
Chandler, AZ 85248
rdarv@amkor.com
Ph. 480-821-2408 ext 5235
Fax. 480-821-6730
Abstract
A generalized solder joint fatigue life model for surface
mount packages was previously published in Refs [1,2]. The
model is based on correlation to measured crack growth data
on BGA joints during thermal cycling. It was subsequently
discovered by Anderson et.al. that the ANSYS
TM
5.2 finite
element code used in the model had an error in its method for
calculating plastic work [3]. It was shown that significant
error in life prediction could result by using a recent version
of the code where the bug has been fixed. The error comes
about since the original crack growth constants were derived
based on plastic work calculations that had the bug.
In this paper, crack initiation and growth constants are re-
calculated using ANSYS
TM
5.6. In addition, several other
model related issues are explored with respect to the crack
growth correlations. For example, 3D slice models were
compared to quarter symmetry models. Anand’s constitutive
model was compared with Darveaux’s constitutive model. It
was shown that the crack growth rate dependence on strain
energy density always had an exponent of 1.10 +/- 0.15. This
is in the range of the original correlation, so the accuracy of
relative predictions should still be within +/- 25%. However,
the accuracy of absolute predictions could be off by a factor
of 7 in the worst case, if the analyst uses a modeling
procedure that is not consistent with that used for the crack
growth correlation. The key to good accuracy is to maintain
consistency in the modeling procedure.
Introduction
Analytical models in engineering have several practical
uses: 1) rapid design optimization during the development
phase of a product, 2) predicting field use limits, and 3)
failure analysis of product returned from the field or failed in
a qualification test. The solder joint fatigue model presented
here was first published in Ref [1]. An outline of the
procedure to predict fatigue life is shown schematically in
Figure 1. The model utilizes finite element analysis to
calculate the inelastic strain energy density accumulated per
cycle during thermal or power cycling. The strain energy
density is then used with crack growth data to calculate the
number of cycles to initiate a cracks, and the number of cycles
to propagate cracks through a joint.
In reference [2], more work was presented regarding
sensitivity of the life prediction to the FEA procedure. As a
result, the procedure was modified slightly to include volume
averaging of the strain energy values near the joint interface.
The model has been successfully applied to TSOP, CQFP,
CBGA, PBGA, and power hybrid packages [1,2,4-9].
Calculate Strain Energy Density
Accumulated per Cycle
Calculate Number of Cycles
to Crack Initiation
Calculate Crack Growth Rate
Calculate Fatigue Life
Based on Joint Length
Figure 1. Solder joint fatigue life prediction method.
It was subsequently discovered by Anderson et.al. that
ANSYS 5.5.2 and earlier versions of the finite element code
had an error in their method for calculating plastic work [3].
Even though there had generally been good correlation to
measured results, it was shown that significant error in life
prediction could result by using a recent version of the code
where the bug has been fixed (e.g. ANSYS 5.5.3 and later
versions). The error comes about since the original crack
growth constants were derived based on plastic work
calculations that had the bug.
In this paper, crack initiation and growth constants are re-
calculated using ANSYS 5.6. In addition, several other
model related issues are explored with respect to the crack
growth correlations. Several recommendations are made so
the analyst can get accurate results more efficiently.