The effect of model building on the accuracy of fatigue life predictions in electronic packages

Empirical fatigue life models such as the Coffin–Manson rule and its variants are commonly used at the present time to predict the reliability of microelectronic packages. While there have been reports of substantial error in empirical correlations relative to the experiments, this has not been accompanied by a rigorous understanding of the sources of the error. In this paper we systematically explore the various modeling errors in the fatigue life prediction. These errors include those in geometry representation, material behavior, load history and boundary condition application, and in the numerical solution procedure. As part of the study, experimentally validated correlations between temperature cycling and power cycling are developed for a TI 144 chip-scale package. The accuracy of the predicted life under power cycling conditions compared to the experimentally determined life is used as the basis for judging the model accuracy. The criticality of spatial refinement, temporal refinement, and accurate boundary conditions, including the often ignored natural convection boundary conditions, and their effect on predicted life is described in detail. It is shown that model errors can be a significant part of both the constitutive life models and the application models that use the constitutive life models to predict the fatigue life under a given environmental condition. It is also shown that with careful model building, solutions accurate to within 3% can be obtained.