Statistical Variations of Solder Joint Fatigue Life Under Realistic Service Conditions

Abstract

Common damage accumulation rules fail to predict the fatigue life of solder joints under realistic service conditions where cycling amplitudes vary over time. A modification of Miner's rule of linear damage accumulation has been proposed that accounts for effects of amplitude variations in, for example, the vibration of microelectronic assemblies with lead-free solder joints on the average or characteristic fatigue life. We are, however, obviously much more concerned with the first failure across a very large sample set. Prediction of, say, the first failure out of 10000 or a million would require the extrapolation of experimental failure distributions and the assumption of a shape of this distribution. Even qualitative comparisons of accelerated test results and their scatter should account for effects of amplitude variations. We have argued that the long-term life of solder joints in vibration or cyclic bending is limited by the accumulation of inelastic work, and that much can be learned from the low cycle fatigue behavior in shear. Individual ball grid array scale SAC305 and SAC105 solder joints were cycled in shear at room temperature with combinations of two different stress amplitudes. Relying on our modified Miner's rule and the associated understanding of the effects of amplitude variations, we show that the statistical uncertainty in the fatigue life of solder joints under a specific set of realistic service conditions must be significantly greater than measured in fixed amplitude cycling tests. The predicted failure distribution was best fit by a Weibull distribution over a limited range, but we argue that the assumption of such a distribution is likely to be increasingly conservative when it comes to the prediction of earlier failure. Estimates are provided for the potential errors.