Conventionally, the fatigue life of solder joints in semiconductor structures is estimated using Coffin-Manson’s law. However, as the structures have become miniaturized or thinner, accurately estimate fatigue life has become difficult using conventional methods. This is because the fatigue life is strongly affected by crack propagation in miniaturized or thinner joints, and the conventional methods cannot evaluate this phenomenon well. We have therefore developed a new method for evaluating fatigue life that takes into account the influence of crack propagation in micro-solder joints. In micro-solder joints, a solder crack path might propagate not only at the solder and land interface itself, but also near the interface. Many crack-propagation have been proposed, but a model that can reproduce a crack path has yet to be proposed. The fatigue life of a solder in our crack-propagation model is evaluated based on the damage that accumulates during crack propagation, and the crack paths are automatically calculated. Using this model, we analyzed the crack path of a ball grid array (BGA) structure, and we determined that the model could reproduce the above-mentioned characteristic crack paths. When the fatigue life is calculated using a finite element method, one of the most difficult issues is correcting for the effect of element size. We determined the calculated life dependency on element size, and we developed a formula for approximating this dependency in the proposed model. We then used this formula to calculate the fatigue life of three different size BGA solder joints that were subjected to mechanical fatigue testing. The calculated lives were found to correspond with the measured lives. Furthermore, we applied this method to evaluate the differences in the fatigue life of a solder-mask-defined (SMD) structure and a non-solder-mask-defined (NSMD) structure. Both are typical structures of BGA solder joints. We determined that the fatigue life of the NSMD structure was longer than that of the SMD structure. The main cause for this difference is that the crack-propagation life of the NSMD structure was longer than that of the SMD structure, even though the crack-initiation lives of both structures were the same.

This content is only available via PDF.
You do not currently have access to this content.