The effect of process-induced voids on the durability of Sn-Pb and Pb-free solder interconnects in electronic products is not clearly understood and researchers have reported conflicting findings. Studies have shown that depending on the size and location, voids are not always detrimental to reliability, and in fact, may sometimes even increase the durability of joints. This debate is more intensified in Pb-free solders; since voids are more common in Pb-free joints.

Results of experimental studies are presented in this study to empirically explore the influence of voids on the durability of Ball Grid Array (BGA) Pb-free solder joints. In order to quantify the detailed influence of size, location, and volume fraction of voids, extensive modeling is conducted, using a continuum damage model (Energy Partitioning model), rather than the existing approaches, such as fracture mechanics, reported in the literature. The E-P approach is modified in this study by use of a successive initiation method, since depending on their location and size; voids may influence either the time to initiate cyclic fatigue damage or time to propagate fatigue damage, or both. Modeling results show competing interactions between void size and location, that results in a non-monotonic relationship between void size and durability. It also suggests that voids in general are not detrimental to reliability except when a large portion of the damage propagation path is covered with either a large void or with many small voids.

In addition, this dissertation also addresses several fundamental issues in solder fatigue damage modeling. One objective is to use experimental data to identify the correct fatigue constants to be used when explicitly modeling fatigue damage propagation in Pb-free solders. Explicit modeling of damage propagation improves modeling accuracy across solder joints of vastly different architectures, since the joint geometry may have a strong influence on the ratio of initiation-life to propagation-life. Most solder fatigue models in the literature do not provide this capability since they predict failure based only on the damage accumulation rates during the first few cycles in the undamaged joint.

Another objective is to incorporate into cyclic damage propagation models, the effect of material softening caused by cyclic micro-structural damage accumulation in Pb-free solder materials. In other words the model constants of the solder viscoplastic constitutive model are continuously updated with the help of experimental data, to include this cyclic softening effect as damage accumulates during the damage-propagation phase. The ability to model this damage evolution process increases the accuracy of durability predictions, and is not available in most current solder fatigue models reported in the literature. This mechanism-based microstructural damage evolution model, called the Energy Partitioning Damage Evolution (EPDE) model is developed and implemented in Finite Element Analysis of solder joints with the successive initiation technique and the results are provided here. Experimental results are used as guidance to calibrate the Energy Partitioning fatigue model constants, for use in successive initiation modeling with and without damage evolution.

FEA results show 15% difference between the life predicted by averaging technique and successive initiation. This difference could significantly increase in the case of long joints such as thermal pads or die-attach, hence validating the use of successive initiation in these cases. The difference between using successive initiation with and without damage evolution is about 10%. Considering the small amount of effort that has to be made to update the constitutive properties for progressive degradation, it is recommended that softening be included whenever damage propagation needs to be explicitly modeled. However the damage evolution exponents and the corresponding E-P model constants obtained in this study, using successive initiation with damage evolution, are partially dependent on the specimen geometry. Hence, these constants may have to be re-calibrated for other geometries.