Iterative Machine Learning-Aided Framework Bridges Between Fatigue and Creep Damages in Solder Interconnections

Costly and time-consuming approaches for solder joint lifetime estimation in electronic systems along with the limited availability and incoherency of data challenge the reliability considerations to be among the primary design criteria of electronic devices. In this article, an iterative machine learning framework is designed to predict the useful lifetime of the solder joint using a set of self-healing data that reinforce the machine learning predictive model with thermal loading specifications, material properties, and geometry of the solder joint. The self-healing dataset is iteratively injected through a correlation-driven neural network (CDNN) to fulfill the data diversity. Outcomes show a very significant enhancement in lifetime prediction accuracy of the solder joint within a very short time. The effects of solder alloy and solder layer geometry are separately evaluated on the creep-fatigue damage evolution of the solder joint. The results reveal that Sn-Ag-Cu-based solder alloy generally has a better performance. Moreover, the creep and fatigue damage evolutions are found dominant, respectively, in Sn-Pb- and Sn-Ag-Cu-based solder alloys. The proposed framework offers a tool allowing for the reliability-driven design of electronic devices in the early stage of fabrication.

Automated summary

In this article, an iterative machine learning framework is designed to predict the useful lifetime of solder joints. The effects of solder alloy and solder layer geometry on the creep-fatigue damage evolution of the solder joint are evaluated separately. The proposed framework significantly improves the accuracy of solder joint lifetime prediction in a short time.