Predicting the Life of the Solder Joints in Electronic Assemblies using Physics-informed Data-driven Methodology

The accurate prediction of the life of solder joints is critical to the design of electronic devices. However, accelerated life tests used to estimate the reliability of solder joints are costly and time-consuming due to the multiple implementations of tests under different conditions. Furthermore, it is impractical to incorporate various factors affecting life into a single physics- based model. This paper proposes a physics-informed data-driven methodology to evaluate the impacts of different factors on the life of solder joints and improve prediction accuracy. Various factors (such as aging conditions, component configuration, board substrate material, solder paste material, solder sphere alloy, and stencil thickness) and their effects on life have been explored. Machine learning methods, including linear regression, decision trees, and artificial neural network (ANN), are used for life prediction because of their distinct advantages in accuracy and interpretability. For instance, the ANN method shows the highest prediction accuracy (R2 = 92%). This physics-informed data- driven methodology will be a starting point for future efforts in leveraging physics and experimental data in electronics reliability analysis.