Machine Learning Modelling for Electronic Reliability Analysis in Solder Joints

Solder joints are critical components in electronic assemblies, influencing the overall reliability, performance, and safety of devices across industries such as consumer electronics, automotive, and medical systems. These joints are exposed to various stresses, including thermal cycling, mechanical vibrations, and humidity, which necessitate a comprehensive understanding of their behavior under these conditions. This study addresses the challenges of calculating plastic strain values in solder joints, particularly lifetime relevant characteristic plastic strains per load cycle, using Finite Element Modeling. Finite Element simulations were conducted by varying loading and design parameters such as temperature, vibration amplitude, printed circuit board thickness, chip thickness, and solder volume to generate a dataset for machine learning models. Recurrent neural networks with different architectures were trained and optimized through hyperparameter tuning to predict characteristic plastic strain values. The models were evaluated using statistical metrics, including mean squared error which was prioritized for evaluation, and root mean squared error. The best-performing model demonstrated high predictive accuracy and generalization potential with a mean square error of $2.276 \cdot$ $e^{-9}$ and a root mean square error of $4.770 \cdot e^{-5}$, offering a robust approach predicting solder joint stress behavior in diverse operational conditions while minimizing the calculation time.