Physics-Informed Machine Learning for Solder Design and Reliability Prediction for Electronics

Abstract

We propose a physics-informed machine learning framework for solder materials design and reliability prediction for electronics. It includes a comprehensive set of factors that impact solder reliability and can work as a solder-agnostic framework for solder design and reliability prediction. This framework is built on a deep modular artificial neural network (ANN), with its structure imitating the relationships among elements and processes in electronics manufacturing, and its ANN modules model the elements and processes in electronic product manufacturing. It includes physics understandings of solder materials and their performance and provides a suitable machine learning framework with interpretability in understanding the solder design and reliability in electronics. A preliminary case study showed the superiority of the proposed framework in reliability prediction accuracy and interpretability.