Machine learning inverse design of high-strength mid-temperature Ag-based solders

Abstract

Traditional trial-and-error experimentation and computational methods are often inefficient for designing solders with specific properties, revealing the need for more effective design strategies. This work presents a novel inverse design framework for accelerating the discovery of mid-temperature (400–600 °C) Ag-based solders. A Wasserstein Autoencoder (WAE) generates candidate compositions, targeting melting temperatures within the 400–600 °C range through a Gaussian Mixture Model and neural network classifier. Yield strength is predicted using a stacking ensemble learning model, combining Multilayer Perceptron and Gradient Boosted Decision Trees with a Decision Tree meta-learner, achieving high accuracy, which was confirmed by experimental validation of four selected alloys. This data-driven approach demonstrates significant potential for the efficient design of high-performance solder materials.