Abstract
Before being introduced to the marketplace, electronic packaging products typically undergo an Accelerated Thermal Cycling (ATC) test. During the design phase, extensive repetitive experiments can lead to significant waste of both time and resources. Therefore, applying machine learning with finite element analysis to predict the reliability of advanced packaging can greatly enhance the efficiency of electronic packaging product development. To reduce costs, the validated finite element models will replace real experiments to generate a small dataset for training the machine learning models. The smaller the required dataset, the lower the time cost for simulation; however, this also increases the difficulty of training the high-performance machine learning models. To address challenges associated with training on small datasets, this study uses Wafer-Level Packaging (WLP) as an example and applies ensemble learning with multiple machine learning algorithms to predict the reliability for packaging.