Utilizing Ensemble Learning on Small Database for Predicting the Reliability Life of Wafer-Level Packaging

Efficiently and accurately predicting the fatigue life of solder joints in wafer-level packaging is a high-interest topic. This is because it can significantly accelerate the update and iteration cycle of electronic products. Nowadays, with the continuous development of computer hardware and computer software, the concept of Artificial Intelligence Plus (AI +) has been widely applied across various industries. Machine learning, as one of the key methods for implementing AI+ has made significant advancements in the prediction of advanced packaging reliability. Generally, machine learning is based on two fundamental considerations: data sampling and algorithm selection. The quality and quantity of training data will directly impact the performance of the machine learning model. Considering the cost of data acquisition, this paper will use the validated two-dimensional finite element model to generate database through simulation. At the same time, to balance the prediction accuracy of the AI model with the computational cost of executing the finite element model, this study will design a standard process for generating the small database and introduce ensemble learning to achieve accurate predictions of package reliability. In terms of algorithm selection, artificial neural networks (ANNs) will be utilized.