Machine learning-driven optimization of the cure cycles of self-polymerizing epoxy molding compounds for semiconductor packaging applications

Abstract

A significant obstacle to both the performance and production efficiency of semiconductor packages is warpage, which is generated by the coefficient of thermal expansion (CTE) mismatch caused by high-temperature bonding between the epoxy molding compound (EMC) and the substrate. In this study, we propose an autoencoder-based machine learning model for predicting the curvature and cure cycle required to obtain the curvature of a semiconductor package based on the curvature data with respect to the cure cycle involving ramp, rapid cooling, and reheating. The curing reaction was measured according to the steps of the cure cycle using differential scanning calorimetry, and was analyzed to investigate the effect of heating rate on curvature and self-polymerization during the rapid cooling step. Furthermore, the bonding point was determined based on the curvature of the semiconductor package and the degree of cure. Using these data, the cure cycle for the target curvature was predicted using the machine learning model. The predicted values by the trained machine learning model were validated through experiments.