Automatic Modeling of Logic Device Performance Based on Machine Learning and Explainable AI

In this paper, we propose a machine learning framework for predicting performances of semiconductor devices that can automatically reflect modifications in process conditions. While standard TCAD simulators require intensive modeling and calibration works to capture new process conditions, our proposed framework can learn these conditions from data efficiently and directly. Furthermore, by applying recently attention-getting explainable AI techniques, important factors that affecting device performances can be discovered automatically from the proposed model. Specifically, our model quantifies significance of each process step as the game-theoretic Shapley value, that cannot be achieved by TCAD simulators.