Digital Design for High-Performance IGBT Devices Using Machine Learning and Multi-Physics Simulation

Abstract:
Digital design for power electronics packaging enables comprehensive simulations to evaluate electrical, thermal, and mechanical behavior. This approach both minimizes the need for repetitive prototyping and enhances the understanding of package behavior under real operating conditions. However, the on-state resistance of power semiconductor devices varies depending on bias conditions, significantly influencing both electrical and thermal performance. To address this problem, this study proposes a digital design framework that integrates machine learning and finite element analysis (FEA). By estimating the equivalent resistivity and equivalent thermal conductivity of a semiconductor chip under various bias conditions, the simulation can more accurately predict and adjust material properties, thereby bringing them closer to actual operating conditions. The results of this study are expected to lay the foundation for the development of digital twin models in the future.