Hybrid Physics-Based Data-Driven Thermal Model for Multichip Power Modules

With the rapid advancement of power electronics technology, the power density of power modules has significantly increased, making the junction temperature a critical parameter for ensuring reliable operation. Therefore, accurately predicting junction temperature is crucial for reliable thermal design and condition monitoring of power modules. This paper proposes a hybrid physics-based data-driven (HPDD) thermal model for multi-chip power modules. Firstly, the chip's self-thermal temperature rise (STR) is predicted by the Foster model. Then, based on the information of STR, the chip's coupling-thermal temperature rise (CTR) is predicted by the data-driven model eXtreme Gradient Boosting (XGBoost). Finally, the junction temperature can be predicted based on the STR and CTR. The proposed HPDD thermal model utilizes the XGBoost model to calculate complex CTR, reducing the modeling and computational complexity while significantly decreasing the number of finite element methods (FEM) simulations and parameter fittings. In addition, it boosts the prediction robustness by reducing the parameter uncertainty of the XGBoost model based on the physical information in STR obtained from the FEM simulation. Simulations and experiments are conducted, and the results confirm the effectiveness of proposed model.