Intelligent Thermal Model of SiC Power Modules Using Convolutional Neural Network

The thermal behavior of silicon carbide (SiC) power modules is greatly influenced by the material properties of the SiC chip and packaging materials under high-temperature conditions. This paper proposes a novel intelligent thermal model for multi-chip SiC modules using a convolutional neural network (CNN), accounting for temperature-dependent material properties. Taking advantage of convolutional kernels to extract inter-data features, the intelligent thermal model (ITM) can capture the thermal cross-coupling (TCC) effect between chips, providing greater accuracy compared to traditional neural networks. Moreover, ITM is developed by training on a dataset generated through finite element model (FEM) low-cost simulations that consider temperature-dependent effects. Experimental validation demonstrates that the ITM significantly enhances computational efficiency, with calculation errors within 1 °C, showing superior performance in estimating junction temperature at high-temperature conditions.