A High Efficient Design Method for Electrothermal Coupling Performance of Power MOSFETs Based on Iterative Multiple Artificial Neural Networks

Electrothermal coupling has a critical impact on the steady-state losses and junction temperature of power MOSFETs due to their strong thermal dependent on-resistance. Existing analytical loss calculation methods for power MOSFETs, which are based on an assumed operating temperature, are not accurate, whereas iterative electrothermal simulations offer high accuracy. However, in automatic design of converters based on artificial intelligence (AI), where circuit parameters and heat sink structural configurations vary dynamically, obtaining MOSFET losses through iterative simulations is very time costing. To address this challenge, the article proposes a high efficient design method for electrothermal coupling performance of power MOSFETs based on iterative multiple artificial neural networks (ANN). Two ANN surrogate models are respectively developed for the power losses and thermal distribution based on LTspice and COMSOL with limited number of simulations. Further, an iterative calculation is performed based on these two ANN models which significantly reduce the calculation time compared to the traditional iterative electrothermal simulations. The proposed method is validated in three-phase LLC converter in this article. Compared with iterative simulations, the proposed method achieves a 93.8-fold acceleration in data collection and over a 107-fold speedup in evaluation. Experimental validation at multiple power ratings shows high accuracy, underscoring the method\u2019s engineering applicability.