电子元件与材料2025，Vol.44Issue(9)：1087-1097,11.DOI:10.14106/j.cnki.1001-2028.2025.0278
基于改进型热敏感电参数的SiC IGBT级联神经网络结温预测模型
Cascade neural network junction temperature prediction model for SiC IGBTs based on an improved temperature-sensitive electrical parameters
摘要
Abstract
Silicon carbide insulated gate bipolar transistor(SiC IGBT)demonstrates promising application potential in high-voltage,high-temperature,and high-power domains.Effective control of the junction temperature(Tj)is a critical technology to ensure the safe operation of these devices.Therefore,accurately obtaining Tj information plays a vital role in the reliable operation of SiC IGBT.Typical combined thermally sensitive electrical parameters(TSEPs),such as on-state voltage drop(VCEon)and collector current(Ic),are expected to be utilized for Tj prediction in SiC IGBT.However,a complex nonlinear relationship existed between the combined TSEP(VCEon and Ic)and Tj,which led to unsatisfactory performance of Tj prediction models.To address this issue,a novel TSEP(the rate of change of Ic with respect to VCEon,dIc/dVCEon)was explored.It was found that integrating this TSEP for improving the combined TSEP can effectively mitigate the complex nonlinear relationship between the combined TSEP and Tj.Subsequently,a cascaded neural network model consisting of the sparrow search algorithm optimized back propagation neural network(SSA-BPNN)and a generalized regression neural network(GRNN)was constructed based on the improved combined TSEP for Tj prediction of SiC IGBT.Finally,the prediction accuracy of the model was evaluated using Silvaco TCAD simulation data.The results indicate that the proposed model achieves high-precision Tj prediction,with small error fluctuation and an average absolute error as low as 0.1 ℃.关键词
碳化硅/绝缘栅双极晶体管/结温/热敏感电参数(TSEP)/神经网络/级联Key words
SiC/IGBT/junction temperature/thermally sensitive electrical parameter(TSEP)/neural network/cascade分类
电子信息工程引用本文复制引用
黄玲琴,李乾坤,刘新超,师威鹏,谷晓钢..基于改进型热敏感电参数的SiC IGBT级联神经网络结温预测模型[J].电子元件与材料,2025,44(9):1087-1097,11.基金项目
国家自然科学基金(62074071) （62074071）
