一种基于神经网络优化预测模型的SiC MOSFET阻断电压确定方法  

A Method of Determining Blocking Voltage of SiC MOSFET Based on the Neural Network Predict Model

作　　者：沈俊 刘东[1] 唐茂森 葛兴来[1] 叶峻涵 周荣斌 SHEN Jun;LIU Dong;TANG Maosen;GE Xinglai;YE Junhan;ZHOU Rongbin(School of Electrical Engineering,Southwest Jiaotong University,Chengdu,Sichuan 611756,China)

机构地区：[1]西南交通大学电气工程学院,四川成都611756

出　　处：《机车电传动》2021年第5期150-155,共6页Electric Drive for Locomotives

摘　　要：功率器件设计者常需要通过器件仿真进行器件的设计与优化,但由于仿真结果的未知性,设计者需要逐步调整相关参数,使仿真结果不断逼近目标值,这需要耗费大量的时间,而目前又没有有效的解决办法。为解决这个问题,提出了一种基于神经网络的优化预测模型,以确定SiC MOSFET的阻断电压。将测试温度、第一场限环间距、环间距变化步长、场限环个数和漂移区浓度作为自变量,代入器件仿真软件中进行仿真,从而得到VDMOSFET阻断电压(作为因变量),将其分别代入BP神经网络和RBF神经网络中进行预测,并对二者的预测误差进行对比。结果表明,使用LM算法的BP神经网络预测模型能够很好地预测VDMOSFET的正向阻断电压,可为设计者节约大量时间。Power device designers often need to design and optimize devices by device simulation, however, due to the simulation results are unknown before the simulation, designers need to gradually adjust the relevant parameters to make the simulation results constantly approach the target value, which costs much time. There is no effective solution to this problem. In order to solve the problem, an optimal prediction model based on neural network was proposed to determine the blocking voltage of SiC MOSFET. The test temperature, the first field limiting ring spacing, the ring spacing change step, the number of field limiting rings and the drift zone concentration were regarded as independent variable, and brought into the device simulation software for simulation. The blocking voltage of VDMOSFET was obtained as the dependent variable, which was brought into BP neural network and RBF neural network for prediction respectively, and the prediction errors of the two were compared. The results show that the BP neural network prediction model by using LM algorithm can well predict the forward blocking voltage of VDMOSFET,and save a lot of time for designers.

关 键 词：阻断电压 SiC MOSFET 场限环 神经网络 预测模型 仿真 

分 类 号：TN3[电子电信—物理电子学]