China Mechanical Engineering ›› 2026, Vol. 37 ›› Issue (4): 959-966.DOI: 10.3969/j.issn.1004-132X.2026.04.020

Previous Articles     Next Articles

A Novel SiC MOSFET Lifetime Prediction

HU Yawei1,2(), FANG Xiang2, YIN Chuanan3, LIN Zijun1, LIN Xiaowei1   

  1. 1.Zhejiang Guangli Engineering Machinery Co. ,Ltd. ,Lishui,Zhejiang,323700
    2.School of Electrical Engineering and Automation,Anhui University,Hefei,230601
    3.JAC Motors Technology Center,Hefei,231299
  • Received:2025-10-28 Online:2026-04-25 Published:2026-05-11
  • Contact: HU Yawei

一种新型的SiC金属氧化物半导体场效应管的寿命预测

胡娅维1,2(), 方响2, 尹传安3, 林子俊1, 林小卫1   

  1. 1.浙江广力工程机械有限公司, 丽水, 323700
    2.安徽大学电气工程与自动化学院, 合肥, 230601
    3.江淮汽车技术中心, 合肥, 231299
  • 通讯作者: 胡娅维
  • 作者简介:胡娅维*(通信作者),女,1989年生,博士研究生。研究方向为性能可靠评估与智能运维、可持续设计与制造。E-mail:yaweihu@126.com
  • 基金资助:
    国家自然科学基金(52377035);安徽省先进电力电子与电能变换工程研究中心开放课题(APE202504)

Abstract:

To address the reliability challenges faced by SiC MOSFETs in high-frequency, high-temperature, and high-power density applications, a novel lifetime prediction method was proposed that integrated a CNN, an ECA mechanism, and a BiLSTM. This method used the drain-source on-state voltage drop as the core degradation feature, incorporated preprocessing strategies such as outlier removal, normalization, and exponential smoothing, and reconstructed the degradation time series through a sliding window to achieve effective modeling under small sample conditions. Comparative experimental results demonstrate that the proposed method offers significant advantages in prediction accuracy, stability, and robustness.

Key words: SiC metal-oxide-semiconductor field-effect transistors(MOSFET), lifetime prediction, convolutional neural network(CNN), bidirectional long short-term memory(BiLSTM), efficient channel attention(ECA), power cycling

摘要:

为应对碳化硅金属氧化物半导体场效应管(SiC MOSFET)在高频、高温和大功率密度应用中面临的可靠性挑战,提出了一种新型的融合卷积神经网络、高效通道注意力机制与双向长短期记忆网络的SiC MOSFET寿命预测方法。该方法以漏源极导通电压为核心退化特征,结合异常值剔除、归一化和指数平滑等预处理策略,并通过滑动窗口对退化时间序列进行重构,实现小样本条件下的有效建模。实验对比结果表明,所提方法在预测精度、稳定性和鲁棒性方面均具有明显优势。

关键词: 碳化硅金属氧化物半导体场效应管, 寿命预测, 卷积神经网络, 双向长短期记忆网络, 高效通道注意力, 功率循环

CLC Number: