欢迎访问《兵工学报》官方网站,今天是

兵工学报 ›› 2021, Vol. 42 ›› Issue (6): 1265-1274.doi: 10.3969/j.issn.1000-1093.2021.06.017

• 论文 • 上一篇    下一篇

基于长短时记忆算法的热应力下半导体器件故障预测模型

张明宇1, 王琦1,2, 于洋1   

  1. (1.沈阳工业大学 信息科学与工程学院, 辽宁 沈阳 110870; 2.辽宁工业大学, 辽宁 锦州 121001)
  • 上线日期:2021-07-19
  • 作者简介:张明宇(1987—),女,博士研究生。E-mail: jiabingde@126.com
  • 基金资助:
    中国航空工业创新基金项目(sh2012-18)

LSTM-based Fault Prediction Model of Semiconductor Device under Thermal Stress

ZHANG Mingyu1, WANG Qi1,2, YU Yang1   

  1. (1.School of Information Science and Engineering, Shenyang University of Technology, Shenyang 110870, Liaoning, China;2.Liaoning University of Technology, Jinzhou 121001, Liaoning, China)
  • Online:2021-07-19

摘要: 针对半导体器件故障主要由时间和应力导致的问题,从多源数据入手,提出基于长短时记忆(LSTM)算法的热应力下半导体器件故障预测模型,研究半导体器件状态随热应力等级和应力累计时长的变化趋势,预测半导体器件故障时间。该模型利用LSTM算法具备长期记忆能力的优势搭建多源数据堆叠结构,提高模型拟合半导体器件状态曲线的能力;通过加权滑动平均滤波法滤除高低频噪音;基于1阶预测器数据压缩算法处理连续缓变的特征向量;采用实验数据对模型进行测试。结果表明:所提模型可较好地反映热应力作用下半导体器件状态的变化趋势,5次实验的预测误差均在1.7%以内,具有较高的准确性;模型能够提前预测故障时间,验证了模型的可行性和有效性。

关键词: 半导体器件, 热应力, 故障预测, 长短时记忆

Abstract: For the semiconductor device failure caused by time and stress, starting with multi-source data, a fault predicition model of semiconductor device under thermal stress based on long short-term memory(LSTM) algorithm is proposed to study the change of semiconductor device status with thermal stress level and the cumulative duration of stress, and predict the failure time of semiconductor device. The proposed model uses the advantage of long-term memory ability of LSTM algorithm to build multi-source data stack structure and improve the ability of model fitting the state curve of semiconductor device. The high and low frequency noises are filtered by using the weighted moving average filtering method. The first-order predictor data compression algorithm is used to deal with the feature vectors with continuous and slow variation. And the experimental data is used to test and verify the model. The results show that the model can better reflect the variation trend of semiconductor device status under the action of thermal stress, and the prediction errors of five experiments are within 1.7%, which has high accuracy. The model can predict the failure time in advance, which verifies the feasibility and effectiveness of the proposed model.

Key words: semiconductordevice, thermalstress, faultprediction, longshort-termmemory

中图分类号: