基于CNN-LSTM的晶圆良率预测  

Wafer Yield Prediction Method Based on CNN-LSTM

作　　者：吴立辉[1,2] 周秀 张中伟 WU Lihui;ZHOU Xiu;ZHANG Zhongwei(School of Mechatronics Engineering,Henan University of Technology,Zhengzhou 450001,China;School of Mechanical Engineering,Shanghai Institute of Technology,Shanghai 201418,China)

机构地区：[1]河南工业大学机电工程学院,郑州450001 [2]上海应用技术大学机械工程学院,上海201418

出　　处：《组合机床与自动化加工技术》2023年第7期142-146,151,共6页Modular Machine Tool & Automatic Manufacturing Technique

基　　金：国家自然科学基金项目(U1704156);上海应用技术大学引进人才科研启动项目(YJ2022-33);河南省科技攻关计划项目(212102210357);河南省高等学校重点科研项目(23A460003)。

摘　　要：晶圆良率是衡量半导体制造系统加工能力的关键指标,对其精准预测有利于排查晶圆制程工艺缺陷、提高晶圆生产率、控制企业生产成本。基于晶圆允收测试(wafer acceptance test,WAT)大数据,提出了基于卷积神经网络和长短期记忆网络(convolutional neural networks and long short-term memory,CNN-LSTM)的晶圆良率预测方法。该方法对WAT数据进行缺失、异常与归一化预处理;构建CNN模型对复杂WAT参数的关键特征进行识别;考虑相邻晶圆间的时序相关性,设计长短期记忆网络进行回归分析,从而实现晶圆良率的准确预测。以某工厂晶圆允收测试过程中采集的实际生产数据进行实验,并与其他传统晶圆良率预测方法的结果进行对比分析,从而验证所提方法的有效性。Wafer yield is a key indicator of the processing capability of a semiconductor manufacturing system and its accurate prediction can help identify wafer process defects,improve wafer productivity,and control production costs.In this paper,we propose a wafer yield prediction method based on convolutional neural networks and long short-term memory(CNN-LSTM)based on wafer acceptance test(WAT)data.method.The method performs missing,abnormal and normalized pre-processing of WAT data;constructs CNN models to identify key features of complex WAT parameters;considers the temporal correlation between adjacent wafers and designs a long and short-term memory network for regression analysis to achieve accurate prediction of wafer yield.The actual production data collected during WAT in a factory are used for experiments,and the results are compared and analyzed with those of other traditional wafer yield prediction methods to verify the effectiveness of the proposed method.

关 键 词：晶圆良率 晶圆允收测试 预测 卷积神经网络 长短期记忆网络 

分 类 号：TH161[机械工程—机械制造及自动化] TG68[金属学及工艺—金属切削加工及机床]