基于通道注意力机制和多路径深度卷积的混合型晶圆缺陷分类  

Hybrid Wafer Defect Classification Based on Channel Attention Mechanism and Multi-Path Depth Convolution

作　　者：范胜娇 王红成[1] FAN Shengjiao;WANG Hongcheng(School of Electrical Engineering&Intelligentization,Dongguan University of Technology,Dongguan 523808,China;School of Computer Science and Technology,Dongguan University of Technology,Dongguan 523808,China)

机构地区：[1]东莞理工学院电信工程与智能化学院,广东东莞523808 [2]东莞理工学院计算机科学与技术学院,广东东莞523808

出　　处：《东莞理工学院学报》2024年第5期50-57,共8页Journal of Dongguan University of Technology

基　　金：东莞市社会发展科技项目(20231800940532);松山湖科技特派员项目(20234373-01KCJ-G)。

摘　　要：晶圆图缺陷的准确分类对改进制造工艺具有至关重要的作用。相比于单一缺陷,混合缺陷具有特征复杂、种类繁多的特点,更加符合真实工业制造情况。为了有效识别并分类混合缺陷,提出一种结合通道注意力机制和多路径深度卷积神经网络的方法。此方法在多路径深度卷积神经网络支路上增加通道注意力机制,以关注混合型晶圆图的详细特征。在38类缺陷真实数据集上的实验结果表示,模型在精度方面优于一些现有的深度学习模型,其平均正确率高达97.67%,可以有效分类晶圆图混合缺陷。Accurate classification of wafer map defects plays a crucial role in improving manufacturing processes.Compared with single defects,mixed defects have complex characteristics and various types,which are more consistent with real industrial manufacturing conditions.In order to effectively identify and classify hybrid defects,a method combining channel attention mechanism and multi-path deep convolutional neural network is proposed.This method adds a channel attention mechanism to the multi-path deep convolutional neural network branch to focus on the detailed features of the hybrid wafer map.Experimental results on a real data set of 38 types of defects show that the model is superior to some existing deep learning models in terms of accuracy,with an average accuracy rate of up to 97.67%,and can effectively classify mixed defects in wafer images.

关 键 词：计算机视觉 晶圆缺陷识别 深度学习 通道注意力机制 多路径深度卷积 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] TP183[自动化与计算机技术—计算机科学与技术]