基于改进的稠密神经网络的晶圆缺陷分类方法研究  

The Studyof Wafer Defect Classification Method Based on Improved Dense Neural Network

作　　者：邹佳霖 王红成[1] ZOU Jialin;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43-49,共7页Journal of Dongguan University of Technology

基　　金：东莞市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：针对现有的晶圆缺陷分类模型并行度不高、无法很好地学习全局信息等问题,提出一种基于卷积神经网络和稠密神经网络(DenseNet)的深度学习模型,对晶圆缺陷进行了分类。在卷积神经网络的基础上,引入卷积注意力模块,在同时考虑通道维度和空间维度的特征,提升模型收敛效果和构建改进的稠密神经网络基础上,实现对晶圆缺陷的分类。结果显示:此方法在MIR-WM811K数据集上平均准确率为98.9%,F1值为92.7%,平均准确率相较DenseNet提升约2%。A deep learning model based on convolutional neural networks and dense neural networks(DenseNet)is proposed to address the issues of low parallelism and inability to learn global information well in existing wafer defect classification models,and wafer defects are classified.On the basis of convolutional neural networks,a convolutional attention module is introduced to consider the characteristics of channel and spatial dimensions,improve the convergence effect of the model,and construct an improved dense neural network to achieve the classification of wafer defects.We achieved an average accuracy of 98.9%and an F1 value of 92.7%on the MIR-WM811K dataset,with an average accuracy improvement of approximately 2%compared to DenseNet.

关 键 词：晶圆缺陷分类 深度学习 卷积神经网络 

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]