摘要
针对晶圆检验时扫描电镜图像的缺陷检测和缺陷分类问题,利用ZFNet卷积神经网络对晶圆缺陷进行分类,并在此基础上,设计基于块的卷积神经网络缺陷检测算法。为提高准确率和加快速度,通过改进Faster RCNN分类器,提出另一种检测算法。实验结果表明,2种检测算法都能通过学习已标记位置和类型的缺陷数据,从扫描电镜图像中准确检测并分类多种类型缺陷。
For defect detection and defect classification problems of Scanning Electron Microscope( SEM) images during wafer inspection,this paper applies a Convolutional Neural Network( CNN) called ZFNet to classify wafer defects. On this basis,a patch-based CNN defect detection algorithm is proposed. For better accuracy and higher speed,another detection algorithm is proposed by modifying Faster RCNN classifier. Experimental results show that,by learning from the defects data marked with locations and types,the two detection algorithms can both detect and classify defects of multiple types on SEM images.
作者
邡鑫
史峥
FANG Xin,SHI Zheng(Institute of VLSI Design, Zhejiang University, Hangzhou 310027, Chin)
出处
《计算机工程》
CAS
CSCD
北大核心
2018年第8期218-223,共6页
Computer Engineering
基金
国家自然科学基金(61474098
61674129)