电子测量与仪器学报

史浩琛,金致远,唐文婧,王静,蒋楷,夏伟.基于深度学习的高精度晶圆缺陷检测方法研究[J].电子测量与仪器学报,2022,36(11):79-90

基于深度学习的高精度晶圆缺陷检测方法研究

Research on high precision wafer defect detection based on deep learning

DOI：

英文关键词:deep learning wafer defect detection silicon carbide wafer YOLOv4

基金项目:国家自然科学基金(62005094)、山东省自然科学基金(ZR2021MF128)、济南市引进创新团队项目(2018GXRC011)、山东省工业技术研究院协同创新中心共建项目(CXZX2019007)资助

作者	单位
史浩琛	1.济南大学物理科学与技术学院
金致远	1.济南大学物理科学与技术学院
唐文婧	1.济南大学物理科学与技术学院
王静	1.济南大学物理科学与技术学院
蒋楷	1.济南大学物理科学与技术学院
夏伟	1.济南大学物理科学与技术学院

Author	Institution
Shi Haochen	1.School of Physics and Technology, University of Jinan
Jin Zhiyuan	1.School of Physics and Technology, University of Jinan
Tang Wenjing	1.School of Physics and Technology, University of Jinan
Wang Jing	1.School of Physics and Technology, University of Jinan
Jiang Kai	1.School of Physics and Technology, University of Jinan
Xia Wei	1.School of Physics and Technology, University of Jinan

摘要点击次数: 2104

全文下载次数: 1629

中文摘要:

为了解决半导体制造领域缺陷检测中出现的检测效率低、错误率高、结果不稳定、成像精度低下导致无法精确地检测出不同种类的缺陷等问题,本文利用定制的 CCD 工业相机搭配高倍率的光学显微镜采集晶圆表面的扫描图像,结合改进的 YOLOv4 算法,实现了基于深度学习的高精度晶圆缺陷检测方法。实验表明,对于碳化硅晶圆缺陷,提出的方法模型可以识别各种复杂条件下的不同种类缺陷,具有良好的鲁棒性。对缺陷的平均识别精度达到 99. 24%,相较于 YOLOv4-Tiny 和原 YOLOv4 分别提升 10. 08%和 1. 92%。对缺陷的平均每图识别时间达到 0. 028 3 s,相较于基于 Halcon 软件方法和 OpenCV 模板匹配方法分别提升 93. 42%和 90. 52%,优于其他常规的晶圆缺陷检测方法,已实现在自主设计的验证系统和应用平台上稳定运行。

英文摘要:

In order to solve the semiconductor manufacturing defect detection with low efficiency, the error rate is high, the result is not stable, imaging accuracy is low and cannot accurately detect the problem such as different kinds of defects. In this paper, by using a custom CCD industrial camera with a high ratio of optical microscope scan images on the surface of the wafer, combined with the improved YOLOv4 algorithm, a high precision wafer defect detection method based on deep learning is implemented. Experimental results show that the proposed model can identify different kinds of silicon carbide wafer defects under various complex conditions and has good robustness. The average accuracy of defect identification is 99. 24%, which is about 10. 08% and 1. 92% higher than that of YOLOV4-Tiny and original YOLOv4, respectively. Compared with the Halcon-based method and OpenCV template matching method, the average recognition time of defects per graph reaches 0. 028 3 s, which is about 93. 42% and 90. 52% higher than other conventional wafer defect detection methods and has realized stable operation in independently designed verification systems and application platform.

查看全文查看/发表评论下载PDF阅读器