摘要
针对检测中晶圆台可能影响检测精度的问题,提出了一种基于双目视觉测量芯片晶圆台的测量方法。由于没有现成的六边形芯片数据集,因此采集的样本主要来自由CAD软件绘制生成的理想数据集,然后对采集的数据集进行训练,训练结果值符合检测识别的要求,且准确识别了六边形芯片。最后,利用YOLOv5s网络框架和全局立体匹配算法来对同一直线边缘芯片进行匹配测量,通过求差,计算是否含有高度差,由高度差可以判断出晶圆台是否在同一水平面上。多次实验的结果表明,此方法可以准确检测到六边形芯片晶圆台的高度差。
Aiming at the problem that the wafer stage may affect the detection accuracy in the detection,this paper proposes a measurement method based on binocular vision measurement chip wafer stage.Because there is no ready-made hexagonal chip data set.Therefore,the collected samples are mainly drawn from CAD software to generate an ideal data set,and then the collected data set is trained.The training result value meets the requirements of detection and recognition,and the hexagonal chip is accurately identified.Finally,the YOLOv5s network framework and the global stereo matching algorithm are used to match and measure the same linear edge chip.By calculating the difference,whether there is a height difference is calculated.From the height difference,it can be judged whether the wafer stage is on the same horizontal plane.The experimental results show that this method can accurately detect the height difference of the hexagonal chip wafer stage.
作者
沈立博
孟新宇
单峤
SHEN Li-bo;MENG Xin-yu;SHAN Qiao(School of Mechanical Engineering,Shenyang University of Technology,Shenyang 110870,China)
出处
《机械工程与自动化》
2024年第4期129-131,共3页
Mechanical Engineering & Automation