Computer Vision Based Defect Detection of Die Bonding in IC Package Process

• Main Authors: Si-Ming Chen, 陳錫明
• Other Authors: 詹永寬
• Format: Others
• Language: zh-TW
• Published: 2017
• Online Access: http://ndltd.ncl.edu.tw/handle/60353342686900804699

碩士 === 國立中興大學 === 資訊管理學系所 === 105 === There are some inspection stages in integrated circuit packaging process for meeting the quality standards and keeping the quality stable. We need so many persons using a microscope to check whether the quality defects exist or not after wire bonding (3RD inspection stage). The Inspection by human’s eyes not only consumes so much manpower cost and wastes more production cycle-time but also causes the under-killed abnormal product moving to the next production stage because of visual fatigue. In this thesis, we purposed an inspection method “Defect Die Bonding Detector in IC Package Process” (DDB Detector). We collected the defect pictures of abnormal products at 3RD inspection stage and then recognized the abnormal products after die bonding process stage by computer vision methods. We hope that we can achieve automated inspection so that the manpower cost, the production cycle-time and the under-killed rate of abnormal products can be decreased. We divide the DDB Detector procedure of computer vision methods into two parts. In the first part, we focus on incoming PCB defects. Segment the input image into foreground and background by Otsu’s threshold selection method, label every block in the image by connected-component labeling and calculate the area of block image. We distinguish the defective PCB by comparing the value of block image area. In second part, we focus on the defects of die bonding, for instance, missing IC die, shifted and rotated IC. We find the corner point of lead finger on PCB by global searching method, segment IC image by horizontal projection method, find IC baseline by Canny edge detection and Hough line transform, determine IC rotation by calculate the angle between IC baseline and lead finger. In this study, we input good product images and abnormal product images into the two parts of defect recognition procedure respectively for calculating the recognition accuracy. From our experimental results, the recognition accuracy of abnormal PCB can reach 100% in first part of procedure, and the recognition accuracy of abnormal die bonding can reach 92.06% in second part of procedure.