STUDY ON WAFER IDENTIFICATION RECOGNITION

• Main Authors: Chou-Nan Chen, 陳周楠
• Other Authors: Chau-Yun Hsu
• Format: Others
• Language: en_US
• Published: 2006
• Online Access: http://ndltd.ncl.edu.tw/handle/09418891297880706352

碩士 === 大同大學 === 通訊工程研究所 === 94 === It is important for Assembly/Test process of semiconductor manufacturing to link wafer Identification (wafer ID) generated by Laser scribed with wafer Database. The wafer ID like a card plate license has a unique number and has much process information such as wafer test result which has been uploaded to wafer Database for successive processes such as dicing saw, die mount, wire bond, or final test. This thesis proposed a recognizing wafer ID with the Template matching method to recognize the wafer ID engraved by Laser. The wafer ID recognition system consists of pre-processing, image processing, locating wafer ID, segmenting characters from wafer ID, extracting character from wafer ID, and wafer ID output. In pre-processing, adjust a wafer with location pin to get less than 1 degree slope, focus on wafer ID window to simply process ,and rotate the wafer 180 degrees when the wafer is made by vendors; in the image processing, we convert a RGB image into a gray image which has been normalized to 480 by 640 resolution and then use Otsu’s method to find an adaptive threshold value which could automatically convert a gray image into binarized image, instead of trial and error to get a threshold value, finally use the Sobel edge detector to identify the edge of wafer ID; in locating wafer ID, first we use vertical projection method to decide the height of wafer ID string, and then use horizontal projection method to identify the width of wafer ID string; in segmenting characters, we use Template matching method to recognize the character of wafer ID image; and then display matching result from the recognize the character. This system is evaluated by 53 wafer ID images from manufacturing. The successful rate of recognizing wafer ID is 81.1% which is not high rate but improving productivity. The average recognition time of each image is 2.2 seconds.