A new automatic recognition technology research on semiconductor wafer defect detection

• Main Authors: Wen-Tsung Chang, 張文聰
• Other Authors: Leh Luoh
• Format: Others
• Language: zh-TW
• Published: 2013
• Online Access: http://ndltd.ncl.edu.tw/handle/08797060477724885937

碩士 === 中華大學 === 電機工程學系碩士班 === 102 === In Taiwan, flourishing electronics industry requires a lot of electronic components, Therefore, Electronic components made of semiconductor wafers is also a large and fast production, More than hundred level of the production process, The most common problems is the surface defect, for avoid surface defects of production process because there is not detected of defect caused by the waste of resources and production decline, Therefore, the relevant defect detection technology to become a very important goal ,Previous standard defect detection process is divided into macroscopic detection (Macro) and microscopic detection (Micro), The difference is that Macro detect is use large numbers of people use visually to make quick detection of wafer surface, however Micro detect is more than ten times magnification using a microscope detection wafers of circuit of the wafer(die), In order to accelerate the detection process and reduce the cost of based on computer vision related research is popular research objectives. Among the many studies, Order to increase the defect identification stability and reduce the difficulty of identification often use micro detection as the research topics, But micro defects require a lot of time to complete a wafer defect identification, Therefore ,This study attempts to use macro detection to complete defect identification, And propose a method based on Image Parameter Reference Model to Accomplish the defect detection, first ,we take the wafer image into equal-sized blocks and then converted into characteristic parameters for image characteristic, Finally use the CART method to Identification defect ,it be able to quickly and efficiently make identification defect and classification, Under normal circumstances, the recognition rate of 85% computing time is about 23 seconds ,Therefore, this study for the current Macro defect identification provides an effective identification method.