Research On Computer Vision Based Automated Optical Inspection Method And Application

Automated optical inspection (AOI) system has become more and more important in the field of wafer production. In order to be competitive in the semiconductor manufacturing industry, defect inspection becomes a critical issue. The goal of wafer defect inspection is to detect defective dies and discard them. However, these existing methods have limitations when dealing with the high speed and high accuracy applications. In this paper, the previous research works and literatures of the key algorithms in automated optical inspection. Some novel algorithms with high speed, high accuracy are proposed and the system of defect inspection is designed aims the existing method’s shortages.This dissertation firstly analyzed the image used in an automated optical inspection system. The image acquirement system which is the most relevant factors to the image quality is analyzed. A novel bright-dark-field light source design is proposed based on the research of the previous light source design to solve the insufficient of camera dynamic range.The algorithms for image segmentation have been studied in details. The existing thresholding, geometric primitives and template matching methods have been summarized and compared. A novel image segmentation algorithm based on compressed sensing principle is proposed. This algorithm is efficient and robust. Then this algorithm is used in a wafer defect inspection system with template matching algorithm to get an efficient and accurate inspection result. The feature extraction and defect classification are the key technique of the automated optical inspection. For feature extraction of wafer defect, the traditional methods are low efficiency, poor classification, and sensitive to lighting and noise. In this paper, the line detection algorithm based on phase congruency is improved to get high efficiency and accuracy. This novel algorithm can be also used in corner detection. Then a new feature extraction algorithm based on scale-space principle and Harris corner detector is proposed. This algorithm has accuracy classification results in back-propagation neural network test.Finally, an automated wafer defect inspection system based on computer vision was presented. The algorithms and methods in this paper were verified and compared using this system. The experiment results have shown that the proposed algorithms and methods can inspect the wafer defect with high speed and accuracy. The inspection time of one wafer is12minutes, the missing rate is0%and the false rate is less than10%.