The Machine Vision System Of Defect Detection For Deep-submicrometer IC

As the circuit density of IC improved, the manufacture technology of chip is more and more advanced, and the quality of IC is very crucial. Taken into account the effect of factors such as technology and cost, inspection for IC become a step which can not be omitted in IC industry. The defect detection strategy and arithmetic is the key parts of IC inspection equipment.This article roots in Guangdong Province 2004 Annual Technology Projects "Based On Digital Image Processing IC Wafer Microsurgery Automatic Detection System". This machine vision system is composed of computer and embedded system. Computer takes charge image processing, defect detection and database management; embedded system takes charge control of working platform and image acquisition.Defect detection method with high speed and high precision is based on image acquisition with high quality, it is necessary to rectify the image distortion. This article studies for the two following problems:One is the rectification for image distortion. Many standard griding model and point array model were made by Auto CAD, and printed by laser printer. A lot of image sample with distortion gained by CCD image sensor. A few typical image sample was select. The fact coordinate (with distortion) and ideal coordinate (without distortion) of were gained. The distortion coefficient was calculated by MATLAB. Then, realized with VC++ 6.0 . The experiment show that there are a good effect of distortion rectification.Another is a defect detection method for memory, CCD image sensor and LCD. Every unit of this kind of IC has the same pattern. Based on understood the character of IC surface image pattern, a defect detection method of self- compare template matching was developed. After IC micro image was filtered and binarized, which was projected by horizontal and vertical. A image of two dimensions changed into two image of one dimension with noise. The calculation cost reduced obviously. Applying ESPRIT arithmetic, the periods of two image of one dimension were gained. As a result, the size of the building block is known. We can reconstruct the building block by shifting a window of proper size throughout the image and adding the corresponding pixel values. By averaging over all of the blocks in the image, the amount of noise and the effect of defects are reduced considerably. If the periods are not an integer number of pixels, it is necessary to apply double linear interpolation. Standard models are extended by standard building block. Standard models compare with original image, and if the difference is larger than a threshold, that point may be identified as a possible defect. The result of inspection and the information of inspection process were saved into database.