Study On Surface Defect Dectection Technology For IC Wafer

With the rapid development of IC manufacturing technology, the continuous decrease of feature size makes the defects even smaller. Surface defects of wafer have become a major obstacle to yield. How to detect defects of wafer automatically and accurately is a complex and challenging task. Defects detection technology is the key technology in IC industry. The thesis has deeply researched the surface defect detection technology.In this thesis, we use image processing technology and super-resolution technology to detect the defects of IC wafer. Image processing includes image filtering, feature extraction and image matching algorithm. Super resolution technique can produce a high-resolution(HR) image from a set of low-resolution(LR) images with complementary information。The main results and conclusions of this paper are summaried as followsing:First of all,this thesis discussed the source of defects during wafer fabrication. According to the morphological characteristics of defects, defects can be divided into redundancy material defect, crystal defect and mechanical damage.The main factors affecting the accuracy were analyzed in defect detection, such as the choice of light source and beam modulation. Choosing shorter wavelengths of laser and using optical interferometric microscopy were effective measures to improve the accuracy.Secondly, using OpenCV can greatly improve the efficiency of the algorithm and enhance system scalability. This paper implemented linear filter and median filter by OpenCV. Median filter reserved more image detail than the linear filter. Hough transform was used to extract the linear feature and the Harris algorithm was used to extract the angular point of the wafer. An improved algorithm of Harris was presented. Image matching based on the gray and image matching based on feature are two different ways of image matching. Template matching used normalized autocorrelation function. SIFT algorithm was realized and had acquired a very good effect. The experimental results proved that SIFT algorithm was efficient and accurate for image matching.Finally, the observation model of Super-resolution image was established. By thinking of the cause of degrading image quality, the paper analyzed how to realize the super-resolution restoration and reconstruction.POCS(Projection Onto Convex Sets) and MAP(Maximum A Posteriori) are elucidated clearly in the paper.Experimental results showed that both of the algorithms achieved good results.If motion estimation was reliable, POCS could achieve better results and its noise reduction capability would be stronger.