Research On Wafer Surface Defect Detection Algorithm Based On Machine Vision

With the rapid development of the semiconductor industry,the demand of semiconductor wafer manufacturing is increasing day by day,but in the production process,it will inevitably lead to a variety of wafer defects,which will affect the quality of semiconductor chip products.For wafer quality detection is particularly important,manual detection is prone to misjudgment and slow speed,so the introduction of wafer surface detection method based on machine vision has become a hot spot.Therefore,this paper studies the wafer surface defect detection algorithm based on machine vision.In order to extract and segment thousands of grains in wafer,this paper uses template matching algorithm based on gray level to match grains.Defects may cause the edge of the grain to fracture,but it has little effect on the global gray information of the grain image.Therefore,the accuracy of the gray based template matching algorithm is better than that of the contour based template matching algorithm in grain segmentation.In order to further reduce the extraction time of grains,the non maximum suppression of template matching results is carried out to improve the efficiency of multi-objective matching grains.For the problem of noise,after the extraction and segmentation of the grain,the complex noise contained in the grain is avoided to be mistakenly detected as defects.In this paper,the adaptive median filter and wavelet transform fusion algorithm are used to denoise the noise in the grain to improve the accuracy of defect detection.For the global impulse noise in the image,the adaptive median filter can completely save the image edge and other details when removing the noise.In order to remove the aliasing noise of the contour edge in the image,the wavelet transform algorithm pays more attention to the details of the image,and the denoising effect of the detail noise is better.In order to highlight the defect,the filtered image is enhanced to make the defect contour and edge become obvious.For the filtered and enhanced images,an improved subtraction defect detection algorithm is proposed.In the template selection of subtraction operation,each wafer image uses the static method to generate the template.Aiming at the speed problem of defect detection,Gaussian pyramid is used to downsampling the template and the image to be detected.The characteristic information of the image is retained while the number of pixels is reduced,so as to reduce the amount of calculation of difference operation in the corresponding pixels of two images,reduce the processing time of the algorithm,and improve the detection speed.In order to verify the accuracy and speed of the wafer surface defect detection algorithm based on machine vision.Due to the high precision of the wafer,the hardware platform of this experiment is based on the wafer detection platform designed by the research group,and the wafer images are collected on the platform.The experimental results show that the accuracy of wafer defect detection is 97.08%,and the detection time of each single grain is less than 100ms.The detection accuracy and speed are in line with the actual industrial requirements.