Analysis And Design Of Wafer Defect Detection System Based On Machine Vision

Wafers play an important role in semiconductor manufacturing.If a wafer containing defects is packaged,the performance of later integrated circuit will be affected.As semiconductor sizes become smaller and smaller,there may be problems in that the acquired wafer image is difficult to be sharp,the wafer is difficult to locate when corner missing or occlusion,and the wafer defect shape is similar to the background geometric pattern.In order to reduce the influence of wafer defects on semiconductor manufacturing,an on-line automatic detection technology for wafer surface defects based on machine vision is designed.The content is mainly divided into three parts: image acquisition system,image processing algorithm and client software.In the image acquisition system part: according to the project requirements,the machine vision image acquisition system is designed,using a CCD camera with a pixel of 3384×2710,a telecentric lens,and the illumination scheme is a white ring light source with front lighting,dark field lighting and one-way lighting.The imaging scheme can make the imaging of the inner target area of the wafer sample clear,but suppresses the non-target area,thereby enhancing the contrast of the image.In the image processing algorithm part: we use the industrial machine vision software Halcon framework to write algorithms.Firstly,a contour matching method is designed to locate the wafer in order to solve the problem of corner missing or occlusion in a single wafer.Secondly,for the problem that the background of the wafer has a regular pattern of geometric patterns and defects that may be similar to the background geometric pattern,a geometric pattern contour affine transformation and sub-area detection method is designed by using the principle of affine transformation.On this basis,we cut the inner and outer regions of the geometric pattern of the wafer,and perform threshold segmentation and morphological processing respectively to extract the defects;then intersect the defects to obtain the total defects,and record the coordinates of the defective wafer.In the client software part: we use the C++ code exported by the Halcon framework and the MFC framework for software development.Firstly,we analyze software functional requirements.including graphical visualization interface,communication data with detection devices,camera debugging,parameter debugging,logging,automatic detection and so on.Secondly,a simple and easy-to-use software interface based on functional requirements is designed.Finally,the MFCbased client image processing software is compiled,and the software is experimentally verified and analyzed.The specific steps of the automatic detection process are as follows: first open the automatic detection process,camera snapshot thread,TCP communication thread.When the system receives the snapshot signal,it starts to take pictures with the camera and performs wafer detection defects,and simultaneously sends the detection result to the robot.The experimental results show that the detection algorithm can effectively solve the problems of unmatched positioning caused by corner missing or severe occlusion and defects that may be similar to the background geometric pattern,and can effectively detect various defects.The detection speed of a single wafer is about 430 ms.The average detection defect rate is 99.81%,and the average detection defect accuracy rate is 98.80%.The algorithm has fast detection speed,high detection accuracy and good effect,and meets the requirements of industrial projects.