Research On Chip Defect Detection Technology Based On Machine Vision

In recent years,the rapid growth of artificial intelligence and big data technology has made chips an indispensable part of the modern industry.Nevertheless,due to the intricacy of the chip production process and the hidden nature of flaws,the existence and impact of chip defects is still a widespread problem.Automatic defect detection of chip image using machine vision and other technologies can effectively increase the production efficiency of enterprises.Therefore,this paper aims to delve into the practicality of machine vision in detecting chip defects.By adopting advanced image processing technology,an defect detection system for wafer chips and chip frames has been successfully developed.The main purpose of this paper is to study the algorithms of chip image preprocessing and chip defect detection.The main work is as follows:(1)Research on preprocessing algorithms.Aiming at the clear edge contour of wafer images,a set of preprocessing segmentation algorithms for wafer images is designed,including image stitching,tilt correction,image filtering,edge detection,and contour extraction.Finally,a single chip image is segmented using the minimum circumscribed rectangular coordinate information of the single chip contour,with a segmentation accuracy of 100%;A chip frame image segmentation algorithm based on template matching is proposed for chip frame images with complex backgrounds,blurred boundaries,and chip connectivity.Firstly,multiple region modules are pre segmented for the entire chip frame image;Then,a template matching algorithm is used to match the single chip image based on the region module image;Finally,the single chip image is segmented by merging the overlapping matching frames of the single chip and recording the coordinate information of the merging frames.The experimental results show that by selecting appropriate templates and thresholds through experiments,the segmentation accuracy of this algorithm can reach 100%,and it saves at least 45.76% of the segmentation time compared to segmentation algorithms not based on region module matching,meeting the requirements of high-precision and high-speed chip frame segmentation.(2)Research on defect detection algorithms.Through analyzing the defect type characteristics of wafer chips and chip frames,defect feature extraction methods are designed for each defect type for defect detection.For empty Die defects,template matching algorithms are used to identify empty Die defect chips;Aiming at foreign object defects,median filtering combined with Gaussian filtering is proposed to blur the chip image and extract foreign object defect features through threshold segmentation;Aiming at the defect of chip contamination,a band-pass filter was proposed to enhance contaminants in the frequency domain image,and a threshold segmentation algorithm was used to extract contaminants;For defects such as solder joint deviation,excessive solder ball size,and high silver paste creep,the threshold operator in the Halcon tool is used to extract the target and measure the relevant dimensions to achieve defect detection for the chip frame.The experimental results show that the error detection rate of the defect detection algorithm designed in this paper is less than 5%,and the detection accuracy reaches more than 90%.(3)Design of defect detection system.The design of imaging system hardware,especially the selection and installation process of camera,lens,light source and other components are comprehensively summarized.In addition,the software operation process and implementation details are also described.The software of the system is specially designed to implement the algorithm proposed in this study to achieve chip segmentation and chip defect detection.