Research On Visual Inspection Method Of Dispensing Quality In Photoelectric Imaging Chip Packaging Process

Whether it is traditional semiconductor chip or photoelectric imaging chip,packaging is an essential link in its production process.Packaging can protect the chip from the physical and chemical influence of the surrounding environment,protect the chip and enhance the thermal conductivity.Too much or too little glue during packaging will erode the chip or reduce the service performance,or the cover plate will fall off due to weak adhesion,affecting the durability,etc,Therefore,it is particularly important to control the adhesive quality.This paper studies the adhesive quality in the packaging process of photoelectric imaging chip,and detects the adhesive quality combined with vision technology.(1)Through the comprehensive analysis of adhesive layer defects during packaging,this paper uses machine vision and image processing algorithm to classify and detect the packaging quality of adhesive layer,selects a set of appropriate hardware equipment through calculation,and independently builds an integrated machine vision framework to accommodate the whole hardware system.(2)The software algorithm of vision system is studied.Through the global threshold segmentation of the clear image collected by the hardware system,the foreground image where the chip is located is cut out,and the detection range is reduced.Then,a two-dimensional power-law transform image enhancement algorithm is proposed to enhance the foreground image,so as to solve the problem that the edge of the image is not clear due to the height drop,For the enhanced image,combined with morphological processing and a kind of straight sub-pixel contour extraction algorithm proposed in this paper to improve the fitting rate,the sub-pixel contour of the chip glue containing area is extracted,and the tolerance is creatively set in combination with the sub-pixel contour area to obtain the inner and outer detection edges of glue overflow and less glue,which improves the generalization of detection.(3)The correlation template matching algorithm is studied,and A multi-resolution guided Zero mean normalized cross correlation template matching algorithm is proposed,which is combined with the pose transformation positioning detection algorithm to solve the problem of unknown rotation scale of the glue containing area of the image to be detected.Then,the template matching algorithm is applied to the sub-pixel glue area,and the glue area is extracted by threshold segmentation,Binary large object analysis,area link and morphological processing,By judging the characteristics of the adhesive layer area,we can judge whether it is normal,broken,overflowed,less or no adhesive.(4)Through the comprehensive analysis of the test data,it can be seen that the application effect in the adhesive layer defect detection is good.The detection accuracy of this detection system has reached 99.2% and the average detection time is 394.2ms.Compared with the commonly used algorithm processing system,the accuracy and efficiency of this adhesive layer defect quality detection system and the generalization of the detection system have been greatly improved.In order to achieve the purpose of visual detection,The business logic is sorted out and the corresponding UI detection interface is written.