X-ray Detection System For Solder Defects Of Ceramic Packaging Chips Based On Object Detection

With the development of microminiaturization and high integration of chips,the requirements for chip packaging technology are higher and higher,and the defects that are difficult to detect are more likely to appear in the packaging process.Due to the strong penetration of X-ray,it can detect various defects(such as solder defects,internal short circuit and open circuit,etc.),so X-ray detection technology is widely used in the detection of internal defects of chip packaging.However,chip solder defects have the characteristics of different size,changeable shape,and a large number of small and narrow defects,resulting in the accuracy and efficiency of the traditional detection method.In view of the difficulties in the detection of solder defects of ceramic packaging chips,this paper studies the detection method of solder defects of ceramic packaging chips based on target detection algorithm,and on this basis,develops a set of intelligent detection system of ceramic packaging chips solder defects.The specific work content is as follows:(1)This paper introduces the overall scheme of X-ray detection system for solder defects of ceramic packaging chips,analyzes the features of X-ray images of solder defects of ceramic packaging chips,and selects the target detection algorithm as the detection method according to the features.The target detection data sets of 1600 X-ray images of ceramic packaging chip were manually marked,and the CLAHE algorithm was used to enhance the defect features in view of the complex background,low contrast and obscure features of the X-ray images of ceramic packaging chip.(2)In view of the characteristics of chip solder defects with different sizes,changeable shapes and a large number of small and narrow areas,an algorithm for solder defect detection of ceramide packaging chips,YOLO-STPN(Swin Transformer Pyramid Network),based on improved YOLOv5,is proposed.In order to improve the detection effect of the algorithm for small defects,a shallow prediction head is added.On this basis,a characteristic Pyramid Network STPN(Swin Transformer Pyramid Network)combining Swin Transformer and CBAM is proposed,which greatly improves the detection effect of the network on dense defects and large defects.In order to further improve the detection effect of the network on narrow and long defects,the positive sample matching mechanism and CIo U loss of the original YOLOv5 were improved to make the model focus on narrow and long defects and improve the detection effect of the model on narrow and long defects.Ten different types of ceramic packaging chips were used as test sets for experiments.Experimental results show that the proposed YOLOSTPN model can quickly and accurately detect solder defects of different sizes and shapes.(3)Further,as the target Detection algorithm is difficult to accurately measure extreme narrow and long defects,a method DASNet(Detection and Segmentation Network)based on multi-task network is proposed for chip X-ray image solder defect detection.A semantic segmentation decoder is introduced to recognize extremely narrow and long defects and improve the recognition accuracy.In order to improve the detection effect of the model on minor defects,a new positive sample allocation strategy was introduced to solve the problems of uneven label distribution and missing positive samples.By adding focal Dice loss to the segmentation loss,the imbalance between the foreground and background of the defect of the narrow and long solder was alleviated.DASNet proposed in this paper can quickly and accurately detect chip solder defects of different sizes and shapes.(4)To apply the algorithm to actual production,a software system of X-ray detection of solder defects in ceramic packaging chip was designed and developed.The software system integrates many functions,including defect identification,soldering surface identification,chip alignment,judging whether the chip is qualified and manual marking measurement.After a batch of chips are tested,the qualified rate of the batch of chips can be calculated and the test report can be generated.In order to continuously optimize the algorithm model,a set of software which can label the new data and update the detection model is developed to solve the problem of poor detection effect of the new chip models added later.