Research On Chip Surface Defect Image Detection Method

After the bare chip is finished bonding,there are often defects such as staining,scratches and chipping on its surface,which can seriously affect the reliability of the chip.Usually the defects on the bare chip are detected by manual visual inspection,which has the disadvantages of poor real-time,high cost,low accuracy,and non-uniform standards,and is gradually replaced by automatic inspection technology.Machine vision inspection technology has the advantages of short time,low cost,and can be used for real-time inspection,etc.More and more chip manufacturers are applying machine vision-based technology to defect detection in all aspects of chip manufacturing.Although machine vision-based chip defect detection technology has made great progress in printing defect detection,pin defect detection,etc.,the detection and classification of bare chip defects are still in their infancy.Therefore,this study builds a system of bare chip surface defect recognition technology based on image processing key technology of machine vision in bare chip defect detection,based on image pre-processing,image alignment,image segmentation,feature extraction and classification recognition.In view of the small number of defective samples of a certain type of chip and the high resolution of the acquired images,this paper proposes a chip defect recognition algorithm applicable to small data sets and high resolution images,and its principle is to use template matching to generate differential images to complete the bare chip defect recognition.The main research objective of this paper is to detect the defects that appear after the bare chip is finished bonding and to classify the common defects.Taking a certain model of bare chip as an example,the three defects of staining,scratching and chipping that appear more frequently in this model are detected and classified,and the main research contents are as follows:(1)Align the chip image to be tested with the template image.Firstly,the chip sample without defects is used as the template chip,and the template chip image is collected and processed with corresponding cropping and noise reduction to complete the template image production;the SURF algorithm is used to roughly align the sample image to be tested with the template image,and for the shortcomings of the feature point-based image alignment method with low accuracy,this paper proposes an image precision matching method using the sparrow search algorithm to optimize the single-response matrix;After finishing the image alignment,the chip image to be measured is transformed,cropped and subtracted from the template image to obtain the grayscale difference image.(2)Image segmentation of the chip grayscale difference image.The extraction of defective regions is accomplished by threshold segmentation and denoising of the grayscale differential image.Firstly,the Otsu algorithm based on genetic algorithm optimization is used to obtain the binary differential image by threshold segmentation of the grayscale differential image;the binary differential image is processed by mathematical morphology to remove noise;the indentation marks on the electrical performance test points of the bare chip are removed by applying a mask to the differential image.(3)Extraction of features of individual defect regions.The DBSCAN algorithm is used to cluster the defect regions in the binary differential image to complete the instance segmentation of defects and obtain the number and range of defects;22 feature vectors such as texture features,geometric features and grayscale features of individual defect regions are extracted and constitute the feature matrix.(4)Support vector machine optimized by genetic algorithm is used to classify the defects.In order to reduce data redundancy,reduce the dimensionality of the feature matrix,and speed up the defect classification and recognition,the NCA algorithm is used to calculate the weights of each feature and select the combination of features with the highest recognition rate so as to realize data dimensionality reduction;the support vector machine optimized by genetic algorithm is used to triple-classify the defects on the chip surface to complete the classification and recognition of defects.In this paper,we adopt the method based on template matching to generate differential images for chip defect recognition,use manual extraction of defect features to generate feature matrix,and use support vector machine optimized by genetic algorithm to complete the defect classification task,and use MATLAB GUI development platform to develop chip surface defect detection system to realize bare chip surface defect detection,which provides some theoretical support and methodological guidance for the defect detection link of chip intelligent manufacturing.It provides some theoretical support and methodological guidance for defect detection in chip smart manufacturing.