Research On Key Techniques For Integrated Circuit Chip Surface Defects Vision Detction

Integrated circuit chips has been widely used in many areas,however,defects produced in the manufacturing process directly affect the life span and reliability.Traditional detection method is manual inspection,which is time-consuming,labor-intensive and with high error detection rate.As a result,traditional detection is unable to meet production requirements.Machine vision inspection uses machine vision method to analyze products and inspect whether the products meet the quality requirements which can ensure the quality of products and plays a key role on increasing the rate of qualified products.Therefore,with the support from the National Natural Science Foundation(Grant No.50805012)and the Transformation of Scientific and Technological Achievements Special Fund of Jiangsu(Grant No.BA2010093),this thesis does the researches on key techniques for integrated circuit chip surface defects vision detection.The main research are as follows:(1)Multilevel segmentation for integrated circuit chip surface defects imageFor the characteristics of the IC chip surface defects images,a defect image segmentation method with multilevel thresholds of two-dimensional entropy based on firefly algorithm(FA)is proposed.Firstly,expand two-dimensional entropy thresholding method to two-dimensional entropy multilevel thresholding method.Secondly,analyse the biomimetic and optimization process of firefly algorithm.Finally,take the two-dimensional entropy as the objective function of firefly algorithm to search the thresholds.Experiment results show that this method can segment the IC chip surface defects images efficiently;the running speed is much faster than that of the exhaustive method;in the terms of accuracy,running speed and peak signal to noise ratio,the proposed method performs better than multilevel thresholding of two-dimensional entropy based on particle swarm optimization.However,the problem of weak real-time performance still exists in the method.There are some problems such as large amount of calculation,and long computing time existing in multilevel thresholding.In order to solve the problem,a defects image segmentation using multilevel thresholding based on firefly algorithm with opposition-learning is proposed.Firstly,the Otsu thresholding method is expanded to the Otsu multilevel thresholding method.Secondly,firefly algorithm with opposition-learning(OFA)is proposed.The core idea of the opposition-learning algorithm is in some cases,the opposition solution is closer to the optimum solution than the current solution.In OFA,the opposition-learning is introduced to the firefly algorithm.The opposite fireflies are generated to increase the diversity of fireflies and improve the global search ability.Finally,the variance between classes is taken as the objective function of OFA to search the threshold.Experiment results show that the OFA based multilevel Otsu thresholding method performs better than the exhaustive method,PSO based Otsu multilevel thresholding method and FA based Otsu multilevel thresholding method.However,when the thresholds are four,in some cases,the OFA multilevel defect image segmentation cannot search the precise thresholds.In order to segment the IC chip surface defects image,a multilevel image segmentation based on an improved firefly algorithm is proposed.According to the unbalance phenomenon of global search and local search in firefly algorithm,an improved firefly algorithm(IFA)is proposed.In IFA,two strategies(i.e.,diversity enhancing strategy with Cauchy mutation and neighborhood strategy)are proposed and adaptively chosen according to different stagnation stations.The variance between classes is taken as the objective function of IFA to search the threshold.Experiment results show that the IFA based Otsu multilevel thresholding method can segment defect images efficiently.Moreover,in terms of accuracy,computational time,convergence and stability,the IFA multilevel defect image segmentation has the better performance than the Darwinian particle swarm optimization,hybrid differential evolution optimization,FA and OFA based Otsu multilevel thresholding methods.(2)Integrated circuit chip surface defects extractionThe IC chip surface defects images are captured with the bright field illumination and the dark field illumination.In the bright field images,there are a lot of noise points which interfere the defects extraction.In order to extract defects on the bright field images,a defects extraction based on mathematical morphology and modified region growing is proposed.Firstly,according to the image level,obtain the shallow defects image and the deep defect image after the bright field images segmenting with four thresholds.Secondly,for the different characteristics of the shallow defects image and the deep defects image,use different mathematical morphology operators to remove noise pixels and locate the connected region of defects.Moreover,apply the modified region growing method to extracting the defects from IC chip surface.Experiment results show that defects extraction based on mathematical morphology and modified region growing can extract defects from the bright filed images efficiently.In the dark field images,there are textures produced in plastic process which interfere the defects extraction.In order to extract defects on the dark field images,a defects extraction based on texture detection and defects region selection is proposed.Firstly,an algorithm for inspect the direction of chip surface textures is proposed.Secondly,according to the image level,the bright defects image and the dark defect images are obtained after the dark field images segmenting with four thresholds.Finally,for different characteristics of the bright defects image and the dark defects image,based on the principle that the texture direction of defects is different from that of the IC chip,the defects region selection is applied to reserving defects regions and removing other regions.Experiment results show that defects extraction based on texture detection and defects region selection can extract defects from the dark filed images efficiently.(3)Integrated circuit chip surface defects feature extraction and dimension reductionIn order to extract the integrated circuit chip surface defect feature,32 features are extracted from geometry feature,texture feature and gray feature.Geometry features includes area,perimeter,compactness,barycentric coordinates,squareness,duty cycle,eccentricity and seven Hu invariant moments.Texture features are 14 gray level co-occurrence matrix features.Gray features are gray mean value,gray variance and gray entropy.The principal component analysis(PCA)and the sequential floating forward selection based on KNN(KSFFS)are respectively used to reduce feature dimension.According to the principal that the cumulative contribution rate of eigenvalues should be greater than 90%,PCA algorithm reduces the feature dimensions to 6.The KSFFS algorithm takes the KNN classification performance of every feature as evaluation function to select features and reduces the feature dimensions to 10.(4)Integrated circuit chip surface defects classificationIn order to classify the IC chip surface defects,BP neural network and RBF neural network are analyzed.Moreover,in order to improve the classification performance,support vector machine based on the improved firefly algorithm(IFA-SVM)is proposed.In IFA-SVM,the algorithm takes the classification accuracy as objective function to search the penalty parameter and kernel function parameter.Put the features obtained from PCA and the features obtained from KSFFS into BP neural network,RBF neural network and IFA-SVM,then form six classifier.Experiment results show that when the features selected by KSFFS are taken as input,the classification accuracy of IFA-SVM is 91.367%,higher than other five classifiers.The thesis does the researches on key techniques for integrated circuit chip surface defects vision detection,which has made certain achievements in theoretical research and technology development.Therefore,the thesis provides both theoretical guidance and technical support for integrated circuit chip surface defects vision inspection.