Research On IC Chip Appearance Defect Recognition Algorithms Based On Deep Learning

In “Made-in-China 2025”,integrated circuit(IC chip)is listed as the first in the development and reform of information technology industry.Since the 21 st century,IC chips have been widely used in various industries.As the core of the information industry,the key to industrial upgrading and optimization lies in the operational efficiency and life of IC chips.Improving the detection quality of high-end IC chips can ensure the high quality of the chips.In the manufacturing and packaging process of IC chips,various defects may occur on the surface of IC chips due to the processing technology.At present,in the IC chip manufacturing and testing workshop,most of the appearance inspection processes are carried out by the inspectors with high power microscopy.There are some shortcomings such as easy fatigue,low detection accuracy,low efficiency and so on.Machine vision detection technology has many advantages,such as high accuracy,high stability,fast non-contact,etc.It has replaced manual detection in many detection fields.In the current market of IC chip appearance defect detection equipment,the main self-developed semi-automatic detection equipment in China is to complete IC chip surface defect detection through automatic feeding and unloading and manual visual inspection.Fully automated testing equipment needs to be imported from abroad.There are many problems,such as high price,long order period,and difficult for suppliers to provide timely maintenance services.In the process of IC chip appearance defect detection,there are many kinds of defects.The background changes will lead to random changes in defect features and location.The recognition method based on general feature extraction can't effectively identify defects.In-depth learning can extract and combine the underlying features of the sample target,which is often used to find the scene of the hidden features of the sample.Therefore,this paper studies IC chip appearance defect recognition algorithm based on in-depth learning,mainly including IC chip image preprocessing algorithm,region of interest(ROI)precise location algorithm,chip defect recognition algorithm and other key links.This topic has important theoretical research and application value for the development and application of IC chip defect identification technology in engineering practice.The main contents of this paper are as follows:(1)IC chip image preprocessing method.Aiming at the problem of more interference in IC chip image,a filtering algorithm suitable for IC chip image is proposed by comparing various spatial and frequency domain image enhancement algorithms.To solve the problem of multiple IC chips in plastic cover/pin image,a robust segmentation algorithm is proposed to extract ROI from single chip.Aiming at the problem of chip local rotation caused by mechanical conveying and positioning as well as manufacturing errors of disc,an automatic rotation correction and optimization algorithm is proposed to realize rough positioning of chip ROI.In order to solve the problem of uneven gray level in the chip image(caused by pin frame,welding wire or defect),an adaptive threshold segmentation algorithm based on brightness and shade coefficients is proposed to realize the threshold segmentation of chip image.(2)Accurate location algorithm for IC chip plastic encapsulation area.The appearance defects of IC chips exist in the plastic sealing area of ROI chips.The random deformation of lead frame,welding wire and uneven distribution of epoxy resin will interfere with the defect extraction in plastic sealing area.A new location algorithm based on projection feature is proposed to provide a reliable basis for subsequent defect feature extraction.(3)IC chip defect recognition algorithm based on multi-feature.Analysis of the actual IC chip detection technology requirements and precise positioning results of plastic encapsulation area,summarize and determine the typical characteristics of different defects.This paper classifies and analyses chip defects,extracts typical features and features of Convolutional Neural Networks(CNN),and proposes a multi-feature fusion algorithm for IC chip defect recognition,which improves the robustness of recognition.Experiments show that the method can effectively improve the recognition rate compared with single feature for defect recognition.(4)IC chip defect recognition algorithm based on CNN classifier.Aiming at the accuracy of large sample training network and the minimum requirement of dirty samples,an automatic data sample cleaning algorithm based on prior knowledge is proposed.The validity of the proposed algorithm is verified by MINST handwritten data set.Aiming at the shortcomings of BP neural network such as local minimum in the actual training process,the classical convolutional neural network Alexnet model is constructed.The defect image set of IC chip is cleaned by data,and a good defect sample set is trained to complete defect recognition of IC chip.By adjusting activation parameters and convolution layers of classical Alexnet,the effects of different parameters on IC chip appearance defect recognition were tested.Aiming at the technical requirements of IC chip appearance defect recognition,this paper studies the image processing and defect recognition algorithm of IC chip,completes the software of IC chip appearance defect recognition system,and realizes the instruction and automatic detection of IC chip appearance defect.The experimental results show that the algorithm of IC chip appearance defect recognition based on deep learning has good stability and detection accuracy,and the missed detection rate is less than 0.05%.It has strong theoretical research significance and engineering practice value.