Research On Multiscale Defect Detection Technology Of Wafer Surface Based On Gabor And RCNN

With the rapid development of semiconductor industry,the quality requirements of semiconductor wafer are becoming higher and higher.The defects on the wafer surface can easily lead to the decrease of wafer yield and the conductivity of wafers,which may cause the scrapping of semiconductor and further affect the normal operation and instrument damage of the whole system.Moreover,in traditional template matching,it is difficult to acquire wafer templates and low efficiency,which leads to longer test time of wafers.This will reduce the production efficiency of the entire production line and increase the operation cost of the enterprise.To solve these problems,a multi-scale defect detection technology based on area convolution neural network is proposed and applied to wafer image detection system,which verifies the effectiveness of the proposed method.The first chapter introduces the development process of semiconductor industry and wafer materials,and analyzes the research status of image feature extraction and dimension reduction technology and commonly used wafer defect detection technology at home and abroad,and expounds the research contents and organization framework of this paper.In the second chapter,the technology of wafer fabrication and the types and reasons of the defects are described in detail.In view of the defects of the wafer in different scales,the difficulties and solutions are described in detail.The preprocessing of wafer detection is to filter the wafer in the macro scale and to enhance the image set of the wafer in the micro scale.The third chapter focuses on the detection algorithms of wafer defects in the macro scale.First of all,reducing the influence of wafer reflection due to reflection,and combining with the pattern of the wafer surface,most of them are periodic patterns.The texture detection method is proposed to detect the wafer surface defects at macro scale.The Gabor wavelet transform is used to get the texture feature of the wafer surface,and the high-dimensional feature is projected and dimensionality reduced.Finally,the two valued algorithm is used to detect the defect location quickly and accurately.The fourth chapter is to use the method of regional convolution neural network to detect the defect detection of wafer in micro scale.By selecting deep network to extract feature map,a neural network algorithm structure for micro wafer data set is designed,and based on it,activation function and regularization method are improved.The training process of the model is expounded in detail,and the results are proved by comparing the experimental results of other neural network algorithms.The fifth chapter is a software detection system for wafer defects,which is realized by the theory of the above algorithm.The platform structure and software development tools of the software system are introduced.Combined with the actual characteristics of wafer detection,modularization design of software function is carried out,and then the function and design plan of each module are elaborated in detail.Combined with concrete examples,the superiority of the algorithm is described in detail.The sixth chapter is a summary of the results of the paper,and points out the shortcomings of the scheme,and expects that the technology of the wafer can be greatly developed in the future.