Recognition And Quantification Of Semiconductor's SEM Image

The process of semiconductor's growth consists of two points of manufacture and monitoring,the process of monitoring takes charge of the growth's condition and is responsible for improving the technology of growth.The SEM is one of the usual equipment used to reflecting the microstructure of semiconductor's growth,which will do much favor to analyzing the feature of semiconductor's growth.The thesis aims at the feature of SEM image of growing semiconductor's microstructure,in order to expanding the traditional and manual way of analysis,which is limited to be qualitative,some methods of image processing are proposed to give quantitative analysis,which is in order to achieve semiconductor growth's real-time control intelligently.The thesis presents the flow of adaptive process systems,after a lot of testing on contrast adjustment,pseudo portrait removing,image segmentation,optimize imaging,quantitative analysis,etc.The flow is drawn out as follows which includes image preprocessing,image segmentation and optimizing and image quantitative statistics.At first,the thesis expounds preprocessing algorithms about image enhancement for semiconductor's SEM images.The preprocessing includes fuzzy enhancement,pseudo color enhancement and gray morphology method.There is a comparison between the pseudo-color coding based on rainbow-code and optimization algorithm coding,the second one turns out to be ascendant.Then,it expounds semiconductor's image segmentation.Two practical algorithms are listed out,one is based on K-means,the other is based on Ostu method,and a genetic algorithm is brought in to find the best optimal solution for Ostu,the second algorithm is better than the first in identifying robustness and edge smooth,so the second one is recommended for intricate images.Based on the binary image,two kinds of quantization methods are presented.The first is nc-Si image with pseudo portrait,the morphology is brought in to solve this problem.The second is serried-distribution nanorod image,in response to this problem,the study takes advantage of corner detection algorithm.The thesis presents all of the research contents' algorithms and simulation result.In the part of quantization,the accuracy can reach 97% at least in comparison with manual method,the result that is under lots of tests can meet the practical requirements.