Theoretical And Experimental Research On Submicrion Critical Dimension Measurement Based On Computer Micro-vision

Micro-, nano-manipulation and manufacturing need submicron measurement with fastspeed and high precision. The microstructure and the surface topography measurementtechniques can reach nano precision based on probe or scanning methods. However, thesetechniques are costly and slower, as well as difficult to be integrated for submicronmeasurement. The computer micro-vision based method is applying microscopic imagemetrology and contemporary image processing technology to measure submicron criticaldimensions fastly and precisely. Because of bigger non-contact measuring range, higherprecision and longer working distance, the computer micro-vision measuring system seems tobe more accordance with the requirement of fast on-line measurement and easy of integrationwith low cost. This thesis focuses on the submicron linewidth measurement and studies themicro-vision image processing technology based on the microscopic metrology to improvethe measuring precision. The main contributions of this thesis are listed as follows:First, the microscopic imaging characteristics of the measuring system and thesub-micron measuring principle are investigated. The traditional image processing technologyis used to check the measuring precision and put forward the important researching interestsfor precision improvement.Second, the enhancement technology for micro-vision image is studied. Computermicro-vision imaging system has high magnification, uneven illumination and low reflectinglight density. Thus the pixels on image are overall darker. Traditional image enhancementtechnology can not improve contrast and save the microstructure details. In order to make itconvinent for observing and selecting the region of interest a sophisticated imageenhancement should be invented by which the blurry microscopic image can becomeclearness and has fidelity. By frequency analysis the thesis studies characteristics of theillumination component, the microstructure component and the noise. Based on the lightreflection model and the theory of Retinex, an image enhancement optimization standard isdefined with the mean and variance of the contrast to put forward the algorithm of "scaleoptimized Retinex image enhancement”. Experimental investigations show that this algorithmcan enhance microscopic image with satisfying quality. The fidelity between the enhancementimage and the original image of the0.8μm lines reach to0.55, which is bigger than that of thehistogram equalization method or the homomorphic filtering method.Third, the strip center localization and the calibration method for the measurement system with micron grade view field are studied. Because of noise and the roof like fuzzytransition of the pixel profile along the direction orthogonal to edge, the border based medialaxes transform method can’t meet the requirement of precision, and the Blob analysis istime-consuming. Therefore the Gaussian convolution is used to put forward “the ridge profiletransform of strip” method and moment invariants are used to compute the center coordinateand the orientation angle of the strip. The novel strip center localization is avoided ofinterference by noise and illumination changing. The proposed method is used to measure thecenter distance of156pixels apart line pairs getting the spread uncertainty as small as0.003pixels, as well as reaching1.5%relative error for nominal distance of10μm. Based on thestrip center localization method the calibration for the measuring system with view field smallto micron grad realized by use of microscope resolution testing board.Finally, a novel subpixel edge position method is proposed which is used for themeasurement of submicron linewidth. Because of the low pass filtering effect of the PSF andthe interference from optical diffraction, the edge pixel intensity distribution of the submicronlinewidth become distortion and is overlaid with noise, which reduce the edge localizationaccuracy. According to the direction information measure and the theory of invariant momentthe edge detector of “sum and difference of neighborhoods along axis (SDNAA)” is defined,and the formula of high order SDNAA is induced to construct subpixel edge localizationmethod. Experimental investigations show that this algorithm has strong ability to resist noiseand high precision for edge localization. By use of the algorithm the linewidth of2μm on thephotomask is measured with the erros lower than0.1μm, and the precision is higher than thatof the edge straight fitting method.