Intelligent Analysis Of Spectral Ellipsometry Data Based On Deep Learning

Spectral ellipsometry data analysis is based on optical model.Firstly establishing an optical model is to simulate the measured sample.Secondly comparing the ellipsometry measurement data with the data generated by the model,optimizing the model parameters until the two match well.The key is to build an optical model so that the optical model can characterize the sample.The most important part of building an optical model is to construct the dielectric function model of the material.For the absorption materials,the dielectric function model is more complicated.It is usually composed of several general oscillator models at different wavelengths.The existing ellipsometry data analysis depends on the operator.The ability of the operator to build an accurate optical model requires an understanding of the characteristics of different dispersion models,familiarity with many types of materials,and data analysia experience.This paper focuses on the ellipsometric data analysis,combined with traditional ellipsometric data analysis,we propose a method of constructing a dielectric function model based on neural network.The main content and innovations include several aspects.Firstly,based on the traditional ellipsometric data analysis,we divide the construction of the dielectric function model into classification and regression problems.A scheme for intelligently constructing the dielectric function model is proposed to standardize and intelligentize the ellipsometric data analysis process.Secondly,for the oscillator classification and parameter regression task,we design the convolutional neural network.For network optimization,optimization methods such as weight initialization,regularization,and deepening the network layers are used.Finally,we apply the method proposed in this research on the data analysis of actual samples using classification and regression neural network,which proves that this method is feasible and works well.