Deep Learning for Rapid Analysis of Spectroscopic Ellipsometry Data

Abstract

High-throughput experimental approaches to rapidly develop new materialsrequire high-throughput data analysis methods to match. Spectroscopic ellips-ometry is a powerful method of optical properties characterization, but forunknown materials and/or layer structures the data analysis using traditionalmethods of nonlinear regression is too slow for autonomous, closed-loop, high-throughput experimentation. Herein, three methods (termed spectral, piecewise,and pointwise) of spectroscopic ellipsometry data analysis based on deeplearning are introduced and studied. After initial training, the incremental time forinferring optical properties can be a thousand times faster than traditionalmethods. Results for multilayer sample structures with optically isotropicmaterials are presented, appropriate for high-throughput studies of thinfilms ofphase-change materials such as Ge─Sb─Te (GST) alloys. Results for studies onhighly birefringent layered materials are also presented, exemplified by thetransition metal dichalcogenide MoS2. How the materials under test and theexperimental objectives may guide the choice of analysis methods are discussed.The utility of our approach is demonstrated by analyzing data measured on acomposition spread of Ge─Sb─Te phase-change alloys containing 177 distinctcompositions, and identifying the composition with optimal phase-changefigureof merit in only 1.4 s of analysis time.