Machine Learning Approach for Optical Non-Destructive Nanometrology of Nano (Micro) Scale Rough Surfaces

A comprehensive approach for non-contact, non-destructive, and high-precision analysis of surface inhomogeneity distributions has been proposed, enabling reconstruction of the surface height distribution with coefficients of determination (R 2 ) ranging from 0.86 to 0.90, based on the photoluminescence intensity distribution of surface-deposited perovskite nanoparticles.Perovskite nanoparticles of about 5 nm were employed as functional markers, reproducing the surface topographic profile after deposition.A machine learning model based on graph convolutional networks (GCN) was developed to analyze and reconstruct the height distribution of surfaces with nano-and micro-inhomogeneities.The model was trained using experimental data obtained by atomic force microscopy to determine the surface height profile, together with measurements of the luminescence intensity of the deposited nanoparticles.