Characterization of thin films by neural networks and analytical approximations

Abstract

In this work we focus on the characterization of thin films by Electrostatic Force Microscopy (EFM). We use the estimations made by Artificial Neural Networks (ANNs) trained by numerical results from the Generalized Image Charge Method (GICM). The ANN outputs suggest that an effective dielectric constant can be defined for any thin film sample. The definition of an effective dielectric constant allows us to include complex thin film samples in analytical approximations previously developed for much simpler surfaces. From the new analytical formulation we can conclude that the dielectric thin films interact in a complex way.Moreover, the metallic plate below the surface changes the effective dielectric constant, suggesting that its contribution cannot be isolated from the one of the dielectric plates.

Photoacoustic Characterization of TiO2 Thin-Films Deposited on Silicon Substrate Using Neural Networks

“…The network designated as NN4 has three neurons in the output layer that simultaneously predict all three mentioned parameters. The bases formed for the training of the first three networks were made individually (Base 1, Base 2 and Base 3), while the training base NN4 (Base 4) was made by merging all three individual bases [ 67 , 68 , 69 , 70 ].…”

Determining optical constants of 2D materials with neural networks from multi-angle reflectometry data

“…With the availability of open-source and easy to use libraries [1,2] and graphics processing units at affordable prices, researchers from various disciplines of science and engineering are using artificial neural networks to learn from and make predictions on data in various forms. Optical material characterization based on reflectometry (or ellipsometry) data is one of these applications, where deep learning has been implemented to identify two-dimensional (2D) nanostructures [3][4][5][6] and to obtain optical constants of particles [7], thin films [8,9], solutions [10], tissues [11], and soils [12]. This work focuses on determining optical constants of atomically thin layered materials as follows.…”

“…Hence, the method used for optical property extraction should be not only accurate but also efficient. As previously mentioned [3][4][5][6][7][8][9][10][11], deep learning is a promising field for this very purpose, which can be used in two different ways: regression [8] or classification [12,19]. Since previous regression based Original content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence.…”

“…approaches have relatively high prediction errors, e.g. 5% in [8], we follow the second approach to predict a continuous output from discrete classes. This paper is organized as follows.…”