Investigation and predictive modeling of the optical behavior of chalcogenide thin film using different artificial neural network techniques

Abstract

The \({\text{Te}}_{72}{\text{Ge}}_{24}{\text{As}}_{4}\) samples were recently created in our laboratory in bulk form using the traditional melt-quench method. For its optical characterization. The studied thin film samples have been created using physical vapor deposition. By selecting the 400 nm to 2500 nm spectral range of wavelength, the spectral of the experimental transmission T(λ) and reflectance R(λ) for the studied film samples have been employed to examine optical characteristics. First, we have determined the extinction coefficient (\(k\)) and refraction index (\(n\)) indices and their spectral distribution of them. Using Tauc's theory, we then computed the optical band gap \({E}_{\text{opt}}\). Urbach energy \({E}_{r}\) is determined from the linear dependence of photon energy on the absorption coefficient which was taken as an indicator to identify the disorder degree in the films. The additional variables, like the dissipation and quality factors, the dielectric constant in complex form, optical, thermal, and electrical conductivity, and volume/surface energy were measured. A comprehensive analysis and predictive modeling using various artificial neural networks (ANNs) techniques were applied to examine the optical behavior of the film samples studied. Materials made of chalcogenide are well-known for having special optical properties, making them appropriate for applications in photonics and optoelectronics. We employed multiple architectures, including Feedforward Neural Networks (FNN) and Recurrent Neural Networks (RNN), to model the extinction coefficient (\(k\)) and the refractive index (\(n\)) of these films using experimental data. The performance of each model was evaluated using metrics such as mean squared error MSE and correlation coefficients R². The optical parameters relevant to absorbance, refractive indices, and dielectric coefficients are computed rely on the modeling results and compared with those computed based on experimental measurements. Results demonstrate that FNN and RNN effectively capture the complex relationships between the optical parameters and exhibit small error rates. FFN shows superior accuracy in prediction. That highlights the potential of ANN techniques for advancing the understanding of chalcogenide materials and their applications in modern technology.