Design and Optimization of High-performance Multilayer Coatings in the Far-ultraviolet Region Based on Machine Learning Techniques

We have developed a neural network (NN) for the fast and accurate prediction and determination of the thickness of individual layers in multilayer coating designs in the far ultraviolet (FUV) region with a working wavelength of 100–190 nm. Within this range of wavelengths, the absorption coefficient of materials is typically extremely high, making it challenging to achieve high-performance coating designs, such as high-reflectance films or wideband anti-reflective films, using conventional quarter-wavelength optical thickness designs. To address this challenge, we have developed forward and inverse neural networks. By training on a large dataset of simulated or experimentally measured spectra, the model can predict optimal layer configurations that minimize the discrepancy between the designed coating's spectral response and the target spectrum. This methodology ensures alignment between the optical properties of the coating design and the specific requirements of the application.