Modeling optical spectra of disordered metallic nanostructures with feature-based machine learning

Abstract

Disordered nanostructures have numerous optical applications, including optical absorbers and photonic crystals. Multiple scattering generated in disordered nanostructures produces a remarkable optical behavior under illumination. However, conventional numerical methods have difficulty modeling disordered nanostructures because of their time-consuming processes and unpredictable geometry. In this study, feature-based machine learning (ML) with a multilayer perceptron was used to model disordered silver nanostructures. These nanostructures were fabricated on silicon substrates by varying silver deposition and annealing conditions. The extracted spatial features and size distributions of the disordered silver nanostructures were used as inputs for training the ML model, and their measured reflection spectra were used as the output. The ML model, constructed using the forward-propagation algorithm, can acquire optical interactions between the nanostructural features and reflection spectra. The validation results indicated that the coefficient of determination between the predicted and actual values exceeded 0.9. Moreover, the proposed model can realize a highly accurate reflectance spectrum for disordered metallic nanostructures within a short time (less than 30 s). This study holds significant potential for the rapid prediction of optical properties in disordered metallic nanostructures.