Modeling of All-Dielectric Metasurfaces Using Deep Neural Networks

This paper presents a novel approach using deep neural networks to model all-dielectric meta-atoms over a wide spectrum. A deep neural network is constructed and fully trained in a supervised way to predict the performance of dielectric meta-atoms. The proposed approach has been validated by comparing the obtained results with full-wave simulation results.