Inverse Design of Dual-resonant Absorption Photonic Structure based on Deep Learning

Abstract

Deep learning has made great progress in the field of inverse design of photonic structures, but the general artificial neural network has the problem of falling into a local minimum in inverse design. We introduce adaptive BN to solve the problem of difficult convergence and large error in a small sampling space. Using this method to predict the photonic structure parameters of graphene corresponding to the double resonance perfect absorption spectrum, a higher prediction accuracy is obtained., showing the superiority of the adaptive BN artificial neural network, and realizing the photonic structure of the on-demand spectral response anti-design.