Conditional Wasserstein Generative Adversarial Network for Metasuface Design

Abstract

As a method for inverse design of metasurfaces, we propose a conditional Wasserstein generative adversarial network that combines the design idea of a conditional generative adversarial network with the optimization method of a Wasserstein generative adversarial network with gradient penalty. After being trained, the proposed network immediately generated desirable designs with large degrees of freedom for given intended spectra. The proposed network’s ability for metasurface design was demonstrated by designing a metasurface absorber. This inverse design approach can serve as an efficient tool to accelerate the design of complex metasurfaces or photonic structures that have desired optical properties.