Fast Design of Plasmonic Metasurfaces Enabled by Deep Learning

Authors

Abhishek Mall, Abhijeet Patil, Dipesh Tamboli, Amit Sethi, and Anshuman Kumar

Abstract

Metasurfaces are ultra-thin structures composed of sub-wavelength antennae that manipulate light, enabling miniaturized optical elements. However, designing or optimizing a metasurface for a desired functionality traditionally requires costly, iterative trial-and-error. We propose a bidirectional autoencoder deep learning architecture that builds a Geometry and Parameter Space Library (GPSL) of metasurface designs with their optical responses. This GPSL serves as a universal design–response space for rapid template-based search. As an example, we demonstrate inverse design of a multi-band gap-plasmon half-wave plate metasurface, showing our network converges to multiple topologies with low error between desired and actual optical responses. Our approach addresses the one-to-many mapping in inverse design and enables fast, accurate design of metasurfaces with diverse functionalities.