Next-Generation Multi-layer Metasurface Design: Hybrid Deep Learning Models for Beyond-RGB Reconfigurable Structural Colors

Abstract

Metasurfaces are key to the development of flat optics and nanophotonic devices, offering significant advantages in creating structural colors and high-quality factor cavities. Multi-layer metasurfaces (MLMs) further amplify these benefits by enhancing light-matter interactions within individual nanopillars. However, the numerous design parameters involved make traditional simulation tools impractical and time-consuming for optimizing MLMs. This highlights the need for more efficient approaches to accelerate their design. In this work, we introduce NanoPhotoNet, an AI-driven design tool based on a hybrid deep neural network (DNN) model that combines convolutional neural networks (CNN) and Long Short-Term Memory (LSTM) networks. NanoPhotoNet enhances the design and optimization of MLMs, achieving a prediction accuracy of over 98.3% and a speed improvement of 50,000x compared to conventional methods. The tool enables MLMs to produce structural colors beyond the standard RGB region, expanding the RGB gamut area by 163%. Furthermore, we demonstrate the generation of tunable structural colors, extending the metasurface functionality to tunable color filters. These findings present a powerful method for applying NanoPhotoNet to MLMs, enabling strong light-matter interactions in applications such as tunable nanolasers and reconfigurable beam steering.