NanoPhotoNet: AI-enhanced design tool for reconfigurable and high-performance multi-layer metasurfaces

Metasurfaces are crucial in advancing flat optics and nanophotonics, offering unique advantages in creating vibrant structural colors and high-Q factor cavities. Multi-layer metasurfaces (MLMs) take this further by enhancing light-matter interactions inside the single meta-atom at the nanoscale. However, optimizing MLM designs is challenging due to the complex interplay of many parameters, making traditional simulation methods slow and inefficient. In this work, we introduce NanoPhotoNet, an advanced AI-powered design tool that leverages a hybrid deep neural network (DNN) combining convolutional neural networks (CNN) and Long Short-Term Memory (LSTM) models. NanoPhotoNet significantly accelerates the design process for MLMs, achieving over 98.3 % prediction accuracy and a 50,000x speed improvement compared to conventional techniques. This enables the creation of structural colors far beyond the standard RGB range, increasing the RGB gamut area up to 163 %. Additionally, NanoPhotoNet facilitates tunable color generation, extending the capabilities of MLMs to advanced applications like tunable color filters, nanolasers, and reconfigurable beam steering. This approach represents a transformative progress in metasurface design, unlocking new possibilities for high-performance, tunable nanophotonic devices.