AI for designing miniaturised optics for communication and energy harvesting

Abstract

Metasurfaces are revolutionary optical devices that can manipulate wave properties such as amplitude, phase, and polarisation with extreme precision, which can be used in applications such as beam steering and focusing beam scanning that are useful for LiDAR applications. However, crafting an optimal solution to design such a sophisticated device presents many challenges. This project aims to find an optimal solution to design a metasurface through the integration of numerical simulation and machine learning, specifically focusing on the implementation of Graph Neural Networks (GNN). The study focuses on understanding the interactions between the metasurface structures and their electric field interactions, aiming to develop an optimized design that maximises transmission for the first-order of diffraction. By incorporating TORCWA for structure generation and utilising GNN for evaluation, this study not only focuses on improving the design process more efficiently but also offers an ‘inverse design’ model for future optical device innovations.