Advanced deep learning approaches in metasurface modeling and design: A review

Abstract

Nanophotonic devices have marked a significant advance in light control at the subwavelength level, achieving high efficiency and multifunctionality. However, the precision and functionality of these devices come with the complexity of identifying suitable meta-atom structures for specific requirements. Traditionally, designing metasurface devices has relied on time-consuming trial-and-error methods to match target electromagnetic (EM) responses, navigating an extensive array of possible structures. Recently, deep learning (DL) has emerged as a potent alternative, streamlining the forward modeling and inverse design process of nanophotonic devices. This review highlights recent strides in deep-learning-based photonic modeling and design, focusing on the fundamentals of various algorithms and their specific applications, and discusses the emerging research opportunities and challenges in this field.