Inverse Design of One-Dimensional Topological Photonic Systems Using Deep Learning

Abstract

We demonstrate a novel approach to inversely design one-dimensional (1D) photonic stubbed systems with targeted topological properties by leveraging the power of deep learning. The process involves developing a data-driven model to accurately predict the geometric parameters of the photonic system based on a label vector that encodes the targeted topological properties. A tandem network comprising an inverse network connected to a pre-trained forward network is trained to efficiently learn the intricate relationship between the system’s topological properties and the corresponding geometry. After training, the model is shown to effectively perform the inverse design task. The study’s outcomes give new perspectives for the design of topological photonic systems.