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

We present a novel method for the inverse design of one-dimensional (1D) photonic stubbed systems with specific topological properties using deep learning techniques. Our approach involves building a data-driven model that predicts the geometric parameters of the photonic system from a label vector representing the desired topological features. A tandem neural network, consisting of an inverse network linked to a pre-trained forward network, is trained to capture the complex relationship between the system's topology and its geometry. Once trained, the model successfully performs inverse design, offering new insights for designing topological photonic systems.