General Characterization of Intelligent Metasurfaces with Graph Coupling Network

Abstract

Intelligent metasurfaces, as the next‐generation of metasurfaces, have emerged as a versatile artificial electromagnetic (EM) medium capable of adaptively manipulating wave‐matter interactions, especially in the construction of EM space integration and the analogy of wave‐based neural networks. However, current computational landscape for intelligent metasurfaces relies either on time‐consuming full‐wave numerical simulations with excessive computational complexity or on application‐limited physical models that are difficult to consider the coupling effects. Here, a universal graph neural network (GNN) framework is introduced, incorporating the elusive coupling effects inside metasurfaces, enabling rapid and precise characterization with arbitrary‐large size. This framework exhibits exceptional compatibility with physical models, thereby significantly expanding the realm of potential design scenarios. By harnessing the principles of diffraction theory and near‐to‐far transformation algorithms, highly accurate modeling of the scattered fields emanating from metasurfaces is achieved. Through microwave experiments on intelligent metasurfaces, the efficacy of GNN in real‐world scenarios is effectively demonstrated. The utilization of topological strategies to characterize intelligent metasurfaces marks a major leap toward the next‐generation metasurfaces, promising transformative advancements across various applications.