Total: 1
Metasurface coupling constitutes a fundamental yet intricate electromagnetic interaction that occurs within a lattice of artificial subwavelength unit cells. Despite its prevalence, such coupling is typically ignored in conventional metasurface design frameworks due to the high characterization complexity, leading to suboptimal device performance. Here, we reveal a distinctive long-range coupling that exceeds an order of magnitude compared with the interaction range of evanescent waves, substantially enriching the metasurface design landscapes. This coupling exhibits pronounced graph topological features, and we design a graph neural network (GNN) to accurately abstract its inherent physics. Through strategic enhancement of the coupling effects, the discrete metasurface responses are transformed into continuous states, thereby unlocking diverse multiplexing channels. By further integrating the GNN into an inverse design agent, we tailor the multi-channel global response of metasurface to support simultaneous multiplexing across angle, frequency, and polarization domains. Experimentally, we demonstrate a compact metasurface multiplexer with eight independent channels, showcasing its potential for next-generation vehicular networks. This work establishes a new paradigm for highly integrated multifunctional metasurfaces, with promising prospects for high-density optical storage, information encryption, and high-capacity wireless communication.