Intelligent Adaptive Broadband Metasurface with Multilayer Generative Network

Abstract

Adaptive manipulation of electromagnetic scattering is useful to many applications, such as invisibility cloak, wireless communications, and optical imaging. In spite of great advancements brought by metamaterials and metasurfaces over the past decade, the existing works still face great challenges in broadband adaptability. Here, this study presents an intelligent metasurface with a novel structure driven by a conditional data integrated generative network, CVAE (conditional generative autoencoder), capable of autonomous optimization of electromagnetic scattering from 10 to 14 GHz. The proposed condition‐fused generative framework resolves multi‐solution ambiguities by synergistically evolving geometric configurations through sampling electromagnetic simulation‐guided latent space, achieving over 90% accuracy in scattering matrices while suppressing parasitic resonances. By co‐integrating tunable metasurfaces with real‐time intelligent control, this platform enables precise control of arbitrary scattering, from specular reflection to diffuse scattering, under varying illumination conditions. This work establishes a universal paradigm for adaptive electromagnetic engineering, with transformative implications for next‐generation cognitive radar, intelligent wireless management, and reconfigurable sensing.