AI-enabled electromagnetic metasurfaces for wireless communication and invisibility cloak

Programmable metasurfaces have evolved into dynamic electromagnetic (EM) interfaces capable of manipulating wavefronts across spatial, temporal, spectral, and information domains. This review summarizes recent advances in programmable metasurfaces and their integration with artificial intelligence (AI). We first introduce the fundamental coding mechanisms, including spatial coding, temporal modulation, and space-time coding (STC), which provide the physical basis for multidimensional EM control. We then review AI-enabled methodologies for metasurfaces, covering inverse design, large-scale and STC metasurface synthesis, and closed-loop systems integrating sensing, learning, and real-time EM control. Representative system-level applications are discussed in two directions. For wireless communication, programmable metasurfaces enable radio-environment orchestration, direct information modulation, and integrated sensing and communication. For stealth-oriented applications, they support AI-assisted cloaking design, adaptive invisibility, and EM signature regulation, including Doppler-signature manipulation. Finally, we discuss key challenges in data efficiency, physical consistency, experimental validation, hardware scalability, energy consumption, and closed-loop deployment. This review presents programmable metasurfaces as a unifying platform concept for intelligent EM systems, while recognizing that current system-level demonstrations remain more mature in the microwave and millimeter-wave regimes.