Abstract

The burgeoning synergy between metasurfaces and artificial intelligence has emerged as a compelling research frontier, characterized by a mutually reinforcing paradigm: artificial intelligence enables intelligent nanophotonics, while metasurfaces facilitate optical intelligence. Leveraging its directional computation and extensive generalization capabilities, artificial intelligence serves as a powerful electromagnetic simulator and modeling engine, accelerating the development of self-adaptive meta-devices.

Conversely, metasurface acts as a versatile electromagnetic wave manipulator and calculating platform, offering inherent advantages in ultrafast processing, massive parallelism, and compact integration, thereby driving the evolution of intelligent computing from electronic to photonic domains. This review systematically surveys parallel advances along these two directions. For intelligent nanophotonics, we examine intelligence-driven methodologies, including physical modeling and structural optimization.

For optical intelligence, we discuss metasurface-enabled frameworks, covering optical mathematical computing and optical neural networks. Finally, we outline future roadmap, emphasizing practical deployment scenarios such as chip-scale intelligent visual perception and communication engineering, aiming to further foster this interdisciplinary filed to the next level.