In the past decade, programmable metasurfaces have attracted widespread attention due to their ability to manipulate electromagnetic waves at the subwavelength scale in space. By enabling controllable manipulation of the electromagnetic propagation environment, they not only provide new means for beam steering, coverage enhancement, and spectrum reuse in communication systems, but also demonstrate tremendous application potential in fields such as optical remote sensing, holographic imaging, and antenna integration. However, the design of metasurfaces involves a vast parameter space, high-complexity electromagnetic solutions, and nonlinear spatial coupling. These challenges make traditional electromagnetic simulation and optimization methods inefficient and difficult to achieve real-time control. Meanwhile, artificial intelligence, especially methods such as machine learning, deep neural networks, generative models, and reinforcement learning, has provided new ideas for handling complex nonlinear scenarios. This paper aims to comprehensively review the applications of these artificial Intelligence (AI) methods in metasurface structure design and reconfigurable intelligent surface (RIS) communication control, presenting rich literature cases to illustrate how various AI algorithms drive progress in this interdisciplinary field and providing the authors’ personal analysis and suggestions on the basis of this summary.
Zhan et al. (Tue,) studied this question.