Qian Ma, Zi Rui Feng, Junming Hou, Che Liu, Yu Ming Ning, Rui Si Li, Qian Wen Wu, Yi Su Huang, Yi Zhang, Ze Gu, Jian Wei You, Tie Jun Cui
Abstract The rapid development of artificial intelligence (AI) has revolutionized traditional design and optimization approaches for metamaterials. By leveraging AI techniques, the time required for the metamaterial design process is significantly reduced, thus improving the overall efficiency. On the other hand, the advancement of AI has placed extremely high demands on computational power. In exploring high-speed and energy- efficient next-generation computing hardware, metasurface-based intelligent computing technologies such as diffractive neural networks have attracted widespread attention. In this work, we focus on two main aspects: intelligent design of metasurfaces and intelligent computing based on metasurfaces. In the field of intelligent design, we analyze how deep learning-based inverse design methods overcome the efficiency bottleneck of traditional electromagnetic simulations and enable high-precision and automated generation of subwavelength structures. In intelligent computing, we comprehensively explore the implementation mechanisms of optical diffractive neural networks and microwave programmable neural networks based on metamaterials. By comparing the application performance of physical neural networks with different architectures in scenarios such as image recognition and wireless communication, we reveal the crucial role of deep integration of the metasurfaces and AI in driving new computing paradigms, providing a theoretical framework and technological roadmap for next-generation intelligent electromagnetic systems.