Deep-learning-assisted inverse design of dual-spin/frequency metasurface for quad-channel off-axis vortices multiplexing

Abstract. Recently, the metasurfaces for independently controlling the wavefront and amplitude of two orthogonal circularly polarized electromagnetic (EM) waves have been demonstrated to open a way toward spin-multiplexing compact metadevices. However, these metasurfaces are mostly restricted to a single operation frequency band. The main challenge to achieving multiple frequency manipulations stems from the complicated and time-consuming design caused by multifrequency cross talk. To solve this problem, we propose a deep-learning-assisted inverse design method for designing a dual-spin/frequency metasurface with flexible multiplexing of off-axis vortices. By analyzing the cross talk between different spin/frequency channels based on the deep-learning method, we established the internal mapping relationship between the physical parameters of a meta-atom and its phase responses in multichannels, realizing the rapid inverse design of the spin/frequency multiplexing EM device. As a proof of concept, we demonstrated in the microwave region a dual-frequency arbitrary spin-to-orbit angular momentum converter, a dual-frequency off-axis vector vortex multiplexer, and a large-capacity (16-channel) vortex beam generator. The proposed method may provide a compact and efficient platform for the multiplexing of vortices, which may further stimulate their applications in wireless communication and quantum information science.