Deep neural network-aided design of terahertz bifunctional metasurface

Based on the insulator-to-metal phase transition of vanadium dioxide (VO2), we present a bifunctional meta-surface (MS), which can be switched between the two working states, i.e., a broadband absorber and a broadband polarization converter. When the top VO2 is in the metallic state, the MS can work as an absorber, which has an absorption bandwidth of 3.53 THz with an >= 90% absorption rate; and conversely, when the top VO2 is in the insulating state, the structure can function as a polarization converter, which exhibits a bandwidth of 3.0 THz with a >= 90% conversion efficiency. The overlapped bandwidth of the two states is the widest when compared with the other bifunctional counterparts. Meanwhile, this bifunctional MS has good angular and parametric tolerance characteristics. In the MS's design, the deep neural network (DNN) is also utilized to assist us in optimizing the structural parameters efficiently. The proposed structure and design method of the bifunctional MS may provide a valuable reference for new multifunctional terahertz devices.