Deep Learning Based Multi-layer Metallic Metasurface Design

An innovative deep-learning based design approach for multi-layer metallic metasurfaces has been presented in this paper. By leveraging the strength of deep-learning networks, the higher order degree of freedom for metasurface design is unlocked. The structure and performance of the proposed deep neural network (DNN) are discussed. It is demonstrated that it can predict the possible amplitude and phase coverage of metasurface unit-cells under consideration. The physical insight derived from the proposed DNN can be utilized towards achieving a more time efficient and accurate metasurface design.