Fast and Accurate Modeling and Optimization of Large Metasurface Considering Near Couplings

When considering near couplings, the design and optimization of metasurfaces would consume huge time and memory costs. Deep neural networks provide unprecedented computational efficiency and accuracy for solving complex electromagnetic calculation problems. In this paper, two deep neural networks (DNN) are designed to solve the fast design and optimization problems of surface-enhanced Raman scattering (SERS) and focusing with metasurface respectively. The numerical calculation results prove the effectiveness of the work in this paper.