Prediction of strong coupling in resonant perovskite metasurfaces by deep learning

Resonant metasurfaces are often used to achieve strong coupling, and numerical simulations are the common method for designing and optimizing structural parameters of metasurfaces, while their calculation process takes a lot of time and occupies more computing resources. In this work, the deep learning strategy is proposed to simulate the strong coupling phenomenon in resonant perovskite metasurfaces. The designed fully connected neural network is constructed based on the deep learning algorithm that is used to predict transmission spectra, multipole decomposition spectral lines, and anti-cross phenomena of a perovskite metasurface. Through comparison of numerical simulation results, it can be seen that the neural network can efficiently and accurately predict the strong coupling phenomenon. Compared with the traditional design process, the proposed deep learning model can guide the design of the resonant metasurface more quickly, which significantly improves the feasibility of the design in complex metasurface structures.