Terahertz Metasurface Spectrum Prediction Based On Deep Learning

Metasurface, a two-dimensional periodic artificial electromagnetic material, usually composed of metal/medium or all medium. It has become a popular research area due to its unique electromagnetic properties. However, due to the complex modeling and electromagnetic simulation, the study of the electromagnetic characteristics of terahertz metasurface is time-consuming. To solve this problem, we introduced deep learning method to explore and optimize the terahertz metasurface design process. We constructed a terahertz metasurface structure, and obtained a small part of spectral data in the whole sample space through software simulation, and then put these data into the neural network for learning. By verifying the trained neural network, the spectral information of terahertz mteasurface with different structural parameters can be predicted. This method can help us to simplify the design process of metasurface.