Inverse engineering of electromagnetically induced transparency in terahertz metamaterial via deep learning
Wei Huang, Ziming Wei, Benying Tan, Shan Yin, Wentao Zhang
Abstract
Abstract In this paper, we apply the deep learning network to the inverse engineering of electromagnetically induced transparency (EIT) in terahertz metamaterial. We take three specific points of the EIT spectrum with six inputs (each specific point has two physical values with frequency and amplitude) into the deep learning model to predict and inversely design the geometrical parameters of EIT metamaterials. We propose this algorithm for the general inverse design of EIT metamaterials, and we demonstrate that our method is functional by taking one example structure. Our deep learning model exhibits a mean square error of 0.0085 in the training set and 0.014 in the test set. We believe that this finding will open a new approach for designing geometrical parameters of EIT metamaterials, and it has great potential to enlarge the applications of the THz EIT metamaterial.