The unique properties of electromagnetically induced transparency (EIT) metamaterials have caused a lot of concern in the field of terahertz wave regulation, but the traditional design methods of metamaterials have the problems of long design cycles and high trial and error costs. Applying the deep learning method to the inverse design process of terahertz metamaterials can greatly reduce the design complexity so that the EIT metamaterial structure can be quickly designed according to the requirements. This paper constructs a generative adversarial network (GAN) model for EIT metamaterial structure design, which realizes the mapping relationship between the target spectrum and metamaterial structure parameters. The proposed GAN model can accurately predict structure parameters of the EIT metamaterial according to the target spectrum, and the error between the generated and the real parameters is less than 1μm. Moreover, by introducing fuzzy processing, the proposed GAN model can accurately generate multiple sets of metamaterial structures according to the same target spectrum, providing more options for designers. This model offers a novel and efficient design method for EIT metamaterials.

Deep learning,Inverse design,Electromagnetically induced transparency metamaterials