Adaptively reverse design of terahertz metamaterial for electromagnetically induced transparency with generative adversarial network

Abstract

Metamaterials for electromagnetically induced transparency (EIT) have promoted prosperous development of terahertz (THz) devices due to their counterintuitive manipulation rules on the electromagnetic responses. However, traditional design rules of EIT metamaterial require prior knowledge of unnatural parameters of geometrical structures. Here, by taking full advantages of unsupervised generative adversarial networks (GANs), we propose an adaptively reverse design strategy to achieve intelligent design of metamaterial structures with the EIT phenomenon. The game theory ingrained in the GAN model facilitates the effective and error-resistant design process of metamaterial structures with preset electromagnetic responses and vice versa. The close match between the preset electromagnetic response and that from the generated structure validates the feasibility of the GAN model. Thanks to high efficiency and complete independence from prior knowledge, our method could provide a novel design technique for metamaterials with specific functions and shed light on their powerful capabilities on boosting the development of THz functional devices.