Prior Knowledge-Guided Deep Learning Algorithms for Metantenna Design (Invited)

This paper presents the recent advances in deeplearning (DL)-enabled methods for metacell and metalens design. On one hand, metamaterials or metasurfaces have been new opportunities for innovative antenna (metantennas) designs by increasing the degrees of freedom (DoF) in design through introducing new physical properties in antenna design. On the other hand, the increased DoF significantly increases the complexity of design procedure, demanding for new design methods. Prior-knowledge-guided deep-learning-enabled (PK-DL) synthesis method has been then proposed to alleviate this challenge. Prior knowledge is introduced into the conditional deep convolutional generative adversarial networks (cDCGAN) algorithm to guide and speed up the proposed synthesis method. Then, two case studies are presented. The PK-DL synthesis method is first utilized to break the phaseshift limit up to 330 degrees from 308 degrees in existing techniques and improve the 1-dB gain bandwidth of metalens antennas by 52%. Secondly, both amplitude and phase responses of metacells are simultaneously synthesized in a single three-layered metasurface for a metalens antenna, realizing the sidelobe level suppression more than 30 dB. The latest progress demonstrates that the applications of artificial intelligence (AI) algorithms to metantenna design and AI technologies provide a new opportunity in synthesizing metantennas with outstanding performances and multiple functions.