A Deep Learning Guided End-to-End Electromagnetic Metasurface Design Framework
Electromagnetic metasurfaces have shown great potential in the field of antenna design thanks to their abilitiesto systematically manipulate electromagnetic waves in extraordinary ways such as the generalized refraction andreflection, impedance matching, polarization control, etc. However, the synthesis of a metasurface based on therequired scattering properties still largely relies on the designer’s intuition and experience as well as manyiterations of full-wave electromagnetic simulations which is an expensive and time-consuming process. The newlyemerged deep-learning neural networks have shown great potential in solving challenging problems across manydifferent disciplines. In this project, we propose a deep learning based metasurface design framework that canguide antenna engineers to go from desired far-field requirements (such as the main beam direction, half-powerbeamwidth, null locations, and side lobe levels) to appropriate physical designs (i.e., copper trace design ondielectric substrates), thus, eliminate the traditional iterative designing process. This project will be used as thecore product of the partnered organization.