Chiral Metasurface Vortex Focusing in Terahertz Band Based on Deep Learning

Abstract

This work demonstrates an on-demand design approach to reverse the design of the chiral metasurface structure by the given target circular dichroism (CD) value, which is based on deep learning. A conditional generation network (RCGN) with a mean absolute error (MAE) of 0.027 was built in this work. Using this network, the metasurface structure with a CD up to 0.471 can be designed, verifying the validity and feasibility of the reverse metasurface design structure. At a frequency of 1.126 THz, the right-hand circularly polarised wave can be phase-modulated by Pancharatnam Berry (PB) to achieve the function of focusing the vortex beam of a metasurface array. The addition of the reverse design-on-demand feature based on deep learning makes the process of designing metasurface structures much faster and more accurate.