A deep learning-driven forward and inverse cooperative network for circular dichroism in chiral metasurfaces

Abstract

Chiral metasurfaces exhibit pronounced circular dichroism (CD), positioning them as highly promising for applications in sensing, communication, and nanophotonic devices. Traditional methods for designing chiral metasurfaces encounter significant challenges in terms of high efficiency and precision. In this study, we propose a letter b-like shaped chiral metasurface, in which a high CD value of ∼0.785 can be achieved in the shortwave infrared band. To improve the efficiency and accuracy of the design process, fully connected forward and inverse collaborative networks (FICN) integrated with artificial neural network (ANN) technology are utilized for rapid and precise parameter selection of the chiral metasurface, achieving a mean squared error value of 1.6 × 10⁻⁴. Through multiple training tests, the average values of mean absolute percentage error and root mean square error can reach 4.79% and 1.532 × 10⁻², respectively, surpassing those of other classical machine learning algorithms. Our research results are anticipated to promote high precision and high efficiency of the CD responses for chiral metasurfaces.