Deep learning for circular dichroism of nanohole arrays

Abstract

Chiral metasurfaces with nanohole structures have a strong circular dichroism (CD) response and are easy to prepare. Therefore, they are widely used in many fields, such as biological monitoring and analytical chemistry. In this work, a deep learning (DL) framework based on the convolutional neural network (CNN) is proposed to predict the CD response of chiral metasurfaces. A dataset containing many data values is used to predict CD values, which are found to be highly consistent with those obtained from COMSOL Multiphysics simulation. Results show that the proposed CNN-based DL model is about a thousand of times faster than conventional finite element methods. It can accurately map chiral metasurfaces and predict their optical response with negligible loss functions. The insights gained from this research may be helpful in the study of complex optical chirality and the design of highly sensitive sensing systems in DL networks.