Wei Ma and colleagues from Northeastern University, USA have developed an accurate, efficient deep-learning-based model for designing three-dimensional chiral metamaterials. The model consists of two bidirectional neural networks, with a forward combiner and an inverse combiner. It is trained heuristically with multiple functions for fast prototyping, optimization and inverse design. The unit cell of the chiral metamaterial under study consists of two stacked gold split-ring resonators twisted at a certain angle and separated by two spacing dielectric layers with a continuous gold reflector at the bottom. The team demonstrated that the model can significantly shorten the prediction time for generating chiral dichroism and can solve the design-on-demand inverse problem accurately and efficiently, retrieving geometric parameters of the metamaterial from specific requirements of its optical response.
Author information
Authors and Affiliations
Corresponding author
Rights and permissions
About this article
Cite this article
Won, R. Deep-learning boost. Nature Photon 12, 443 (2018). https://doi.org/10.1038/s41566-018-0231-3
Published:
Issue Date:
DOI: https://doi.org/10.1038/s41566-018-0231-3