Abstract:Chiral plasmonic metasurfaces, which exhibit distinct spectral responses for left-handed circularly polarized (LCP) and right-handed circularly polarized (RCP) incident light, are promising platforms for sensing chiral molecules, thanks to their great potential of enhancing light-matter interaction. A so-called Born-Kuhn analog for 3D chiral plasmonics, consisting of corner-stacked orthogonally arranged gold nanorods with C4 rotational symmetry, is a classical and valuable candidate for understanding and tailoring the enhanced interactions with chiral light. In spite of relatively clear physical picture behind, it is still very time consuming and may not be sufficient to achieve the optimum chiroptical responses by simply scanning dimensional parameters. Intelligent algorithms (IA) have emerged as efficient tools for nanophotonic design with remarkable time efficiency and achieved performances. Here we mainly develop an IA, by combining an adaptive genetic algorithm (AGA) and finite-difference time-domain (FDTD) simulation, aiming to achieve the most prominent circular dichroism (CD) at a specific wavelength (customized at 1035 nm, corresponding to our laser wavelength for sensing applications). The CD at 1035 nm reaches −0.52 for left-handed (LH) enantiomer and −0.46 for right-handed (RH) enantiomer assisted by our intelligent algorithm, 41% more remarkable compared with the results obtained by parametric scanning (5-fold more time needed). Besides, deep learning algorithms have recently attracted great attention in nanophotonic inverse design. With the spectra are fed to the input, deep learning networks (DNN) can directly output the structural parameters within short time on the order of milliseconds. Here we show some preliminary results that transfer learning could help to reduce the training time and dataset size when predicting dimensional parameters of the 3D chiral metasurfaces with opposite handedness, after transferring the parameters of an existing model used for training a DNN for left-handed enantiomers. We envision that our developed algorithms could provide time-effective routes for optimizing not only chiral metasurfaces but also other nanophotonic devices.

Deep Learning for the Design of 3D Chiral Plasmonic Metasurfaces

Efficient Design of 3D Chiral Plasmonic Metasurfaces Assisted by Intelligent Algorithms

Design of chiral plasmonic metamaterials based on interpretable deep learning

Neural Network-Based Surrogate Model for Inverse Design of Metasurfaces

Multitask Deep-Learning-based Design of Chiral Plasmonic Metamaterials

Deep-Learning-Enabled On-Demand Design of Chiral Metamaterials.

Intelligent Design of the Chiral Metasurfaces for Flexible Targets: Combining a Deep Neural Network with a Policy Proximal Optimization Algorithm

On-demand design based on deep learning and phase manipulation of all-silicon terahertz chiral metasurfaces

Optical Circular Dichroism Engineering in Chiral Metamaterials Utilizing a Deep Learning Network.

Chiral Metasurface for Near-Field Imaging and Far-Field Holography Based on Deep Learning

Deep-learning-assisted Designing Chiral Terahertz Metamaterials with Asymmetric Transmission Properties

Exploiting Deep Learning Network in Optical Chirality Tuning and Manipulation of Diffractive Chiral Metamaterials

Graphic-processable Deep Neural Network for the Efficient Prediction of 2D Diffractive Chiral Metamaterials.

Deep‐Learning Empowered Customized Chiral Metasurface for Calibration‐Free Biosensing

Data Enhanced Iterative Few-Sample Learning Algorithm-Based Inverse Design of 2D Programmable Chiral Metamaterials

Plasmonic nanostructure design and characterization via Deep Learning

Plasmonic Nanoparticle Simulations And Inverse Design Using Machine Learning

Flexible design of chiroptical response of planar chiral metamaterials using deep learning

Deep learning for circular dichroism of nanohole arrays

Inverse Design of Generic Metasurfaces for Multifunctional Wavefront Shaping Based on Deep Neural Networks

Efficient Inverse Design of Plasmonic Patch Nanoantennas using Deep Learning