AI for accelerated simulations of plasmonic nanostructures
Abstract
The spectral responses of large aperiodic plasmonic metasurfaces were observed to hold essential importance for the practical and theoretical studies. Recent developments in machine-learning approaches have proved effective in determining the optical properties of nanostructures. In particular, a neural network has been designed and trained to predict the spectral response and near-field distribution of complex multiscale patterns of plasmonic nanostructures. This technique neatly accounts for spectral coupling in the far- or near-field domain with a computational saving (x100) compared to traditional Maxwell solvers. As such, the neural-network representation provides a promising path forward for fast design and optimization of plasmonic nanostructures and metasurfaces.