Deep neural network for designing near- and far-field properties in plasmonic antennas

Abstract

The electromagnetic response of plasmonic nanostructures is highly sensitive to their geometric parameters. In multi-dimensional parameter space, conventional full-wave simulation and numerical optimization can consume significant computation time and resources. It is also highly challenging to find the globally optimized result and perform inverse design for a highly nonlinear data structure. In this work, we demonstrate that a simple multi-layer perceptron deep neural network can capture the highly nonlinear, complex relationship between plasmonic geometry and its near- and far-field properties. Our deep learning approach proves accurate inverse design of near-field enhancement and far-field spectrum simultaneously, which can enable the design of dual-functional optical sensors. Such implementation is helpful for exploring subtle, complex multifunctional nanophotonics for sensing and energy conversion applications.