Plasmonic Hybrid Resonances Inverse Design in Asymmetric Core-Shell Silver Nanowires with Artificial Neural Networks

Artificial neural network (ANN) based modeling is becoming a new efficient tool in the field of nanophotonic simulation with capability of efficient inverse design. In this paper, we show that fully-connected feed-forward neural network with proper architecture can learn mapping from multiple resonance spectrum to core-shell nanowire structure geometry. When fed as input a set of desired wavelengths and their Q-factors, the network outputs candidate geometry parameters. Dataset for training is generated from semi-analytical solution of the Helmholtz equation of corresponding electromagnetic wave scattering problem.