Machine learning classification for field distributions of photonic modes

TL;DR

This paper introduces a machine learning approach combined with finite element simulations to classify and analyze three-dimensional photonic modes, aiding optimization in nanophotonics applications.

Contribution

It presents a novel clustering method that reduces complex field distribution data to prototypes, enhancing analysis of photonic modes for various nanophotonics applications.

Findings

Successfully identified modes with large field energies and specific spatial properties

Demonstrated predictive power using fluorescence enhancement data

Applicable to diverse optimization tasks involving 3D field data

Abstract

Machine learning techniques can reveal hidden structure in large data amounts and can potentially extent or even replace analytical scientific methods. In nanophotonics, modes can increase the light yield from emitters located inside the nanostructure or near the surface. Optimizing such systems enforces to systematically analyze large amounts of three-dimensional field distribution data. We present a method based on finite element simulations and machine learning for the identification of modes with large field energies and specific spatial properties. By clustering we reduce the field distribution data to a minimal subset of prototypes. The predictive power of the approach is demonstrated using an analysis of experimentally measured fluorescence enhancement of quantum dots on a photonic crystal surface. The clustering method can be used for any optimization task that depends on three-dimensional field data, and is therefore relevant for biosensing, quantum dot solar cells or photon upconversion.