Training deep neural networks for the inverse design of nanophotonic structures

TL;DR

This paper introduces a tandem neural network architecture that effectively trains deep models for inverse photonic design despite data non-uniqueness, enabling the creation of complex nanophotonic structures.

Contribution

It proposes a novel tandem architecture combining forward modeling and inverse design to address data inconsistency in training deep neural networks for nanophotonic inverse design.

Findings

Overcomes non-uniqueness in inverse scattering problems

Enables training with large, non-unique datasets

Facilitates design of complex photonic structures

Abstract

Data inconsistency leads to a slow training process when deep neural networks are used for the inverse design of photonic devices, an issue that arises from the fundamental property of non-uniqueness in all inverse scattering problems. Here we show that by combining forward modeling and inverse design in a tandem architecture, one can overcome this fundamental issue, allowing deep neural networks to be effectively trained by data sets that contain non-unique electromagnetic scattering instances. This paves the way for using deep neural networks to design complex photonic structures that requires large training sets.