Design of Nanofabrication-Robust Metasurfaces Through Deep Learning-Augmented Multiobjective Optimization

Research output: Chapter in Book/Report/Conference proceedingChapter

Abstract

Recent developments in optimization and inverse-design techniques have repeatedly increased the achievable performance ceiling of optical metasurfaces, demonstrating that high-efficiency nanophotonic devices operating at optical wavelengths are feasible through freeform meta-element parameterizations. However, devices which achieve very high performance in simulation often lose their edge during the nanofabrication process due to certain common defects. In this chapter, we show how augmenting a freeform metaelement inverse-design process with deep learning makes it possible to not only characterize performance loss due to these fabrication errors, but also how to control for them. Through the proposed technique, exhaustive tolerance analysis of metasurface supercells is integrated into a multiobjective optimization so that tradeoffs between nominal performance and fabrication robustness can be directly analyzed. Moreover, by introducing deep learning into the optimization process, such studies are made tractable, with runtimes on the order of days rather than months.

Original languageEnglish (US)
Title of host publicationAdvances in Electromagnetics Empowered by Artificial Intelligence and Deep Learning
Publisherwiley
Pages253-279
Number of pages27
ISBN (Electronic)9781119853923
ISBN (Print)9781119853893
DOIs
StatePublished - Jan 1 2023

All Science Journal Classification (ASJC) codes

  • General Engineering

Cite this