A physics-driven neural network framework for end-to-end inverse design of metasurface-based holograms

Wei Wei*, Ping Tang, Jingzhu Shao, Jiang Zhu, Xiangyu Zhao, Chongzhao Wu

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

A novel unsupervised deep neural network framework driven by a physics model is introduced to design metasurface-based holograms. The proposed framework shows perfect reconstructions of holographic images with a shorter prediction time, higher peak signal-to-noise ratio and better structural similarity compared with the conventional Gerchberg-Saxton algorithm. An end-to-end design of metasurface-based holograms without requirements of complete light modulation is demonstrated. The proposed framework opens up a new approach to inverse design of metasurface-based photonic devices.

Original languageEnglish
Title of host publicationIRMMW-THz 2023 - 48th Conference on Infrared, Millimeter, and Terahertz Waves
PublisherIEEE Computer Society
ISBN (Electronic)9798350336603
DOIs
Publication statusPublished - 2023
Externally publishedYes
Event48th International Conference on Infrared, Millimeter, and Terahertz Waves, IRMMW-THz 2023 - Montreal, Canada
Duration: 17 Sept 202322 Sept 2023

Publication series

NameInternational Conference on Infrared, Millimeter, and Terahertz Waves, IRMMW-THz
ISSN (Print)2162-2027
ISSN (Electronic)2162-2035

Conference

Conference48th International Conference on Infrared, Millimeter, and Terahertz Waves, IRMMW-THz 2023
Country/TerritoryCanada
CityMontreal
Period17/09/2322/09/23

Fingerprint

Dive into the research topics of 'A physics-driven neural network framework for end-to-end inverse design of metasurface-based holograms'. Together they form a unique fingerprint.

Cite this