Programmable chalcogenide-based all-optical deep neural networks

Ting Yu Teo; Xiaoxuan Ma; Ernest Pastor; Hao Wang; Jonathan K. George; Joel K.W. Yang; Simon Wall; Mario Miscuglio; Robert E. Simpson; Volker J. Sorger

doi:10.1515/nanoph-2022-0099

Programmable chalcogenide-based all-optical deep neural networks

Ting Yu Teo^*, Xiaoxuan Ma, Ernest Pastor, Hao Wang, Jonathan K. George, Joel K.W. Yang, Simon Wall, Mario Miscuglio, Robert E. Simpson^*, Volker J. Sorger^*

^*Corresponding author for this work

Electronic, Electrical and Systems Engineering

Research output: Contribution to journal › Article › peer-review

8 Citations (Scopus)

42 Downloads (Pure)

Abstract

We demonstrate a passive all-chalcogenide all-optical perceptron scheme. The network’s nonlinear activation function (NLAF) relies on the nonlinear response of Ge₂Sb₂Te₅ to femtosecond laser pulses. We measured the sub-picosecond time-resolved optical constants of Ge₂Sb₂Te₅ at a wavelength of 1500 nm and used them to design a high-speed Ge₂Sb₂Te₅-tuned microring resonator all-optical NLAF. The NLAF had a sigmoidal response when subjected to different laser fluence excitation and had a dynamic range of −9.7 dB. The perceptron’s waveguide material was AlN because it allowed efficient heat dissipation during laser switching. A two-temperature analysis revealed that the operating speed of the NLAF is ≤ 1 ns. The percepton’s nonvolatile weights were set using low-loss Sb2S3-tuned Mach Zehnder interferometers (MZIs). A three-layer deep neural network model was used to test the feasibility of the network scheme and a maximum training accuracy of 94.5% was obtained. We conclude that combining Sb2S3-programmed MZI weights with the nonlinear response of Ge₂Sb₂Te₅ to femtosecond pulses is sufficient to perform energy-efficient all-optical neural classifications at rates greater than 1 GHz.

Original language	English
Pages (from-to)	4073-4088
Number of pages	16
Journal	Nanophotonics
Volume	11
Issue number	17
DOIs	https://doi.org/10.1515/nanoph-2022-0099
Publication status	Published - 25 May 2022

Bibliographical note

Funding Information:
Research funding: The NLAF design, SbS MZI weights, material growth, and thermal modeling were supported by the Agency for Science, Technology and Research (A*STAR) under the Advanced Manufacturing and Engineering (AME) grant #A18A7b0058. The network training was supported from the Presidential Early Career Award for Scientist and Engineers (PECASE) nominated by the Department of Defense through the Air Force Office of Scientific Research under award number FA9550-20-1-0193. The pump–probe measurements were funded by Fundació Cellex, Fundació Mir-Puig, and Generalitat de Catalunya through CERCA. E.P acknowledges the support from IJC2018-037384-I funded by MCIN/AEI/10.13039/501100011033. 2 3

Publisher Copyright:
© 2022 Ting Yu Teo et al., published by De Gruyter, Berlin/Boston.

Keywords

all-optical deep neural network
chalcogenide reconfigurable photonics
ultra-fast dynamic response of phase change material

ASJC Scopus subject areas

Biotechnology
Electronic, Optical and Magnetic Materials
Atomic and Molecular Physics, and Optics
Electrical and Electronic Engineering

Access to Document

10.1515/nanoph-2022-0099Licence: Creative Commons: Attribution (CC BY)

TeoT2022ProgrammableFinal published version, 3.06 MBLicence: Creative Commons: Attribution (CC BY)

Cite this

@article{279e2eb1ab714f1d83586f189a43a57c,

title = "Programmable chalcogenide-based all-optical deep neural networks",

abstract = "We demonstrate a passive all-chalcogenide all-optical perceptron scheme. The network{\textquoteright}s nonlinear activation function (NLAF) relies on the nonlinear response of Ge2Sb2Te5 to femtosecond laser pulses. We measured the sub-picosecond time-resolved optical constants of Ge2Sb2Te5 at a wavelength of 1500 nm and used them to design a high-speed Ge2Sb2Te5-tuned microring resonator all-optical NLAF. The NLAF had a sigmoidal response when subjected to different laser fluence excitation and had a dynamic range of −9.7 dB. The perceptron{\textquoteright}s waveguide material was AlN because it allowed efficient heat dissipation during laser switching. A two-temperature analysis revealed that the operating speed of the NLAF is ≤ 1 ns. The percepton{\textquoteright}s nonvolatile weights were set using low-loss Sb2S3-tuned Mach Zehnder interferometers (MZIs). A three-layer deep neural network model was used to test the feasibility of the network scheme and a maximum training accuracy of 94.5% was obtained. We conclude that combining Sb2S3-programmed MZI weights with the nonlinear response of Ge2Sb2Te5 to femtosecond pulses is sufficient to perform energy-efficient all-optical neural classifications at rates greater than 1 GHz. ",

keywords = "all-optical deep neural network, chalcogenide reconfigurable photonics, ultra-fast dynamic response of phase change material",

author = "Teo, {Ting Yu} and Xiaoxuan Ma and Ernest Pastor and Hao Wang and George, {Jonathan K.} and Yang, {Joel K.W.} and Simon Wall and Mario Miscuglio and Simpson, {Robert E.} and Sorger, {Volker J.}",

note = "Funding Information: Research funding: The NLAF design, SbS MZI weights, material growth, and thermal modeling were supported by the Agency for Science, Technology and Research (A*STAR) under the Advanced Manufacturing and Engineering (AME) grant #A18A7b0058. The network training was supported from the Presidential Early Career Award for Scientist and Engineers (PECASE) nominated by the Department of Defense through the Air Force Office of Scientific Research under award number FA9550-20-1-0193. The pump–probe measurements were funded by Fundaci{\'o} Cellex, Fundaci{\'o} Mir-Puig, and Generalitat de Catalunya through CERCA. E.P acknowledges the support from IJC2018-037384-I funded by MCIN/AEI/10.13039/501100011033. 2 3 Publisher Copyright: {\textcopyright} 2022 Ting Yu Teo et al., published by De Gruyter, Berlin/Boston.",

year = "2022",

month = may,

day = "25",

doi = "10.1515/nanoph-2022-0099",

language = "English",

volume = "11",

pages = "4073--4088",

journal = "Nanophotonics",

issn = "2192-8606",

publisher = "De Gruyter",

number = "17",

}

TY - JOUR

T1 - Programmable chalcogenide-based all-optical deep neural networks

AU - Teo, Ting Yu

AU - Ma, Xiaoxuan

AU - Pastor, Ernest

AU - Wang, Hao

AU - George, Jonathan K.

AU - Yang, Joel K.W.

AU - Wall, Simon

AU - Miscuglio, Mario

AU - Simpson, Robert E.

AU - Sorger, Volker J.

N1 - Funding Information: Research funding: The NLAF design, SbS MZI weights, material growth, and thermal modeling were supported by the Agency for Science, Technology and Research (A*STAR) under the Advanced Manufacturing and Engineering (AME) grant #A18A7b0058. The network training was supported from the Presidential Early Career Award for Scientist and Engineers (PECASE) nominated by the Department of Defense through the Air Force Office of Scientific Research under award number FA9550-20-1-0193. The pump–probe measurements were funded by Fundació Cellex, Fundació Mir-Puig, and Generalitat de Catalunya through CERCA. E.P acknowledges the support from IJC2018-037384-I funded by MCIN/AEI/10.13039/501100011033. 2 3 Publisher Copyright: © 2022 Ting Yu Teo et al., published by De Gruyter, Berlin/Boston.

PY - 2022/5/25

Y1 - 2022/5/25

N2 - We demonstrate a passive all-chalcogenide all-optical perceptron scheme. The network’s nonlinear activation function (NLAF) relies on the nonlinear response of Ge2Sb2Te5 to femtosecond laser pulses. We measured the sub-picosecond time-resolved optical constants of Ge2Sb2Te5 at a wavelength of 1500 nm and used them to design a high-speed Ge2Sb2Te5-tuned microring resonator all-optical NLAF. The NLAF had a sigmoidal response when subjected to different laser fluence excitation and had a dynamic range of −9.7 dB. The perceptron’s waveguide material was AlN because it allowed efficient heat dissipation during laser switching. A two-temperature analysis revealed that the operating speed of the NLAF is ≤ 1 ns. The percepton’s nonvolatile weights were set using low-loss Sb2S3-tuned Mach Zehnder interferometers (MZIs). A three-layer deep neural network model was used to test the feasibility of the network scheme and a maximum training accuracy of 94.5% was obtained. We conclude that combining Sb2S3-programmed MZI weights with the nonlinear response of Ge2Sb2Te5 to femtosecond pulses is sufficient to perform energy-efficient all-optical neural classifications at rates greater than 1 GHz.

AB - We demonstrate a passive all-chalcogenide all-optical perceptron scheme. The network’s nonlinear activation function (NLAF) relies on the nonlinear response of Ge2Sb2Te5 to femtosecond laser pulses. We measured the sub-picosecond time-resolved optical constants of Ge2Sb2Te5 at a wavelength of 1500 nm and used them to design a high-speed Ge2Sb2Te5-tuned microring resonator all-optical NLAF. The NLAF had a sigmoidal response when subjected to different laser fluence excitation and had a dynamic range of −9.7 dB. The perceptron’s waveguide material was AlN because it allowed efficient heat dissipation during laser switching. A two-temperature analysis revealed that the operating speed of the NLAF is ≤ 1 ns. The percepton’s nonvolatile weights were set using low-loss Sb2S3-tuned Mach Zehnder interferometers (MZIs). A three-layer deep neural network model was used to test the feasibility of the network scheme and a maximum training accuracy of 94.5% was obtained. We conclude that combining Sb2S3-programmed MZI weights with the nonlinear response of Ge2Sb2Te5 to femtosecond pulses is sufficient to perform energy-efficient all-optical neural classifications at rates greater than 1 GHz.

KW - all-optical deep neural network

KW - chalcogenide reconfigurable photonics

KW - ultra-fast dynamic response of phase change material

UR - http://www.scopus.com/inward/record.url?scp=85131125436&partnerID=8YFLogxK

U2 - 10.1515/nanoph-2022-0099

DO - 10.1515/nanoph-2022-0099

M3 - Article

AN - SCOPUS:85131125436

SN - 2192-8606

VL - 11

SP - 4073

EP - 4088

JO - Nanophotonics

JF - Nanophotonics

IS - 17

ER -

Programmable chalcogenide-based all-optical deep neural networks

Abstract

Bibliographical note

Keywords

ASJC Scopus subject areas

Access to Document

Fingerprint

Cite this