Hypermultiplexed integrated photonics–based optical tensor processor

Shaoyuan Ou; Kaiwen Xue; Lian Zhou; Chun-ho Lee; Alexander Sludds; Ryan Hamerly; Ke Zhang; Hanke Feng; Yue Yu; Reshma Kopparapu; Eric Zhong; Cheng Wang; Dirk Englund; Mengjie Yu; Zaijun Chen

doi:10.1126/sciadv.adu0228

Hypermultiplexed integrated photonics–based optical tensor processor

Shaoyuan Ou, Kaiwen Xue, Lian Zhou, Chun-ho Lee, Alexander Sludds, Ryan Hamerly, Ke Zhang, Hanke Feng, Yue Yu, Reshma Kopparapu, Eric Zhong, Cheng Wang, Dirk Englund, Mengjie Yu, Zaijun Chen^*

^*Corresponding author for this work

Research output: Journal Publications and Reviews › RGC 21 - Publication in refereed journal › peer-review

7 Citations (Scopus)

5 Downloads (CityUHK Scholars)

Abstract

The escalating data volume and complexity resulting from the rapid expansion of artificial intelligence (AI), Internet of Things (IoT), and 5G/6G mobile networks is creating an urgent need for energy-efficient, scalable computing hardware. Here, we demonstrate a hypermultiplexed tensor optical processor that can perform trillions of operations per second using space-time-wavelength three-dimensional optical parallelism, enabling O(N²) operations per clock cycle with O(N) modulator devices. The system is built with wafer-fabricated III/V micrometer-scale lasers and high-speed thin-film lithium niobate electro-optics for encoding at tens of femtojoules per symbol. Lasing threshold incorporates analog inline rectifier (ReLU) nonlinearity for low-latency activation. The system scalability is verified with machine learning models of 405,000 parameters. A combination of high clock rates, energy-efficient processing, and programmability unlocks the potential of light for low-energy AI accelerators for applications ranging from training of large AI models to real-time decision-making in edge deployment.

Copyright © 2025 The Authors, some rights reserved; exclusive licensee American Association for the Advancement of Science.

Original language	English
Article number	eadu0228
Number of pages	11
Journal	Science Advances
Volume	11
Issue number	23
Online published	4 Jun 2025
DOIs	https://doi.org/10.1126/sciadv.adu0228
Publication status	Published - 6 Jun 2025

Publisher's Copyright Statement

This full text is made available under CC-BY-NC 4.0. https://creativecommons.org/licenses/by-nc/4.0/

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

Access Output

10.1126/sciadv.adu0228

288303696.pdfFinal Published version, 1.4 MBLicence: CC BY-NC

Other Access Options

Check CityUHK Library

Permanent Link

Right click to copy link

Cite this

@article{e41e7301b21b482095c6b8cd175f722e,

title = "Hypermultiplexed integrated photonics–based optical tensor processor",

abstract = "The escalating data volume and complexity resulting from the rapid expansion of artificial intelligence (AI), Internet of Things (IoT), and 5G/6G mobile networks is creating an urgent need for energy-efficient, scalable computing hardware. Here, we demonstrate a hypermultiplexed tensor optical processor that can perform trillions of operations per second using space-time-wavelength three-dimensional optical parallelism, enabling O(N2) operations per clock cycle with O(N) modulator devices. The system is built with wafer-fabricated III/V micrometer-scale lasers and high-speed thin-film lithium niobate electro-optics for encoding at tens of femtojoules per symbol. Lasing threshold incorporates analog inline rectifier (ReLU) nonlinearity for low-latency activation. The system scalability is verified with machine learning models of 405,000 parameters. A combination of high clock rates, energy-efficient processing, and programmability unlocks the potential of light for low-energy AI accelerators for applications ranging from training of large AI models to real-time decision-making in edge deployment.Copyright {\textcopyright} 2025 The Authors, some rights reserved; exclusive licensee American Association for the Advancement of Science.",

author = "Shaoyuan Ou and Kaiwen Xue and Lian Zhou and Chun-ho Lee and Alexander Sludds and Ryan Hamerly and Ke Zhang and Hanke Feng and Yue Yu and Reshma Kopparapu and Eric Zhong and Cheng Wang and Dirk Englund and Mengjie Yu and Zaijun Chen",

year = "2025",

month = jun,

day = "6",

doi = "10.1126/sciadv.adu0228",

language = "English",

volume = "11",

journal = "Science Advances",

issn = "2375-2548",

publisher = "American Association for the Advancement of Science",

number = "23",

}

TY - JOUR

T1 - Hypermultiplexed integrated photonics–based optical tensor processor

AU - Ou, Shaoyuan

AU - Xue, Kaiwen

AU - Zhou, Lian

AU - Lee, Chun-ho

AU - Sludds, Alexander

AU - Hamerly, Ryan

AU - Zhang, Ke

AU - Feng, Hanke

AU - Yu, Yue

AU - Kopparapu, Reshma

AU - Zhong, Eric

AU - Wang, Cheng

AU - Englund, Dirk

AU - Yu, Mengjie

AU - Chen, Zaijun

PY - 2025/6/6

Y1 - 2025/6/6

N2 - The escalating data volume and complexity resulting from the rapid expansion of artificial intelligence (AI), Internet of Things (IoT), and 5G/6G mobile networks is creating an urgent need for energy-efficient, scalable computing hardware. Here, we demonstrate a hypermultiplexed tensor optical processor that can perform trillions of operations per second using space-time-wavelength three-dimensional optical parallelism, enabling O(N2) operations per clock cycle with O(N) modulator devices. The system is built with wafer-fabricated III/V micrometer-scale lasers and high-speed thin-film lithium niobate electro-optics for encoding at tens of femtojoules per symbol. Lasing threshold incorporates analog inline rectifier (ReLU) nonlinearity for low-latency activation. The system scalability is verified with machine learning models of 405,000 parameters. A combination of high clock rates, energy-efficient processing, and programmability unlocks the potential of light for low-energy AI accelerators for applications ranging from training of large AI models to real-time decision-making in edge deployment.Copyright © 2025 The Authors, some rights reserved; exclusive licensee American Association for the Advancement of Science.

AB - The escalating data volume and complexity resulting from the rapid expansion of artificial intelligence (AI), Internet of Things (IoT), and 5G/6G mobile networks is creating an urgent need for energy-efficient, scalable computing hardware. Here, we demonstrate a hypermultiplexed tensor optical processor that can perform trillions of operations per second using space-time-wavelength three-dimensional optical parallelism, enabling O(N2) operations per clock cycle with O(N) modulator devices. The system is built with wafer-fabricated III/V micrometer-scale lasers and high-speed thin-film lithium niobate electro-optics for encoding at tens of femtojoules per symbol. Lasing threshold incorporates analog inline rectifier (ReLU) nonlinearity for low-latency activation. The system scalability is verified with machine learning models of 405,000 parameters. A combination of high clock rates, energy-efficient processing, and programmability unlocks the potential of light for low-energy AI accelerators for applications ranging from training of large AI models to real-time decision-making in edge deployment.Copyright © 2025 The Authors, some rights reserved; exclusive licensee American Association for the Advancement of Science.

U2 - 10.1126/sciadv.adu0228

DO - 10.1126/sciadv.adu0228

M3 - RGC 21 - Publication in refereed journal

C2 - 40465708

SN - 2375-2548

VL - 11

JO - Science Advances

JF - Science Advances

IS - 23

M1 - eadu0228

ER -

Hypermultiplexed integrated photonics–based optical tensor processor

Abstract

Publisher's Copyright Statement

UN SDGs

Access Output

Other Access Options

Permanent Link

Fingerprint

Cite this