Energy-efficient neural network inference with microcavity exciton polaritons
We propose all-optical neural networks characterized by very high energy efficiency and performance density of inference. We argue that the use of microcavity exciton-polaritons allows to take advantage of the properties of both photons and electrons in a seamless manner. This results in strong o...
Main Authors: | , , , , , , , , , , |
---|---|
Other Authors: | |
Format: | Journal Article |
Language: | English |
Published: |
2021
|
Subjects: | |
Online Access: | https://hdl.handle.net/10356/154197 |
_version_ | 1811676633953730560 |
---|---|
author | Matuszewski. M. Opala, A. Mirek, R. Furman, M. Król, M. Tyszka, K. Liew, Timothy Chi Hin Ballarini, D. Sanvitto, D. Szczytko, J. Piętka, B. |
author2 | School of Physical and Mathematical Sciences |
author_facet | School of Physical and Mathematical Sciences Matuszewski. M. Opala, A. Mirek, R. Furman, M. Król, M. Tyszka, K. Liew, Timothy Chi Hin Ballarini, D. Sanvitto, D. Szczytko, J. Piętka, B. |
author_sort | Matuszewski. M. |
collection | NTU |
description | We propose all-optical neural networks characterized by very high energy
efficiency and performance density of inference. We argue that the use of
microcavity exciton-polaritons allows to take advantage of the properties of
both photons and electrons in a seamless manner. This results in strong optical
nonlinearity without the use of optoelectronic conversion. We propose a design
of a realistic neural network and estimate energy cost to be at the level of
attojoules per bit, also when including the optoelectronic conversion at the
input and output of the network, several orders of magnitude below
state-of-the-art hardware implementations. We propose two kinds of nonlinear
binarized nodes based either on optical phase shifts and interferometry or on
polariton spin rotations. |
first_indexed | 2024-10-01T02:24:35Z |
format | Journal Article |
id | ntu-10356/154197 |
institution | Nanyang Technological University |
language | English |
last_indexed | 2024-10-01T02:24:35Z |
publishDate | 2021 |
record_format | dspace |
spelling | ntu-10356/1541972023-02-28T20:00:41Z Energy-efficient neural network inference with microcavity exciton polaritons Matuszewski. M. Opala, A. Mirek, R. Furman, M. Król, M. Tyszka, K. Liew, Timothy Chi Hin Ballarini, D. Sanvitto, D. Szczytko, J. Piętka, B. School of Physical and Mathematical Sciences Science::Physics Artificial-Intelligence Classification We propose all-optical neural networks characterized by very high energy efficiency and performance density of inference. We argue that the use of microcavity exciton-polaritons allows to take advantage of the properties of both photons and electrons in a seamless manner. This results in strong optical nonlinearity without the use of optoelectronic conversion. We propose a design of a realistic neural network and estimate energy cost to be at the level of attojoules per bit, also when including the optoelectronic conversion at the input and output of the network, several orders of magnitude below state-of-the-art hardware implementations. We propose two kinds of nonlinear binarized nodes based either on optical phase shifts and interferometry or on polariton spin rotations. Ministry of Education (MOE) Published version M.M. acknowledges support from National Science Center, Poland Grant No. 2017/25/Z/ST3/03032 under the QuantERA program. A.O. acknowledges support from National Science Center, Poland Grant No. 2016/22/E/ ST3/00045. R.M. acknowledges support from National Science Center, Poland Grant No. 2019/33/N/ST3/02019. B.P. acknowldges support from National Science Center, Poland Grant No. 2020/37/B/ST3/01657. K.T. acknowldges support from National Science Center, Poland Grant No. 2020/04/X/ST7/01379. T.L. acknowledges the support of the Singapore Ministry of Education, via the Academic research fund project MOE2019-T2-1-004. D.B. and D.S. acknowledge support from the project FISR2020- COVID, WaveSense (FISR2020IP_04324), and the PRIN 2017 InPhoPOL. 2021-12-19T08:14:27Z 2021-12-19T08:14:27Z 2021 Journal Article Matuszewski. M., Opala, A., Mirek, R., Furman, M., Król, M., Tyszka, K., Liew, T. C. H., Ballarini, D., Sanvitto, D., Szczytko, J. & Piętka, B. (2021). Energy-efficient neural network inference with microcavity exciton polaritons. Physical Review Applied, 16(2), 024045-. https://dx.doi.org/10.1103/PhysRevApplied.16.024045 2331-7019 https://hdl.handle.net/10356/154197 10.1103/PhysRevApplied.16.024045 2-s2.0-85114415173 2 16 024045 en MOE2019-T2-1-004 Physical Review Applied © 2021 American Physical Society. All rights reserved. This paper was published in Physical Review Applied and is made available with permission of American Physical Society. application/pdf |
spellingShingle | Science::Physics Artificial-Intelligence Classification Matuszewski. M. Opala, A. Mirek, R. Furman, M. Król, M. Tyszka, K. Liew, Timothy Chi Hin Ballarini, D. Sanvitto, D. Szczytko, J. Piętka, B. Energy-efficient neural network inference with microcavity exciton polaritons |
title | Energy-efficient neural network inference with microcavity exciton polaritons |
title_full | Energy-efficient neural network inference with microcavity exciton polaritons |
title_fullStr | Energy-efficient neural network inference with microcavity exciton polaritons |
title_full_unstemmed | Energy-efficient neural network inference with microcavity exciton polaritons |
title_short | Energy-efficient neural network inference with microcavity exciton polaritons |
title_sort | energy efficient neural network inference with microcavity exciton polaritons |
topic | Science::Physics Artificial-Intelligence Classification |
url | https://hdl.handle.net/10356/154197 |
work_keys_str_mv | AT matuszewskim energyefficientneuralnetworkinferencewithmicrocavityexcitonpolaritons AT opalaa energyefficientneuralnetworkinferencewithmicrocavityexcitonpolaritons AT mirekr energyefficientneuralnetworkinferencewithmicrocavityexcitonpolaritons AT furmanm energyefficientneuralnetworkinferencewithmicrocavityexcitonpolaritons AT krolm energyefficientneuralnetworkinferencewithmicrocavityexcitonpolaritons AT tyszkak energyefficientneuralnetworkinferencewithmicrocavityexcitonpolaritons AT liewtimothychihin energyefficientneuralnetworkinferencewithmicrocavityexcitonpolaritons AT ballarinid energyefficientneuralnetworkinferencewithmicrocavityexcitonpolaritons AT sanvittod energyefficientneuralnetworkinferencewithmicrocavityexcitonpolaritons AT szczytkoj energyefficientneuralnetworkinferencewithmicrocavityexcitonpolaritons AT pietkab energyefficientneuralnetworkinferencewithmicrocavityexcitonpolaritons |