Additional Parallelism for WDM-based Photonic Integrated Convolutional Neural Network

Bin Shi; Nicola Calabretta; Ripalta Stabile

Abstract

In this work, we introduce an additional parallelism with combination of space and wavelength domains, provided by cyclic array waveguide grating, to further accelerate the WDM-based convolutional operation on chip. The proposed parallelism improves CNN acceleration by 10s times with respecting to the state-of-the-art photonic integrated convolutional processors, suggesting up to 40 Tera Operations/s for a chip with 64 weighting elements. The operation of the additional parallelism is demonstrated for the handwritten digit classification problem in simulation, with a classification accuracy of 83%.

Original language	English
Number of pages	2
Publication status	Published - 25 Oct 2023
Event	Neuromorphic, Natural and Physical Computing: Interdisciplinary Foundations (NNPC 2023) - The Castle of Herrenhausen, Hannover, Germany Duration: 25 Oct 2023 → 27 Oct 2023 https://nnpc-conference.com/

Conference

Conference	Neuromorphic, Natural and Physical Computing: Interdisciplinary Foundations (NNPC 2023)
Country/Territory	Germany
City	Hannover
Period	25/10/23 → 27/10/23
Internet address	https://nnpc-conference.com/

Cite this

@conference{a9135692fc5f4ff6ba08a1e8b3697860,

title = "Additional Parallelism for WDM-based Photonic Integrated Convolutional Neural Network",

abstract = "In this work, we introduce an additional parallelism with combination of space and wavelength domains, provided by cyclic array waveguide grating, to further accelerate the WDM-based convolutional operation on chip. The proposed parallelism improves CNN acceleration by 10s times with respecting to the state-of-the-art photonic integrated convolutional processors, suggesting up to 40 Tera Operations/s for a chip with 64 weighting elements. The operation of the additional parallelism is demonstrated for the handwritten digit classification problem in simulation, with a classification accuracy of 83\%.",

author = "Bin Shi and Nicola Calabretta and Ripalta Stabile",

year = "2023",

month = oct,

day = "25",

language = "English",

note = "Neuromorphic, Natural and Physical Computing: Interdisciplinary Foundations (NNPC 2023) ; Conference date: 25-10-2023 Through 27-10-2023",

url = "https://nnpc-conference.com/",

}

TY - CONF

T1 - Additional Parallelism for WDM-based Photonic Integrated Convolutional Neural Network

AU - Shi, Bin

AU - Calabretta, Nicola

AU - Stabile, Ripalta

PY - 2023/10/25

Y1 - 2023/10/25

N2 - In this work, we introduce an additional parallelism with combination of space and wavelength domains, provided by cyclic array waveguide grating, to further accelerate the WDM-based convolutional operation on chip. The proposed parallelism improves CNN acceleration by 10s times with respecting to the state-of-the-art photonic integrated convolutional processors, suggesting up to 40 Tera Operations/s for a chip with 64 weighting elements. The operation of the additional parallelism is demonstrated for the handwritten digit classification problem in simulation, with a classification accuracy of 83%.

AB - In this work, we introduce an additional parallelism with combination of space and wavelength domains, provided by cyclic array waveguide grating, to further accelerate the WDM-based convolutional operation on chip. The proposed parallelism improves CNN acceleration by 10s times with respecting to the state-of-the-art photonic integrated convolutional processors, suggesting up to 40 Tera Operations/s for a chip with 64 weighting elements. The operation of the additional parallelism is demonstrated for the handwritten digit classification problem in simulation, with a classification accuracy of 83%.

M3 - Abstract

T2 - Neuromorphic, Natural and Physical Computing: Interdisciplinary Foundations (NNPC 2023)

Y2 - 25 October 2023 through 27 October 2023

ER -

Additional Parallelism for WDM-based Photonic Integrated Convolutional Neural Network

Abstract

Conference

Fingerprint

Cite this