Overcoming crossbar nonidealities in binary neural networks through learning

Mohammed E. Fouda; Jongeun Lee; Ahmed M. Eltawil; Fadi Kurdahi

doi:10.1145/3232195.3232226

Overcoming crossbar nonidealities in binary neural networks through learning

Mohammed E. Fouda, Jongeun Lee, Ahmed M. Eltawil, Fadi Kurdahi

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

11 Scopus citations

Abstract

The crossbar nonidealaties may considerably degrade the accuracy of matrix multiplication operation, which is the cornerstone of hardware accelerated neural networks. In this paper, we show that the crossbar nonidealities especially the wire resistance should be taken into consideration for accurate evaluation. We also present a simple yet highly effective way to capture the wire resistance effect for the inference and training of deep neural networks without extensive SPICE simulations. Different scenarios have been studied and used to show the efficacy of our proposed method.

Original language	English (US)
Title of host publication	Proceedings of the 14th IEEE/ACM International Symposium on Nanoscale Architectures, NANOARCH 2018
Publisher	Association for Computing Machinery, Incacmhelp@acm.org
ISBN (Print)	9781450358156
DOIs	https://doi.org/10.1145/3232195.3232226
State	Published - Jul 17 2018
Externally published	Yes

Access to Document

10.1145/3232195.3232226

Cite this

@inproceedings{f4b7cb24ab324c538b93a9dc382c2a69,

title = "Overcoming crossbar nonidealities in binary neural networks through learning",

abstract = "The crossbar nonidealaties may considerably degrade the accuracy of matrix multiplication operation, which is the cornerstone of hardware accelerated neural networks. In this paper, we show that the crossbar nonidealities especially the wire resistance should be taken into consideration for accurate evaluation. We also present a simple yet highly effective way to capture the wire resistance effect for the inference and training of deep neural networks without extensive SPICE simulations. Different scenarios have been studied and used to show the efficacy of our proposed method.",

author = "Fouda, {Mohammed E.} and Jongeun Lee and Eltawil, {Ahmed M.} and Fadi Kurdahi",

note = "Generated from Scopus record by KAUST IRTS on 2019-11-20",

year = "2018",

month = jul,

day = "17",

doi = "10.1145/3232195.3232226",

language = "English (US)",

isbn = "9781450358156",

booktitle = "Proceedings of the 14th IEEE/ACM International Symposium on Nanoscale Architectures, NANOARCH 2018",

publisher = "Association for Computing Machinery, Incacmhelp@acm.org",

}

Overcoming crossbar nonidealities in binary neural networks through learning. / Fouda, Mohammed E.; Lee, Jongeun; Eltawil, Ahmed M. et al.
Proceedings of the 14th IEEE/ACM International Symposium on Nanoscale Architectures, NANOARCH 2018. Association for Computing Machinery, Incacmhelp@acm.org, 2018.

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

TY - GEN

T1 - Overcoming crossbar nonidealities in binary neural networks through learning

AU - Fouda, Mohammed E.

AU - Lee, Jongeun

AU - Eltawil, Ahmed M.

AU - Kurdahi, Fadi

N1 - Generated from Scopus record by KAUST IRTS on 2019-11-20

PY - 2018/7/17

Y1 - 2018/7/17

N2 - The crossbar nonidealaties may considerably degrade the accuracy of matrix multiplication operation, which is the cornerstone of hardware accelerated neural networks. In this paper, we show that the crossbar nonidealities especially the wire resistance should be taken into consideration for accurate evaluation. We also present a simple yet highly effective way to capture the wire resistance effect for the inference and training of deep neural networks without extensive SPICE simulations. Different scenarios have been studied and used to show the efficacy of our proposed method.

AB - The crossbar nonidealaties may considerably degrade the accuracy of matrix multiplication operation, which is the cornerstone of hardware accelerated neural networks. In this paper, we show that the crossbar nonidealities especially the wire resistance should be taken into consideration for accurate evaluation. We also present a simple yet highly effective way to capture the wire resistance effect for the inference and training of deep neural networks without extensive SPICE simulations. Different scenarios have been studied and used to show the efficacy of our proposed method.

UR - http://dl.acm.org/citation.cfm?doid=3232195.3232226

UR - http://www.scopus.com/inward/record.url?scp=85060733229&partnerID=8YFLogxK

U2 - 10.1145/3232195.3232226

DO - 10.1145/3232195.3232226

M3 - Conference contribution

SN - 9781450358156

BT - Proceedings of the 14th IEEE/ACM International Symposium on Nanoscale Architectures, NANOARCH 2018

PB - Association for Computing Machinery, Incacmhelp@acm.org

ER -

Overcoming crossbar nonidealities in binary neural networks through learning

Abstract

Access to Document

Other files and links

Fingerprint

Cite this