XOR-CiM: An Efficient Computing-in-SOT-MRAM Design for Binary Neural Network Acceleration

Mehrdad Morsali; Ranyang Zhou; Sepehr Tabrizchi; Arman Roohi; Shaahin Angizi

doi:10.1109/ISQED57927.2023.10129322

XOR-CiM: An Efficient Computing-in-SOT-MRAM Design for Binary Neural Network Acceleration

Mehrdad Morsali, Ranyang Zhou, Sepehr Tabrizchi, Arman Roohi, Shaahin Angizi

Electrical and Computer Engineering

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

3 Scopus citations

Abstract

In this work, we leverage the uni-polar switching behavior of Spin-Orbit Torque Magnetic Random Access Memory (SOT-MRAM) to develop an efficient digital Computing-in-Memory (CiM) platform named XOR-CiM. XOR-CiM converts typical MRAM sub-arrays to massively parallel computational cores with ultra-high bandwidth, greatly reducing energy consumption dealing with convolutional layers and accelerating X(N)OR-intensive Binary Neural Networks (BNNs) inference. With a similar inference accuracy to digital CiMs, XOR-CiM achieves ∼4.5× and 1.8× higher energy-efficiency and speed-up compared to the recent MRAM-based CiM platforms.

Original language	English (US)
Title of host publication	Proceedings of the 24th International Symposium on Quality Electronic Design, ISQED 2023
Publisher	IEEE Computer Society
ISBN (Electronic)	9798350334753
DOIs	https://doi.org/10.1109/ISQED57927.2023.10129322
State	Published - 2023
Event	24th International Symposium on Quality Electronic Design, ISQED 2023 - San Francisco, United States Duration: Apr 5 2023 → Apr 7 2023

Publication series

Name	Proceedings - International Symposium on Quality Electronic Design, ISQED
Volume	2023-April
ISSN (Print)	1948-3287
ISSN (Electronic)	1948-3295

Conference

Conference	24th International Symposium on Quality Electronic Design, ISQED 2023
Country/Territory	United States
City	San Francisco
Period	4/5/23 → 4/7/23

All Science Journal Classification (ASJC) codes

Hardware and Architecture
Electrical and Electronic Engineering
Safety, Risk, Reliability and Quality

Keywords

SOT-MRAM
binary neural networks
computing-in-memory

Access to Document

10.1109/ISQED57927.2023.10129322

Cite this

Morsali, M., Zhou, R., Tabrizchi, S., Roohi, A., & Angizi, S. (2023). XOR-CiM: An Efficient Computing-in-SOT-MRAM Design for Binary Neural Network Acceleration. In Proceedings of the 24th International Symposium on Quality Electronic Design, ISQED 2023 (Proceedings - International Symposium on Quality Electronic Design, ISQED; Vol. 2023-April). IEEE Computer Society. https://doi.org/10.1109/ISQED57927.2023.10129322

@inproceedings{8f3371d52df14f6a8252e618238cb1b5,

title = "XOR-CiM: An Efficient Computing-in-SOT-MRAM Design for Binary Neural Network Acceleration",

abstract = "In this work, we leverage the uni-polar switching behavior of Spin-Orbit Torque Magnetic Random Access Memory (SOT-MRAM) to develop an efficient digital Computing-in-Memory (CiM) platform named XOR-CiM. XOR-CiM converts typical MRAM sub-arrays to massively parallel computational cores with ultra-high bandwidth, greatly reducing energy consumption dealing with convolutional layers and accelerating X(N)OR-intensive Binary Neural Networks (BNNs) inference. With a similar inference accuracy to digital CiMs, XOR-CiM achieves ∼4.5× and 1.8× higher energy-efficiency and speed-up compared to the recent MRAM-based CiM platforms.",

keywords = "SOT-MRAM, binary neural networks, computing-in-memory",

author = "Mehrdad Morsali and Ranyang Zhou and Sepehr Tabrizchi and Arman Roohi and Shaahin Angizi",

note = "Publisher Copyright: {\textcopyright} 2023 IEEE.; 24th International Symposium on Quality Electronic Design, ISQED 2023 ; Conference date: 05-04-2023 Through 07-04-2023",

year = "2023",

doi = "10.1109/ISQED57927.2023.10129322",

language = "English (US)",

series = "Proceedings - International Symposium on Quality Electronic Design, ISQED",

publisher = "IEEE Computer Society",

booktitle = "Proceedings of the 24th International Symposium on Quality Electronic Design, ISQED 2023",

address = "United States",

}

Morsali, M, Zhou, R, Tabrizchi, S, Roohi, A & Angizi, S 2023, XOR-CiM: An Efficient Computing-in-SOT-MRAM Design for Binary Neural Network Acceleration. in Proceedings of the 24th International Symposium on Quality Electronic Design, ISQED 2023. Proceedings - International Symposium on Quality Electronic Design, ISQED, vol. 2023-April, IEEE Computer Society, 24th International Symposium on Quality Electronic Design, ISQED 2023, San Francisco, United States, 4/5/23. https://doi.org/10.1109/ISQED57927.2023.10129322

XOR-CiM: An Efficient Computing-in-SOT-MRAM Design for Binary Neural Network Acceleration. / Morsali, Mehrdad; Zhou, Ranyang; Tabrizchi, Sepehr et al.
Proceedings of the 24th International Symposium on Quality Electronic Design, ISQED 2023. IEEE Computer Society, 2023. (Proceedings - International Symposium on Quality Electronic Design, ISQED; Vol. 2023-April).

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

TY - GEN

T1 - XOR-CiM

T2 - 24th International Symposium on Quality Electronic Design, ISQED 2023

AU - Morsali, Mehrdad

AU - Zhou, Ranyang

AU - Tabrizchi, Sepehr

AU - Roohi, Arman

AU - Angizi, Shaahin

PY - 2023

Y1 - 2023

N2 - In this work, we leverage the uni-polar switching behavior of Spin-Orbit Torque Magnetic Random Access Memory (SOT-MRAM) to develop an efficient digital Computing-in-Memory (CiM) platform named XOR-CiM. XOR-CiM converts typical MRAM sub-arrays to massively parallel computational cores with ultra-high bandwidth, greatly reducing energy consumption dealing with convolutional layers and accelerating X(N)OR-intensive Binary Neural Networks (BNNs) inference. With a similar inference accuracy to digital CiMs, XOR-CiM achieves ∼4.5× and 1.8× higher energy-efficiency and speed-up compared to the recent MRAM-based CiM platforms.

AB - In this work, we leverage the uni-polar switching behavior of Spin-Orbit Torque Magnetic Random Access Memory (SOT-MRAM) to develop an efficient digital Computing-in-Memory (CiM) platform named XOR-CiM. XOR-CiM converts typical MRAM sub-arrays to massively parallel computational cores with ultra-high bandwidth, greatly reducing energy consumption dealing with convolutional layers and accelerating X(N)OR-intensive Binary Neural Networks (BNNs) inference. With a similar inference accuracy to digital CiMs, XOR-CiM achieves ∼4.5× and 1.8× higher energy-efficiency and speed-up compared to the recent MRAM-based CiM platforms.

KW - SOT-MRAM

KW - binary neural networks

KW - computing-in-memory

UR - http://www.scopus.com/inward/record.url?scp=85161584056&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85161584056&partnerID=8YFLogxK

U2 - 10.1109/ISQED57927.2023.10129322

DO - 10.1109/ISQED57927.2023.10129322

M3 - Conference contribution

AN - SCOPUS:85161584056

T3 - Proceedings - International Symposium on Quality Electronic Design, ISQED

BT - Proceedings of the 24th International Symposium on Quality Electronic Design, ISQED 2023

PB - IEEE Computer Society

Y2 - 5 April 2023 through 7 April 2023

ER -

Morsali M, Zhou R, Tabrizchi S, Roohi A, Angizi S. XOR-CiM: An Efficient Computing-in-SOT-MRAM Design for Binary Neural Network Acceleration. In Proceedings of the 24th International Symposium on Quality Electronic Design, ISQED 2023. IEEE Computer Society. 2023. (Proceedings - International Symposium on Quality Electronic Design, ISQED). doi: 10.1109/ISQED57927.2023.10129322

XOR-CiM: An Efficient Computing-in-SOT-MRAM Design for Binary Neural Network Acceleration

Abstract

Publication series

Conference

All Science Journal Classification (ASJC) codes

Keywords

Access to Document

Other files and links

Fingerprint

Cite this