Based on 1 or 0 BNN Accelerator with Array Architecture

Zhangkong Xian; Hongge Li; Yuliang Li

doi:10.1109/ICTA48799.2019.9012873

Based on 1 or 0 BNN Accelerator with Array Architecture

Zhangkong Xian
, Hongge Li
, Yuliang Li

School of Electronic and Information Engineering

Beihang University

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

Abstract

In this paper, we introduce a binary convolutional neural network accelerator which using a new binarization method. We propose the weights and the activations to either 1 or 0 instead of +1 or-1, which makes the convolution process simplified and more suitable for hardware implementation. To decrease the data access from off-chip memory, we propose a novel method of the data reuse, which can reduce 58.8% data access. Our accelerator is implemented on the VC709 Evaluation Kit. Experimental results show that the proposed accelerator achieves a throughput of 3378 GOPS and 1624 GOPS/W energy efficiency.

Original language	English
Title of host publication	2019 IEEE International Conference on Integrated Circuits, Technologies and Applications, ICTA 2019 - Proceedings
Publisher	Institute of Electrical and Electronics Engineers Inc.
Pages	79-80
Number of pages	2
ISBN (Electronic)	9781728151670
DOIs	https://doi.org/10.1109/ICTA48799.2019.9012873
State	Published - Nov 2019
Event	2nd IEEE International Conference on Integrated Circuits, Technologies and Applications, ICTA 2019 - Chengdu, China Duration: 13 Nov 2019 → 15 Nov 2019

Publication series

Name	2019 IEEE International Conference on Integrated Circuits, Technologies and Applications, ICTA 2019 - Proceedings

Conference

Conference	2nd IEEE International Conference on Integrated Circuits, Technologies and Applications, ICTA 2019
Country/Territory	China
City	Chengdu
Period	13/11/19 → 15/11/19

UN SDGs

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

SDG 7 Affordable and Clean Energy

Keywords

BNN
data reuse
energy efficiency
throughput

Access to Document

10.1109/ICTA48799.2019.9012873

Cite this

Xian, Z., Li, H., & Li, Y. (2019). Based on 1 or 0 BNN Accelerator with Array Architecture. In 2019 IEEE International Conference on Integrated Circuits, Technologies and Applications, ICTA 2019 - Proceedings (pp. 79-80). Article 9012873 (2019 IEEE International Conference on Integrated Circuits, Technologies and Applications, ICTA 2019 - Proceedings). Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/ICTA48799.2019.9012873

Xian, Zhangkong ; Li, Hongge ; Li, Yuliang. / Based on 1 or 0 BNN Accelerator with Array Architecture. 2019 IEEE International Conference on Integrated Circuits, Technologies and Applications, ICTA 2019 - Proceedings. Institute of Electrical and Electronics Engineers Inc., 2019. pp. 79-80 (2019 IEEE International Conference on Integrated Circuits, Technologies and Applications, ICTA 2019 - Proceedings).

@inproceedings{9c0a50767f0f400195b1f117f42fc5d6,

title = "Based on 1 or 0 BNN Accelerator with Array Architecture",

abstract = "In this paper, we introduce a binary convolutional neural network accelerator which using a new binarization method. We propose the weights and the activations to either 1 or 0 instead of +1 or-1, which makes the convolution process simplified and more suitable for hardware implementation. To decrease the data access from off-chip memory, we propose a novel method of the data reuse, which can reduce 58.8\% data access. Our accelerator is implemented on the VC709 Evaluation Kit. Experimental results show that the proposed accelerator achieves a throughput of 3378 GOPS and 1624 GOPS/W energy efficiency.",

keywords = "BNN, data reuse, energy efficiency, throughput",

author = "Zhangkong Xian and Hongge Li and Yuliang Li",

note = "Publisher Copyright: {\textcopyright} 2019 IEEE.; 2nd IEEE International Conference on Integrated Circuits, Technologies and Applications, ICTA 2019 ; Conference date: 13-11-2019 Through 15-11-2019",

year = "2019",

month = nov,

doi = "10.1109/ICTA48799.2019.9012873",

language = "英语",

series = "2019 IEEE International Conference on Integrated Circuits, Technologies and Applications, ICTA 2019 - Proceedings",

publisher = "Institute of Electrical and Electronics Engineers Inc.",

pages = "79--80",

booktitle = "2019 IEEE International Conference on Integrated Circuits, Technologies and Applications, ICTA 2019 - Proceedings",

address = "美国",

}

Xian, Z, Li, H & Li, Y 2019, Based on 1 or 0 BNN Accelerator with Array Architecture. in 2019 IEEE International Conference on Integrated Circuits, Technologies and Applications, ICTA 2019 - Proceedings., 9012873, 2019 IEEE International Conference on Integrated Circuits, Technologies and Applications, ICTA 2019 - Proceedings, Institute of Electrical and Electronics Engineers Inc., pp. 79-80, 2nd IEEE International Conference on Integrated Circuits, Technologies and Applications, ICTA 2019, Chengdu, China, 13/11/19. https://doi.org/10.1109/ICTA48799.2019.9012873

Based on 1 or 0 BNN Accelerator with Array Architecture. / Xian, Zhangkong; Li, Hongge; Li, Yuliang.
2019 IEEE International Conference on Integrated Circuits, Technologies and Applications, ICTA 2019 - Proceedings. Institute of Electrical and Electronics Engineers Inc., 2019. p. 79-80 9012873 (2019 IEEE International Conference on Integrated Circuits, Technologies and Applications, ICTA 2019 - Proceedings).

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

TY - GEN

T1 - Based on 1 or 0 BNN Accelerator with Array Architecture

AU - Xian, Zhangkong

AU - Li, Hongge

AU - Li, Yuliang

PY - 2019/11

Y1 - 2019/11

N2 - In this paper, we introduce a binary convolutional neural network accelerator which using a new binarization method. We propose the weights and the activations to either 1 or 0 instead of +1 or-1, which makes the convolution process simplified and more suitable for hardware implementation. To decrease the data access from off-chip memory, we propose a novel method of the data reuse, which can reduce 58.8% data access. Our accelerator is implemented on the VC709 Evaluation Kit. Experimental results show that the proposed accelerator achieves a throughput of 3378 GOPS and 1624 GOPS/W energy efficiency.

AB - In this paper, we introduce a binary convolutional neural network accelerator which using a new binarization method. We propose the weights and the activations to either 1 or 0 instead of +1 or-1, which makes the convolution process simplified and more suitable for hardware implementation. To decrease the data access from off-chip memory, we propose a novel method of the data reuse, which can reduce 58.8% data access. Our accelerator is implemented on the VC709 Evaluation Kit. Experimental results show that the proposed accelerator achieves a throughput of 3378 GOPS and 1624 GOPS/W energy efficiency.

KW - BNN

KW - data reuse

KW - energy efficiency

KW - throughput

UR - https://www.scopus.com/pages/publications/85081985774

U2 - 10.1109/ICTA48799.2019.9012873

DO - 10.1109/ICTA48799.2019.9012873

M3 - 会议稿件

AN - SCOPUS:85081985774

T3 - 2019 IEEE International Conference on Integrated Circuits, Technologies and Applications, ICTA 2019 - Proceedings

SP - 79

EP - 80

BT - 2019 IEEE International Conference on Integrated Circuits, Technologies and Applications, ICTA 2019 - Proceedings

PB - Institute of Electrical and Electronics Engineers Inc.

T2 - 2nd IEEE International Conference on Integrated Circuits, Technologies and Applications, ICTA 2019

Y2 - 13 November 2019 through 15 November 2019

ER -

Xian Z, Li H, Li Y. Based on 1 or 0 BNN Accelerator with Array Architecture. In 2019 IEEE International Conference on Integrated Circuits, Technologies and Applications, ICTA 2019 - Proceedings. Institute of Electrical and Electronics Engineers Inc. 2019. p. 79-80. 9012873. (2019 IEEE International Conference on Integrated Circuits, Technologies and Applications, ICTA 2019 - Proceedings). doi: 10.1109/ICTA48799.2019.9012873

Based on 1 or 0 BNN Accelerator with Array Architecture

Abstract

Publication series

Conference

UN SDGs

Keywords

Access to Document

Other files and links

Fingerprint

Cite this