Energy-Efficient CNNs Accelerator Implementation on FPGA with Optimized Storage and Dataflow

Yonghua Zhang; Hongxu Jiang; Xiaobin Li; Rui Miao; Jinyan Nie; Yu Du

doi:10.1109/ISPA-BDCloud-SocialCom-SustainCom52081.2021.00166

Energy-Efficient CNNs Accelerator Implementation on FPGA with Optimized Storage and Dataflow

Yonghua Zhang
, Hongxu Jiang
, Xiaobin Li
, Rui Miao
, Jinyan Nie
, Yu Du

School of Computer Science and Engineering

Beihang University
Beijing Union University

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

1 Scopus citations

Abstract

Convolutional neural networks (CNNs) has been widely used in computer vision and speech processing, and has achieved great success. However, the deployment of large-scale CNN model is limited by computing and memory in the smart embedded system. Through the current high parallel computing paradigm of CNNs accelerator, the computing requirements can be effectively met to achieve high throughput. However, because the communication cost may be higher than the computing cost for small smart platform, the energy consumption is still very high. In order to solve this problem, a new CNNs accelerator based on storage and data flow is proposed, which realizes energy-saving CNNs inference acceleration by minimizing data access and maximizing data reuse. This paper implements the accelerator on the Zynq UltraScale+ MPSoC ZCU102 evaluation board, and evaluates the throughput and energy efficiency of the accelerator for typical vgg16 and tiny Yolo benchmark networks. Compared with other accelerators, the our accelerator improves the system energy efficiency by 6.3×, the system throughput by 41×, and the throughput of a single DSP by 7.63×.

Original language	English
Title of host publication	19th IEEE International Symposium on Parallel and Distributed Processing with Applications, 11th IEEE International Conference on Big Data and Cloud Computing, 14th IEEE International Conference on Social Computing and Networking and 11th IEEE International Conference on Sustainable Computing and Communications, ISPA/BDCloud/SocialCom/SustainCom 2021
Publisher	Institute of Electrical and Electronics Engineers Inc.
Pages	1209-1214
Number of pages	6
ISBN (Electronic)	9781665435741
DOIs	https://doi.org/10.1109/ISPA-BDCloud-SocialCom-SustainCom52081.2021.00166
State	Published - 2021
Event	19th IEEE International Symposium on Parallel and Distributed Processing with Applications, 11th IEEE International Conference on Big Data and Cloud Computing, 14th IEEE International Conference on Social Computing and Networking and 11th IEEE International Conference on Sustainable Computing and Communications, ISPA/BDCloud/SocialCom/SustainCom 2021 - New York, United States Duration: 30 Sep 2021 → 3 Oct 2021

Publication series

Name	19th IEEE International Symposium on Parallel and Distributed Processing with Applications, 11th IEEE International Conference on Big Data and Cloud Computing, 14th IEEE International Conference on Social Computing and Networking and 11th IEEE International Conference on Sustainable Computing and Communications, ISPA/BDCloud/SocialCom/SustainCom 2021

Conference

Conference	19th IEEE International Symposium on Parallel and Distributed Processing with Applications, 11th IEEE International Conference on Big Data and Cloud Computing, 14th IEEE International Conference on Social Computing and Networking and 11th IEEE International Conference on Sustainable Computing and Communications, ISPA/BDCloud/SocialCom/SustainCom 2021
Country/Territory	United States
City	New York
Period	30/09/21 → 3/10/21

UN SDGs

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

SDG 7 Affordable and Clean Energy

Keywords

CNNs Accelerator
Dataflow
Energy-Efficient
FPGA
Storage

Access to Document

10.1109/ISPA-BDCloud-SocialCom-SustainCom52081.2021.00166

Cite this

Zhang, Y., Jiang, H., Li, X., Miao, R., Nie, J., & Du, Y. (2021). Energy-Efficient CNNs Accelerator Implementation on FPGA with Optimized Storage and Dataflow. In 19th IEEE International Symposium on Parallel and Distributed Processing with Applications, 11th IEEE International Conference on Big Data and Cloud Computing, 14th IEEE International Conference on Social Computing and Networking and 11th IEEE International Conference on Sustainable Computing and Communications, ISPA/BDCloud/SocialCom/SustainCom 2021 (pp. 1209-1214). (19th IEEE International Symposium on Parallel and Distributed Processing with Applications, 11th IEEE International Conference on Big Data and Cloud Computing, 14th IEEE International Conference on Social Computing and Networking and 11th IEEE International Conference on Sustainable Computing and Communications, ISPA/BDCloud/SocialCom/SustainCom 2021). Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/ISPA-BDCloud-SocialCom-SustainCom52081.2021.00166

Zhang, Yonghua ; Jiang, Hongxu ; Li, Xiaobin et al. / Energy-Efficient CNNs Accelerator Implementation on FPGA with Optimized Storage and Dataflow. 19th IEEE International Symposium on Parallel and Distributed Processing with Applications, 11th IEEE International Conference on Big Data and Cloud Computing, 14th IEEE International Conference on Social Computing and Networking and 11th IEEE International Conference on Sustainable Computing and Communications, ISPA/BDCloud/SocialCom/SustainCom 2021. Institute of Electrical and Electronics Engineers Inc., 2021. pp. 1209-1214 (19th IEEE International Symposium on Parallel and Distributed Processing with Applications, 11th IEEE International Conference on Big Data and Cloud Computing, 14th IEEE International Conference on Social Computing and Networking and 11th IEEE International Conference on Sustainable Computing and Communications, ISPA/BDCloud/SocialCom/SustainCom 2021).

@inproceedings{67f0ed2eea2a4f85bb3b1e0a063ffb2d,

title = "Energy-Efficient CNNs Accelerator Implementation on FPGA with Optimized Storage and Dataflow",

abstract = "Convolutional neural networks (CNNs) has been widely used in computer vision and speech processing, and has achieved great success. However, the deployment of large-scale CNN model is limited by computing and memory in the smart embedded system. Through the current high parallel computing paradigm of CNNs accelerator, the computing requirements can be effectively met to achieve high throughput. However, because the communication cost may be higher than the computing cost for small smart platform, the energy consumption is still very high. In order to solve this problem, a new CNNs accelerator based on storage and data flow is proposed, which realizes energy-saving CNNs inference acceleration by minimizing data access and maximizing data reuse. This paper implements the accelerator on the Zynq UltraScale+ MPSoC ZCU102 evaluation board, and evaluates the throughput and energy efficiency of the accelerator for typical vgg16 and tiny Yolo benchmark networks. Compared with other accelerators, the our accelerator improves the system energy efficiency by 6.3×, the system throughput by 41×, and the throughput of a single DSP by 7.63×.",

keywords = "CNNs Accelerator, Dataflow, Energy-Efficient, FPGA, Storage",

author = "Yonghua Zhang and Hongxu Jiang and Xiaobin Li and Rui Miao and Jinyan Nie and Yu Du",

note = "Publisher Copyright: {\textcopyright} 2021 IEEE.; 19th IEEE International Symposium on Parallel and Distributed Processing with Applications, 11th IEEE International Conference on Big Data and Cloud Computing, 14th IEEE International Conference on Social Computing and Networking and 11th IEEE International Conference on Sustainable Computing and Communications, ISPA/BDCloud/SocialCom/SustainCom 2021 ; Conference date: 30-09-2021 Through 03-10-2021",

year = "2021",

doi = "10.1109/ISPA-BDCloud-SocialCom-SustainCom52081.2021.00166",

language = "英语",

series = "19th IEEE International Symposium on Parallel and Distributed Processing with Applications, 11th IEEE International Conference on Big Data and Cloud Computing, 14th IEEE International Conference on Social Computing and Networking and 11th IEEE International Conference on Sustainable Computing and Communications, ISPA/BDCloud/SocialCom/SustainCom 2021",

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

pages = "1209--1214",

booktitle = "19th IEEE International Symposium on Parallel and Distributed Processing with Applications, 11th IEEE International Conference on Big Data and Cloud Computing, 14th IEEE International Conference on Social Computing and Networking and 11th IEEE International Conference on Sustainable Computing and Communications, ISPA/BDCloud/SocialCom/SustainCom 2021",

address = "美国",

}

Zhang, Y, Jiang, H, Li, X, Miao, R, Nie, J & Du, Y 2021, Energy-Efficient CNNs Accelerator Implementation on FPGA with Optimized Storage and Dataflow. in 19th IEEE International Symposium on Parallel and Distributed Processing with Applications, 11th IEEE International Conference on Big Data and Cloud Computing, 14th IEEE International Conference on Social Computing and Networking and 11th IEEE International Conference on Sustainable Computing and Communications, ISPA/BDCloud/SocialCom/SustainCom 2021. 19th IEEE International Symposium on Parallel and Distributed Processing with Applications, 11th IEEE International Conference on Big Data and Cloud Computing, 14th IEEE International Conference on Social Computing and Networking and 11th IEEE International Conference on Sustainable Computing and Communications, ISPA/BDCloud/SocialCom/SustainCom 2021, Institute of Electrical and Electronics Engineers Inc., pp. 1209-1214, 19th IEEE International Symposium on Parallel and Distributed Processing with Applications, 11th IEEE International Conference on Big Data and Cloud Computing, 14th IEEE International Conference on Social Computing and Networking and 11th IEEE International Conference on Sustainable Computing and Communications, ISPA/BDCloud/SocialCom/SustainCom 2021, New York, United States, 30/09/21. https://doi.org/10.1109/ISPA-BDCloud-SocialCom-SustainCom52081.2021.00166

Energy-Efficient CNNs Accelerator Implementation on FPGA with Optimized Storage and Dataflow. / Zhang, Yonghua; Jiang, Hongxu; Li, Xiaobin et al.
19th IEEE International Symposium on Parallel and Distributed Processing with Applications, 11th IEEE International Conference on Big Data and Cloud Computing, 14th IEEE International Conference on Social Computing and Networking and 11th IEEE International Conference on Sustainable Computing and Communications, ISPA/BDCloud/SocialCom/SustainCom 2021. Institute of Electrical and Electronics Engineers Inc., 2021. p. 1209-1214 (19th IEEE International Symposium on Parallel and Distributed Processing with Applications, 11th IEEE International Conference on Big Data and Cloud Computing, 14th IEEE International Conference on Social Computing and Networking and 11th IEEE International Conference on Sustainable Computing and Communications, ISPA/BDCloud/SocialCom/SustainCom 2021).

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

TY - GEN

T1 - Energy-Efficient CNNs Accelerator Implementation on FPGA with Optimized Storage and Dataflow

AU - Zhang, Yonghua

AU - Jiang, Hongxu

AU - Li, Xiaobin

AU - Miao, Rui

AU - Nie, Jinyan

AU - Du, Yu

PY - 2021

Y1 - 2021

N2 - Convolutional neural networks (CNNs) has been widely used in computer vision and speech processing, and has achieved great success. However, the deployment of large-scale CNN model is limited by computing and memory in the smart embedded system. Through the current high parallel computing paradigm of CNNs accelerator, the computing requirements can be effectively met to achieve high throughput. However, because the communication cost may be higher than the computing cost for small smart platform, the energy consumption is still very high. In order to solve this problem, a new CNNs accelerator based on storage and data flow is proposed, which realizes energy-saving CNNs inference acceleration by minimizing data access and maximizing data reuse. This paper implements the accelerator on the Zynq UltraScale+ MPSoC ZCU102 evaluation board, and evaluates the throughput and energy efficiency of the accelerator for typical vgg16 and tiny Yolo benchmark networks. Compared with other accelerators, the our accelerator improves the system energy efficiency by 6.3×, the system throughput by 41×, and the throughput of a single DSP by 7.63×.

AB - Convolutional neural networks (CNNs) has been widely used in computer vision and speech processing, and has achieved great success. However, the deployment of large-scale CNN model is limited by computing and memory in the smart embedded system. Through the current high parallel computing paradigm of CNNs accelerator, the computing requirements can be effectively met to achieve high throughput. However, because the communication cost may be higher than the computing cost for small smart platform, the energy consumption is still very high. In order to solve this problem, a new CNNs accelerator based on storage and data flow is proposed, which realizes energy-saving CNNs inference acceleration by minimizing data access and maximizing data reuse. This paper implements the accelerator on the Zynq UltraScale+ MPSoC ZCU102 evaluation board, and evaluates the throughput and energy efficiency of the accelerator for typical vgg16 and tiny Yolo benchmark networks. Compared with other accelerators, the our accelerator improves the system energy efficiency by 6.3×, the system throughput by 41×, and the throughput of a single DSP by 7.63×.

KW - CNNs Accelerator

KW - Dataflow

KW - Energy-Efficient

KW - FPGA

KW - Storage

UR - https://www.scopus.com/pages/publications/85124144442

U2 - 10.1109/ISPA-BDCloud-SocialCom-SustainCom52081.2021.00166

DO - 10.1109/ISPA-BDCloud-SocialCom-SustainCom52081.2021.00166

M3 - 会议稿件

AN - SCOPUS:85124144442

T3 - 19th IEEE International Symposium on Parallel and Distributed Processing with Applications, 11th IEEE International Conference on Big Data and Cloud Computing, 14th IEEE International Conference on Social Computing and Networking and 11th IEEE International Conference on Sustainable Computing and Communications, ISPA/BDCloud/SocialCom/SustainCom 2021

SP - 1209

EP - 1214

BT - 19th IEEE International Symposium on Parallel and Distributed Processing with Applications, 11th IEEE International Conference on Big Data and Cloud Computing, 14th IEEE International Conference on Social Computing and Networking and 11th IEEE International Conference on Sustainable Computing and Communications, ISPA/BDCloud/SocialCom/SustainCom 2021

PB - Institute of Electrical and Electronics Engineers Inc.

T2 - 19th IEEE International Symposium on Parallel and Distributed Processing with Applications, 11th IEEE International Conference on Big Data and Cloud Computing, 14th IEEE International Conference on Social Computing and Networking and 11th IEEE International Conference on Sustainable Computing and Communications, ISPA/BDCloud/SocialCom/SustainCom 2021

Y2 - 30 September 2021 through 3 October 2021

ER -

Zhang Y, Jiang H, Li X, Miao R, Nie J, Du Y. Energy-Efficient CNNs Accelerator Implementation on FPGA with Optimized Storage and Dataflow. In 19th IEEE International Symposium on Parallel and Distributed Processing with Applications, 11th IEEE International Conference on Big Data and Cloud Computing, 14th IEEE International Conference on Social Computing and Networking and 11th IEEE International Conference on Sustainable Computing and Communications, ISPA/BDCloud/SocialCom/SustainCom 2021. Institute of Electrical and Electronics Engineers Inc. 2021. p. 1209-1214. (19th IEEE International Symposium on Parallel and Distributed Processing with Applications, 11th IEEE International Conference on Big Data and Cloud Computing, 14th IEEE International Conference on Social Computing and Networking and 11th IEEE International Conference on Sustainable Computing and Communications, ISPA/BDCloud/SocialCom/SustainCom 2021). doi: 10.1109/ISPA-BDCloud-SocialCom-SustainCom52081.2021.00166

Energy-Efficient CNNs Accelerator Implementation on FPGA with Optimized Storage and Dataflow

Abstract

Publication series

Conference

UN SDGs

Keywords

Access to Document

Other files and links

Fingerprint

Cite this