Energy efficiency optimization of FPGA-based CNN accelerators with full data reuse and VFS

Abstract—While FPGA has been recognized as a promising platform to accelerate Convolutional Neural Networks (CNNs) in embedded computing given its high flexibility and power efficiency, two challenges still have to be addressed to enhance its applicability on the edgecomputing paradigm. First, the p...

Full description

Bibliographic Details
Main Authors: Jiang, Weixiong, Yu, Heng, Liu, Xinzhe, Ha, Yajun
Format: Conference or Workshop Item
Language:English
Published: 2019
Online Access:https://eprints.nottingham.ac.uk/60004/
_version_ 1848799709693476864
author Jiang, Weixiong
Yu, Heng
Liu, Xinzhe
Ha, Yajun
author_facet Jiang, Weixiong
Yu, Heng
Liu, Xinzhe
Ha, Yajun
author_sort Jiang, Weixiong
building Nottingham Research Data Repository
collection Online Access
description Abstract—While FPGA has been recognized as a promising platform to accelerate Convolutional Neural Networks (CNNs) in embedded computing given its high flexibility and power efficiency, two challenges still have to be addressed to enhance its applicability on the edgecomputing paradigm. First, the power and performance of the CNN accelerator are still bounded by memory throughput, and a CNNcustomized architecture is desirable to fully utilize the on-chip storage. Second, power optimization algorithms are insufficiently explored on CNN-targeted platforms. In this paper, we design a novel FPGA-based CNN accelerator architecture that makes full use of the on-chip storage resources leveraging data reuse and loop unrolling strategies. We also present an efficient FPGA-based voltage and frequency scaling (VFS) system that enables VFS of the CNN accelerator for power optimization. We devise a VFS policy that fully exploits the power efficiency potential of the FPGA. Experiment results show up to 40% energy can be saved with our VFS platform and policy.
first_indexed 2025-11-14T20:39:59Z
format Conference or Workshop Item
id nottingham-60004
institution University of Nottingham Malaysia Campus
institution_category Local University
language English
last_indexed 2025-11-14T20:39:59Z
publishDate 2019
recordtype eprints
repository_type Digital Repository
spelling nottingham-600042020-02-28T03:34:20Z https://eprints.nottingham.ac.uk/60004/ Energy efficiency optimization of FPGA-based CNN accelerators with full data reuse and VFS Jiang, Weixiong Yu, Heng Liu, Xinzhe Ha, Yajun Abstract—While FPGA has been recognized as a promising platform to accelerate Convolutional Neural Networks (CNNs) in embedded computing given its high flexibility and power efficiency, two challenges still have to be addressed to enhance its applicability on the edgecomputing paradigm. First, the power and performance of the CNN accelerator are still bounded by memory throughput, and a CNNcustomized architecture is desirable to fully utilize the on-chip storage. Second, power optimization algorithms are insufficiently explored on CNN-targeted platforms. In this paper, we design a novel FPGA-based CNN accelerator architecture that makes full use of the on-chip storage resources leveraging data reuse and loop unrolling strategies. We also present an efficient FPGA-based voltage and frequency scaling (VFS) system that enables VFS of the CNN accelerator for power optimization. We devise a VFS policy that fully exploits the power efficiency potential of the FPGA. Experiment results show up to 40% energy can be saved with our VFS platform and policy. 2019-11-09 Conference or Workshop Item PeerReviewed application/pdf en cc_by https://eprints.nottingham.ac.uk/60004/1/1-merged.pdf Jiang, Weixiong, Yu, Heng, Liu, Xinzhe and Ha, Yajun (2019) Energy efficiency optimization of FPGA-based CNN accelerators with full data reuse and VFS. In: 26th IEEE International Conference on Electronics, Circuits and Systems, ICECS 2019, 27 November 2019 through 29 November 2019, Genoa; Italy. http://dx.doi.org/10.1109/ICECS46596.2019.8964717 10.1109/ICECS46596.2019.8964717 10.1109/ICECS46596.2019.8964717 10.1109/ICECS46596.2019.8964717
spellingShingle Jiang, Weixiong
Yu, Heng
Liu, Xinzhe
Ha, Yajun
Energy efficiency optimization of FPGA-based CNN accelerators with full data reuse and VFS
title Energy efficiency optimization of FPGA-based CNN accelerators with full data reuse and VFS
title_full Energy efficiency optimization of FPGA-based CNN accelerators with full data reuse and VFS
title_fullStr Energy efficiency optimization of FPGA-based CNN accelerators with full data reuse and VFS
title_full_unstemmed Energy efficiency optimization of FPGA-based CNN accelerators with full data reuse and VFS
title_short Energy efficiency optimization of FPGA-based CNN accelerators with full data reuse and VFS
title_sort energy efficiency optimization of fpga-based cnn accelerators with full data reuse and vfs
url https://eprints.nottingham.ac.uk/60004/
https://eprints.nottingham.ac.uk/60004/
https://eprints.nottingham.ac.uk/60004/