Refresh Power Reduction of DRAMs in DNN Systems Using Hybrid Voting and ECC Method

Tsung Fu Hsieh; Jin Fu Li; Jenn Shiang Lai; Chih Yen Lo; Ding Ming Kwai; Yung Fa Chou

doi:10.1109/ITC-Asia51099.2020.00019

Refresh Power Reduction of DRAMs in DNN Systems Using Hybrid Voting and ECC Method

Tsung Fu Hsieh, Jin Fu Li, Jenn Shiang Lai, Chih Yen Lo, Ding Ming Kwai, Yung Fa Chou

Department of Electrical Engineering

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

Abstract

A deep neural network (DNN) system typically needs dynamic random access memories (DRAMs) for the data buffering. In this paper, an error-correction-code (ECC)-based technique is proposed to reduce the refresh power of DRAMs in the DNN system by extending the refresh period. By taking advantage of the characteristics of weight data of DNNs, a hybrid voting and ECC (VECC) method is used to protect the weight data from data retention fault caused by the refresh period extension. Analysis results show that the VECC method can achieve about 93% refresh power saving with about 0.5% accuracy loss and smaller than 0.5% check bit overhead on AlexNet, ResNet, and VGG19 convolutional neural network CNN models trained by ImageNet data set.

Original language	English
Title of host publication	Proceedings - 2020 IEEE International Test Conference in Asia, ITC-Asia 2020
Publisher	Institute of Electrical and Electronics Engineers Inc.
Pages	41-46
Number of pages	6
ISBN (Electronic)	9781728189444
DOIs	https://doi.org/10.1109/ITC-Asia51099.2020.00019
State	Published - Sep 2020
Event	4th IEEE International Test Conference in Asia, ITC-Asia 2020 - Taipei, Taiwan Duration: 23 Sep 2020 → 25 Sep 2020

Publication series

Name	Proceedings - 2020 IEEE International Test Conference in Asia, ITC-Asia 2020

Conference

Conference	4th IEEE International Test Conference in Asia, ITC-Asia 2020
Country/Territory	Taiwan
City	Taipei
Period	23/09/20 → 25/09/20

Keywords

Deep neural network
dynamic random access memory
error-correction code
low power

Access to Document

10.1109/ITC-Asia51099.2020.00019

Robustness and Reliability Enhancement Techniques for Deep Neural Network Systems(2/3)
Li, J.-F. (PI)
1/08/20 → 31/07/21
Project: Research
Reconfigurable Deep Neural Network Techniques for Supervised Learning(2/3)
Li, J.-F. (PI)
1/08/20 → 31/07/21
Project: Research

Cite this

Hsieh, T. F., Li, J. F., Lai, J. S., Lo, C. Y., Kwai, D. M., & Chou, Y. F. (2020). Refresh Power Reduction of DRAMs in DNN Systems Using Hybrid Voting and ECC Method. In Proceedings - 2020 IEEE International Test Conference in Asia, ITC-Asia 2020 (pp. 41-46). Article 9226545 (Proceedings - 2020 IEEE International Test Conference in Asia, ITC-Asia 2020). Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/ITC-Asia51099.2020.00019

Hsieh, Tsung Fu ; Li, Jin Fu ; Lai, Jenn Shiang et al. / Refresh Power Reduction of DRAMs in DNN Systems Using Hybrid Voting and ECC Method. Proceedings - 2020 IEEE International Test Conference in Asia, ITC-Asia 2020. Institute of Electrical and Electronics Engineers Inc., 2020. pp. 41-46 (Proceedings - 2020 IEEE International Test Conference in Asia, ITC-Asia 2020).

@inproceedings{aa03aec8f132498c93a948e54ad104d2,

title = "Refresh Power Reduction of DRAMs in DNN Systems Using Hybrid Voting and ECC Method",

abstract = "A deep neural network (DNN) system typically needs dynamic random access memories (DRAMs) for the data buffering. In this paper, an error-correction-code (ECC)-based technique is proposed to reduce the refresh power of DRAMs in the DNN system by extending the refresh period. By taking advantage of the characteristics of weight data of DNNs, a hybrid voting and ECC (VECC) method is used to protect the weight data from data retention fault caused by the refresh period extension. Analysis results show that the VECC method can achieve about 93% refresh power saving with about 0.5% accuracy loss and smaller than 0.5% check bit overhead on AlexNet, ResNet, and VGG19 convolutional neural network CNN models trained by ImageNet data set.",

keywords = "Deep neural network, dynamic random access memory, error-correction code, low power",

author = "Hsieh, {Tsung Fu} and Li, {Jin Fu} and Lai, {Jenn Shiang} and Lo, {Chih Yen} and Kwai, {Ding Ming} and Chou, {Yung Fa}",

note = "Publisher Copyright: {\textcopyright} 2020 IEEE.; 4th IEEE International Test Conference in Asia, ITC-Asia 2020 ; Conference date: 23-09-2020 Through 25-09-2020",

year = "2020",

month = sep,

doi = "10.1109/ITC-Asia51099.2020.00019",

language = "???core.languages.en_GB???",

series = "Proceedings - 2020 IEEE International Test Conference in Asia, ITC-Asia 2020",

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

pages = "41--46",

booktitle = "Proceedings - 2020 IEEE International Test Conference in Asia, ITC-Asia 2020",

}

Hsieh, TF, Li, JF, Lai, JS, Lo, CY, Kwai, DM & Chou, YF 2020, Refresh Power Reduction of DRAMs in DNN Systems Using Hybrid Voting and ECC Method. in Proceedings - 2020 IEEE International Test Conference in Asia, ITC-Asia 2020., 9226545, Proceedings - 2020 IEEE International Test Conference in Asia, ITC-Asia 2020, Institute of Electrical and Electronics Engineers Inc., pp. 41-46, 4th IEEE International Test Conference in Asia, ITC-Asia 2020, Taipei, Taiwan, 23/09/20. https://doi.org/10.1109/ITC-Asia51099.2020.00019

Refresh Power Reduction of DRAMs in DNN Systems Using Hybrid Voting and ECC Method. / Hsieh, Tsung Fu; Li, Jin Fu; Lai, Jenn Shiang et al.
Proceedings - 2020 IEEE International Test Conference in Asia, ITC-Asia 2020. Institute of Electrical and Electronics Engineers Inc., 2020. p. 41-46 9226545 (Proceedings - 2020 IEEE International Test Conference in Asia, ITC-Asia 2020).

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

TY - GEN

T1 - Refresh Power Reduction of DRAMs in DNN Systems Using Hybrid Voting and ECC Method

AU - Hsieh, Tsung Fu

AU - Li, Jin Fu

AU - Lai, Jenn Shiang

AU - Lo, Chih Yen

AU - Kwai, Ding Ming

AU - Chou, Yung Fa

PY - 2020/9

Y1 - 2020/9

N2 - A deep neural network (DNN) system typically needs dynamic random access memories (DRAMs) for the data buffering. In this paper, an error-correction-code (ECC)-based technique is proposed to reduce the refresh power of DRAMs in the DNN system by extending the refresh period. By taking advantage of the characteristics of weight data of DNNs, a hybrid voting and ECC (VECC) method is used to protect the weight data from data retention fault caused by the refresh period extension. Analysis results show that the VECC method can achieve about 93% refresh power saving with about 0.5% accuracy loss and smaller than 0.5% check bit overhead on AlexNet, ResNet, and VGG19 convolutional neural network CNN models trained by ImageNet data set.

AB - A deep neural network (DNN) system typically needs dynamic random access memories (DRAMs) for the data buffering. In this paper, an error-correction-code (ECC)-based technique is proposed to reduce the refresh power of DRAMs in the DNN system by extending the refresh period. By taking advantage of the characteristics of weight data of DNNs, a hybrid voting and ECC (VECC) method is used to protect the weight data from data retention fault caused by the refresh period extension. Analysis results show that the VECC method can achieve about 93% refresh power saving with about 0.5% accuracy loss and smaller than 0.5% check bit overhead on AlexNet, ResNet, and VGG19 convolutional neural network CNN models trained by ImageNet data set.

KW - Deep neural network

KW - dynamic random access memory

KW - error-correction code

KW - low power

UR - http://www.scopus.com/inward/record.url?scp=85096351040&partnerID=8YFLogxK

U2 - 10.1109/ITC-Asia51099.2020.00019

DO - 10.1109/ITC-Asia51099.2020.00019

M3 - 會議論文篇章

AN - SCOPUS:85096351040

T3 - Proceedings - 2020 IEEE International Test Conference in Asia, ITC-Asia 2020

SP - 41

EP - 46

BT - Proceedings - 2020 IEEE International Test Conference in Asia, ITC-Asia 2020

PB - Institute of Electrical and Electronics Engineers Inc.

T2 - 4th IEEE International Test Conference in Asia, ITC-Asia 2020

Y2 - 23 September 2020 through 25 September 2020

ER -

Hsieh TF, Li JF, Lai JS, Lo CY, Kwai DM, Chou YF. Refresh Power Reduction of DRAMs in DNN Systems Using Hybrid Voting and ECC Method. In Proceedings - 2020 IEEE International Test Conference in Asia, ITC-Asia 2020. Institute of Electrical and Electronics Engineers Inc. 2020. p. 41-46. 9226545. (Proceedings - 2020 IEEE International Test Conference in Asia, ITC-Asia 2020). doi: 10.1109/ITC-Asia51099.2020.00019

Refresh Power Reduction of DRAMs in DNN Systems Using Hybrid Voting and ECC Method

Abstract

Publication series

Conference

Keywords

Access to Document

Fingerprint

Projects

Robustness and Reliability Enhancement Techniques for Deep Neural Network Systems(2/3)

Reconfigurable Deep Neural Network Techniques for Supervised Learning(2/3)

Cite this