Binary Weighted Memristive Analog Deep Neural Network for Near-Sensor Edge Processing

O. Krestinskaya, Alex James Pappachen

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

The memristive crossbar aims to implement analog weighted neural network, however, the realistic implementation of such crossbar arrays is not possible due to limited switching states of memristive devices. In this work, we propose the design of an analog deep neural network with binary weight update through backpropagation algorithm using binary state memristive devices. We show that such networks can be successfully used for image processing task and has the advantage of lower power consumption and small on-chip area in comparison with digital counterparts. The proposed network was benchmarked for MNIST handwritten digits recognition achieving an accuracy of approximately 90%.

Original languageEnglish
Title of host publication18th International Conference on Nanotechnology, NANO 2018
PublisherIEEE Computer Society
ISBN (Electronic)9781538653364
DOIs
Publication statusPublished - Jan 24 2019
Event18th International Conference on Nanotechnology, NANO 2018 - Cork, Ireland
Duration: Jul 23 2018Jul 26 2018

Publication series

NameProceedings of the IEEE Conference on Nanotechnology
Volume2018-July
ISSN (Print)1944-9399
ISSN (Electronic)1944-9380

Conference

Conference18th International Conference on Nanotechnology, NANO 2018
CountryIreland
CityCork
Period7/23/187/26/18

Fingerprint

Backpropagation algorithms
Image processing
Electric power utilization
analogs
Neural networks
sensors
Sensors
Processing
digits
image processing
chips
Deep neural networks

ASJC Scopus subject areas

  • Bioengineering
  • Electrical and Electronic Engineering
  • Materials Chemistry
  • Condensed Matter Physics

Cite this

Krestinskaya, O., & James Pappachen, A. (2019). Binary Weighted Memristive Analog Deep Neural Network for Near-Sensor Edge Processing. In 18th International Conference on Nanotechnology, NANO 2018 [8626224] (Proceedings of the IEEE Conference on Nanotechnology; Vol. 2018-July). IEEE Computer Society. https://doi.org/10.1109/NANO.2018.8626224

Binary Weighted Memristive Analog Deep Neural Network for Near-Sensor Edge Processing. / Krestinskaya, O.; James Pappachen, Alex.

18th International Conference on Nanotechnology, NANO 2018. IEEE Computer Society, 2019. 8626224 (Proceedings of the IEEE Conference on Nanotechnology; Vol. 2018-July).

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Krestinskaya, O & James Pappachen, A 2019, Binary Weighted Memristive Analog Deep Neural Network for Near-Sensor Edge Processing. in 18th International Conference on Nanotechnology, NANO 2018., 8626224, Proceedings of the IEEE Conference on Nanotechnology, vol. 2018-July, IEEE Computer Society, 18th International Conference on Nanotechnology, NANO 2018, Cork, Ireland, 7/23/18. https://doi.org/10.1109/NANO.2018.8626224
Krestinskaya O, James Pappachen A. Binary Weighted Memristive Analog Deep Neural Network for Near-Sensor Edge Processing. In 18th International Conference on Nanotechnology, NANO 2018. IEEE Computer Society. 2019. 8626224. (Proceedings of the IEEE Conference on Nanotechnology). https://doi.org/10.1109/NANO.2018.8626224
Krestinskaya, O. ; James Pappachen, Alex. / Binary Weighted Memristive Analog Deep Neural Network for Near-Sensor Edge Processing. 18th International Conference on Nanotechnology, NANO 2018. IEEE Computer Society, 2019. (Proceedings of the IEEE Conference on Nanotechnology).
@inproceedings{6cfcc41140b44a109e80c1cb3747a7c9,
title = "Binary Weighted Memristive Analog Deep Neural Network for Near-Sensor Edge Processing",
abstract = "The memristive crossbar aims to implement analog weighted neural network, however, the realistic implementation of such crossbar arrays is not possible due to limited switching states of memristive devices. In this work, we propose the design of an analog deep neural network with binary weight update through backpropagation algorithm using binary state memristive devices. We show that such networks can be successfully used for image processing task and has the advantage of lower power consumption and small on-chip area in comparison with digital counterparts. The proposed network was benchmarked for MNIST handwritten digits recognition achieving an accuracy of approximately 90{\%}.",
author = "O. Krestinskaya and {James Pappachen}, Alex",
year = "2019",
month = "1",
day = "24",
doi = "10.1109/NANO.2018.8626224",
language = "English",
series = "Proceedings of the IEEE Conference on Nanotechnology",
publisher = "IEEE Computer Society",
booktitle = "18th International Conference on Nanotechnology, NANO 2018",
address = "United States",

}

TY - GEN

T1 - Binary Weighted Memristive Analog Deep Neural Network for Near-Sensor Edge Processing

AU - Krestinskaya, O.

AU - James Pappachen, Alex

PY - 2019/1/24

Y1 - 2019/1/24

N2 - The memristive crossbar aims to implement analog weighted neural network, however, the realistic implementation of such crossbar arrays is not possible due to limited switching states of memristive devices. In this work, we propose the design of an analog deep neural network with binary weight update through backpropagation algorithm using binary state memristive devices. We show that such networks can be successfully used for image processing task and has the advantage of lower power consumption and small on-chip area in comparison with digital counterparts. The proposed network was benchmarked for MNIST handwritten digits recognition achieving an accuracy of approximately 90%.

AB - The memristive crossbar aims to implement analog weighted neural network, however, the realistic implementation of such crossbar arrays is not possible due to limited switching states of memristive devices. In this work, we propose the design of an analog deep neural network with binary weight update through backpropagation algorithm using binary state memristive devices. We show that such networks can be successfully used for image processing task and has the advantage of lower power consumption and small on-chip area in comparison with digital counterparts. The proposed network was benchmarked for MNIST handwritten digits recognition achieving an accuracy of approximately 90%.

UR - http://www.scopus.com/inward/record.url?scp=85062302345&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85062302345&partnerID=8YFLogxK

U2 - 10.1109/NANO.2018.8626224

DO - 10.1109/NANO.2018.8626224

M3 - Conference contribution

AN - SCOPUS:85062302345

T3 - Proceedings of the IEEE Conference on Nanotechnology

BT - 18th International Conference on Nanotechnology, NANO 2018

PB - IEEE Computer Society

ER -