Mixed Precision Quantization Scheme for Re-configurable ReRAM Crossbars Targeting Different Energy Harvesting Scenarios

Md Fahim Faysal Khan; Nicholas Anton Jao; Changchi Shuai; Keni Qiu; Mehrdad Mahdavi; Vijaykrishnan Narayanan

doi:10.1007/978-3-030-43605-6_12

Mixed Precision Quantization Scheme for Re-configurable ReRAM Crossbars Targeting Different Energy Harvesting Scenarios

Md Fahim Faysal Khan, Nicholas Anton Jao, Changchi Shuai, Keni Qiu, Mehrdad Mahdavi, Vijaykrishnan Narayanan

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

1 Scopus citations

Abstract

Crossbar arrays with non-volatile memory have recently become very popular for DNN acceleration due to their In-Memory-Computing property and low power requirements which makes them suitable for deployment on edge. Quantized Neural Networks (QNNs) enable us to run inference with limited hardware resource and power availability and can easily be ported on smaller devices. On the other hand, to make edge devices self sustainable a great deal of promise has been shown by energy harvesting scenarios. However, the power supplied by the energy harvesting sources is not constant which becomes problematic as a fixed trained neural network requires a constant amount of power to run inference. This work addresses this issue by tuning network precision at layer granularity for variable power availability predicted for different energy harvesting scenarios.

Original language	English (US)
Title of host publication	Internet of Things. A Confluence of Many Disciplines - 2nd IFIP International Cross-Domain Conference, IFIPIoT 2019, Revised Selected Papers
Editors	Augusto Casaca, Srinivas Katkoori, Sandip Ray, Leon Strous
Publisher	Springer
Pages	197-216
Number of pages	20
ISBN (Print)	9783030436049
DOIs	https://doi.org/10.1007/978-3-030-43605-6_12
State	Published - 2020
Event	2nd IFIP International Cross-Domain Conference on Internet of Things, IFIPIoT 2019 - Tampa, United States Duration: Oct 31 2019 → Nov 1 2019

Publication series

Name	IFIP Advances in Information and Communication Technology
Volume	574 IFIP
ISSN (Print)	1868-4238
ISSN (Electronic)	1868-422X

Conference

Conference	2nd IFIP International Cross-Domain Conference on Internet of Things, IFIPIoT 2019
Country/Territory	United States
City	Tampa
Period	10/31/19 → 11/1/19

All Science Journal Classification (ASJC) codes

Information Systems and Management

Access to Document

10.1007/978-3-030-43605-6_12

Cite this

Khan, M. F. F., Jao, N. A., Shuai, C., Qiu, K., Mahdavi, M., & Narayanan, V. (2020). Mixed Precision Quantization Scheme for Re-configurable ReRAM Crossbars Targeting Different Energy Harvesting Scenarios. In A. Casaca, S. Katkoori, S. Ray, & L. Strous (Eds.), Internet of Things. A Confluence of Many Disciplines - 2nd IFIP International Cross-Domain Conference, IFIPIoT 2019, Revised Selected Papers (pp. 197-216). (IFIP Advances in Information and Communication Technology; Vol. 574 IFIP). Springer. https://doi.org/10.1007/978-3-030-43605-6_12

Khan, Md Fahim Faysal ; Jao, Nicholas Anton ; Shuai, Changchi et al. / Mixed Precision Quantization Scheme for Re-configurable ReRAM Crossbars Targeting Different Energy Harvesting Scenarios. Internet of Things. A Confluence of Many Disciplines - 2nd IFIP International Cross-Domain Conference, IFIPIoT 2019, Revised Selected Papers. editor / Augusto Casaca ; Srinivas Katkoori ; Sandip Ray ; Leon Strous. Springer, 2020. pp. 197-216 (IFIP Advances in Information and Communication Technology).

@inproceedings{df18987aa0404bf79b3451906555a2bd,

title = "Mixed Precision Quantization Scheme for Re-configurable ReRAM Crossbars Targeting Different Energy Harvesting Scenarios",

abstract = "Crossbar arrays with non-volatile memory have recently become very popular for DNN acceleration due to their In-Memory-Computing property and low power requirements which makes them suitable for deployment on edge. Quantized Neural Networks (QNNs) enable us to run inference with limited hardware resource and power availability and can easily be ported on smaller devices. On the other hand, to make edge devices self sustainable a great deal of promise has been shown by energy harvesting scenarios. However, the power supplied by the energy harvesting sources is not constant which becomes problematic as a fixed trained neural network requires a constant amount of power to run inference. This work addresses this issue by tuning network precision at layer granularity for variable power availability predicted for different energy harvesting scenarios.",

author = "Khan, {Md Fahim Faysal} and Jao, {Nicholas Anton} and Changchi Shuai and Keni Qiu and Mehrdad Mahdavi and Vijaykrishnan Narayanan",

note = "Publisher Copyright: {\textcopyright} 2020, IFIP International Federation for Information Processing.; 2nd IFIP International Cross-Domain Conference on Internet of Things, IFIPIoT 2019 ; Conference date: 31-10-2019 Through 01-11-2019",

year = "2020",

doi = "10.1007/978-3-030-43605-6_12",

language = "English (US)",

isbn = "9783030436049",

series = "IFIP Advances in Information and Communication Technology",

publisher = "Springer",

pages = "197--216",

editor = "Augusto Casaca and Srinivas Katkoori and Sandip Ray and Leon Strous",

booktitle = "Internet of Things. A Confluence of Many Disciplines - 2nd IFIP International Cross-Domain Conference, IFIPIoT 2019, Revised Selected Papers",

}

Khan, MFF, Jao, NA, Shuai, C, Qiu, K, Mahdavi, M & Narayanan, V 2020, Mixed Precision Quantization Scheme for Re-configurable ReRAM Crossbars Targeting Different Energy Harvesting Scenarios. in A Casaca, S Katkoori, S Ray & L Strous (eds), Internet of Things. A Confluence of Many Disciplines - 2nd IFIP International Cross-Domain Conference, IFIPIoT 2019, Revised Selected Papers. IFIP Advances in Information and Communication Technology, vol. 574 IFIP, Springer, pp. 197-216, 2nd IFIP International Cross-Domain Conference on Internet of Things, IFIPIoT 2019, Tampa, United States, 10/31/19. https://doi.org/10.1007/978-3-030-43605-6_12

Mixed Precision Quantization Scheme for Re-configurable ReRAM Crossbars Targeting Different Energy Harvesting Scenarios. / Khan, Md Fahim Faysal; Jao, Nicholas Anton; Shuai, Changchi et al.
Internet of Things. A Confluence of Many Disciplines - 2nd IFIP International Cross-Domain Conference, IFIPIoT 2019, Revised Selected Papers. ed. / Augusto Casaca; Srinivas Katkoori; Sandip Ray; Leon Strous. Springer, 2020. p. 197-216 (IFIP Advances in Information and Communication Technology; Vol. 574 IFIP).

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

TY - GEN

T1 - Mixed Precision Quantization Scheme for Re-configurable ReRAM Crossbars Targeting Different Energy Harvesting Scenarios

AU - Khan, Md Fahim Faysal

AU - Jao, Nicholas Anton

AU - Shuai, Changchi

AU - Qiu, Keni

AU - Mahdavi, Mehrdad

AU - Narayanan, Vijaykrishnan

PY - 2020

Y1 - 2020

N2 - Crossbar arrays with non-volatile memory have recently become very popular for DNN acceleration due to their In-Memory-Computing property and low power requirements which makes them suitable for deployment on edge. Quantized Neural Networks (QNNs) enable us to run inference with limited hardware resource and power availability and can easily be ported on smaller devices. On the other hand, to make edge devices self sustainable a great deal of promise has been shown by energy harvesting scenarios. However, the power supplied by the energy harvesting sources is not constant which becomes problematic as a fixed trained neural network requires a constant amount of power to run inference. This work addresses this issue by tuning network precision at layer granularity for variable power availability predicted for different energy harvesting scenarios.

AB - Crossbar arrays with non-volatile memory have recently become very popular for DNN acceleration due to their In-Memory-Computing property and low power requirements which makes them suitable for deployment on edge. Quantized Neural Networks (QNNs) enable us to run inference with limited hardware resource and power availability and can easily be ported on smaller devices. On the other hand, to make edge devices self sustainable a great deal of promise has been shown by energy harvesting scenarios. However, the power supplied by the energy harvesting sources is not constant which becomes problematic as a fixed trained neural network requires a constant amount of power to run inference. This work addresses this issue by tuning network precision at layer granularity for variable power availability predicted for different energy harvesting scenarios.

UR - http://www.scopus.com/inward/record.url?scp=85084173062&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85084173062&partnerID=8YFLogxK

U2 - 10.1007/978-3-030-43605-6_12

DO - 10.1007/978-3-030-43605-6_12

M3 - Conference contribution

AN - SCOPUS:85084173062

SN - 9783030436049

T3 - IFIP Advances in Information and Communication Technology

SP - 197

EP - 216

BT - Internet of Things. A Confluence of Many Disciplines - 2nd IFIP International Cross-Domain Conference, IFIPIoT 2019, Revised Selected Papers

A2 - Casaca, Augusto

A2 - Katkoori, Srinivas

A2 - Ray, Sandip

A2 - Strous, Leon

PB - Springer

T2 - 2nd IFIP International Cross-Domain Conference on Internet of Things, IFIPIoT 2019

Y2 - 31 October 2019 through 1 November 2019

ER -

Khan MFF, Jao NA, Shuai C, Qiu K, Mahdavi M , Narayanan V. Mixed Precision Quantization Scheme for Re-configurable ReRAM Crossbars Targeting Different Energy Harvesting Scenarios. In Casaca A, Katkoori S, Ray S, Strous L, editors, Internet of Things. A Confluence of Many Disciplines - 2nd IFIP International Cross-Domain Conference, IFIPIoT 2019, Revised Selected Papers. Springer. 2020. p. 197-216. (IFIP Advances in Information and Communication Technology). doi: 10.1007/978-3-030-43605-6_12

Mixed Precision Quantization Scheme for Re-configurable ReRAM Crossbars Targeting Different Energy Harvesting Scenarios

Abstract

Publication series

Conference

All Science Journal Classification (ASJC) codes

Access to Document

Other files and links

Fingerprint

Cite this