On the Reliability of RRAM-Based Neural Networks

Hassen Aziza; Cristian  Zambelli; Said Hamdioui; Sumit Diware; Rajendra Bishnoi; Anteneh Gebregiorgis

doi:10.1109/VLSI-SoC57769.2023.10321859

On the Reliability of RRAM-Based Neural Networks

Hassen Aziza, Cristian Zambelli, Said Hamdioui, Sumit Diware, Rajendra Bishnoi, Anteneh Gebregiorgis

Computer Engineering

Research output: Chapter in Book/Conference proceedings/Edited volume › Conference contribution › Scientific › peer-review

37 Downloads (Pure)

Abstract

Emerging device technologies such as Resistive RAMs (RRAMs) are under investigation by many researchers and semiconductor companies; not only to realize e.g., embedded non-volatile memories, but also to enable energy-efficient computing making use of new data processing paradigms such as computation-in-memory. However, such devices suffer from various non-idealities and reliability failure mechanisms (e.g., variability, endurance, and retention); these negatively impact the memory robustness and the computation accuracy. This paper discusses the non-idealities and reliability failure mechanisms for RRAM devices, provides an overview on the most popular ones. In addition, it reports detailed anlysis of some of these based on data measurements. Finally, it presents two different mitigation schemes for RRAM based accelerators; one is based on RRAM non-ideality aware quantization and conductance control for neural network accuracy enhancement while the second is based on reliability-aware biased training technique.

Original language	English
Title of host publication	Proceedings of the 2023 IFIP/IEEE 31st International Conference on Very Large Scale Integration (VLSI-SoC)
Publisher	IEEE
Number of pages	8
ISBN (Electronic)	979-8-3503-2599-7
ISBN (Print)	979-8-3503-2600-0
DOIs	https://doi.org/10.1109/VLSI-SoC57769.2023.10321859
Publication status	Published - 2023
Event	2023 IFIP/IEEE 31st International Conference on Very Large Scale Integration (VLSI-SoC) - Dubai, United Arab Emirates Duration: 16 Oct 2023 → 18 Oct 2023 Conference number: 31st

Conference

Conference	2023 IFIP/IEEE 31st International Conference on Very Large Scale Integration (VLSI-SoC)
Country/Territory	United Arab Emirates
City	Dubai
Period	16/10/23 → 18/10/23

Bibliographical note

Green Open Access added to TU Delft Institutional Repository 'You share, we take care!' - Taverne project https://www.openaccess.nl/en/you-share-we-take-care
Otherwise as indicated in the copyright section: the publisher is the copyright holder of this work and the author uses the Dutch legislation to make this work public.

Keywords

RRAM
reliability
neural network
in-memory computing

Access to Document

10.1109/VLSI-SoC57769.2023.10321859

On_the_Reliability_of_RRAM-Based_Neural_NetworksFinal published version, 2.49 MB

Cite this

@inproceedings{3e51d5a16f654c45b5e615999e215f78,

title = "On the Reliability of RRAM-Based Neural Networks",

abstract = "Emerging device technologies such as Resistive RAMs (RRAMs) are under investigation by many researchers and semiconductor companies; not only to realize e.g., embedded non-volatile memories, but also to enable energy-efficient computing making use of new data processing paradigms such as computation-in-memory. However, such devices suffer from various non-idealities and reliability failure mechanisms (e.g., variability, endurance, and retention); these negatively impact the memory robustness and the computation accuracy. This paper discusses the non-idealities and reliability failure mechanisms for RRAM devices, provides an overview on the most popular ones. In addition, it reports detailed anlysis of some of these based on data measurements. Finally, it presents two different mitigation schemes for RRAM based accelerators; one is based on RRAM non-ideality aware quantization and conductance control for neural network accuracy enhancement while the second is based on reliability-aware biased training technique.",

keywords = "RRAM, reliability, neural network, in-memory computing",

author = "Hassen Aziza and Cristian Zambelli and Said Hamdioui and Sumit Diware and Rajendra Bishnoi and Anteneh Gebregiorgis",

note = "Green Open Access added to TU Delft Institutional Repository 'You share, we take care!' - Taverne project https://www.openaccess.nl/en/you-share-we-take-care Otherwise as indicated in the copyright section: the publisher is the copyright holder of this work and the author uses the Dutch legislation to make this work public.; 2023 IFIP/IEEE 31st International Conference on Very Large Scale Integration (VLSI-SoC) ; Conference date: 16-10-2023 Through 18-10-2023",

year = "2023",

doi = "10.1109/VLSI-SoC57769.2023.10321859",

language = "English",

isbn = "979-8-3503-2600-0",

booktitle = "Proceedings of the 2023 IFIP/IEEE 31st International Conference on Very Large Scale Integration (VLSI-SoC)",

publisher = "IEEE",

address = "United States",

}

Aziza, H, Zambelli, C, Hamdioui, S , Diware, S , Bishnoi, R & Gebregiorgis, A 2023, On the Reliability of RRAM-Based Neural Networks. in Proceedings of the 2023 IFIP/IEEE 31st International Conference on Very Large Scale Integration (VLSI-SoC). IEEE, 2023 IFIP/IEEE 31st International Conference on Very Large Scale Integration (VLSI-SoC), Dubai, United Arab Emirates, 16/10/23. https://doi.org/10.1109/VLSI-SoC57769.2023.10321859

TY - GEN

T1 - On the Reliability of RRAM-Based Neural Networks

AU - Aziza, Hassen

AU - Zambelli, Cristian

AU - Hamdioui, Said

AU - Diware, Sumit

AU - Bishnoi, Rajendra

AU - Gebregiorgis, Anteneh

N1 - Conference code: 31st

PY - 2023

Y1 - 2023

N2 - Emerging device technologies such as Resistive RAMs (RRAMs) are under investigation by many researchers and semiconductor companies; not only to realize e.g., embedded non-volatile memories, but also to enable energy-efficient computing making use of new data processing paradigms such as computation-in-memory. However, such devices suffer from various non-idealities and reliability failure mechanisms (e.g., variability, endurance, and retention); these negatively impact the memory robustness and the computation accuracy. This paper discusses the non-idealities and reliability failure mechanisms for RRAM devices, provides an overview on the most popular ones. In addition, it reports detailed anlysis of some of these based on data measurements. Finally, it presents two different mitigation schemes for RRAM based accelerators; one is based on RRAM non-ideality aware quantization and conductance control for neural network accuracy enhancement while the second is based on reliability-aware biased training technique.

AB - Emerging device technologies such as Resistive RAMs (RRAMs) are under investigation by many researchers and semiconductor companies; not only to realize e.g., embedded non-volatile memories, but also to enable energy-efficient computing making use of new data processing paradigms such as computation-in-memory. However, such devices suffer from various non-idealities and reliability failure mechanisms (e.g., variability, endurance, and retention); these negatively impact the memory robustness and the computation accuracy. This paper discusses the non-idealities and reliability failure mechanisms for RRAM devices, provides an overview on the most popular ones. In addition, it reports detailed anlysis of some of these based on data measurements. Finally, it presents two different mitigation schemes for RRAM based accelerators; one is based on RRAM non-ideality aware quantization and conductance control for neural network accuracy enhancement while the second is based on reliability-aware biased training technique.

KW - RRAM

KW - reliability

KW - neural network

KW - in-memory computing

UR - http://www.scopus.com/inward/record.url?scp=85179841555&partnerID=8YFLogxK

U2 - 10.1109/VLSI-SoC57769.2023.10321859

DO - 10.1109/VLSI-SoC57769.2023.10321859

M3 - Conference contribution

SN - 979-8-3503-2600-0

BT - Proceedings of the 2023 IFIP/IEEE 31st International Conference on Very Large Scale Integration (VLSI-SoC)

PB - IEEE

T2 - 2023 IFIP/IEEE 31st International Conference on Very Large Scale Integration (VLSI-SoC)

Y2 - 16 October 2023 through 18 October 2023

ER -

On the Reliability of RRAM-Based Neural Networks

Abstract

Conference

Bibliographical note

Keywords

Access to Document

Other files and links

Fingerprint

Cite this