Thermal Integrity of ReRAM-based Near-Memory Computing in 3D Integrated DNN Accelerators

Abrar Abdurrob; Emre Salman; Jack Lombardi

doi:10.1109/SOCC58585.2023.10256927

Thermal Integrity of ReRAM-based Near-Memory Computing in 3D Integrated DNN Accelerators

Abrar Abdurrob
, Emre Salman
, Jack Lombardi

Electrical Engineering

Stony Brook University
Air Force Research Laboratory

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

2 Scopus citations

Abstract

In this paper, the thermal integrity of near-memory computing in 3D integrated deep neural network (DNN) accelerators is investigated. Both conventional memory technologies (such as SRAM and DRAM) and emerging resistive memory (ReRAM) technology are considered. Through silicon via (TSV) based 3D integration and monolithic inter-tier via (MIV) based 3D technologies are leveraged for near-memory computing. The results demonstrate that monolithic 3D integrated DNN accelerator is thermally more feasible than TSV-based 3D accelerator due to reduced thermal resistance to heat sink. Furthermore, for near-memory computing with three-tier 3D systems, ReRAM based accelerator produces the lowest temperature whereas embedded DRAM (eDRAM) significantly increases the peak temperature.

Original language	English
Title of host publication	Proceedings - 2023 IEEE 36th International System-on-Chip Conference, SOCC 2023
Editors	Jurgen Becker, Andrew Marshall, Tanja Harbaum, Amlan Ganguly, Fahad Siddiqui, Kieran McLaughlin
Publisher	IEEE Computer Society
ISBN (Electronic)	9798350300116
DOIs	https://doi.org/10.1109/SOCC58585.2023.10256927
State	Published - 2023
Event	36th IEEE International System-on-Chip Conference, SOCC 2023 - Santa Clara, United States Duration: Sep 5 2023 → Sep 8 2023

Publication series

Name	International System on Chip Conference
Volume	2023-September
ISSN (Print)	2164-1676
ISSN (Electronic)	2164-1706

Conference

Conference	36th IEEE International System-on-Chip Conference, SOCC 2023
Country/Territory	United States
City	Santa Clara
Period	09/5/23 → 09/8/23

Access to Document

10.1109/SOCC58585.2023.10256927

Cite this

Abdurrob, A., Salman, E., & Lombardi, J. (2023). Thermal Integrity of ReRAM-based Near-Memory Computing in 3D Integrated DNN Accelerators. In J. Becker, A. Marshall, T. Harbaum, A. Ganguly, F. Siddiqui, & K. McLaughlin (Eds.), Proceedings - 2023 IEEE 36th International System-on-Chip Conference, SOCC 2023 (International System on Chip Conference; Vol. 2023-September). IEEE Computer Society. https://doi.org/10.1109/SOCC58585.2023.10256927

Abdurrob, Abrar ; Salman, Emre ; Lombardi, Jack. / Thermal Integrity of ReRAM-based Near-Memory Computing in 3D Integrated DNN Accelerators. Proceedings - 2023 IEEE 36th International System-on-Chip Conference, SOCC 2023. editor / Jurgen Becker ; Andrew Marshall ; Tanja Harbaum ; Amlan Ganguly ; Fahad Siddiqui ; Kieran McLaughlin. IEEE Computer Society, 2023. (International System on Chip Conference).

@inproceedings{745f5545c0724e058ea42124822988d9,

title = "Thermal Integrity of ReRAM-based Near-Memory Computing in 3D Integrated DNN Accelerators",

abstract = "In this paper, the thermal integrity of near-memory computing in 3D integrated deep neural network (DNN) accelerators is investigated. Both conventional memory technologies (such as SRAM and DRAM) and emerging resistive memory (ReRAM) technology are considered. Through silicon via (TSV) based 3D integration and monolithic inter-tier via (MIV) based 3D technologies are leveraged for near-memory computing. The results demonstrate that monolithic 3D integrated DNN accelerator is thermally more feasible than TSV-based 3D accelerator due to reduced thermal resistance to heat sink. Furthermore, for near-memory computing with three-tier 3D systems, ReRAM based accelerator produces the lowest temperature whereas embedded DRAM (eDRAM) significantly increases the peak temperature.",

author = "Abrar Abdurrob and Emre Salman and Jack Lombardi",

note = "Publisher Copyright: {\textcopyright} 2023 IEEE.; 36th IEEE International System-on-Chip Conference, SOCC 2023 ; Conference date: 05-09-2023 Through 08-09-2023",

year = "2023",

doi = "10.1109/SOCC58585.2023.10256927",

language = "English",

series = "International System on Chip Conference",

publisher = "IEEE Computer Society",

editor = "Jurgen Becker and Andrew Marshall and Tanja Harbaum and Amlan Ganguly and Fahad Siddiqui and Kieran McLaughlin",

booktitle = "Proceedings - 2023 IEEE 36th International System-on-Chip Conference, SOCC 2023",

}

Abdurrob, A, Salman, E & Lombardi, J 2023, Thermal Integrity of ReRAM-based Near-Memory Computing in 3D Integrated DNN Accelerators. in J Becker, A Marshall, T Harbaum, A Ganguly, F Siddiqui & K McLaughlin (eds), Proceedings - 2023 IEEE 36th International System-on-Chip Conference, SOCC 2023. International System on Chip Conference, vol. 2023-September, IEEE Computer Society, 36th IEEE International System-on-Chip Conference, SOCC 2023, Santa Clara, United States, 09/5/23. https://doi.org/10.1109/SOCC58585.2023.10256927

Thermal Integrity of ReRAM-based Near-Memory Computing in 3D Integrated DNN Accelerators. / Abdurrob, Abrar; Salman, Emre; Lombardi, Jack.
Proceedings - 2023 IEEE 36th International System-on-Chip Conference, SOCC 2023. ed. / Jurgen Becker; Andrew Marshall; Tanja Harbaum; Amlan Ganguly; Fahad Siddiqui; Kieran McLaughlin. IEEE Computer Society, 2023. (International System on Chip Conference; Vol. 2023-September).

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

TY - GEN

T1 - Thermal Integrity of ReRAM-based Near-Memory Computing in 3D Integrated DNN Accelerators

AU - Abdurrob, Abrar

AU - Salman, Emre

AU - Lombardi, Jack

PY - 2023

Y1 - 2023

N2 - In this paper, the thermal integrity of near-memory computing in 3D integrated deep neural network (DNN) accelerators is investigated. Both conventional memory technologies (such as SRAM and DRAM) and emerging resistive memory (ReRAM) technology are considered. Through silicon via (TSV) based 3D integration and monolithic inter-tier via (MIV) based 3D technologies are leveraged for near-memory computing. The results demonstrate that monolithic 3D integrated DNN accelerator is thermally more feasible than TSV-based 3D accelerator due to reduced thermal resistance to heat sink. Furthermore, for near-memory computing with three-tier 3D systems, ReRAM based accelerator produces the lowest temperature whereas embedded DRAM (eDRAM) significantly increases the peak temperature.

AB - In this paper, the thermal integrity of near-memory computing in 3D integrated deep neural network (DNN) accelerators is investigated. Both conventional memory technologies (such as SRAM and DRAM) and emerging resistive memory (ReRAM) technology are considered. Through silicon via (TSV) based 3D integration and monolithic inter-tier via (MIV) based 3D technologies are leveraged for near-memory computing. The results demonstrate that monolithic 3D integrated DNN accelerator is thermally more feasible than TSV-based 3D accelerator due to reduced thermal resistance to heat sink. Furthermore, for near-memory computing with three-tier 3D systems, ReRAM based accelerator produces the lowest temperature whereas embedded DRAM (eDRAM) significantly increases the peak temperature.

UR - https://www.scopus.com/pages/publications/85174600187

U2 - 10.1109/SOCC58585.2023.10256927

DO - 10.1109/SOCC58585.2023.10256927

M3 - Conference contribution

AN - SCOPUS:85174600187

T3 - International System on Chip Conference

BT - Proceedings - 2023 IEEE 36th International System-on-Chip Conference, SOCC 2023

A2 - Becker, Jurgen

A2 - Marshall, Andrew

A2 - Harbaum, Tanja

A2 - Ganguly, Amlan

A2 - Siddiqui, Fahad

A2 - McLaughlin, Kieran

PB - IEEE Computer Society

T2 - 36th IEEE International System-on-Chip Conference, SOCC 2023

Y2 - 5 September 2023 through 8 September 2023

ER -

Abdurrob A, Salman E, Lombardi J. Thermal Integrity of ReRAM-based Near-Memory Computing in 3D Integrated DNN Accelerators. In Becker J, Marshall A, Harbaum T, Ganguly A, Siddiqui F, McLaughlin K, editors, Proceedings - 2023 IEEE 36th International System-on-Chip Conference, SOCC 2023. IEEE Computer Society. 2023. (International System on Chip Conference). doi: 10.1109/SOCC58585.2023.10256927

Thermal Integrity of ReRAM-based Near-Memory Computing in 3D Integrated DNN Accelerators

Abstract

Publication series

Conference

Access to Document

Other files and links

Fingerprint

Cite this