Scalable three-dimensional optical interconnects for data centers

R. Morris, A. K. Kodi, Ahmed Louri

Research output: Chapter in Book/Report/Conference proceedingChapter

Abstract

Technology scaling has paved the way to integrate hundreds and probably thousands of cores on a single-chip in the future. As the numbers of cores increases, energy efficiency and scalability of Network-on-Chips(NoCs) has become a critical challenge that could potentially hinder the performance of future warehouse scale computers (WSC) or datacenters. Emerging technologies such as silicon photonics and 3D stacking are currently being explored to overcome the drawbacks of traditional metallic interconnects for on-chip networks. In this chapter, we combine 3D stacking and silicon photonics to deliver high on-chip bandwidth and low energy/bit to achieve a high throughput, reconfigurable, energy efficient photonic NoC fabric. We propose to develop a multi-layer photonic interconnect that can dynamically reconfigure without system intervention and allocate channel bandwidth from less utilized links to more utilized communication links. In addition, we propose a Power Reduction Technique (PRT) that dynamically deactivates photonic interconnects that are idle and reduce power dissipation. For 64-core reconfigured network, our simulation results indicate that the performance can be further improved by 10%-25% for Splash-2, PARSEC and SPEC CPU2006 benchmarks, whereas simulation results for a 256-core chip indicate a performance improvement of more than 25% while saving 23% of energy when compared to state-of-the-art on-chip electrical and optical networks.

Original languageEnglish (US)
Title of host publicationOptical Interconnects for Data Centers
PublisherElsevier Inc.
Pages223-246
Number of pages24
ISBN (Electronic)9780081005125
ISBN (Print)9780081005132
DOIs
StatePublished - Nov 14 2016
Externally publishedYes

Fingerprint

optical interconnects
Optical interconnects
Photonics
chips
photonics
Telecommunication links
Bandwidth
Silicon
bandwidth
Warehouses
Fiber optic networks
energy
silicon
Energy efficiency
Scalability
Energy dissipation
Throughput
emerging
dissipation
simulation

Keywords

  • Data center
  • Network-on-Chips (NoCs)
  • Photonic interconnects
  • Reconfigurable network
  • Router microarchitecture

ASJC Scopus subject areas

  • Engineering(all)
  • Physics and Astronomy(all)

Cite this

Morris, R., Kodi, A. K., & Louri, A. (2016). Scalable three-dimensional optical interconnects for data centers. In Optical Interconnects for Data Centers (pp. 223-246). Elsevier Inc.. https://doi.org/10.1016/B978-0-08-100512-5.00009-7

Scalable three-dimensional optical interconnects for data centers. / Morris, R.; Kodi, A. K.; Louri, Ahmed.

Optical Interconnects for Data Centers. Elsevier Inc., 2016. p. 223-246.

Research output: Chapter in Book/Report/Conference proceedingChapter

Morris, R, Kodi, AK & Louri, A 2016, Scalable three-dimensional optical interconnects for data centers. in Optical Interconnects for Data Centers. Elsevier Inc., pp. 223-246. https://doi.org/10.1016/B978-0-08-100512-5.00009-7
Morris R, Kodi AK, Louri A. Scalable three-dimensional optical interconnects for data centers. In Optical Interconnects for Data Centers. Elsevier Inc. 2016. p. 223-246 https://doi.org/10.1016/B978-0-08-100512-5.00009-7
Morris, R. ; Kodi, A. K. ; Louri, Ahmed. / Scalable three-dimensional optical interconnects for data centers. Optical Interconnects for Data Centers. Elsevier Inc., 2016. pp. 223-246
@inbook{6aeed740567e46f6b6d81624ae19d5f4,
title = "Scalable three-dimensional optical interconnects for data centers",
abstract = "Technology scaling has paved the way to integrate hundreds and probably thousands of cores on a single-chip in the future. As the numbers of cores increases, energy efficiency and scalability of Network-on-Chips(NoCs) has become a critical challenge that could potentially hinder the performance of future warehouse scale computers (WSC) or datacenters. Emerging technologies such as silicon photonics and 3D stacking are currently being explored to overcome the drawbacks of traditional metallic interconnects for on-chip networks. In this chapter, we combine 3D stacking and silicon photonics to deliver high on-chip bandwidth and low energy/bit to achieve a high throughput, reconfigurable, energy efficient photonic NoC fabric. We propose to develop a multi-layer photonic interconnect that can dynamically reconfigure without system intervention and allocate channel bandwidth from less utilized links to more utilized communication links. In addition, we propose a Power Reduction Technique (PRT) that dynamically deactivates photonic interconnects that are idle and reduce power dissipation. For 64-core reconfigured network, our simulation results indicate that the performance can be further improved by 10{\%}-25{\%} for Splash-2, PARSEC and SPEC CPU2006 benchmarks, whereas simulation results for a 256-core chip indicate a performance improvement of more than 25{\%} while saving 23{\%} of energy when compared to state-of-the-art on-chip electrical and optical networks.",
keywords = "Data center, Network-on-Chips (NoCs), Photonic interconnects, Reconfigurable network, Router microarchitecture",
author = "R. Morris and Kodi, {A. K.} and Ahmed Louri",
year = "2016",
month = "11",
day = "14",
doi = "10.1016/B978-0-08-100512-5.00009-7",
language = "English (US)",
isbn = "9780081005132",
pages = "223--246",
booktitle = "Optical Interconnects for Data Centers",
publisher = "Elsevier Inc.",
address = "United States",

}

TY - CHAP

T1 - Scalable three-dimensional optical interconnects for data centers

AU - Morris, R.

AU - Kodi, A. K.

AU - Louri, Ahmed

PY - 2016/11/14

Y1 - 2016/11/14

N2 - Technology scaling has paved the way to integrate hundreds and probably thousands of cores on a single-chip in the future. As the numbers of cores increases, energy efficiency and scalability of Network-on-Chips(NoCs) has become a critical challenge that could potentially hinder the performance of future warehouse scale computers (WSC) or datacenters. Emerging technologies such as silicon photonics and 3D stacking are currently being explored to overcome the drawbacks of traditional metallic interconnects for on-chip networks. In this chapter, we combine 3D stacking and silicon photonics to deliver high on-chip bandwidth and low energy/bit to achieve a high throughput, reconfigurable, energy efficient photonic NoC fabric. We propose to develop a multi-layer photonic interconnect that can dynamically reconfigure without system intervention and allocate channel bandwidth from less utilized links to more utilized communication links. In addition, we propose a Power Reduction Technique (PRT) that dynamically deactivates photonic interconnects that are idle and reduce power dissipation. For 64-core reconfigured network, our simulation results indicate that the performance can be further improved by 10%-25% for Splash-2, PARSEC and SPEC CPU2006 benchmarks, whereas simulation results for a 256-core chip indicate a performance improvement of more than 25% while saving 23% of energy when compared to state-of-the-art on-chip electrical and optical networks.

AB - Technology scaling has paved the way to integrate hundreds and probably thousands of cores on a single-chip in the future. As the numbers of cores increases, energy efficiency and scalability of Network-on-Chips(NoCs) has become a critical challenge that could potentially hinder the performance of future warehouse scale computers (WSC) or datacenters. Emerging technologies such as silicon photonics and 3D stacking are currently being explored to overcome the drawbacks of traditional metallic interconnects for on-chip networks. In this chapter, we combine 3D stacking and silicon photonics to deliver high on-chip bandwidth and low energy/bit to achieve a high throughput, reconfigurable, energy efficient photonic NoC fabric. We propose to develop a multi-layer photonic interconnect that can dynamically reconfigure without system intervention and allocate channel bandwidth from less utilized links to more utilized communication links. In addition, we propose a Power Reduction Technique (PRT) that dynamically deactivates photonic interconnects that are idle and reduce power dissipation. For 64-core reconfigured network, our simulation results indicate that the performance can be further improved by 10%-25% for Splash-2, PARSEC and SPEC CPU2006 benchmarks, whereas simulation results for a 256-core chip indicate a performance improvement of more than 25% while saving 23% of energy when compared to state-of-the-art on-chip electrical and optical networks.

KW - Data center

KW - Network-on-Chips (NoCs)

KW - Photonic interconnects

KW - Reconfigurable network

KW - Router microarchitecture

UR - http://www.scopus.com/inward/record.url?scp=85032374689&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85032374689&partnerID=8YFLogxK

U2 - 10.1016/B978-0-08-100512-5.00009-7

DO - 10.1016/B978-0-08-100512-5.00009-7

M3 - Chapter

SN - 9780081005132

SP - 223

EP - 246

BT - Optical Interconnects for Data Centers

PB - Elsevier Inc.

ER -