Toggle Main Menu Toggle Search

Open Access padlockePrints

Efficient surface-wave-based network-on-chip architecture for spiking-neural-network

Lookup NU author(s): Ammar Karkar, Professor Alex Yakovlev

Downloads

Full text for this publication is not currently held within this repository. Alternative links are provided below where available.


Abstract

© 2017 IEEE. Networks-on-chip (NoC) has been proven to satisfy different on-chip communication requirements in terms of costs, performance, and reliabilities. This has been proven true especially for spiking neural network (SNN) with its high multicast communication demands. However, metal-wires that form the foundation for regular NoCs face a set of challenges since metal-wires have been stretched to their physical limits with the relentless technology scaling. Thus, this paper tackles SNN communication issues on two fronts: Firstly, the physical interconnect layer by attempting to prototyping a new and promising type of wireless interconnects called surface-wave interconnects (SWI). Secondly, the architectural level by proposing NoC architecture enabled by SWI and further more effective for multicast communication. A surface-wave has been demonstrated for range 14-25 GHz with improvements over free-space propagation in S21 up to 20 dB. Furthermore, based on this technology a hybrid wire and surface-wave architecture is developed and evaluated to demonstrate its potentials for SNN, which found to perform 95% better than related-work. Consequently, this emerging technology would provide the high efficiency that are required for future NoC-based SNN systems.


Publication metadata

Author(s): Karkar AJM, Mak T, Yakovlev A

Publication type: Conference Proceedings (inc. Abstract)

Publication status: Published

Conference Name: 2017 Annual Conference on New Trends in Information and Communications Technology Applications, NTICT 2017

Year of Conference: 2017

Pages: 322-327

Online publication date: 13/07/2017

Acceptance date: 02/04/2016

Publisher: Institute of Electrical and Electronics Engineers Inc.

URL: https://doi.org/10.1109/NTICT.2017.7976117

DOI: 10.1109/NTICT.2017.7976117

Library holdings: Search Newcastle University Library for this item

ISBN: 9781538629628


Actions

Link to this publication


Share