SpinLink: An interconnection system for the SpiNNaker biologically inspired multi-computer
SpinLink: An interconnection system for the SpiNNaker biologically inspired multi-computer
SpiNNaker is a large-scale biologically-inspired multi-computer designed to model very heavily distributed problems, with the flagship application being the simulation of large neural networks. The project goal is to have one million processors included in a single machine, which consequently span many thousands of circuit boards. A computer of this scale imposes large communication requirements between these boards, and requires an extensible method of connecting to external equipment such as sensors, actuators and visualisation systems. This paper describes two systems that can address each of these problems.
Firstly, SpinLink is a proposed method of connecting the SpiNNaker boards by using time-division multiplexing (TDM) to allow eight SpiNNaker links to run at maximum bandwidth between two boards. SpinLink will be deployed on Spartan-6 FPGAs and uses a locally generated clock that can be paused while the asynchronous links from SpiNNaker are sending data, thus ensuring a fast and glitch-free response. Secondly, SpiNNterceptor is a separate system, currently in the early stages of design, that will build upon SpinLink to address the important external I/O issues faced by SpiNNaker. Specifically, spare resources in the FPGAs will be used to implement the debugging and I/O interfacing features of SpiNNterceptor.
0701702451
52-58
Dugan, Kier J.
8e29c558-b709-466f-96dc-1a26f8f91362
Reeve, Jeff S.
dd909010-7d44-44ea-83fe-a09e4d492618
Brown, Andrew D.
5c19e523-65ec-499b-9e7c-91522017d7e0
30 August 2012
Dugan, Kier J.
8e29c558-b709-466f-96dc-1a26f8f91362
Reeve, Jeff S.
dd909010-7d44-44ea-83fe-a09e4d492618
Brown, Andrew D.
5c19e523-65ec-499b-9e7c-91522017d7e0
Dugan, Kier J., Reeve, Jeff S. and Brown, Andrew D.
(2012)
SpinLink: An interconnection system for the SpiNNaker biologically inspired multi-computer.
UK Electronics Forum 2012, Newcastle upon Tyne, United Kingdom.
30 - 31 Aug 2012.
.
Record type:
Conference or Workshop Item
(Paper)
Abstract
SpiNNaker is a large-scale biologically-inspired multi-computer designed to model very heavily distributed problems, with the flagship application being the simulation of large neural networks. The project goal is to have one million processors included in a single machine, which consequently span many thousands of circuit boards. A computer of this scale imposes large communication requirements between these boards, and requires an extensible method of connecting to external equipment such as sensors, actuators and visualisation systems. This paper describes two systems that can address each of these problems.
Firstly, SpinLink is a proposed method of connecting the SpiNNaker boards by using time-division multiplexing (TDM) to allow eight SpiNNaker links to run at maximum bandwidth between two boards. SpinLink will be deployed on Spartan-6 FPGAs and uses a locally generated clock that can be paused while the asynchronous links from SpiNNaker are sending data, thus ensuring a fast and glitch-free response. Secondly, SpiNNterceptor is a separate system, currently in the early stages of design, that will build upon SpinLink to address the important external I/O issues faced by SpiNNaker. Specifically, spare resources in the FPGAs will be used to implement the debugging and I/O interfacing features of SpiNNterceptor.
Text
UKEF-20120524.pdf
- Author's Original
More information
Submitted date: 25 May 2012
Published date: 30 August 2012
Venue - Dates:
UK Electronics Forum 2012, Newcastle upon Tyne, United Kingdom, 2012-08-30 - 2012-08-31
Organisations:
EEE
Identifiers
Local EPrints ID: 344374
URI: http://eprints.soton.ac.uk/id/eprint/344374
ISBN: 0701702451
PURE UUID: 38c962ee-a7dc-41d1-bd7c-aefcd7ff38b0
Catalogue record
Date deposited: 23 Oct 2012 11:09
Last modified: 14 Mar 2024 12:12
Export record
Contributors
Author:
Kier J. Dugan
Author:
Jeff S. Reeve
Author:
Andrew D. Brown
Download statistics
Downloads from ePrints over the past year. Other digital versions may also be available to download e.g. from the publisher's website.
View more statistics