Enabling Low-Power, Fault-Resilient Communications in a Million-Core Neural Processing Architecture

Javier Navaridas, Mikel Luján, Luis A. Plana, Stephen Furber

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

The SpiNNaker neuromimetic architecture is a biologically-inspired massively-parallel architecture based on a custom-made multicore System-on-Chip (SoC). In order to deal with the challenging and atypical communication demands of spiking neural networks, SpiNNaker features a specialised, ad hoc communication infrastructure based around a custom-made multicast router. This paper summarizes the main research work
around the peculiarities of SpiNNaker's architecture and
interconnects focusing on the most exceptional features of the
platform. We derived the main topological properties of the
network, analyzed the effects of failures congestion and traffic
burstiness in the stability of the interconnect and, finally,
proposed a collection of multicast routing algorithms.
Original languageEnglish
Title of host publicationEmerging Technologies'16
Place of PublicationBarcelona
Pages20-23
Publication statusPublished - 2 Jun 2016

Fingerprint

Dive into the research topics of 'Enabling Low-Power, Fault-Resilient Communications in a Million-Core Neural Processing Architecture'. Together they form a unique fingerprint.

Cite this