Understanding the Interconnection Network of a Massively Parallel Real-Time Neural Net Simulator

Javier Navaridas, Mikel Lujan, Jose Miguel-Alonso, Luis Plana, Steve Furber

Abstract

SpiNNaker is a massively parallel architecture designed to model large-scale sp iking neural networks in (biological) real-time. Its design is based around ad hoc multi-core System-on-Chips which are interconnect ed using a two-dimensional toroidal triangular mesh. Neurons are modeled in software and their spikes generate packets that pr opagate through the on- and inter-chip communication fabric relying on custom-made on-chip multicast routers. This pap er models and evaluates large-scale instances of its novel interconnect, focusing on real-time features and fault tolerance. Fir st we derive analytically the topological characteristics of the network which are later confirmed by experimental work. With the computational model developed, we investigate the topology of SpiNNaker, and compare it with a standard 3-dimensi onal torus. The ad hoc emergency routing mechanism, implemented within the routers, allows the topology of Spinnaker to be more robust than the 3-dimensional torus, regardless of the latter having better topological characteristics. Furthermore , we search for optimal values of two router parameters related with livelock and deadlock avoidance mechanisms.