SpiNNaker is a computing system composed of over a million ARM cores, embedded in a bespoke asynchronous communication fabric. The physical realization of the system consist of 57600 nodes (a node is a silicon die), each node containing 18 ARM cores and a routing engine. The communication infrastructure allows the cores to communicate via short, fixed-length (40- or 72-bit), hardware-brokered packets. The packets find their way through the network in a sequence of hops, and the specifics of each route are held (distributed) in the route engines, not unlike internet routing. On arrival at a target core, a hardware-triggered interrupt invokes code to handle the incoming packet. Within this computing model, the state of the system-under-simulation is distributed, held in memory local to the cores, and the topology is also distributed, held in the routing engine internal tables. The message passing is non-deterministic and non-transitive, there is no memory coherence between the core local memories, and there is no global synchronization. This paper shows how such a system can be used to simulate large systems of neurons using discrete event-based techniques. More notably, the solution time remains approximately constant with neural system size as long as sufficient hardware cores are available.
IOS Press, Inc.
6751 Tepper Drive
Clifton, VA 20124
Tel.: +1 703 830 6300
Fax: +1 703 830 2300 firstname.lastname@example.org
(Corporate matters and books only) IOS Press c/o Accucoms US, Inc.
For North America Sales and Customer Service
West Point Commons
Lansdale PA 19446
Tel.: +1 866 855 8967
Fax: +1 215 660 5042 email@example.com