A GALS infrastructure for a massively parallel multiprocessor

The Spinnaker (Spiking Neural Network Architecture) system for large-scale neural modeling is based on a scalable processor chip containing multiple ARM cores. Using a globally asynchronous, locally synchronous (GALS) approach allows custom, off-the-shelf IP to be readily integrated without significant timing-closure design effort. The ARM processors are used to simulate neurons, and generated neural events are carried over an on-chip, packet-switched fabric. This self-timed interconnect is also extended off chip to a provide chip-to-chip interconnect that scales to networks of thousands of chips. © 2007 IEEE.