updated 05:00 pm EDT, Mon July 11, 2011
ARM-based SpiNNaker project to simulate brain
Steve Furber, who was the main designer of the 32-bit ARM RISC processor in the 1980s and is now the ICL professor at the University of Manchester in the UK, believes an ARM core can substitute 1,000 spiking neurons in the human brain, The Register reported recently. This is done as part of Furber's SpiNNaker project (PDF), which is being performed in conjunction with Andrew Brown of the University of Southampton. As part of the project, the two are trying to build a supercomputer cluster with one million processors to simulate the activities of nearly one billion neurons.
This translates to only about percent of the human brain, as it's generally believed the human brain has between 80 and 90 billion neurons. Through their undertaking, the pair hopes to better understand how the brain works. They aim to prove the speculation that data in the brain is encoded in the order in which populations of neurons fire, among other things.
The SpiNNaker system uses an embedded version of the 32-bit ARM processor Furber designed. He did so because of his familiarity with the design, the amount of available tools to customize it and its energy efficiency.
The team has thus far made processors with between four and eight ARM968E-S cores on a single die, with simulator work making them confident they can put up to 20 cores on a single die. One core from such a design would be a monitor processor to control how code is loaded on the remaining 19 cores. Each core will sport 64KB of data cache and 32KB of instruction cache, a communications controller that simulates neural spikes using packets, and a memory controller to link to 1GB of DDR1 main memory located on the chip package.
The simulated neurons can fire off a pulse to any other simulated neuron in the million-core system in about one millisecond, or about as fast as neurons.
SpiNNaker ARM-based system on chip
SpiNNaker's 2D mesh interconnect