One million ARM cores to simulate brain at Manchester
UK scientists aim to model 1% of a human brain with up to one million ARM cores.
“They will form the system architecture for a massive computer, called SpiNNaker [spiking neural network architecture], which aims to map-out the brain’s individual functions,” said the University of Manchester. “SpiNNaker could be a vital tool for neuroscientists, psychologists and doctors to help them understand complex brain injuries, diseases and conditions, and identify the most effective therapies.”
Manchester is getting £2.5m from the EPSRC for designing the architecture, with the Universities of Southampton, Cambridge and Sheffield sharing another £2.5m for further work towards the computer.
“The key challenge is developing and understanding the information processing of the brain and the extremely high connectivity of the brain cells,” said the University of Manchester. “There are 100 billion neurons with 1,000 million connections in the human brain.”
Neurons communicate with spikes of electricity, which SpiNNaker will model as data packets.
“This packet is sent to all connected neurons. Neurons are represented by simple equations which are solved in real-time by software running on the ARM processors,” said the University. “The electronic connections in SpiNNaker convey these spikes much quicker than the biological connections in the brain, hence SpiNNaker can transmit spikes as effectively and quickly with many fewer connections.”
The researchers is led by Professor Steve Furber of Manchester, who was lead hardware designer of both the BBC Micro computer and the original ARM core.
SpiNNaker will be constructed using custom chips, each with 18 ARM cores, designed in Manchester and manufactured in Taiwan. Each processor chip is mounted with a memory chip in a 19x19mm 3D system-in-package from Unisem Europe. The package is claimed to deliver the “computing power of a PC” and dissipate 1W.
“We hope the machine will be able to model important functions of the human brain and help us gain key insights into their principles of operation,” said Furber. “Developing and understanding information processing in the brain is the key. We are actively engaging with neuroscientists and psychologists, both here at the University and elsewhere.”
Furber pointed out that psychologists already have neural networks on which they can reproduce the clinical pathologies, and they use these to test alternative therapies.
“At present they are limited in the fidelity they can achieve with these networks by the available computer power, but we hope that SpiNNaker will raise that bar a lot higher,” he said.
No one is claiming to understand the brain.
“We hope that our machine will enable significant progress towards understanding how the brain works as an information-processing system,” said Furber.
ARM was approached in May 2005 to participate in SpiNNaker and agreed give the project processor intellectual property along with a cell library to aid design and manufacturing. The agreement extends to Manchester making enough chips for a computer with a million cores.