Million-processor computer to model the human brain
Up to a million ARM processors are to be linked together to create a massive computer, called SpiNNaker (Spiking Neural Network architecture) that can simulate the workings of the brain.
SpiNNaker, under development by a group of British universities, could help neuroscientists, psychologists and doctors understand complex brain injuries and diseases, and identify the most effective therapies.
Despite the enormous number of processors, though, it will only be able to model less than one percent of the human brain, which contains 100 billion neurons with 1,000 million connections. In SpiNNaker, each impulse is modelled as a packet of data, which is then sent to all connected neurons. Neurons are represented by small simple equations which are solved in real-time by software running on the ARM processors.
The electronic connections in SpiNNaker convey these spikes much more quickly than the biological connections in the brain, meaning that SpiNNaker can transmit spikes as effectively and quickly with many fewer connections.
The bespoke microchips are integrated in a single 19mm square package with a second microchip that provides substantial memory using 3D System-in-Package technology from Unisem Europe. This package delivers the computing power of a PC in a tiny space and for around one watt of electrical power.
"This could ultimately be of great help for patients, for example, who have presented with reading problems caused by strokes or similar brain injuries. Psychologists have already developed neural networks on which they can reproduce the clinical pathologies," says Manchester University's professor Steve Thurber.
"They then use these networks to test alternative therapies, to identify which is most effective in treating the patient's symptoms. 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."