SpiNNaker neuromorphic supercomputer reaches one million cores
The SpiNNaker machine can model more biological neurons in real-time than any other machine on the planet, according to the University of Manchester, where the supercomputer has been developed.
With each processor chip within the machine containing 100 million transistors, the SpiNNaker system is capable of emulating regions of the human brain and the developers have the ultimate goal of being able to emulate one billion neurons or about 1 percent of the human brain,
SpiNNaker, which is based on spiking neural network modelling, is the brain child of Professor Steve Furber, one of two original designers of the Advanced RISC Machine or ARM processor. The project dates back to 2005 in construction and the present instantiation is composed of 57,600 processors each with 18 ARM 32bit cores and 128Mbytes of SDRAM totalling 1,036,800 cores and more than 7 terabytes of RAM.
So far the project has cost £15 million (about $20 million) and has been 20 years in conception and 12 years in construction. The project is now supported by the European Human Brain Project, a multinational European Union co-funded institution.
One of the fundamental uses of Spinnaker is to help neuroscientists understand how the human brain works by running large-scale simulations. Examples include an 80,000 neuron model of part of the cortex, the outer layer of the brain and the basal ganglia, the area affected by Parkinson’s disease.
“Neuroscientists can now use SpiNNaker to help unlock some of the secrets of how the human brain works by running unprecedentedly large scale simulations. It also works as real-time neural simulator that allows roboticists to design large scale neural networks into mobile robots so they can walk, talk and move with flexibility and low power,” said Professor Furber, in a statement.
Related links and articles:
News articles:
Eta adds spiking neural network support to MCU
BrainChip launches spiking neural network SoC
Neuromorphic computing runs on asynchronous Cortex-M3
Self-timed logic is Eta Compute’s low-power secret
