Issue link: https://htpgraphics.uberflip.com/i/433487
SpiNNaker 62 RESEARCH PROJECT The brain leads... can computers follow? SpiNNaker is a research project that demonstrates the continuous renewal and leading-edge development in computer science at the University. Led by Professor Steve Furber, SpiNNaker started in conception in 1998 and in implementation, in 2005. Its ultimate goal is to model and thereby understand the biological functions of the human brain. SpiNNaker and the Human Brain project are looking for the 'intelligence antidote' to the paradox that the amount of computing power which can be delivered in small mobile packages has increased exponentially while the fundamentals of computing have not changed. Traditional super-computers can reliably repeat specific instructions at speed but do this in a deterministic manner. Biological systems, on the other hand, use massive numbers of neurons which use weighting to make judgements and to control how many neurons "fire" in response to a message. Looking to emulate such 'decision making', Furber's research has focused on the use of neural networks. These are based on a combination of hardware and software developments to produce the massive connectivity which can model biological systems. Through this research project, SpiNNaker chips are now available. The principal innovation lies in the communications architecture which enables the software implementation of the 'neurons', when they fire, to communicate with everywhere they need to go – which may be 100,000 targets out of 1,000,000,000 in under 0.001 seconds – the equivalent to biological real-time. The essence of the SpiNNaker chip is the simplicity with which this is achieved. Equivalent real-time capability cannot be achieved with a standard high- performance computer because while such machines have impressive communications bandwidth, they cannot cope with the very large number of independent small packets of information required for real-time brain modelling. The latest SpiNNaker implementation achieves more real- time capability than is available anywhere in the world and this has been demonstrated at the leading international workshops on cognitive neuromorphic engineering. At demonstrations Furber notes that "… over the past three years we've noticed that pretty much all that actually work are SpiNNaker-based". "Computers are just becoming faster and faster idiots". Steve Furber, 2013