Electronic components that simultaneously store and compute data like biological neurons could pave the way for entirely novel computers able to learn and adapt.
The team at Exeter University used ‘phase-change alloys’ that move from an amorphous to fully crystallised state when subject to a current or light pulse.
‘What we are doing is trying to build electronic systems that mimic, in a simple way, the functionality of the basic building blocks of mammalian brains — namely neurons and synapses,’ project lead Prof David Wright of Exeter told The Engineer.
In conventional computers memory and processing units are physically separate, and data has to be continually shunted between the two, creating ‘bottlenecks’.
‘This slows everything down and wastes a lot of power and is the main reason chip manufacturers have moved to multi-core processors,’ Wright said.
The team turned to neurons for inspiration, noting that they make no real distinction between memory and computation. Looking for possible artificial substitutes the researchers came across so-called ‘phase-change materials’ that flip between amorphous and crystal states, in doing so inducing an electrical conductivity difference of up to five orders of magnitude and a large refractive index change.
Using laser pulses to induce the phase changes in germanium-antimony-tellurium (GeSbTe) and silver-indium-antimony-tellurium (AgInSbTe), the team was able to perform basic arithmetic and data storage.
‘A very simple model of a neuron is known as the “integrate and fire” model in which the neuron integrates, or accumulates, excitations applied to its input and fires a pulse along its output after a certain threshold has been passed.
‘We’ve shown that phase-change materials have a natural accumulation and threshold property, which makes them a good candidate for simple implementation of a hardware neuron,’ Wright said.
The upshot of the work is that these phase-change components could potentially be connected in networks via structures akin to synapses — potentially opening up an entirely novel way of computing.
‘The strength of these synaptic connections is altered by learning and experience — a common phrase to describe this is “neurons that fire together, wire together”.
‘We think that phase-change devices could also be used to make synapses — since the electrical resistance, or optical reflectivity, of phase-change materials depends on their excitation history,’ Wright said.
Indeed, shortly after publication of the Exeter work, a team from Stanford University in the US headed by Prof Philip Wong demonstrated just that, creating a nanoscale device with interconnected phase-change components.