The researchers are studying phase-change materials - such as GeSbTe or AgInSbTe alloys - that have properties allowing them to process and store.
“Currently the storage is in one place and the processing is in another and that means sending the information back and forth,” David Wright, lead scientist on the project said “If you could make calculations and store the results in the same place using the same device, it would speed things up a lot.”
The scientists used lasers to change the state of the materials in order to perform calculations and said the development could have far-reaching implications.
“If you can do logic and arithmetic, you could make a microprocessor using these phase-change cells rather than a standard silicon approach,” Wright said.
Phase-change materials hold potential because of the different characteristics in each state, and are already widely used in computing, but not for processing.
“If you use rewritable Blu-ray disks, the active material in there is a phase-change material that switches between amorphous and crystal states using a laser,” said Wright.
Brain power
While simple processors might be a starting point, Wright believes the new approach could lead to dramatic improvements in human-like thinking in computers, because multitasking cells are more reflective of the way brains work, with all neurons able to both process and store information.
“Phase-change cells have behaviour similar to a neuron and it can also act as a synapse,” Wright said.
“Because of the way the phase-change cells perform, potentially you could build an artificial brain with the phase-change cells acting as both neuron and synapses all linked together in a three-dimensional arrangement,” Wright said.
“We're planning on making a small demonstrator with 10-100 of these cells connected together to work on pattern recognition or maze solving.”
While working protoypes of the technology remains five years away, the potential is highlighted by the fact that a relatively simple device based on the technology could replicate natural thought simulations that currently require supercomputers.
“If you want to simulate brain-type computing, then currently it takes supercomputers,” said Wright. “IBM had a nice simulation of a cat brain where it used 150,000 CPUs and 150TB of memory and that's because they were mimicking the actions of neurons and synapses in software,” said Wright.
“If you can do that in hardware you can do it more quickly and more energy efficiently.”
This article originally appeared at pcpro.co.uk
