Scientists create brain-like computing component

Researchers at Exeter University have used phase-change materials to demonstrate a component that can store and compute data in the same place, much like the brain's neurons
Written by Jack Clark, Contributor

Scientists from Exeter University have successfully mimicked the brain's ability to remember and compute data in the same place, demonstrating a technique that could eventually help make processors more power efficient.

Exeter brain experiment

Scientists from Exeter University have successfully mimicked the brain's ability to remember and compute data in the same place. Photo credit: Tim Pestridge

The technique, described in a paper published in Advanced Materials, relies upon phase-change materials to create a component more akin to a neuron, which both stores and computes data, than a common transistor, which can only do one of these at a time.

"Our findings have major implications for the development of entirely new forms of computing, including 'brain-like' computers. We have uncovered a technique for potentially developing new forms of 'brain-like' computer systems that could learn, adapt and change over time," Professor David Wright, who led the research, said in the announcement on Thursday.

The researchers said they hope computers using phase-change materials will eventually be significantly more power efficient than transistor-based ones, as less power would be expended on shuttling data to the processor and vice versa.

"Phase-change materials... offer a promising route to the practical realisation of new forms of general-purpose and biologically-inspired computing," the researchers wrote in the paper. "Here we provide, for the first time, an experimental proof-of-principle of such a phase-change material-based 'processor'."

A phase-change material is one that can shift from one state to another. In the Exeter experiment, researchers used alloys that move from an amorphous state into a fully crystallised state when subjected to forces. As the materials shift state, their ability to reflect light changes. The changes in reflectivity indicate the output of a calculation.

The scientists pounded the phase-change alloys — germanium-antimony-tellurium and silver-indium-antimony-tellurium — with pulses of laser light (though they could have also used an electrical current) to set a threshold level at which their state changed. This level also determined the number base in which the researchers calculated the experiments. They then used a combination of the threshold and an accumulative system to perform calculations.

Our findings have major implications for the development of entirely new forms of computing, including 'brain-like' computers.
– Professor David Wright

"[Data] is stored in the fractional state," Exeter team leader Wright told ZDNet UK. "It's stored by the precise state of the partial crystallisation of the material."

In the paper, the researchers demonstrated addition, subtraction and division. Because the phase-change material shifts into a new state according to the input energy levels and pulses, researchers can make additions by bombarding it with pulses and seeing how its reflectivity changes to read off the value of their calculation.

Similarly, for division, the threshold is set as the divisor rather than the base. For example, to calculate 14 divided by 10, the researchers set 10 (the divisor) as the threshold. They then hit the material with 14 laser pulses. Once the material had been hit 10 times, it reset to zero, then was hit by four more pulses. The material's reflectivity equated to a value of four, with the reset counted once by the laser, giving the result of the calculation: 1.4.

Next stop, neural networks

Following the success of this experiment, the scientists plan to create a basic neural network with between 10 and 100 interconnected cells mimicking neurons and synapses, for tasks such as image recognition and processing.

"The accumulation property we used to do arithmetic can also be used to provide a basic 'neuron' kind of hardware mimic," Wright said. "All arithmetic is a simple accumulative process, like counting on fingers. A neuron also works in a simplistic way as an accumulator, so you can build a neuron out of a simple phase-change cell.

"If you want to mimic a neuron in silicon, you need quite a few silicon gates and it's a complex circuit," he added.

Additionally, the researchers feel phase-change materials could become "replacement flash memory for USB cards and such", Wright said.

Chip specialist Intel is known to have research schemes around phase-change memory. In 2008 it said it had, in conjunction with STMicroelectronics, been able to store two bits of data per cell by keeping the material in one of four possible states.

Memristors, which also remember their previous state, are another form of technology in development that bears similarities to neurons. In May, HP's labs claimed a breakthrough when they said they now understood the chemical and physical properties of the technology.

Get the latest technology news and analysis, blogs and reviews delivered directly to your inbox with ZDNet UK's newsletters.
Editorial standards