An article published in the journal “Nature Nanotechnology” describes the creation by IBM of artificial neurons using phase change materials to record and process data. The company created a chip that contains the devices that constitute the “heart” of these neurons on a chip and there may be even millions of them. The aim is to mimic biological neurons to achieve progress in the development of neuromorphic techniques to be applied for example to cognitive computing.
There are several projects to develop chips that mimic the biological functions of neurons and IBM has been developing various of them too. Almost exactly two years ago, the company unveiled the project SyNAPSE chip that emulates the human brain but in this case it went beyond it. In fact the IBM Zurich scientists designed artificial neurons which consist of phase change materials, including germanium antimony telluride, an alloy also known as GeSbTe or GST.
The characteristic of this type of materials is to show two stable states: an amorphous one, without a clearly defined structure, and a crystalline one, which instead is structured. These are materials of the type used for rewritable Blu-Ray. The difference is that artificial neurons don’t record digital information as they’re analog, just like the synapses and neurons in biological brains.
In the published demonstration, electrical pulses were applied to the artificial neurons with the result of obtaining the progressive crystallization of the phase change material and eventually the activation of the neurons. Those are the foundations of event-based processing, which has mechanisms similar to those of biological neurons with stimuli reactions.
These artificial neurons have a stochastic nature, meaning that there’s a degree of randomness in the results as after an event they go back to an amorphous state which is always slightly different from the previous one. Again there’s a similarity with biological neurons, whose state is influenced by environmental variables such as temperature.
This type of chip could be ideal to perform computing related to issues related to activities in which the stochastic element has a strong influence such as various human activities, the weather or cognitive computing. The tests showed that IBM chip can process large amounts of data using very limited amounts of energy.
As always in these cases, we’re talking about prototypes and the road to practical use can be very long. There’s also the problem of writing software that takes advantage of the features of IBM’s artificial neurons. The tests performed by IBM are promising, we must wait to see if the promises will be kept.
