A research team from Stanford University, Groningen University and others have devised an 'artificial synaps', which more closely mimics the human synapses than current computer chips, being able to both store and process information simultaneously, and appears to be more energy efficient than it's biological counterpart.
The breakthrough is expected to make neural networks more effective, and energy efficient.
Neural networks are an essential part of artificial intelligence, and are used for pattern recognition, for example in assessing traffic situation by autonomous cars, and speech recognition.
The problem is, neural networks rely on current computer technology. Even though modern computers excel in processing numbers, they are not as good at pattern recognition, which is where the human brain is still superior.
The best pattern recognition currrently is achieved simulating neural networks with powerful graphical processors. Even though those are getting pretty good at approaching human-like levels of achievement, it comes at a huge energy cost for storing, and processing data.
For this reason, computer scientist have been working on computer chips that are more similar to the human brain.
The design created by the combined research team, which they published yesterday in Nature Materials, is derived from an organic battery.
Their artificial synaps consists of two thin, flexible layers of polymere, with an electrolyte with 3 contact points in between.
By slightly loading, or unloading the electrolyte, the resistance in the lower polymere layer rises or falls.
When the process of loading / unloading is stopped, the layer retains it's resistance, creating a solid memory state.
"The difference with a regular transistor is that a memory value in this synaps isn't limited to one or zero, but can take any value", says lead researched Yoeri van de Burght, who has lead the research in Stanford, and is now teaching at the University of Eindhoven.
With the protoype synaps, the researchers ran a simulation in a virtual neural network. They conclude that what they observed has to be reproducable in practice.
In a follow up research, van den Burght wants to link several dozens of their synaps, to form a small neural network that is capable of simple pattern recognition.
Professor in nano-technology at Twente University, Wilfred van der Wiel, says "it is fascinating, especially since the researchers claim that their system is more energy efficient than it's biological counterpart".
He does add that he thinks it would have been even more convincing if the researchers had waited with publishing until they had tested a neural net in pratice, instead of only a simulation. "But I understand that if they were under pressure (competition) to publish results fast, they chose to do so now"
http://www.volkskrant.nl/wetenschap/computer-die-functioneert-als-menselijk-brein-mogelijk-stapje-dichterbij~a4464938/http://www.nature.com/nmat/journal/vaop/ncurrent/full/nmat4870.html?WT.feed_name=subjects_physics