Scientists from the Kurchatov Institute, the Moscow Institute for Physics and Technology, University of Parma (Italy), the MSU and St. Petersburg State University have created an artificial neural network on the basis of polymeric memristors.
"Such developments can find application in engineering the systems of automated vision, machine hearing and other sense organs, as well as intelligent control systems of various devices, including stand-alone robots", the press release of the MIPT reads.
The memristor is an electric element which is analog of the ordinary resistor. Its difference from the classical element is that the electric resistance of a memristor depends on the charge that passed through it and so it constantly changes its properties under the influence of an external signal. Thanks to it memristors are analogs of synapses, connections of two neurons in the brain that can change the efficiency of signal transmission between neurons under the influence of this transmission itself. Therefore the memristor makes it possible to put a “real” neural network into practice, and physical properties of memristors enables experts to make them as tiny as usual microchips.
The authors of the new research have made memristors of polyaniline polymer and for the first time connected them into a network and carried out experiments on teaching it. The training of a neural network consists in supplying electric impulses in random order. If the network gives a wrong answer in response to it, a special correcting impulse is sent and thus after a certain number of repetitions all the internal parameters of the device (namely the resistance of memristors) are adjusted – i.e. trained – in a required way. The scientists have demonstrated that their memristor network is capable to carry out basic logical operations after just fifteen attempts.
So far the devices created by researchers are too large in size and react to downstream signals for too long so it is too early to speak about their practical application. However some estimates show that the size of a memristor can be reduced to ten nanometers, and the technologies used in production of experimental prototypes allow scaling it to the level of mass production, the press release points out.
Results of the work have been published in the Organic Electronics magazine.