Artificial synapses

The path to the creation of "intelligent" biological computers

DailyTechInfo based on Stanford News: Stanford researchers create a high-performance, low-energy artificial synapse for neural network computing

The human brain is the most efficient and very powerful computer of natural origin. And it is not surprising that many researchers are engaged in the development of computers, the principles of which are based on the principles of the brain. Neural networks, artificial intelligence systems capable of self-learning, are the closest models to the brain that we have today. And scientists from Stanford University, also working in this direction, went a slightly different way, they created an organic artificial synapse, the appearance of which makes us one step closer to the appearance of "intelligent" biological computers.

In the bowels of the brain, neurons exchange electrical signals passing through synapses. The ion channel formed at the junction of synapses from two neurons becomes thicker and more electrically conductive every time a signal passes through it. This strengthening of the electrical connection allows you to spend less energy on transmitting information and on the basis of this effect, a human memory system works, which allows him to learn and accumulate experience.

In most cases, artificial neural networks simulate the above processes programmatically. And as the nodes of neural networks increase, the volume of operational data increases, which grows even more as experience accumulates. This approach to building neural networks has already demonstrated its effectiveness on numerous examples, defeating Li Sedol, the world champion in the Chinese game of Go, creating musical compositions, paintings and much more. Despite this, computer systems running artificial neural networks are very far from the efficiency of a living brain in terms of the amount of energy used for this.

Stanford researchers, instead of modeling a neural network, decided to make a real neural network. And the first step towards this was the creation of an artificial synapse, an organic neuromorphic device that can simultaneously process and store information. The structure of the device created by them resembles the structure of a transistor, an artificial synapse has three electrodes, the conductivity between which is provided by a salt solution with a certain concentration. Electrical signals pass from one electrode to another, and the signal at the third electrode controls it all.

A drawing from the article by Yoeri van de Burgt et al. A non-volatile organic electrochemical device as a low-voltage artificial synapse for neuromorphic computing, published in the journal Nature Materials – VM.

First, the researchers studied the work of the synapse by passing various electrical signals through it, which allowed them to find out the value of the voltages that cause the synapse to switch to a certain electrical state. Unlike a transistor, which can be in two states, on and off, an artificial synapse can be in one of 500 discrete states, which increases its computing power exponentially.

However, switching an artificial synapse from one state to another still requires energy 10 thousand times more than it takes to switch the state of an ordinary living synapse. Nevertheless, this achievement in itself is a big step in the "right" direction. In their further research, Stanford scientists are already planning to use smaller devices, which should increase their efficiency.

Using a single artificial synapse, the scientists conducted an extensive series of experiments and extrapolated all the data they collected for their use in a model of a more complex system consisting of a certain number of synapses. The created model of a fairly simple neural network coped with the task of recognizing handwritten images of numbers from 0 to 9, giving the correct answer in 97 percent of cases. And in the near future, scientists plan to build a real artificial neural network, which is the embodiment of the simulated one, in order to conduct comparative studies.

Portal "Eternal youth" http://vechnayamolodost.ru  22.03.2017