To create a robotic brain, all that remains is to assemble memristors with neuristors
The first artificial neuron is being created by young electronics scientists from MIPT. The device is necessary for the development of a computer that simulates the human brain.
Model of a neuron. Illustration by MIPT
Currently, neural network programs run on ordinary computers that require a lot of energy. To run the same programs on more compact devices, for example, in UAVs, you need to come up with another type of computer — minimally energy-consuming, reminiscent of the human brain.
Scientists have already taken the first step in this direction – they have learned to make memristors – electronic elements that play the role of synapses (transmitters of impulses between neurons). However, to create a full-fledged artificial brain – the physical embodiment of a neural network, they lack the electronic neurons themselves.
Young scientists from the Laboratory of Functional Materials and Devices for Nanoelectronics have undertaken to develop an electronic analogue of a nerve cell – a neurostor.
The way a natural neural network works is as follows: a neuron transmits energy to another neuron through an axon. Each nerve cell receives many such impulses from neighboring neurons (through incoming channels — dendrites), which causes the charge inside it to increase for some time. As soon as it exceeds a certain threshold value, the firing mechanism is triggered, and the impulse is transmitted further.
To create a neuron simulator, scientists need three elements: a capacitance — where the charge accumulates, a threshold trigger mechanism at a certain voltage, and thirdly , a charge leakage mechanism if this threshold is not passed.
To trigger such a mechanism, scientists used conventional capacitors available today, and “old Soviet gas-discharge lamps” were used as a threshold switch.
As soon as the electronic analogue of a nerve cell is successfully tested, researchers will connect it with existing memristors and obtain a chip capable of running neural network programs.
Experts say that such a mechanism will make it possible to create neuromorphic computers that will make all kinds of robots much more perfect.