Physicists from the Moscow Institute of Physics and Technology have constructed a second-order memristor and thus have come one step closer to creating an analog neurocomputer. A legendary, albeit still hypothetical device that copies the work of the human brain using other mechanisms. Earlier, Russian physicists proved the inconsistency of first-order memristors, but high hopes are pinned on the second type.
A memristor is a device for simulating the work of a synapse, which stores part of the information in the form of communication with other synapses. But if you do not use it, do not pass the signal, the connection weakens to the point of complete disappearance. In addition, the work of the synapse can be destroyed under the influence of other factors, which opens the way to modeling and studying such a phenomenon as forgetting memories. In the future neurocomputer, memristors perform the function of basic "building blocks" from which its architecture is built.
Synapse and memristor
First-order memristors were built on the basis of nano-sized conducting bridges, which turned out to be short-lived - they spontaneously collapsed during operation. For the second generation, they took ferroelectric hafnium oxide, which, under the influence of electrical impulses, goes into states of low or high resistance. The most difficult thing was to choose the correct thickness of the plate - experimentally, scientists have established that you need 4 nanometers and not a single nanometer more.
The second-order memristor successfully passed 100 billion endurance tests, and did not continue testing further. But in addition to durability, it has another interesting property - due to imperfect technologies, defects remain at the interface between hafnium oxide and silicon substrate. While the impulses pass frequently, they break through this barrier, but when inactive, the conductivity begins to weaken, which allows simulating the property of the synapse to "forget" information through the destruction of the connection.
So far, there is little practical use of second-order memristors. For example, scientists argue about what the mechanism for switching memristors should be. Then you need to learn how to create memory cells based on them, with which it will be convenient to work. And, finally, to develop the architecture of a neurocomputer and work out the issues of its integration with existing electronics.