A device that “acts as a synapse in the living brain,” capable of storing and forgetting information when it is no longer used, was built by a group of researchers from the Moscow Institute of Physics and Technology (MIPT).
Neuroinformatics has made great progress in recent years, even if it imitates the complexity of a brain, even if it is only to store information on a substrate, there is still something unrealizable at this time. However, the new study, published in ACS Applied Materials & Interfaces, shows that it is possible to use new devices, called memristors, primarily memory cells that also act as resistors, to design analog neurocomputers that at least mimic the ways in which the biological brain performs learning.
One of the many problems facing neuroinformatics scientists is the so-called synaptic plasticity: the synapses themselves do not always remain the same, they tend to strengthen and change over time. Previous attempts to mimic this behavior in artificial devices have also subjected the latter to rupture after prolonged use.
Synaptic plasticity is precisely the characteristic that Russian researchers, led by Anastasia Chouprik and Vitalii Mikheev of MIPT’s Neurocomputing Systems Lab, tried to inoculate on their device, as the same researcher explains: “The mechanism that we used to implement plasticity synaptic is more robust. After the system’s status had changed 100 billion times, it still worked normally, so my colleagues stopped the resistance test.”
The researchers used hafnium oxide, a ferroelectric material that retains the obtained polarization after removal of the electric field.