A computer neural network on the basis of which is built the artificial intelligence, is designed at its core as well as their anatomical ancestors. In order for the AI learned something new, we need to strengthen old and create new relationships between elements of the neural network. At the current level of technology development to build the capacity of neural networks becomes more difficult. But help can come from new memristor developed by the American Institute of Physics (AIP).
Memristor is an element capable of changing the resistance depending on the pass-through charge, so that it can act as a data warehouse, in a very simplified form reminiscent of the work of neurons and synapses of the brain. And the name of the element comes from the merger of two words-memory and resistor.
According to the newspaper on EurekAlert, a group of researchers from the API has developed a new type of “electronic synapse”, which consists of a memristor based on boron nitride with a thickness of only 1 atom. According to the author Ivan Sanchez Esqueda,
“Now, there is great interest in the use of new types of materials for memristor. We have shown that our device can work well in the field of neuromorphic computing applications.”
In fact, the decision to move the memristor in subnanometric level was dictated by the problem of energy saving. The fact that the array of microscopic memristor was 10,000 times more energy efficient than any existing analogues.
“It turns out that if you start to increase the number of parallel operating devices — you will receive a significant increase in the accuracy of the calculations, while retaining the same level of consumption.”
Now a team of scientists wants to use a new kind of “electronic synapses” to perform various tasks, including pattern recognition and images. Also in the future not excluded their application in the field of deep machine learning.
This and other news you can discuss in our chat in Telegram.