Golunoid Project / Проект Голуноид

2022-05-25 09:16:44 Science
SFU has taken a step to build neural networks in the form of a chip

Research by Vadim Avilov, Associate Professor at the Institute of Nanotechnology, Electronics and Instrumentation, Southern Federal University, is aimed at creating and developing in Russia new technologies for the design and production of a promising element-component base for integrated nanoelectronics and artificial intelligence.

Neural networks are currently experiencing another wave of scientific interest. Many services can already be found on the internet that employ neural network computing for various tasks such as photo animation, image and speech generation, pattern recognition, and much more. However, the most popular area for neural networks is robotics.

Such tasks as moving in space in the presence of external influences, dynamic route construction, interaction with a human - cannot be implemented as conventional algorithms, while neural network algorithms, in which information processing is similar to the work of neurons in the brain, can perfectly cope with such solutions.

The main problem, however, is that all these neural network calculations are implemented as programs for standard computing devices that are not optimised for this class of computation. The solution is to fabricate a neural network in the form of a chip, where all the computations are done via artificial synapses. The use of such neural processors could lead to significant breakthroughs in many fields such as robotics, bionic prosthetics, autonomous control and more.

"My current project "Development of structural and technological solutions for crossbar formation of titanium oxide nanostructures for neuromorphic processor elements of bionic, robotic systems and artificial intelligence" focuses on the instrumental implementation of a neural network based on the memristor effect, i.e. the ability of some materials to significantly change their resistance," said Vadim Avilov, PhD in Engineering, Associate Professor of INEP SFU.

During his research, the scientist plans to achieve the implementation of neural network algorithms in the form of a chip based on titanium oxide memristors. These structures are "smart" materials and can change their resistance in a wide range under the influence of an electric field. It is this property that allows the function of artificial neural network synapses to be fully realised. Therefore, the primary goal of the project is to investigate the resistance switching patterns of memristors to further predict the modes of operation of artificial synapses in a neural network.

"Our research team has already done a lot of work and the research in my project is a continuation. We have studied the influence of technological synthesis parameters on the nanostructures being formed, developed a physico-chemical model that allows us to calculate the specifics of nanostructure synthesis leading to membrane switching in them.

A number of works were carried out on the fabrication and investigation of a mock-up of ReRAM resistive memory based on such memristor structures and the possibility of fabrication of multilevel memory was shown. It was the demonstrated multilevel switching of memristors that led to a shift of scientific research into the field of artificial synapses and neural network," shared Vadim Avilov.

According to the scientist, the development of structural and technological solutions for the creation of synaptic structures will stimulate the development of new industrial technologies in the field of neuromorphic processor manufacturing. The results of the project will form the basis for the production of neural processors - individual chips that implement the neural network algorithm for information processing for tasks in robotics, bionic applications and artificial intelligence. In contrast to software solutions for neural network computing, such processors will be optimised.

You can discuss this news at VK or Telegram

Новости на другие темы

Current section news:

Science - 2022-08-16

Science - 2022-05-13

Science - 2022-04-21

Other news:

Technology - 2015-07-20

Space - 2021-12-30

Mars - 2022-02-22

Back to top

Last news

© 2011-2022 Golunoid
Design & Development: 2004-2022 Comrasoft