Engineering of memristor-based artificial neural networks with preset fault tolerance

  • Сергей Николаевич Данилин Murom Institute (branch) Federal state budgetary Educational Institution of Higher Education "Vladimir State University named after Alexader Grigoryevich and Nickolay Grigoryevich Stoletovs"
  • Сергей Андреевич Щаников Murom Institute (branch) Federal state budgetary Educational Institution of Higher Education "Vladimir State University named after Alexader Grigoryevich and Nickolay Grigoryevich Stoletovs"
  • Антон Дмитриевич Зуев Murom Institute (branch) Federal state budgetary Educational Institution of Higher Education "Vladimir State University named after Alexader Grigoryevich and Nickolay Grigoryevich Stoletovs"
  • Илья Алексеевич Борданов Murom Institute (branch) Federal state budgetary Educational Institution of Higher Education "Vladimir State University named after Alexader Grigoryevich and Nickolay Grigoryevich Stoletovs"
  • Александр Евгеньевич Сакулин Murom Institute (branch) Federal state budgetary Educational Institution of Higher Education "Vladimir State University named after Alexader Grigoryevich and Nickolay Grigoryevich Stoletovs"

Abstract

Publications of leading foreign and Russian experts and researchers manifest that the most promising elements of the technical aids for the implementation of next-generation supercomputers that are being created lately are nanomemristors. The main reason for this phenomenon is that the new hardware components enable to make ultra-high-performance computing aids with technical, power and economic features that exceed the existing ones by a huge ratio. Vital and new stage of work is mastering new algorithms and approaches to their system engineering according to the leading experts in this field. The paper defines and applies general approach to engineering nanomemristor-based artificial neural network-based (NMANN) with preset fault tolerance, (FT). It is shown that problem solution of ensuring the preset FT for NMANN is related to problem solution of ensuring the preset accuracy of their functioning at all levels of the structural and functional hierarchy.


The given practice-oriented examples testify contrary to the statement of some researchers and potentially high FT for NMANN is not provided automatically, and depends on many factors and requires the implementation of special technologies at the stages of creating a numerical model, engineering, manufacturing and operation of NMANN. A modified version of the quantitative criterion of FT is proposed, on the basis of which it is possible to construct NMANN reliability diagrams, to calculate and optimize their reliability according to the current interstate and Russian standards. To raise FT for NMANN it is necessary either to increase the gap between admissible and achieved indicator values in operation quality when learning, or to reduce the indicator variation in NMANN operation quality in case of failure of each structural element (its physical and (or) information component) by reducing the indicator value of corresponding influence coefficient of destabilizing factor on functional parameters. The quantitative criterion of fault tolerance provides additional information about NMANN properties and it can be recommended for use in both theoretical research and engineering practice. The research work is carried out with assistance of RFBR grant 19-07-01215.

Author Biographies

Сергей Николаевич Данилин, Murom Institute (branch) Federal state budgetary Educational Institution of Higher Education "Vladimir State University named after Alexader Grigoryevich and Nickolay Grigoryevich Stoletovs"

Candidate of  Technical Sciences, Associate Professor,  Department of Computer-aided Designing System of Electronic Devices, Murom Institute (Branch) “Vladimir State University named after A.G. and N.G. Stoletov”.

Сергей Андреевич Щаников, Murom Institute (branch) Federal state budgetary Educational Institution of Higher Education "Vladimir State University named after Alexader Grigoryevich and Nickolay Grigoryevich Stoletovs"

Candidate of Technical Sciences, Associate Professor, Dean of Information Technologies Department, Murom Institute (Branch) “Vladimir State University named after A.G. and N.G. Stoletov”

Антон Дмитриевич Зуев, Murom Institute (branch) Federal state budgetary Educational Institution of Higher Education "Vladimir State University named after Alexader Grigoryevich and Nickolay Grigoryevich Stoletovs"

Engineer, Department of Computer-aided Designing System of Electronic Devices, Murom Institute (Branch) “Vladimir State University named after A.G. and N.G. Stoletov”

Илья Алексеевич Борданов, Murom Institute (branch) Federal state budgetary Educational Institution of Higher Education "Vladimir State University named after Alexader Grigoryevich and Nickolay Grigoryevich Stoletovs"

Fourth-year Undergraduate Student in major “Applied Information Science”, Murom Institute (Branch) “Vladimir State University named after A.G. and N.G. Stoletov

Александр Евгеньевич Сакулин, Murom Institute (branch) Federal state budgetary Educational Institution of Higher Education "Vladimir State University named after Alexader Grigoryevich and Nickolay Grigoryevich Stoletovs"

Publications of leading foreign and Russian experts and researchers manifest that the most promising elements of the technical aids for the implementation of next-generation supercomputers that are being created lately are nanomemristors. The main reason for this phenomenon is that the new hardware components enable to make ultra-high-performance computing aids with technical, power and economic features that exceed the existing ones by a huge ratio. Vital and new stage of work is mastering new algorithms and approaches to their system engineering according to the leading experts in this field. The paper defines and applies general approach to engineering nanomemristor-based artificial neural network-based (NMANN) with preset fault tolerance, (FT). It is shown that problem solution of ensuring the preset FT for NMANN is related to problem solution of ensuring the preset accuracy of their functioning at all levels of the structural and functional hierarchy.

The given practice-oriented examples testify contrary to the statement of some researchers and potentially high FT for NMANN is not provided automatically, and depends on many factors and requires the implementation of special technologies at the stages of creating a numerical model, engineering, manufacturing and operation of NMANN. A modified version of the quantitative criterion of FT is proposed, on the basis of which it is possible to construct NMANN reliability diagrams, to calculate and optimize their reliability according to the current interstate and Russian standards. To raise FT for NMANN it is necessary either to increase the gap between admissible and achieved indicator values in operation quality when learning, or to reduce the indicator variation in NMANN operation quality in case of failure of each structural element (its physical and (or) information component) by reducing the indicator value of corresponding influence coefficient of destabilizing factor on functional parameters. The quantitative criterion of fault tolerance provides additional information about NMANN properties and it can be recommended for use in both theoretical research and engineering practice. The research work is carried out with assistance of RFBR grant 19-07-01215.

Published
2019-12-31
How to Cite
ДАНИЛИН, Сергей Николаевич et al. Engineering of memristor-based artificial neural networks with preset fault tolerance. Radioengineering and telecommunication systems, [S.l.], n. 2, p. 41-49, dec. 2019. ISSN 2221-2574. Available at: <https://rts-md.mivlgu.ru/jornalRTS/article/view/160>. Date accessed: 18 nov. 2025.
Citation Formats
Issue
No 2 (2019): Radio and telecommunication systems
Section
Network and telecommunication