Multi-level memristive memory for neural networks

Aidana Irmanova, Serikbolsyn Myrzakhmet, Alex James Pappachen

Research output: Chapter in Book/Report/Conference proceedingChapter

Abstract

Analog memory is of great importance in neuromorphic engineering as it enables scalable neural network design and energy efficient implementation of computationally expensive operations. With the advent of memristors, the realization of the analog memory became possible due to the intrinsic properties of memristors such as nanoscale size, non-volatility, and energy efficiency. In hardware implementations of neural networks, memristors store the values of synaptic weights and operate similarly to the synapses that are reinforced with the application of external stimuli. Memristors that are ideally continuum memories, currently are at the early stage of the development, which causes several issues in neuromorphic circuit design. Device level and architecture level issues force memory engineers to approach memristive memory design in different ways. In this chapter device-level problems: restricted number of resistance states, stochastic switching and architecture level problem: sneak paths will be discussed, and their state of the art solutions will be presented.

Original languageEnglish
Title of host publicationModeling and Optimization in Science and Technologies
PublisherSpringer Verlag
Pages103-116
Number of pages14
DOIs
Publication statusPublished - Jan 1 2020

Publication series

NameModeling and Optimization in Science and Technologies
Volume14
ISSN (Print)2196-7326
ISSN (Electronic)2196-7334

Fingerprint

Memristors
Neural Networks
Neural networks
Data storage equipment
Equipment and Supplies
Analogue
Synapses
Circuit Design
Synapse
Hardware Implementation
Network Design
Efficient Implementation
Energy Efficiency
Energy Efficient
Energy efficiency
Efficiency
Weights and Measures
Continuum
Engineering
Hardware

ASJC Scopus subject areas

  • Modelling and Simulation
  • Medical Assisting and Transcription
  • Applied Mathematics

Cite this

Irmanova, A., Myrzakhmet, S., & James Pappachen, A. (2020). Multi-level memristive memory for neural networks. In Modeling and Optimization in Science and Technologies (pp. 103-116). (Modeling and Optimization in Science and Technologies; Vol. 14). Springer Verlag. https://doi.org/10.1007/978-3-030-14524-8_8

Multi-level memristive memory for neural networks. / Irmanova, Aidana; Myrzakhmet, Serikbolsyn; James Pappachen, Alex.

Modeling and Optimization in Science and Technologies. Springer Verlag, 2020. p. 103-116 (Modeling and Optimization in Science and Technologies; Vol. 14).

Research output: Chapter in Book/Report/Conference proceedingChapter

Irmanova, A, Myrzakhmet, S & James Pappachen, A 2020, Multi-level memristive memory for neural networks. in Modeling and Optimization in Science and Technologies. Modeling and Optimization in Science and Technologies, vol. 14, Springer Verlag, pp. 103-116. https://doi.org/10.1007/978-3-030-14524-8_8
Irmanova A, Myrzakhmet S, James Pappachen A. Multi-level memristive memory for neural networks. In Modeling and Optimization in Science and Technologies. Springer Verlag. 2020. p. 103-116. (Modeling and Optimization in Science and Technologies). https://doi.org/10.1007/978-3-030-14524-8_8
Irmanova, Aidana ; Myrzakhmet, Serikbolsyn ; James Pappachen, Alex. / Multi-level memristive memory for neural networks. Modeling and Optimization in Science and Technologies. Springer Verlag, 2020. pp. 103-116 (Modeling and Optimization in Science and Technologies).
@inbook{4a20adcdb6c343208c79a64a484ef24c,
title = "Multi-level memristive memory for neural networks",
abstract = "Analog memory is of great importance in neuromorphic engineering as it enables scalable neural network design and energy efficient implementation of computationally expensive operations. With the advent of memristors, the realization of the analog memory became possible due to the intrinsic properties of memristors such as nanoscale size, non-volatility, and energy efficiency. In hardware implementations of neural networks, memristors store the values of synaptic weights and operate similarly to the synapses that are reinforced with the application of external stimuli. Memristors that are ideally continuum memories, currently are at the early stage of the development, which causes several issues in neuromorphic circuit design. Device level and architecture level issues force memory engineers to approach memristive memory design in different ways. In this chapter device-level problems: restricted number of resistance states, stochastic switching and architecture level problem: sneak paths will be discussed, and their state of the art solutions will be presented.",
author = "Aidana Irmanova and Serikbolsyn Myrzakhmet and {James Pappachen}, Alex",
year = "2020",
month = "1",
day = "1",
doi = "10.1007/978-3-030-14524-8_8",
language = "English",
series = "Modeling and Optimization in Science and Technologies",
publisher = "Springer Verlag",
pages = "103--116",
booktitle = "Modeling and Optimization in Science and Technologies",
address = "Germany",

}

TY - CHAP

T1 - Multi-level memristive memory for neural networks

AU - Irmanova, Aidana

AU - Myrzakhmet, Serikbolsyn

AU - James Pappachen, Alex

PY - 2020/1/1

Y1 - 2020/1/1

N2 - Analog memory is of great importance in neuromorphic engineering as it enables scalable neural network design and energy efficient implementation of computationally expensive operations. With the advent of memristors, the realization of the analog memory became possible due to the intrinsic properties of memristors such as nanoscale size, non-volatility, and energy efficiency. In hardware implementations of neural networks, memristors store the values of synaptic weights and operate similarly to the synapses that are reinforced with the application of external stimuli. Memristors that are ideally continuum memories, currently are at the early stage of the development, which causes several issues in neuromorphic circuit design. Device level and architecture level issues force memory engineers to approach memristive memory design in different ways. In this chapter device-level problems: restricted number of resistance states, stochastic switching and architecture level problem: sneak paths will be discussed, and their state of the art solutions will be presented.

AB - Analog memory is of great importance in neuromorphic engineering as it enables scalable neural network design and energy efficient implementation of computationally expensive operations. With the advent of memristors, the realization of the analog memory became possible due to the intrinsic properties of memristors such as nanoscale size, non-volatility, and energy efficiency. In hardware implementations of neural networks, memristors store the values of synaptic weights and operate similarly to the synapses that are reinforced with the application of external stimuli. Memristors that are ideally continuum memories, currently are at the early stage of the development, which causes several issues in neuromorphic circuit design. Device level and architecture level issues force memory engineers to approach memristive memory design in different ways. In this chapter device-level problems: restricted number of resistance states, stochastic switching and architecture level problem: sneak paths will be discussed, and their state of the art solutions will be presented.

UR - http://www.scopus.com/inward/record.url?scp=85064728858&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85064728858&partnerID=8YFLogxK

U2 - 10.1007/978-3-030-14524-8_8

DO - 10.1007/978-3-030-14524-8_8

M3 - Chapter

T3 - Modeling and Optimization in Science and Technologies

SP - 103

EP - 116

BT - Modeling and Optimization in Science and Technologies

PB - Springer Verlag

ER -