Back to search

Democritus University of Thrace (DUTH)

Harnessing memristor circuits and device variability in emergent computing applications

Abstract

dc:description

Contemporary electronic systems demonstrate unprecedented capabilities in information storage and processing by utilizing the modern nanoscale complementary metal-oxide-semiconductor (CMOS) technology and the general-purpose von Neumann computer architecture. Nevertheless, with the increasing amount of information collection and the necessity for its computationally-demanding processing, conventional technology and computing architecture struggle to keep pace, as they meet their respective limitations either from physical or practical standpoints, i.e. miniaturization limits, increasing costs, heat and energy consumption. Novel approaches to both technology and computing architecture have been intensively investigated in recent decades, aiming at the development of innovative electronic systems with advanced energy efficiency, speed of operation, and compact integration area.In this vein, the memristor, a novel electronic component with nanometer dimensions, comes as a promising candidate to push the limits of recent technology and computing. Materialized roughly a decade ago, memristor devices provide unprecedented features for a single electronic device, such as non-volatile analog information storage with fast and low power operation. Utilizing this feature, memristors are currently used in emerging memory technologies, as well as constituting the leading technology for future artificial neural networks and neuromorphic computing hardware implementations. However, technologically speaking, memristor devices are still in their infancy, as they are still facing major issues that hinder their wide commercialization. Firstly, the modeling of their conduction mechanisms and dynamical behavior is still being investigated, and we are far from a standard memristor model that accurately captures the characteristics of fabricated memristor devices. In addition, the existing memristor models that achieve an accurate description of the device dynamics struggle with numerical stability issues that affect their proper utilization in circuit simulators. Secondly, due to the memristor's nature, the adjustment of its conductivity arises from intrinsically stochastic processes, e.g. ion migration, which result in high variability during the operation of the device, also affecting its functionality within a circuit.Aiming to deal with the aforementioned issues in the context of this Ph.D. thesis, memristor device modeling has been addressed in a two-fold approach. On one hand, working on an existing physics-based memristor model, the necessary mathematical transformations are performed to derive an analytical and transcendental form of the device's dynamical behavior under constant positive and negative stimulus, respectively, enabling the analytical study of the memristor's programming. Beyond the deterministic model, a probabilistic mathematical framework for stochasticity-aware memristor modeling has been developed based on the master equations of Markov jump processes. The proposed framework captures the probabilistic switch of memristor devices that originates from the device's intrinsic stochasticity. Apart from binary stochastic switching, this approach allows multiple finite states, for the first time, providing versatility and further accuracy in probabilistic memristor modeling.From a circuit design perspective, the memristor's stochasticity imposes detrimental variability on the programming of memristor devices. To tackle this problem, I adopted a counter-intuitive approach where noise is added to the programming signals to enhance the memristor's switching ability. Such an approach is based on the nonlinear system phenomenon called Stochastic Resonance (SR), which postulates that a properly selected noisy signal can improve the performance of a nonlinear system. Originally, SR was used to enhance the resistance window of a memristor. Then, in this thesis, that phenomenon was studied for a wider variety of memristor models, including device variability, which was also demonstrated with experimental measurements on single memristor devices. A detailed study on the noise-aided programming of a memristor in ReRAM memory cells, either individually or in an array, has been performed, showcasing the power-efficiency of the proposed noise-aided approach as the nominal programming voltage amplitudes can be reduced in the presence of noise without damaging the programming accuracy.Beyond facing the memristor's stochasticity as a detrimental characteristic, its beneficial exploitation within a novel computing paradigm has been investigated during this thesis. Considering a powerful emerging computing setup known as Cellular Automata (CA), memristor devices are incorporated into the CA array as part of the simplistic locally interacting computing units, named CA cells, that it is composed of. Originally proposed as a deterministic tool, CA's abilities are expanded after the memristor's probabilistic switching is embedded into the CA cells, establishing the novel memristor-based Probabilistic CA (MemPCA). Given its simplest form, the Elementary CA, where each cell contains a binary state and the range of cell interaction is limited to the minimum possible, i.e. the two adjacent cells in a 1D grid, the effect of the memristor's probabilistic switching has been studied for all the cell interaction rules, resulting in the improvement of system entropy for a certain amount of switching probability. Moreover, the first transistor-level implementation of ECA with the all-memristor approach has been performed, with memristor devices participating in both cell and rule modules. The proposed novel MemPCA implementation is able to achieve high operation speed and low area requirements, while also incorporating an ultra-fast source of entropy per cell, i.e. the memristor device. The functionality of the system is showcased in both deterministic and probabilistic operations, which can be adjusted by externally selected parameters without any modification to the design. Lastly, the proposed MemPCA approach also provides a reconfigurable implementation of the rule module that enables both spatial and temporal rule inhomogeneity.

Degree

thesis:*
Grantor dc:publisher
Democritus University of Thrace (DUTH)
Year dc:date
2022

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Ntinas, Vasileios
  • Ντίνας, Βασίλειος

Subjects

dc:subject × 12

Rights

Language dc:language
eng

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/51458

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Ntinas, Vasileios; Ντίνας, Βασίλειος. Harnessing memristor circuits and device variability in emergent computing applications. Democritus University of Thrace (DUTH), 2022. http://hdl.handle.net/10442/hedi/51458