State Characterisation of Self-Directed Channel Memristive Devices
Dániel Hajtó 1 , Waleed El-Geresy 2 , Deniz Gündüz 2 , György Cserey 1 · 2025
Read State Characterisation of Self-Directed Channel Memristive Devices by Dániel Hajtó 1 , Waleed El-Geresy 2 , Deniz Gündüz 2 , György Cserey 1 (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-15757.txt.
Introduction
Daniel Hajto, Waleed El-Geresy, Deniz Gunduz, and Gyorgy Cserey tackled memristor state measurement in twenty twenty-five. Memristors are resistors with memory. They are candidate building blocks for storage, neural accelerators, and neuromorphic chips. To use them well, you need to know their state inside. That state cannot be read in a direct way from the terminals.
The authors focus on self-directed channel memristors. They set out a general model of their state, drawn from physics. They then show how to pin down that state from measured voltage-current pairs. Their estimation method is aware of measurement noise by design. It is built on minimum variance principles. The work is foundational. It gives a sound way to say what state a device is in. That step must come first for trustworthy memory and computing uses.
Catalog record
| CREATOR | Dániel Hajtó 1 , Waleed El-Geresy 2 , Deniz Gündüz 2 , György Cserey 1 |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 49 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2505-15757
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
State Characterisation of Self-Directed Channel Memristive Devices Dániel Hajtó 1 , Waleed El-Geresy 2 , Deniz Gündüz 2 , György Cserey 1 Abstract Knowing how to reliably use memristors as information storage devices is crucial not only to their role as emerging memories, but also for their application in neural network acceleration and as components of novel neuromorphic systems. In order to better understand the dynamics of information storage on memristors, it is essential to be able to characterise and measure their state. To this end, in this paper we propose a general, physics-inspired modelling approach for characterising the state of SDC ( SDC ) memristors. Additionally, to enable the identification of the proposed state from device data, we introduce a noise-aware approach to the minimum-variance estimation of the state from voltage and current pairs. I Introduction Memristors are a class of passive electronic device usually characterised by their ability to switch resistive state. First theoretically formulated by Leon Chua in 1971 [ 1 ] , they are devices whose resistive state can be modified by applying a voltage (voltage-controlled memristors) or a current (current or charge -controlled memristors). Memristors were originally hypothetical devices, theorised as a class of passive electronic device with a non-differential relationship between the electromagnetic flux in the device and the electronic charge. This formulation was based on the devices being the missing piece needed to complete a “periodic-table” of passive electronic devices, consisting of resistors, capacitors, and inductors, each defined as connecting a pair of electronic state variables among v , \frac{dv}{dt} , i , and \frac{di}{dt} , through a non-differential relationship. Since their theoretical formulation, the term memristor has been used to describe a number of electronic devices [ 2 ] , despite not necessarily strictly satisfying the original definition, and would be more correctly referred to as “memristive systems” [ 3 ] . This loose use of the term memristor is prevalent in the literature, with the term “memristor” often used to refer to any kind of resistive switching memory. Indeed, in 2011, Chua lent support to the idea of redefining memristors to include such devices [ 4 ] . This marked a transition from a mechanistic definition of memristors, to one based on their behaviour - namely, the characteristic pinched hysteresis loop indicative of a voltage or current-modulated instantaneous resistance - this instantaneous resistance often thought of as the state variable of the devices. For the purposes of simplicity, we will use the term “memristor” to refer to such resistive switching memories in this paper, since no real device satisfying the requirements of the original theoretical framework has yet been discovered. One key feature of memristors - a consequence of the resistive switching property - is their ability to co-locate the processing and storage of information, circumventing the Von-Neumann bottleneck [ 5 ] . This has the potential to enable for more energy-efficient storage and computation. They have hence emerged as promising candidates for a variety of applications including multilevel analogue data storage, where they offer potential benefits in terms of power consumption and storage density; energy-efficient deep neural network acceleration; and - having been shown early on to be theoretically capable of modelling the dynamics of biological neurons [ 3 ] - as neuromorphic computing elements. For all these applications, it is important to have an understanding of the nature of information storage on the devices. The simplistic assumption that the memristor behaves as a resistor with memory is often unreasonable for complex devices that may exhibit non-linear characteristics. For example, non-linear VI characteristics can be evident [ 6 ] , and in general, the relationship between the device state and the instantaneous resistance is expressible through a (potentially non-linear) resistance readout equation [ 7 ] . A fundamental problem is therefore that of identifying stable state variables that can be used to describe the devices and characterise their state at each instant, without restrictions on the magnitude of voltage or current input variables, by making use of measurements. Different states can also exhibit differing degrees of metastability, with some states being more stable than others, which can be modelled through the phenomenon of stochastic resistive drift [ 7 , 8 ] . Various state models and associated state transition models have been proposed for different types of memristive devices, with each model having a different level of generality and physical grounding. Early on, [ 2 ] proposed a simple width-modulated switching model, with Ohmic conduction, to describe the behaviour of titanium oxide memristors, which assumed that the state variable was equivalent to the instantaneous resistance. Late models made use of exponential state VI characteristics. [ 9 ] described the conduction behaviour in the titanium oxide memristor as being modelled by two terms: a \sinh term based on trap-assisted electron-tunnelling through a thin electronic barrier and a non-linear Schottky diode term. A further challenge for state modelling is presented by the fact that different types of memristors may have different physical characteristics, making distinct state characterisation approaches necessary for each. A number of types of memristors exist, usually being split across several broad categories. These include electronic devices such as Mott insulators [ 10 ] and devices based on electron trapping [ 11 ] ; devices based on a redox reaction occurring in an active layer of the device, including SDC memristors; PCM ( PCM ) devices [ 12 ] ; and other types of memristive device, such as organic memristors [ 13 ] , and magnetic (or spin-transfer torque/spintronic) memristors [ 14 ] . There is a trade-off between tailoring a model to make it effective for a particular physical device, and its performance when applied to a wider range of devices that may not have the same physical characteristics or switching mechanisms. A further, less obvious, problem associated with state models is also present; although a range of state models for different memristive devices exist, the process of identifying the memristive state from straightforward voltage and current measurements may be difficult, and the uncertainty associated with state estimates may be unclear, with noiseless measurements often being assumed, which may not be the case. In this paper, we study the state characterisation of SDC redox memristive devices, with the aim of identifying a stable state variable using a general approach that is motivated both by the physics of real SDC devices, but also general enough to be applicable to a variety of memristive technologies, more accurately characterising the VI relation and associated state than in previous works. In an effort to make the model useful in practical scenarios, we also aim to improve the process of identifying the proposed state variable from measurements, by introducing a state estimation approach which can estimate our proposed state from noisy voltage and current pairs, quantifying the uncertainty associated with the estimates. We derive our state model based on the physics of thin film devices and apply our state characterisation protocol to data collected from SDC memristive devices. Notation We use zero-indexed square bracket notation to denote indexing of a vector, for example x[0] denotes the first element of the vector x . Subscript notation X_{i} is used to distinguish multiple i.i.d. random variables