Sneak, discharge, and leakage current issues in a high-dimensional 1T1M memristive crossbar
- PDF / 1,766,195 Bytes
- 11 Pages / 595.276 x 790.866 pts Page_size
- 103 Downloads / 148 Views
Sneak, discharge, and leakage current issues in a high‑dimensional 1T1M memristive crossbar V. A. Demin1 · I. A. Surazhevsky1 · A. V. Emelyanov1,2 · P. K. Kashkarov1,2,3 · M. V. Kovalchuk1,2,3
© Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract Memristive crossbar arrays are believed to be the future of high-density nonvolatile memory and neuromorphic systems. However, significant challenges related to the passive crossbar architecture, for example, the sneak current issue, impose limitations on their performance. One of the well-known ways to overcome this problem is to use a one-transistor one-memristor (1T1M) scheme. Nevertheless, for a sufficiently large crossbar, even with a 1T1M architecture, problems appear not only with sneak currents but also with leakage through the gates of the transistors and the discharge of their capacitances. These effects are analyzed herein by simulations and analytically to determine their influence on the performance of a 1T1M crossbar, depending on its dimensions. Numerical results are presented for the examples of (CoFeB)x (LiNbO3 )100−x nanocomposite and ZrO2(Y)-based memristive structures. The results reveal that the sneak, discharge, and (to a lesser extent) leakage currents can severely degrade the performance of even a not very large ( < 103 × 103 ) 1T1M crossbar. Finally, analytical estimates are used to reveal how a well-known, simple special scheme for switching and reading can fix these negative effects, even for a 1T1M memristive crossbar with rather large dimensions ( ∼ 106 × 106 ), taking into account its plausible geometrical size and the scaling dependence of its constituent elements. Keywords Neuromorphic hardware · Memristor · Crossbar array · 1T1M crossbar · Sneak current · Leakage current · Discharge current · Memristive crossbar performance
1 Introduction Computation using artificial neural networks is nowadays experiencing a renaissance, as the rapid development of the Internet of Things and the availability of large computing power have made it possible to achieve great results in areas such as image, speech, and text recognition, the development of self-driving cars and drones [1–5], etc. However, with their increasing performance, the operation of such systems also requires growing amounts of additional power. * I. A. Surazhevsky [email protected] V. A. Demin [email protected] 1
National Research Center “Kurchatov Institute”, Moscow, Russia 123182
2
Moscow Institute of Physics and Technology (State University), Dolgoprudny, Moscow Region, Russia 141700
3
Faculty of Physics, Lomonosov Moscow State University, Moscow, Russia 119991
Currently, specialized neuromorphic processors are being developed around the world to overcome this problem [6–10]. It is supposed that they will bring neural computation to a fundamentally new level of performance while reducing the associated energy consumption. One of the promising hardware approaches in this field is the use of memristors. Their ability to change conductivi
Data Loading...
