-
Multilayer spintronic neural networks with radio-frequency connections
Authors:
Andrew Ross,
Nathan Leroux,
Arnaud de Riz,
Danijela Marković,
Dédalo Sanz-Hernández,
Juan Trastoy,
Paolo Bortolotti,
Damien Querlioz,
Leandro Martins,
Luana Benetti,
Marcel S. Claro,
Pedro Anacleto,
Alejandro Schulman,
Thierry Taris,
Jean-Baptiste Begueret,
Sylvain Saïghi,
Alex S. Jenkins,
Ricardo Ferreira,
Adrien F. Vincent,
Alice Mizrahi,
Julie Grollier
Abstract:
Spintronic nano-synapses and nano-neurons perform complex cognitive computations with high accuracy thanks to their rich, reproducible and controllable magnetization dynamics. These dynamical nanodevices could transform artificial intelligence hardware, provided that they implement state-of-the art deep neural networks. However, there is today no scalable way to connect them in multilayers. Here w…
▽ More
Spintronic nano-synapses and nano-neurons perform complex cognitive computations with high accuracy thanks to their rich, reproducible and controllable magnetization dynamics. These dynamical nanodevices could transform artificial intelligence hardware, provided that they implement state-of-the art deep neural networks. However, there is today no scalable way to connect them in multilayers. Here we show that the flagship nano-components of spintronics, magnetic tunnel junctions, can be connected into multilayer neural networks where they implement both synapses and neurons thanks to their magnetization dynamics, and communicate by processing, transmitting and receiving radio frequency (RF) signals. We build a hardware spintronic neural network composed of nine magnetic tunnel junctions connected in two layers, and show that it natively classifies nonlinearly-separable RF inputs with an accuracy of 97.7%. Using physical simulations, we demonstrate that a large network of nanoscale junctions can achieve state-of the-art identification of drones from their RF transmissions, without digitization, and consuming only a few milliwatts, which is a gain of more than four orders of magnitude in power consumption compared to currently used techniques. This study lays the foundation for deep, dynamical, spintronic neural networks.
△ Less
Submitted 7 November, 2022;
originally announced November 2022.
-
Micromagnetics of magnetic chemical modulations in soft-magnetic cylindrical nanowires
Authors:
L. Álvaro-Gómez,
S. Ruiz-Gómez,
C. Fernández-González,
M. Schöbitz,
N. Mille,
J. Hurst,
D. Tiwari,
A. De Riz,
I. M. Andersen,
J. Bachmann,
L. Cagnon,
M. Foerster,
L. Aballe,
R. Belkhou,
J. C Toussaint,
C. Thirion,
A. Masseboeuf,
D. Gusakova,
L. Pérez,
O. Fruchart
Abstract:
We analyze the micromagnetics of short longitudinal modulations of a high-magnetization material in cylindrical nanowires made of a soft-magnetic material of lower magnetization such as permalloy, combining magnetic microscopy, analytical modeling, and micromagnetic simulations. The mismatch of magnetization induces curling of magnetization around the axis in the modulations, in an attempt to scre…
▽ More
We analyze the micromagnetics of short longitudinal modulations of a high-magnetization material in cylindrical nanowires made of a soft-magnetic material of lower magnetization such as permalloy, combining magnetic microscopy, analytical modeling, and micromagnetic simulations. The mismatch of magnetization induces curling of magnetization around the axis in the modulations, in an attempt to screen the interfacial magnetic charges. The curling angle increases with modulation length, until a plateau is reached with nearly full charge screening for a specific length scale~$Δ_\mathrm{mod}$, larger than the dipolar exchange length of any of the two materials. The curling circulation can be switched by the Oersted field arising from a charge current with typical magnitude $10^{12} A/m^{2}$ for a diameter of $\sim$100 nm, and reaching a maximum for $Δ_\mathrm{mod}$.
△ Less
Submitted 6 September, 2022; v1 submitted 13 May, 2022;
originally announced May 2022.
-
Convolutional Neural Networks with Radio-Frequency Spintronic Nano-Devices
Authors:
Nathan Leroux,
Arnaud De Riz,
Dédalo Sanz-Hernández,
Danijela Marković,
Alice Mizrahi,
Julie Grollier
Abstract:
Convolutional neural networks are state-of-the-art and ubiquitous in modern signal processing and machine vision. Nowadays, hardware solutions based on emerging nanodevices are designed to reduce the power consumption of these networks. Spintronics devices are promising for information processing because of the various neural and synaptic functionalities they offer. However, due to their low OFF/O…
▽ More
Convolutional neural networks are state-of-the-art and ubiquitous in modern signal processing and machine vision. Nowadays, hardware solutions based on emerging nanodevices are designed to reduce the power consumption of these networks. Spintronics devices are promising for information processing because of the various neural and synaptic functionalities they offer. However, due to their low OFF/ON ratio, performing all the multiplications required for convolutions in a single step with a crossbar array of spintronic memories would cause sneak-path currents. Here we present an architecture where synaptic communications have a frequency selectivity that prevents crosstalk caused by sneak-path currents. We first demonstrate how a chain of spintronic resonators can function as synapses and make convolutions by sequentially rectifying radio-frequency signals encoding consecutive sets of inputs. We show that a parallel implementation is possible with multiple chains of spintronic resonators to avoid storing intermediate computational steps in memory. We propose two different spatial arrangements for these chains. For each of them, we explain how to tune many artificial synapses simultaneously, exploiting the synaptic weight sharing specific to convolutions. We show how information can be transmitted between convolutional layers by using spintronic oscillators as artificial microwave neurons. Finally, we simulate a network of these radio-frequency resonators and spintronic oscillators to solve the MNIST handwritten digits dataset, and obtain results comparable to software convolutional neural networks. Since it can run convolutional neural networks fully in parallel in a single step with nano devices, the architecture proposed in this paper is promising for embedded applications requiring machine vision, such as autonomous driving.
△ Less
Submitted 9 November, 2021;
originally announced November 2021.
-
Time-resolved imaging of Œrsted field induced magnetization dynamics in cylindrical magnetic nanowires
Authors:
M. Schöbitz,
S. Finizio,
A. De Riz,
J. Hurst,
C. Thirion,
D. Gusakova,
J. -C. Toussaint,
J. Bachmann,
J. Raabe,
O. Fruchart
Abstract:
Recent studies in three dimensional spintronics propose that the Œrsted field plays a significant role in cylindrical nanowires. However, there is no direct report of its impact on magnetic textures. Here, we use time-resolved scanning transmission X-ray microscopy to image the dynamic response of magnetization in cylindrical Co$_{30}$Ni$_{70}$ nanowires subjected to nanosecond Œrsted field pulses…
▽ More
Recent studies in three dimensional spintronics propose that the Œrsted field plays a significant role in cylindrical nanowires. However, there is no direct report of its impact on magnetic textures. Here, we use time-resolved scanning transmission X-ray microscopy to image the dynamic response of magnetization in cylindrical Co$_{30}$Ni$_{70}$ nanowires subjected to nanosecond Œrsted field pulses. We observe the tilting of longitudinally magnetized domains towards the azimuthal Œrsted field direction and create a robust model to reproduce the differential magnetic contrasts and extract the angle of tilt. Further, we report the compression and expansion, or breathing, of a Bloch-point domain wall that occurs when weak pulses with opposite sign are applied. We expect that this work lays the foundation for and provides an incentive to further studying complex and fascinating magnetization dynamics in nanowires, especially the predicted ultra-fast domain wall motion and associated spin wave emissions.
△ Less
Submitted 12 April, 2021; v1 submitted 5 February, 2021;
originally announced February 2021.
-
Mechanism of current-assisted Bloch-point wall stabilization for ultra fast dynamics
Authors:
A. De Riz,
J. Hurst,
M. Schöbitz,
C. Thirion,
J. Bachmann,
J. C. Toussaint,
O. Fruchart,
D. Gusakova
Abstract:
Two types of domain walls exist in magnetically soft cylindrical nanowires: the transverse-vortex wall (TVW) and the Bloch-point wall (BPW). The latter is expected to prevent the usual Walker breakdown, and thus enable high domain wall speed. We showed recently [M. Schöbitz \etal, Phys. Rev. Lett. 123, 217201 (2019)] that the previously overlooked OErsted field associated with an electric current…
▽ More
Two types of domain walls exist in magnetically soft cylindrical nanowires: the transverse-vortex wall (TVW) and the Bloch-point wall (BPW). The latter is expected to prevent the usual Walker breakdown, and thus enable high domain wall speed. We showed recently [M. Schöbitz \etal, Phys. Rev. Lett. 123, 217201 (2019)] that the previously overlooked OErsted field associated with an electric current is a key in experiments to stabilize the BPW and reach speed above 600 m/s with spin-transfer. Here, we investigate in detail this situation with micromagnetic simulations and modeling. The switching of the azimuthal circulation of the BPW to match that of the OErsted field occurs above a threshold current scaling with $1/R^3$ ($R$ is the wire radius), through mechanisms that may involve the nucleation and/or annihilation of Bloch points. The domain wall dynamics then remains of a below-Walker type, with speed largely determined by spin-transfer torque alone.
△ Less
Submitted 30 November, 2020;
originally announced November 2020.
-
Theoretical study of current-induced domain wall motion in magnetic nanotubes with azimuthal domains, including Œrsted field and spin-transfer torques
Authors:
Jérôme Hurst,
Arnaud De Riz,
Staňo,
Jean-Christophe Toussaint,
Olivier Fruchart,
Daria Gusakova
Abstract:
We report a theoretical overview of the magnetic domain wall behavior under an electric current in infinitely-long nanotubes with azimuthal magnetization, combining the $1$D analytic model and micromagnetic simulations. We highlight effects that, besides spin-transfer torques already largely understood in flat strips, arise specifically in the tubular geometry: the \OErsted field and curvature-ind…
▽ More
We report a theoretical overview of the magnetic domain wall behavior under an electric current in infinitely-long nanotubes with azimuthal magnetization, combining the $1$D analytic model and micromagnetic simulations. We highlight effects that, besides spin-transfer torques already largely understood in flat strips, arise specifically in the tubular geometry: the \OErsted field and curvature-induced magnetic anisotropy resulting both from exchange and material growth. Depending on both the geometry of the tube and the strength of the azimuthal anisotropy, Bloch or Néel walls arise at rest, resulting in two regimes of motion largely dominated by either spin-transfer torques or the \OErsted field. We determine the Walker breakdown current in all cases, and highlight the most suitable parameters to achieve high domain wall speed.
△ Less
Submitted 5 October, 2020;
originally announced October 2020.
-
Domain wall pinning in a circular cross-section wire with modulated diameter
Authors:
A. De Riz,
B. Trapp,
J. A. Fernandez-Roldan,
Ch. Thirion,
J. -Ch. Toussaint,
O. Fruchart,
D. Gusakova
Abstract:
Domain wall propagation in cylindrical nanowires with modulations of diameter is a key phenomenon to design physics-oriented devices, or a disruptive three-dimensional magnetic memory. This chapter presents a combination of analytical modelling and micromagnetic simulations, with the aim to present a comprehensive panorama of the physics of pinning of domain walls at modulations, when moved under…
▽ More
Domain wall propagation in cylindrical nanowires with modulations of diameter is a key phenomenon to design physics-oriented devices, or a disruptive three-dimensional magnetic memory. This chapter presents a combination of analytical modelling and micromagnetic simulations, with the aim to present a comprehensive panorama of the physics of pinning of domain walls at modulations, when moved under the stimulus of a magnetic field or a spin-polarized current. For the sake of considering simple physics, we consider diameters of a few tens of nanometers at most, and accordingly domain walls of transverse type. Modeling with suitable approximations provides simple scaling laws, while simulations are more accurate, refining the results and defining the range of validity of the models. While pinning increases with the relative change of diameter, a key feature is the much larger efficiency of pinning at an increase of diameter upon considering current rather than field, due to the drastic decrease of current density related to the increase of diameter.
△ Less
Submitted 18 June, 2019;
originally announced June 2019.
-
Fast domain walls governed by Œrsted fields in cylindrical magnetic nanowires
Authors:
M. Schöbitz,
A. De Riz,
S. Martin,
S. Bochmann,
C. Thirion,
J. Vogel,
M. Foerster,
L. Aballe,
T. O. Menteş,
A. Locatelli,
F. Genuzio,
S. Le Denmat,
L. Cagnon,
J. C. Toussaint,
D. Gusakova,
J. Bachmann,
O. Fruchart
Abstract:
Since its proposal, the idea to vastly increase data storage density with a magnetic non-volatile 3D shift-register has sustained interest in current-induced domain wall (DW) motion. So far, experimental efforts have focused on flat nanostrips, which exhibit a wide range of noteworthy effects, yet suffer from intrinsic DW instabilities limiting their mobility. In contrast, ferromagnetic cylindrica…
▽ More
Since its proposal, the idea to vastly increase data storage density with a magnetic non-volatile 3D shift-register has sustained interest in current-induced domain wall (DW) motion. So far, experimental efforts have focused on flat nanostrips, which exhibit a wide range of noteworthy effects, yet suffer from intrinsic DW instabilities limiting their mobility. In contrast, ferromagnetic cylindrical nanowires (NWs) can host a novel type of magnetic DW, namely the Bloch-point wall (BPW), which due to its specific 3D topology should not experience the same fundamental issue. This could give rise to DW velocities over ~1000 m/s and fascinating new physics including coupling to magnetic spin waves, however, experimental evidence of DW dynamics in NWs is lacking until now. Here we report experimental results on current-induced DW motion in NWs with velocities >600 m/s, quantitatively consistent with predictions. Furthermore, our results indicate that although previously disregarded, the OErsted field induced by the current plays instead a crucial and valuable role in stabilising exclusively BPWs.
△ Less
Submitted 10 January, 2020; v1 submitted 20 March, 2019;
originally announced March 2019.
-
Modeling magnetic-field-induced domain wall propagation in modulated-diameter cylindrical nanowires
Authors:
J. A. Fernandez-Roldan,
A. De Riz,
B. Trapp,
C. Thirion,
J. -C. Toussaint,
O. Fruchart,
D. Gusakova
Abstract:
Domain wall propagation in modulated-diameter cylindrical nanowires is a key phenomenon to be studied with a view to designing three-dimensional magnetic memory devices. This paper presents a theoretical study of transverse domain wall behavior under the influence of a magnetic field within a cylindrical nanowire with diameter modulations. In particular, domain wall pinning close to the diameter m…
▽ More
Domain wall propagation in modulated-diameter cylindrical nanowires is a key phenomenon to be studied with a view to designing three-dimensional magnetic memory devices. This paper presents a theoretical study of transverse domain wall behavior under the influence of a magnetic field within a cylindrical nanowire with diameter modulations. In particular, domain wall pinning close to the diameter modulation was quantified, both numerically, using finite element micromagnetic simulations, and analytically. Qualitative analytical model for gently sloping modulations resulted in a simple scaling law which may be useful to guide nanowire design when analyzing experiments. It shows that the domain wall depinning field value is proportional to the modulation slope.
△ Less
Submitted 17 July, 2018;
originally announced July 2018.
