Magnetic Architectures for Reservoir Computing Hardware (MARCH)
Lead Research Organisation:
University of Sheffield
Department Name: Materials Science and Engineering
Abstract
Abstracts are not currently available in GtR for all funded research. This is normally because the abstract was not required at the time of proposal submission, but may be because it included sensitive information such as personal details.
Publications
Ababei RV
(2021)
Neuromorphic computation with a single magnetic domain wall.
in Scientific reports
Allwood D
(2023)
A perspective on physical reservoir computing with nanomagnetic devices
in Applied Physics Letters
Ellis M
(2023)
Machine learning using magnetic stochastic synapses
Manneschi L
(2021)
Exploiting Multiple Timescales in Hierarchical Echo State Networks
in Frontiers in Applied Mathematics and Statistics
Manneschi L
(2023)
SpaRCe: Improved Learning of Reservoir Computing Systems Through Sparse Representations.
in IEEE transactions on neural networks and learning systems
Manneschi L
(2022)
Signal neutrality, scalar property, and collapsing boundaries as consequences of a learned multi-timescale strategy.
in PLoS computational biology
Ozdemir A
(2021)
EchoVPR: Echo State Networks for Visual Place Recognition
Ozdemir A
(2022)
EchoVPR: Echo State Networks for Visual Place Recognition
in IEEE Robotics and Automation Letters
Venkat G
(2023)
Magnetic domain walls: types, processes and applications
in Journal of Physics D: Applied Physics
Vidamour I
(2023)
Reconfigurable reservoir computing in a magnetic metamaterial
in Communications Physics
Vidamour IT
(2022)
Quantifying the computational capability of a nanomagnetic reservoir computing platform with emergent magnetisation dynamics.
in Nanotechnology
Welbourne A
(2021)
Voltage-controlled superparamagnetic ensembles for low-power reservoir computing
in Applied Physics Letters
Description | (1) We have now clearly established the feasibility of performing reservoir computing using the emergent behaviour of domain walls in rings of nanoscale magnetic rings. We have demonstrated how data can be inputted into devices using rotating applied magnetic fields and data outputted by measuring the resistance of the devices. Using these techniques through a computational framework known as "Reservoir Computing" has allowed us to experimentally perform a number of challenging tasks including the transformation of signals, the recognition of spoken digits, and the prediction of complex time series. (2) We have developed a comprehensive understanding of how domain wall interactions within nanoscale magnetic ring arrays results in emergent behaviour, and used this understanding to create a new kind of model that allows accurate simulation of these systems, which are too complex to simulate by more conventional means. These models are highly useful for further developing and exploring how better to use magnetic ring devices for computation. (3) We have explored how magnetic nanoring-based reservoir computers with a single input and output can be made to have different computational strengths by inserting them into different neural network topologies, which can be virtualised by the way time is inputted to the device and outputted from it. This is highly important for understanding how best to harness them for computation. (4) We have shown in simulation that electric currents could be used to directly drive magnetic nanoring-based reservoir computers, instead of using magnetic fields. This is likely to make future devices more power efficient. Experiments to realise these effects real devices are currently ongoing. (5) We have demonstrated that making slight changes to the geometry of the ring arrays (e.g. changing ring widths, geometric tiling of the rings) can change their magnetic behaviour and that these changes produce changes in the computational properties of the ring arrays as expressed by task independent metrics. However, we have also shown that these effects are limited by the low input-output dimensionality of our devices suggesting that finer grain readout of magnetic states will be required to fully exploit these changes in behaviour. (6) We have demonstrated that heterogeneous arrays containing many different geometries of ring can be measured by fabricating multiple electrical contacts for resistance measurements. Our results suggest that the heterogeneity of the ring arrays results in enhancement in computational properties. (7) We have shown that performing measurements of the spin wave spectra of the nanoring arrays provide a highly informative characterisation of their magnetic states, and that this can be used as a multidimensional output for computation. We have also demonstrated that spin orbit torque FMR measurements can provide a output suitable for measuring device-scale arrays which will make this approach more applicable to real devices. (8) We have shown that the temperature response of the magnetic ring arrays allows them to be used as intelligent sensors that not only sense the state of their environment, but also make decisions/perform inference based on these measurements. (9) We have developed Neural ODE techniques to create "digital twins" of real devices. These allow us to explore computational ideas in simulation prior to transferring them to real devices. (10) We have shown that creating interconnected networks where the ring arrays act as "complex neurons" produce significant enhancements in computational behaviour over individual devices, and allows them to tackle significantly harder problems, including neuro-prosthetics tasks. |
Exploitation Route | Neuromorphic computation with magnetic devices is currently a hot topic, and we think that our results provide valuable information about how best to harness these for new devices. This will likely inspire more academic research in the short term, but in the longer term may contribute to the creation of new, energy efficient computers. More specifically, the applications of our devices as intelligent sensors is novel and may open up new applications in edge computing. Our digital twining approach could be applied to an enormous range of systems to expeditated exploration of possible devices. Finally, our demonstrations of the computational potential of "complex neuron networks" illustrate new ways to extend the potential of simple reservoir computing devices to tackle very challenging real world problems. |
Sectors | Aerospace Defence and Marine Digital/Communication/Information Technologies (including Software) Electronics Energy |
Description | Royce Materials Challenge Accelerator Programme - Magnets that Think and Feel |
Amount | £56,023 (GBP) |
Funding ID | MCAP016 |
Organisation | Henry Royce Institute |
Sector | Academic/University |
Country | United Kingdom |
Start | 11/2022 |
End | 04/2023 |
Title | Quantifying the Computational Capability of a Nanomagnetic Reservoir Computing Platform with Emergent Magnetisation Dynamics |
Description | Data used in the production of the publication 'Quantifying the Computational Capability of a Nanomagnetic Reservoir Computing Platform with Emergent Magnetisation Dynamics'. |
Type Of Material | Database/Collection of data |
Year Produced | 2021 |
Provided To Others? | Yes |
URL | https://figshare.shef.ac.uk/articles/dataset/Quantifying_the_Computational_Capability_of_a_Nanomagne... |
Description | Collaboration with Imperial College London on Reservoir Computing with Magnetic Nanostructures. |
Organisation | Imperial College London |
Department | Imperial College Trust |
Country | United Kingdom |
Sector | Charity/Non Profit |
PI Contribution | We have been working with the group of Dr Will Branford and Dr Jack Gartside on neuromorphic computing with magnetic nanostructures. Our collaborations have involved the Sheffield Functional Magnetic Materials Group performing focused magneto-optic Kerr measurements on spin ice arrays made at ICL. There has also been collaboration between the Sheffield Machine Learning Group and ICL on emulating devices where multiple reservoirs are connected in series. |
Collaborator Contribution | The ICL group have performed FMR measurements on magnetic ring arrays made fabricated at the University of Sheffield. |
Impact | None so far. |
Start Year | 2021 |
Description | Collaboration with Universities of Durham and Poznan |
Organisation | Durham University |
Country | United Kingdom |
Sector | Academic/University |
PI Contribution | We performed ferromagnetic resonance measurements on CoFe/Pt thin films with transition metal delta layers in order to determine their gilbert damping constants. |
Collaborator Contribution | Samples were prepared at Poznan University, with initial characterisation taking place at the University of Durham. |
Impact | None so far |
Start Year | 2023 |
Description | Collaboration with Universities of Durham and Poznan |
Organisation | Poznan University of Technology |
Country | Poland |
Sector | Academic/University |
PI Contribution | We performed ferromagnetic resonance measurements on CoFe/Pt thin films with transition metal delta layers in order to determine their gilbert damping constants. |
Collaborator Contribution | Samples were prepared at Poznan University, with initial characterisation taking place at the University of Durham. |
Impact | None so far |
Start Year | 2023 |
Description | Collaboration with the University of Sherbrooke |
Organisation | University of Sherbrooke |
Country | Canada |
Sector | Academic/University |
PI Contribution | We have begun to collaborate with the group of Prof. Julien Sylvestre at the University of Sherbrooke. Thus, far this has resulted in discussion of how our group can collaborate to use networks of reservoir computers to control soft robotic machines. |
Collaborator Contribution | See above. |
Impact | None so far |
Start Year | 2022 |
Description | MARCH interaction with University of York |
Organisation | University of York |
Department | Department of Computer Science |
Country | United Kingdom |
Sector | Academic/University |
PI Contribution | We collaborate with research groups at the University of York that have expertise in reservoir computing (a type of neuromorphic computing) and robotics. This is part of an ongoing collaboration, with the York team funded via EPSRC grant EP/V006029/1. The Sheffield teams contribute through their expertise in nanoscale magnetism and machine learning theory. |
Collaborator Contribution | The University of York collaborators are contributing to the joint research project by providing guidance and analysis of what physical designs of magnetic nanostructures might make good reservoir computers, and in performing large-scale measurements of devices made by us. |
Impact | So far, we have only one publication that acknowledges this partnership: Stepney S. (2021) Non-instantaneous Information Transfer in Physical Reservoir Computing. In: Kostitsyna I., Orponen P. (eds) Unconventional Computation and Natural Computation. UCNC 2021. Lecture Notes in Computer Science, vol 12984. Springer, Cham. https://doi.org/10.1007/978-3-030-87993-8_11 - although this only has one author, who is based at the University of York. |
Start Year | 2021 |