Magnetic device

The magnetic device with ferromagnetic thin-film nanomagnetic structures and a spin current generation layer addresses high energy and computational costs in AI technologies by enabling efficient all-electrical data input and readout, making it suitable for low-power edge AI chips.

GB2640493AInactive Publication Date: 2025-10-29IMPERIAL COLLEGE INNVOATIONS LTD
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
GB2024004830
Authority / Receiving Office
GB · GB
Patent Type
Applications
Current Assignee / Owner
Filing Date
2024-04-04
Publication Date
2025-10-29
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

Existing machine learning and artificial intelligence technologies, particularly those using CPUs and GPUs, incur high energetic and computational costs, making them unsuitable for deployment in remote locations without access to mains power, and existing AI chips designed for data centers also consume too much power for remote use.

Method used

A magnetic device with ferromagnetic thin-film nanomagnetic structures and a spin current generation layer that converts electrical current into a spin current to control magnetic states, allowing all-electrical data input and readout without individual electrical connections, and is scalable and less complex to operate.

Benefits of technology

The magnetic device supports efficient all-electrical data input and readout, is scalable, and operates at low power, enabling deployment in edge AI chips for computationally demanding tasks.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 00000000_0000_ABST
    Figure 00000000_0000_ABST
Patent Text Reader

Abstract

Magnetic device 1, 11 for neuromorphic in-memory computing has spaced ferromagnetic thin-film nanomagnetic structures 2 (may be surface coated with ionic material), each with magnetic state, on surface 4 of a spin current generation layer 3 (may be heavy metal) configured to convert electrical current [jc, figure 2] into spin current [js, figure 2] for controlling a magnetic state of the ferromagnetic structures. Spin current may exert spin torque on the ferromagnetic structures or spin current generation layer. Subsets of ferromagnetic structures may be sized with different resonant frequencies and may form an artificial spin ice. The device may be part of crossbar array and integrated circuit. Reading a state may include spin torque ferromagnetic resonance measurements indicating frequencies corresponding to spin wave modes, including sensing a rectified voltage. Inputting data may include driving AC current through the spin current generation layer with frequency near a resonant frequency of a ferromagnetic structure and with amplitude for inducing ferromagnetic resonance, and exposing the device to a magnetic field, causing magnetic state switching. Device operation may include applying voltage across the ionic material layer, causing ion migration into / from a ferromagnetic structure; or training by determining an improvement in performance metric / loss function / error value.
Need to check novelty before this filing date? Find Prior Art

Description

Field of the Invention The present invention relates to a magnetic device. In particular, the present invention relates to a magnetic device for neuromorphic in-memory computing and / or data storage. Background Many existing approaches to machine learning and artificial intelligence (AI) are implemented using CPUs and GPUs. As the neural network models used in these approaches tend to be large, the process of training has a high energetic and computational cost. As such, these approaches are not suitable for deployment in the field and instead require devices in the field to transmit data to and from large data centres in which neural network models are implemented. Specialist AI chips are an emerging field. At present, many specialist AI chips such as IBM's Loihi and Graphcore's IPU are designed for data centre use. Again, their power consumption prevents them from being deployed in remote locations where access to mains power is limited. Hailo AI chips are an example of edge AI chips. Edge AI chips are specialized hardware designed to execute AI algorithms locally on devices, rather than relying on remote servers or the cloud. Summary According to a first aspect of the present invention, there Is provided a magnetic device. The magnetic device includes ferromagnetic thin-film nanomagnetic structures spaced apart from each other, each having a respective magnetic state. The magnetic device also Includes a spin current generation layer having a surface on which the ferromagnetic thin-film nanomagnetic structures are disposed. The spin current generation layer is configured to convert an electrical current driven through the spin current generation layer parallel to the surface into a spin current capable of controlling the magnetic state(s) of one or more of the ferromagnetic thin-film nanomagnetic structures. In other words, the spin current generation layer is configured to convert an electrical current driven through the spin current generation layer parallel to the surface into a spin current capable of modifying the magnetic state(s) of one or more of the ferromagnetic thin-film nanomagnetic structures. Thus, all-electrical data input and data readout from all the ferromagnetic thin-film nanomagnetic structures can be supported. Moreover, a need for each ferromagnetic thin-film nanomagnetic structure to be addressed by individual electrical connections can be negated, rendering the magnetic device less complex to operate and more scalable than some other magnetic devices. The spin current may be capable of controlling the magnetic state(s) of one or more of the ferromagnetic thin-film nanomagnetic structures by exerting a spin torque on the ferromagnetic thin-film nanomagnetic structures. The spin current may be capable of controlling the magnetic state(s) of one or more of the ferromagnetic thin-film nanomagnetic structures by being incident on the ferromagnetic thin-film nanomagnetic structures so as to exert a spin torque on the ferromagnetic thin-film nanomagnetic structures. The spin current generation layer may be ferromagnetic. The spin current generation layer may be magnetically coupled to the ferromagnetic thin-film nanomagnetic structures. The spin current may be capable of controlling the magnetic state(s) of one or more of the ferromagnetic thin-film nanomagnetic structures by exerting a spin torque on the spin current generation layer. The spin current generation layer may be magnetically coupled to the ferromagnetic thin-film nanomagnetic structures by dipolar coupling. The ferromagnetic thin-film nanomagnetic structures may be magnetically coupled to each other by dipolar coupling. The spin current generation layer may be patterned to include voids in areas adjacent to the ferromagnetic thin-film nanomagnetic structures. The ferromagnetic thin-film nanomagnetic structures may be wholly disposed on the spin current generation layer. A spacer layer may be interposed between each ferromagnetic thin-film nanomagnetic structure and the spin current generation layer. In other words, each ferromagnetic thin-film nanomagnetic structure can be spaced apart from the spin current generation layer. The spacer layer may be formed from an electrically conducting material for conducting the spin current to the ferromagnetic thin-film nanomagnetic structures. Alternatively, the spacer layer may be formed from an electrically insulating material. Each ferromagnetic thin-film nanomagnetic structure may be separated from the spin current generation layer by the same spacer layer. In other words, the spacer layer may be continuous (for example, taking the form of a sheet), or patterned so as to consist of portions which are directly connected to each other (for example, forming a connected lattice or a mesh). The spacer layer may be patterned to include voids in areas adjacent to the ferromagnetic thin-film nanomagnetic structures. Each of the ferromagnetic thin-film nanomagnetic structures may be wholly disposed on the spacer layer. The spin current generation layer may comprise a heavy metal material and / or a magnetic material. The spin current generation layer may consist essentially or consist of the heavy metal material. Herein, consisting essentially of may mean the same as mostly comprising, for example, comprising in an amount of 50 wt% or more. A heavy metal material is a material which consists essentially or consists of any one or a combination of elements selected from the group consisting of Ta, Pt, Pd, Au, Bi, Zn, Hf, Os, Pb, Tl, W, Tb, Gd, Ho, Dy, Pr, Sm, Nd, V, Ti, Te, and Ir. Preferably, the heavy metal material consists essentially or consists of any one or a combination of elements selected from the group consisting of W, Ta, Pt, and Au. More preferably, the heavy metal material consists essentially or consists of Pt. The spin current generation layer may consist essentially or consist of the magnetic material. The magnetic material may have planar anisotropy. The magnetic material may be a NiFe alloy or a CoFeB alloy. The magnetic material may have perpendicular anisotropy. The magnetic material may be a CoPt alloy, or an FePt alloy, or an FePd alloy, or a Co / Ni stack, or a CO / AI2O3 stack. The magnetic material may consist essentially or consist of one or more materials that are not heavy metal materials. The magnetic material may be selected from the group consisting of Fe, Co, Ni, FeCo, FeCoB, NiFe, NdFeB, WCoFeB, TaCoFeB. Preferably, the magnetic material consists essentially or consists of NiFe. The spin torque may be a spin orbit torque. The spin current generation layer may be a spin-Hall material layer, that is, a layer formed from a material which exhibits the spin-Hall effect either by virtue of having high spin orbit coupling or by virtue of having a non-trivial band structure topology. The spin current generation layer may be formed from a material which exhibits one or more of the spin Hall effect, the magnetization-dependent spin Hall effect, the anomalous spin Hall effect, the altermagnetic spin splitting effect, and the Rashba-Edelstein effect. The spin current generation layer may comprise, consist essentially of, or consist of a topological insulator material. Herein, consisting essentially of may mean the same as comprising substantially. The topological insulator material may comprise one or a combination of compounds selected from the group consisting of Bi2Ses, Bi2Te3, and Sb2Tes. The spacer layer may comprise, consist essentially of, or consist of a non-magnetic material. The non-magnetic material may comprise, consist essentially of, or consist of any one or a combination of elements selected from the group consisting of Cu, SI, Al, Mg and C. Preferably, the spacer layer comprises or consists of Al. The ferromagnetic thin-film nanomagnetic structures may comprise or consist of an alloy, the alloy comprising nickel and iron, optionally permalloy; and / or cobalt, iron, and boron, for example CoFeB; and / or yttrium, iron, and oxygen, for example yttrium iron garnet; and / or cobalt and platinum, for example CoPt. Preferably, the ferromagnetic thin-film nanomagnetic structures consist of permalloy. The spin current generation layer may be magnetic. The spacer layer may be nonmagnetic. Each ferromagnetic thin-film nanomagnetic structure may be magnetically coupled to the spin current generation layer. Each ferromagnetic thin-film nanomagnetic structure may have a respective main surface coated with an ionic material layer. Herein, the term main surface may mean a surface oriented parallel to a principal plane or major plane of a body to which it belongs. That is, the term main surface may mean an upper surface or a lower surface of a layer to which it belongs, as opposed to a side or edge surface of that layer which has a relatively small cross-sectional area. The spacer layer may be an ionic material layer. An ionic material layer may be interposed between each thin-film nanomagnetic structure and the spacer layer. Each ionic material layer may be wholly disposed on a respective ferromagnetic thin-film nanomagnetic structure. A first subset of the ferromagnetic thin-film nanomagnetic structures may be sized to have a first resonant frequency. A second subset of the ferromagnetic thin-film nanomagnetic structures may be sized to have a second resonant frequency which is different to the first resonant frequency. A resonant frequency of a ferromagnetic thin-film nanomagnetic structure may be equal to or greater than 0.1 GHz and / or equal to or less than 20 GHz. A resonant frequency of a ferromagnetic thin-film nanomagnetic structure may be between 0.1 GHz and 20 GHz. A resonant frequency of a ferromagnetic thin-film nanomagnetic structure may be equal to or greater than 1 GHz and / or equal to or less than 10 GHz. A resonant frequency of a ferromagnetic thin-film nanomagnetic structure may be between 1 GHz and 10 GHz. The ferromagnetic thin-film nanomagnetic structures may be arranged to form an artificial spin ice. The artificial spin ice may be a square artificial spin ice. The magnetic states can be non-volatile. The magnetic states may have history dependent dynamics. In other words, the magnetic states may have hysteresis. The ferromagnetic thin-film nanomagnetic structures may form a network having fading memory. The ferromagnetic thin-film nanomagnetic structures may be spaced apart from each other so as to be magnetically separated from each other. The ferromagnetic thin-film nanomagnetic structures may be magnetically isolated from each other. The ferromagnetic thin-film nanomagnetic structures may be not magnetically coupled to each other by dipolar coupling. The magnetic device may be implemented in a crossbar array. The crossbar array may comprise first wiring layers running along a first direction and second wiring layers running along a second direction. The first direction may be orthogonal to the second direction. The crossbar array may comprise first and second wiring layers taking the form of the spin current generation layer, and the ferromagnetic thin-film nanomagnetic structures may be disposed on the spin current generation layer at vertices of the first and second wiring layers. In this arrangement, the ferromagnetic thin-film nanomagnetic structures may be individually addressed by selecting which of the first and second wires of the first and second wiring layers to apply current to. Alternatively, the ferromagnetic thin-film nanomagnetic structures may be interposed between the first and second wiring layers at vertices. According to a second aspect of the present invention, there is provided a method of reading a state of the magnetic device of the first aspect. Reading the state of the device may comprise performing spin-torque ferromagnetic resonance measurements and / or performing ferromagnetic resonance measurements and / or performing magneto-optical Kerr effect measurements and / or performing electrical measurements such as magnetoresistance measurements, and / or performing Hall effect measurements. According to a third aspect of the present invention, there is provided a method of reading a state of the magnetic device of the first aspect. The method comprises performing spin torque ferromagnetic resonance measurements on the magnetic device, the spin torque magnetic resonance measurements indicating characteristic frequencies corresponding to spin wave modes in the magnetic device. The step of performing the spin torque ferromagnetic resonance measurements comprises driving an AC electrical current through the spin current generation layer of the magnetic device, exposing the magnetic device to a magnetic field and / or driving a DC electrical current through the spin current generation layer of the magnetic device, and sensing a rectified voltage across the spin current generation layer. The AC electrical current may have a constant amplitude. The AC electrical current may have a frequency of at least 1 GHz. The step of driving the AC electrical current through the spin current generation layer may comprise sweeping the frequency of the AC electrical current from a first frequency to a second frequency. The step of exposing the magnetic device to a magnetic field may comprise sweeping the magnetic field from a first magnetic field strength to a second magnetic field strength. The step of driving the DC electrical current through the spin current generation layer may comprise sweeping the DC electrical current from a first magnitude to a second magnitude. The step of performing the spin torque ferromagnetic measurements may be performed using an AC electrical current having a first constant amplitude, then again using an AC electrical current having a second constant amplitude. According to a fourth aspect of the present invention, there is provided a method of inputting data to the magnetic device of the first aspect by switching one or more of the magnetic states of the ferromagnetic thin-film nanomagnetic structures and / or exerting control over spin-wave dynamics of one or more of the magnetic states of the ferromagnetic thin-film nanomagnetic structures. According to a fifth aspect of the present invention, there is provided a method of inputting data to the magnetic device of the first aspect. The method comprises driving an AC electrical current through the spin current generation layer, the AC electrical current having a frequency at or near a resonant frequency of the one or more of the ferromagnetic thin-film nanomagnetic structures and an amplitude sufficient to induce ferromagnetic resonance in the one or more of the ferromagnetic thin-film nanomagnetic structures. The method further comprises driving a DC electrical current through the spin current generation layer to cause only the magnetic state(s) of the one or more of the ferromagnetic thin-film nanomagnetic structures to switch, or exposing the magnetic device to a magnetic field to cause only the magnetic state(s) of the one or more of the ferromagnetic thin-film nanomagnetic structures to switch. The magnetic field may be a DC magnetic field or a magnetic field having a time-varying amplitude. Data to be input to the ferromagnetic thin-film nanomagnetic structures may be encoded into the AC electrical current. Data to be input to the ferromagnetic thin-film nanomagnetic structures may be encoded into the frequency and amplitude of the AC electrical current. The AC electrical current may comprise a plurality of components each having a respective amplitude and frequency. According to a sixth aspect of the present invention, there is provided a method of inputting data to the magnetic device of the first aspect. The method comprises driving a DC electrical current through the spin current generation layer to cause only the magnetic state(s) of the one or more of the ferromagnetic thin-film nanomagnetic structures to switch, and / or exposing the magnetic device to a magnetic field to cause only the magnetic state(s) of the one or more of the ferromagnetic thin-film nanomagnetic structures to switch. According to a seventh aspect of the present invention, there is provided a method of operating the magnetic device of the first aspect. A spacer layer is interposed between each ferromagnetic thin-film nanomagnetic structure and the spin current generation layer and the spacer layer is an ionic material layer; or a spacer layer is interposed between each ferromagnetic thin-film nanomagnetic structure and the spin current generation layer and an ionic material layer is interposed between each thin-film nanomagnetic structure and the spacer layer; or each ionic material layer is wholly disposed on a respective ferromagnetic thin-film nanomagnetic structure. The ferromagnetic thin-film nanomagnetic structures are arranged to form an artificial spin ice. The method comprises applying a voltage across the ionic material layer(s) to cause ions to migrate from the Ionic material layer(s) Into one or more ferromagnetic thin-film nanomagnetic structures or to cause ions to migrate from one or more ferromagnetic thin-film nanomagnetic structures into the ionic material layer(s). According to an eighth aspect of the present invention, there is provided a method of operating the magnetic device of the first aspect. The method comprises the method of the fourth aspect, the method of the fifth aspect, or the method of the sixth aspect. The method further comprises the method of the second aspect or the method of the third aspect. The method may be a method of training the magnetic device of the first aspect. According to a ninth aspect of the present invention, there is provided a method of operating the magnetic device of the first aspect. The method comprises inputting data to the magnetic device by switching the magnetic state(s) of one or more of the ferromagnetic thin-film nanomagnetic structures and / or exerting control over spinwave dynamics of the magnetic state(s) of one or more of the ferromagnetic thin-film nanomagnetic structures, optionally by the method of the fifth aspect. The method further comprises reading a state of the magnetic device, optionally by the method of the third aspect. The method further comprises training the magnetic device. Training the magnetic device comprises determining whether a performance metric and / or a loss function and / or an error value of the device is improved for a given task after one or more magnetic states are switched. Training the magnetic device further comprises, in response to a positive determination of improvement of the performance metric or the loss function or the error value, keeping said one or more magnetic states switched, and / or in response to a negative determination of improvement of the performance metric or the loss function or the error value, switching back said one or more magnetic states. The method may be performed using an algorithm that is random or recursive, for example by using backpropagation or equilibrium propagation or by using a gradient descent technique or by using an evolutionary or genetic search algorithm and / or a regression algorithm including ridge or linear regression. According to a tenth aspect of the present invention, there is provided an integrated circuit comprising the magnetic device of the first aspect. The integrated circuit may further comprise an electrical current source. The electrical current source may be configured to drive an AC electrical current and / or a DC electrical current through the spin current generation layer. The integrated circuit may comprise a magnetic field source. The magnetic field source may be integral to the integrated circuit and configured to expose the magnetic device to a magnetic field. The integrated circuit may comprise a voltage sensor. The voltage sensor may be configured to sensing the rectified voltage across the spin current generation layer. The integrated circuit may further comprise a voltage source. The voltage source may be configured to apply the voltage across the ionic material layer(s). The integrated circuit may further comprise a controller. The controller may be configured to control one or more of the electrical current source, the magnetic field source integral to the integrated circuit, the voltage sensor, and the voltage source. The controller may be configured to encode data to be input to the ferromagnetic thin-film nanomagnetic structures into the AC electrical current. The controller may be configured to perform the step of training the magnetic device. The integrated circuit may be operable at a power of 10 W or less, 5 W or less, 2 W or less, 1 W or less, 500 mW or less, 200 mW or less, 100 mW or less, 50 mW or less, 20 mW or less, 10 mW or less, 5 mW or less, 2 mW or less, or 1 mW or less. The integrated circuit may be operable to perform time-series processing and / or forecasting of time series, where functionality is achieved by writing the magnetisations of the ferromagnetic thin-film nanomagnetic structures. The integrated circuit may be operable to perform recurrent neural network computation. The integrated circuit may be for neuromorphic computing, and / or inference-based computing, and / or logic-based computing (for example, for use as a logic gate). According to an eleventh aspect of the present invention, there is provided an apparatus comprising the integrated circuit of the tenth aspect and an external magnetic field source. The external magnetic field source may be external to the integrated circuit and configured to expose the magnetic device to a magnetic field. The apparatus may further comprise an external controller configured to control the external magnetic field source. According to a twelfth aspect of the present invention, there is provided a system comprising the integrated circuit of the tenth aspect or the apparatus of the eleventh aspect. The system further comprises an off-grid power source configured to power the integrated circuit or the apparatus. The off-grid power source may be a battery, a solar panel, or the like. The off-grid power source may not be connected to an established power distribution grid. The off-grid power source may be portable. According to a thirteenth aspect of the present invention, there is provided a method of fabricating the magnetic device of the first aspect. The method comprises providing a body having a planar main surface on which the magnetic device is to be fabricated. The method further comprises providing a lithographically patterned layer of resist on the planar main surface. The lithographically patterned layer of resist exposes areas of the planar main surface on which material is to be deposited. The lithographically patterned layer of resist has an undercut profile. The method further comprises, while rotating the body, depositing a first one or more layers including the spin current generation layer. The method further comprises, after depositing the spin current generation layer and while not rotating the body, depositing a second one or more layers including the ferromagnetic thin-film nanomagnetic structures on the first one or more layers. The method may further comprise, after the deposition steps are complete, removing the patterned layer of resist and any layers deposited thereon. The first one or more layers may include the spacer layer. The ionic material layer may be included in either the first one or more layers or the second one or more layers. The one or more gate electrodes may be deposited on the magnetic device in a subsequent lithography step. Brief Description of the Drawings Certain embodiments of the present invention will now be described, by way of example, with reference to the accompanying drawings in which: Figure 1 is a schematic perspective view of a first magnetic device; Figure 2A is a schematic side view of a first magnetic device showing control of one or more magnetic states; Figure 2B is a schematic side view of a modified version of the first magnetic device having an interlayer interposed between the spin current generation layer and a ferromagnetic thin-film nanomagnetic structure showing control of one or more magnetic states; Figure 3 is a schematic plan view of a second magnetic device; Figure 4A is a schematic cross-section along the line A-A' in Figure 3; Figure 4B is a schematic cross-section along the line B-B' in Figure 3; Figure 4C is a schematic cross-section along the line C-C' in Figure 3; Figure 5 is a schematic plan view of a third magnetic device; Figure 6A is a schematic cross-section along the line D-D' in Figure 5; Figure 6B is a schematic cross-section along the line E-E' in Figure 5; Figure 6C is a schematic cross-section along the line F-F' in Figure 5; Figure 7 is a scanning electron microscopy micrograph showing a schematic plan view of a fourth magnetic device; Figure 8 is a schematic plan view of a fourth magnetic device; Figure 9A is a schematic cross-section along the line G-G' In Figure 8; Figure 9B is a schematic cross-section along the line G-G' for a modified version of the fourth magnetic device; Figure 9C is a schematic cross-section along the line G-G' for a differently modified version of the fourth magnetic device; Figure 10 is a schematic perspective view of a sixth magnetic device having differently sized ferromagnetic thin-film nanomagnetic structures; Figure 11 is a scanning electron microscopy micrograph showing a schematic plan view of a seventh magnetic device having differently sized ferromagnetic thin-film nanomagnetic structures; Figure 12 is a schematic plan view of an eighth magnetic device implemented in a crossbar array; Figure 13 is a schematic plan view of a ninth magnetic device implemented in a crossbar array; Figure 14 is a process flow diagram of a method of performing spin torque ferromagnetic resonance measurements for a magnetic device; Figure 15 is a process flow diagram of a method of inputting data to a magnetic device; Figure 16A is a process flow diagram of a method of operating a magnetic device; Figure 16B is a process flow diagram of a method of training a magnetic device; Figure 17 is a schematic block diagram of an integrated circuit comprising a magnetic device; Figure 18 is a schematic block diagram of an apparatus comprising the integrated circuit of Figure 17; Figure 19 is a schematic block diagram of a system comprising the integrated circuit of Figure 17 or the apparatus of Figure 18; Figure 20A is a schematic cross-section showing rotational deposition; Figure 20B is a schematic cross-section showing static deposition; Figure 21 is a process flow diagram of a method of fabricating a magnetic device; Figure 22 is a spin-torque ferromagnetic resonance spectra of a magnetic device having a 20 pm x 20 pm array of ferromagnetic thin-film nanomagnetic structures; Figure 23A is a scanning electron microscopy micrograph showing a schematic plan view of a magnetic device having a connected chevron array of ferromagnetic thin-film nanomagnetic structures; Figure 23B is a spin-torque ferromagnetic resonance spectra of the magnetic device shown in Figure 23A as a function of magnetic field at an AC electrical current power of 20 dBm showing two switching events labelled Hei and HC2 respectively; Figure 23C Is a spin-torque ferromagnetic resonance spectra of the magnetic device shown in Figure 23A as a function of magnetic field at an AC electrical current power of 26 dBm showing two switching events labelled Hei and HC2 respectively; and Figure 23D is experimental data showing the dependence of Hcz on AC electrical current frequency and showing that when a 26 dBm AC electrical current is applied at a single frequency for around 0.1 s, HC2 reduces from 35 mT to 18 mT with reductions corresponding to resonance frequencies. Detailed Description of Certain Embodiments In the following, like parts are denoted by like reference numerals. Introduction Herein, magnetic devices are described. The magnetic devices include a spin current generation layer on which ferromagnetic thin-film nanomagnetic structures are disposed. The spin current generation layer can allow for all-electrical data input and data readout from all the ferromagnetic thin-film nanomagnetic structures, and thus negate any need to provide individual electrical connections to each of the ferromagnetic thin-film nanomagnetic structures. In this way, the spin current generation layer can render the magnetic devices less complex to operate and more scalable than other magnetic devices implemented with memristors or spin torque oscillators, since in those other magnetic devices memristors and spin torque oscillators require individual electrical connections to be selectively or individually operable. The magnetic devices can have fading memory functionality and thus be capable of relatively computationally demanding tasks. The magnetic devices can be operable at sufficiently low powers to allow for their use in edge artificial intelligence (AI) chips. First magnetic device 1, li Referring to Figure 1, a first magnetic device 1, li is shown. The first magnetic device 1, li includes ferromagnetic thin-film nanomagnetic structures 2 spaced apart from each other and each having a respective magnetic state, and a spin current generation layer 3 having a surface 4 on which the ferromagnetic thin-film nanomagnetic structures 2 are disposed. The spin current generation layer 3 can convert an electrical current jc driven through the spin current generation layer 3 parallel to the surface 4 into a spin current js. In other words, in response to receiving the electrical current jc, the spin current generation layer 3 can generate the spin current js in the spin current generation layer 3. The spin current js can control the magnetic state(s) of one or more of the ferromagnetic thin-film nanomagnetic structures 2. Processes by which the spin current js can control the magnetic state(s) of one or more of the ferromagnetic thin-film nanomagnetic structures 2 are described in more detail hereinafter with reference to Figures 2A and 2B. As all the ferromagnetic thin-film nanomagnetic structures 2 are disposed on the surface 4 of the spin current generation layer 3, the spin current generation layer 2 can provide a way to modify all the magnetic states of the ferromagnetic thin-film magnetic structures. Thus, the first magnetic device 1, li can support all-electrical data input and data readout from the ferromagnetic thin-film nanomagnetic structures 2. Moreover, in this way, a need to individually contact each of the ferromagnetic thin-film nanomagnetic structures 2 to address the ferromagnetic thin-film nanomagnetic structures 2 can be negated. Thus, the first magnetic device 1, li can be less complex to operate and more scalable than other magnetic devices which require such individual contacting, for example, other magnetic devices based on memristors or spin-torque oscillators. In the first magnetic device 1, li, the spin current generation layer 3 is disposed on a main surface 5 of a substrate 6. The substrate 6 may take the form of a wafer. The wafer may be a single-crystal wafer formed from silicon, silicon dioxide, or another suitable material. Alternatively, the wafer may be a silicon-on-insulator (SOI) wafer or a silicon-on-sapphire (SoS) wafer or other, similar type of wafer. The spin current generation 3 need not, however, be disposed on the main surface 5 of a substrate 6. Instead, the ferromagnetic thin-film nanomagnetic structures 2 may be supported within a volume, or the spin current generation layer 3 may be a self-supporting layer. In the first magnetic device 1, li, the ferromagnetic thin-film nanomagnetic structures 2 are arranged to define an array. This need not be the case, however, and the ferromagnetic thin-film nanomagnetic structures 2 may be arranged randomly, that Is, without an underlying order or uniform arrangement. Though the array is defined by the arrangement of ferromagnetic thin-film nanomagnetic structures 2, it is to be appreciated that not every site (alternatively point) in the array is required to be occupied by a ferromagnetic thin-film nanomagnetic structure 2 having a switchable magnetic state. For example, at some sites in the array there may instead be ferromagnetic thin-film nanomagnetic structure(s) 2 having a different type of magnetic ordering, or ferromagnetic thin-film nanomagnetic structure(s) 2 in which the magnetic state is 'pinned' and not switchable into other energetically stable state(s), or indeed no ferromagnetic thin-film nanomagnetic structure(s) 2 having a switchable magnetic state at all. In the first magnetic device 1, li, the array is ordered. The array need not be ordered, however, and the ferromagnetic-film nanomagnetic structures 2 may instead define an array which is disordered. Although Figure 1 shows the ferromagnetic thin-film nanomagnetic structures 2 defining an ordered array according to a square lattice with a motif including one ferromagnetic thin-film nanomagnetic structure 2, the array may be according to any one of the five 2D Bravais lattices, with a motif including one or more ferromagnetic thin-film nanomagnetic structures 2. Likewise, although Figure 1 shows the ferromagnetic thin-film nanomagnetic structures 2 being square shaped, the ferromagnetic thin-film nanomagnetic structures 2 may have any suitable shape. In particular, the ferromagnetic thin-film structures 2 can be bar shaped, optionally with rounded corners or rounded ends. The ferromagnetic thin-film nanomagnetic structures 2 may be arranged to form an artificial spin ice. Artificial spin ices are composed of interacting magnetic dipole moments located on frame structures designed to generate frustrated spin systems. Frame structures may take the form of the ferromagnetic thin-film nanomagnetic structures 2. The artificial spin Ice may be according to any one of the five 2D Bravais lattices, with a motif including one or more ferromagnetic thin-film nanomagnetic structures. In particular, the artificial spin ice may be a square artificial spin ice. Alternatively, the artificial spin ice may be a pinwheel artificial spin ice, a Kagome artificial spin ice, or a Penrose artificial spin ice. Magnetic state control mechanisms The spin current js can control the magnetic state(s) of one or more of the ferromagnetic thin-film nanomagnetic structures 2 by various processes. Referring to Figure 2A, a first process by which the spin current js can control the magnetic state(s) of one or more of the ferromagnetic thin-film nanomagnetic structures 2 is shown. In the first process, the spin current js is generated transverse to the direction of the electrical current jc, for example by the spin Hall effect, so as to be incident on the ferromagnetic thin-film nanomagnetic structures 2 and exert a spin torque on the ferromagnetic thin-film nanomagnetic structures 2. Referring to Figure 2B, a second process by which the spin current js can control the magnetic state(s) of one or more of the ferromagnetic thin-film nanomagnetic structures 2 is shown. The second process relies on the spin current generation layer 3 being ferromagnetic and magnetically coupled to the ferromagnetic thin-film nanomagnetic structures 2, namely by dipolar coupling. In the second process, the spin current js is generated and exerts a spin torque on a magnetisation Mscgl-fm of the spin current generation layer 3 which, in turn, exerts a torque on a magnetisation Mftns of each ferromagnetic thin-film nanomagnetic structure 2 due to the magnetic coupling. By the ferromagnetic thin-film nanomagnetic structures 2 being magnetically coupled to the spin current generation layer 3, that is, by the spin current generation layer 3 being influenced by dipolar fields from the ferromagnetic thin-film nanomagnetic structures 2, the magnetic state of the spin current generation layer 3 can be modified. Spin-torque effects can be used to read these modifications and to infer the magnetic states of the ferromagnetic thin-film nanomagnetic structures 2 without requiring any spin current to be incident on the ferromagnetic thin-film nanomagnetic structures 2. Although Figure 2B shows a ferromagnetic thin-film nanomagnetic structure 2 being spaced apart from the spin current generation layer 3 by an interlayer 7, this need not be the case and the thin-film nanomagnetic structures 2 may be directly disposed on the spin current generation layer 3. The interlayer 7 may be non-magnetic. The spin current js can control the magnetic state(s) of one or more of the ferromagnetic thin-film nanomagnetic structures 2 by exerting a spin torque. The spin torque may be a spin orbit torque, and may be given rise to by any one or combination of various effects. In particular, by one or more of the spin Hall effect, the giant spin Hall effect, the magnetisation-dependent spin Hall effect, the anomalous spin Hall effect, the altermagnetic spin splitting effect, and the Rashba-Edelstein effect. Although Figures 2A and 2B depict generation of a spin current js which is a transverse spin current, the spin current js need not necessarily be a transverse spin current. For example, spin accumulation effects where certain spin orientations are caused to accumulate at the surface 4 can give rise to switching of the magnetic states of the ferromagnetic thin-film nanomagnetic structures 2. In other words, in the case that spins oriented in the opposite direction to a given ferromagnetic thin-film nanomagnetic structure 2 accumulate at a surface of that ferromagnetic thin-film nanomagnetic structure 2, the spin torque resulting from that accumulation can be sufficient to switch that ferromagnetic thin-film nanomagnetic structure 2. Materials A variety of materials can be used to form the spin current generation layer 3. The spin current generation layer 3 may comprise, consist essentially of, or consist of a heavy metal material. A heavy metal material is a material which consists essentially or consists of any one or a combination of elements selected from the group consisting of: Ta, Pt, Pd, Au, Bi, Zn, Hf, Os, Pb, Tl, W, Tb, Gd, Ho, Dy, Pr, Sm, Nd, V, Ti, Te, and Ir. Preferably, the heavy metal material consists essentially or consists of any one or a combination of elements selected from the group consisting of: W, Ta, Pt, and Au. More preferably, the heavy metal material consists essentially or consists of Pt. The spin current generation layer 3 may comprise, consist essentially of, or consist of a magnetic material. The magnetic material may have planar anisotropy, for example, the magnetic material may be a NIFe alloy or a CoFeB alloy. The magnetic material may have perpendicular anisotropy, for example, the magnetic material may be a CoPt alloy, or an FePt alloy, or an FePd alloy, or a Co / Ni stack, or a CO / AI2O3 stack. The magnetic material may consist essentially or consist of one or more materials that are not heavy metal materials. The magnetic material may be selected from the group consisting of Fe, Co, Ni, FeCo, FeCoB, NiFe, NdFeB, WCoFeB, TaCoFeB. Preferably, the magnetic material consists essentially or consists of NiFe. The spin current generation layer 3 may be a spin-Hall material layer, that is, a layer formed from a material which exhibits the spin-Hall effect either by virtue of having high spin orbit coupling or by virtue of having a non-trivial band structure topology. The spin current generation layer 3 need not, however, be magnetic, and may instead be non-magnetic. The spin current generation layer 3 may comprise, consist essentially of, or consist of a topological insulator material. The topological insulator material may comprise one or a combination of compounds selected from the group consisting of: Bi2Ses, Bi2Te3, and Sb2Tes. The ferromagnetic thin-film nanomagnetic structures 2 may comprise, consist essentially of, or consist of an alloy, the alloy comprising: nickel and iron, optionally permalloy; and / or cobalt, iron, and boron, for example CoFeB; and / or yttrium, iron, and oxygen, for example yttrium iron garnet; and / or cobalt and platinum, for example CoPt. Preferably, the ferromagnetic thin-film nanomagnetic structures consist of permalloy. Alternatively, the ferromagnetic thin-film nanomagnetic structures 2 may comprise, consist essentially of, or consist of one or more other suitable ferromagnetic materials. The ferromagnetic thin-film nanomagnetic structures 2 may be electrically conductive or electrically insulating. Second magnetic device 1, I2 Referring to Figure 3, a second magnetic device 1, I2 is shown. The second magnetic device 1, I2 is similar to the first magnetic device 1, li in that it comprises the same layers arranged in the same order in the out-of-plane (or z) direction, but is different to the first magnetic device 1, li, in that its ferromagnetic thin-film nanomagnetic structures 2 are bar shaped and arranged to form a square artificial spin ice. The second magnetic device 1, 12 is also different to the first magnetic device 1, li, in that its spin current generation layer 3 is patterned to include voids in areas adjacent to the ferromagnetic thin-film nanomagnetic structures 2. By the volume of the spin current generation layer being reduced in this way, the current required for allelectrical data input and readout can be reduced. The ferromagnetic thin-film nanomagnetic structures 2 are wholly disposed on the spin current generation layer 3. By the whole area of the ferromagnetic thin-film nanomagnetic structures 2 being disposed on the spin current generation layer 3, the spin current generation layer 3 can more effectively control the magnetic states of the ferromagnetic thin-film nanomagnetic structures 2. Additionally, due to the ferromagnetic thin-film nanomagnetic structures 2 not extending over edges of the spin current layer generation 3, distortion in the magnetisation texture of the ferromagnetic thin-film nanomagnetic structures 2 can be avoided. The ferromagnetic thin-film nanomagnetic structures 2 need not be wholly disposed on the spin current generation layer 3, however, and may instead extend over edges of the spin current generation layer 3. Referring also to Figures 4A to 4C, portions of the second magnetic device 1, lz are shown. Figure 4A is a schematic cross-section along the line A-A' in Figure 3. The line A-A' spans the extent of one ferromagnetic thin-film nanomagnetic structure 2, disposed on the spin current generation layer 3 which is in turn disposed on the substrate 6. Figure 4B is a schematic cross-section along the line B-B' in Figure 3. The line B-B' spans a distance which traverses one ferromagnetic thin-film nanomagnetic structure 2 disposed on the spin current generation layer 3 which is in turn disposed on the substrate 6, and also spans a void 5 in the spin current generation layer 3 adjacent to the one ferromagnetic thin-film nanomagnetic structure 2. Figure 4C is a schematic cross-section along the line C-C' in Figure 3. The line C-C' spans a distance which does not traverse any void 5 or ferromagnetic thin-film nanomagnetic structure 2, and instead just shows the spin current generation layer 3 disposed on the substrate 6. Third magnetic device 1, la Referring to Figure 5, a third magnetic device 1, I3 is shown. The third magnetic device 1, I3 is similar to the second magnetic device 1, I2 in that it comprises the same layers arranged in the same order in the z direction, but it is different to the second magnetic device 1, I2 in that it also includes a spacer layer 9 interposed between each ferromagnetic thin-film nanomagnetic structure 2 and the spin current generation layer 3, in other words spacing apart each ferromagnetic thin-film nanomagnetic structure 2 from the spin current generation layer 3. Although Figure 5 shows each ferromagnetic thin-film nanomagnetic structure 2 as being separated from the spin current generation layer 3 by the same spacer layer 9, this need not be the case and each ferromagnetic thin-film nanomagnetic structure 2 may be spaced apart from the spin current generation layer 3 by a respective spacer layer 9 instead. Each ferromagnetic thin-film nanomagnetic structure 2 as being separated from the spin current generation layer 3 by the same spacer layer 9 can help to simplify fabrication of the third magnetic device 1, I3. The spacer layer 9 may be patterned to include voids in areas adjacent to the ferromagnetic thin-film nanomagnetic structures 2. By the volume of the spacer layer 9 being reduced in this way, the spin current js can be effectively conducted from the spin generation layer 3 to the ferromagnetic thin-film nanomagnetic structures 2. This need not be the case, however, and the spacer layer 9 may instead take the form of a continuous sheet. In the third magnetic device 1, I3, each of the ferromagnetic thin-film nanomagnetic structures 2 is wholly disposed on the spacer layer 9. By the whole area of the ferromagnetic thin-film nanomagnetic structures 2 being disposed on the spacer layer, the spin current generation layer 3 can more effectively control the magnetic state(s) of one or more of the ferromagnetic thin-film nanomagnetic structures 2. Additionally, due to the ferromagnetic thin-film nanomagnetic structures 2 not extending over edges of the spacer layer 9, distortion in the magnetisation texture of the ferromagnetic thin-film nanomagnetic structures 2 can be avoided. The ferromagnetic thin-film nanomagnetic structures 2 need not be wholly disposed on the spacer layer 9, however, and may instead extend over edges of the spacer layer 9. The spacer layer 9 may be formed from an electrically conducting material for conducting the spin current to the ferromagnetic thin-film nanomagnetic structures 2. Alternatively, the spacer layer 9 may be formed from an electrically insulating material. The spacer layer 9 may comprise, consist essentially of, or consist of a non-magnetic material. The non-magnetic material may comprise, consist essentially of, or consist of copper, silicon, aluminium, magnesium, or carbon. The non-magnetic material may comprise of any one or a combination of these elements. The non-magnetic material may comprise mostly of one of these elements, for example, the non-magnetic material may comprise greater than or equal to 50% of aluminium. Preferably, the spacer layer 9 comprises or consists of Al. Referring also to Figures 6A to 6C, portions of the third magnetic device 1, Is are shown. Figure 6A is a schematic cross-section along the line D-D' in Figure 5. The line D-D' spans the extent of one ferromagnetic thin-film nanomagnetic structure 2, disposed on the spacer layer 9 which is in turn disposed on the spin current generation layer 3 which is in turn disposed on the substrate 6. Figure 6B is a schematic cross-section along the line E-E' in Figure 5. The line E-E' spans a distance which traverses one ferromagnetic thin-film nanomagnetic structure 2 disposed on the spacer layer 9 which is in turn the spin current generation layer 3 which is in turn disposed on the substrate 6, and also spans a void 5 in the spin current generation layer 3 adjacent to the one ferromagnetic thin-film nanomagnetic structure 2. Figure 6C is a schematic cross-section along the line F-F' in Figure 5. The line F-F' spans a distance which does not traverse any void 5 or ferromagnetic thin-film nanomagnetic structure 2, and instead just shows the spacer layer 9 disposed on the spin current generation layer 3 which is in turn disposed on the substrate 6. Fourth magnetic device 1, 14 Referring to Figure 7, a fourth magnetic device 1, 14 is shown. The fourth magnetic device 1, 14 is similar to the third magnetic device in that it comprises the same features in a similar arrangement, but is different from the third magnetic device 1, 13 in that in that its ferromagnetic thin-film nanomagnetic structures 2 are bar shaped with rounded ends. In the fourth magnetic device 1, I4, the spin current generation layer 3 is magnetic, the spacer layer 9 is entirely disposed on the spin current generation layer 3 and nonmagnetic, and each ferromagnetic thin-film nanomagnetic structure 2 is magnetically coupled to the spin current generation layer 3. By each ferromagnetic thin-film nanomagnetic structure 2 being able to magnetically couple across the spacer layer 9 to the spin current generation layer 3, the microstate space available for the magnetic states can be Increased. This can help to provide enhanced collective dynamics useful for neuromorphic memcomputing (also referred to herein as "neuromorphic inmemory computing"). Fifth magnetic device 1, Is Referring to Figure 8, a fifth magnetic device 1, Is is shown. The fifth magnetic device 1, Is is similar to the third magnetic device 1, I3 in that it comprises the same layers arranged in the same order in the z direction, but is different to the third magnetic device 1, I3 in that it also includes an ionic material layer 20 interposed between the spin current generation layer 3 and each ferromagnetic thin-film nanomagnetic structure 2, and gate electrodes 21 electrically contacting the ferromagnetic thin-film nanomagnetic structures. More precisely, each of the ferromagnetic thin-film nanomagnetic structures 2 has a respective main surface coated with the ionic material layer 20. The ionic material layer 20 can, in response to a gate voltage being applied between the spin current generation layer 3 and a gate electrode 21 disposed on one or more of the ferromagnetic thin-film nanomagnetic structures 2, cause ions to migrate from the ionic material layer 20 into the ferromagnetic thin-film nanomagnetic structures 2 or vice versa. By applying the gate voltage in this way, the magnetic anisotropy and magnetic state of said one or more of the ferromagnetic thin-film nanomagnetic structures 2 can be controllably changed 2. Although Figure 8 shows gate electrodes 21 patterned in parallel strips, the one or more gate electrodes 21 may be differently patterned, or indeed not be patterned at all. Referring also to Figure 9A, a portion of the fifth magnetic device 1, Is is shown. Figure 9A is a schematic cross-section along the line G-G' in Figure 8. The line G-G' spans the extent of one ferromagnetic thin-film nanomagnetic structure 2, disposed on the spacer layer 9 which is in turn disposed on the spin current generation layer 3 which is in turn disposed on the substrate 6. A gate electrode 21 is disposed on part of the one ferromagnetic thin-film nanomagnetic structure 2. Although Figures 8 and 9A show the ionic material layer 20 as directly disposed on the spin current generation layer 3 and interposed between the spin current generation layer 3 and the ferromagnetic thin-film nanomagnetic structures 2, this need not be the case as will be described hereinafter in more detail with reference to Figures 9B and 9C. Preferably, in the fifth magnetic device 1, Is shown in Figures 8 and 9A, the spin current generation layer 3 is formed from a magnetic material such as permalloy. Referring also to Figures 9B and 9C, portions of modified versions of the fifth magnetic device 1, Is are shown. In the modified version of the fifth magnetic device 1, Is- shown in Figure 9B, a respective ionic material layer 20 is interposed between each ferromagnetic thin-film nanomagnetic structure 2 and a gate electrode 21. In particular, Figure 9B shows a gate electrode 21 disposed on an ionic material layer 20 which is in turn disposed on a ferromagnetic thin-film nanomagnetic structure 2 which is in turn disposed on a spin current generation layer 3 which is in turn disposed on a substrate 6. Preferably, In the modified version of the fifth magnetic device 1, Is- shown in Figure 9B, the spin current generation layer 3 is formed from a heavy metal material. In the modified version of the fifth magnetic device 1, Is- shown in Figure 9C, a respective ionic material layer 20 is interposed between each ferromagnetic thin-film nanomagnetic structure 2 and a gate electrode 21, and a spacer layer 9 is interposed between each ferromagnetic thin-film nanomagnetic structure 2 and the spin current generation layer 3. In particular, Figure 9C shows a gate electrode 21 disposed on an ionic material layer 20 which is In turn disposed on a ferromagnetic thin-film nanomagnetic structure 2 which is in turn disposed on a spacer layer 9 which is in turn disposed on a spin current generation layer 3 which is in turn disposed on a substrate 6. Preferably, in the modified version of the fifth magnetic device 1, Is shown in Figure 9C, the spin current generation layer 3 is formed from a magnetic material such as permalloy, and the spacer layer 9 may be formed from a non-magnetic material (in other words, the spacer layer 9 may be non-magnetic). In the fifth magnetic device 1, Is or any one of the modified versions thereof described hereinbefore, the ionic material layer(s) 20 can be driven by a voltage applied between the spin current generation layer 3 and one or more gate electrodes 21 to cause ions to migrate from the ionic material layer 20 into the ferromagnetic thin-film nanomagnetic structures 2 or vice versa, and thus can controllably change, with a degree of local selectivity, the magnetic anisotropy and magnetic state of one or more of the ferromagnetic thin-film nanomagnetic structures 2. The ionic material layer 20 comprises, consists essentially of, or consists of a material which allows ion migration. The material which allows ion migration may be a Gd-based material such as GdOx, a LiPON-based material, or a Hf-based material. The fifth magnetic device 1, Is or any one of the modified versions thereof described hereinbefore can be operated by applying, between the spin current generation layer 3 and one or more gate electrodes 21, a voltage across the ionic material layer(s) 20 to cause ions to migrate from the ionic material layer(s) 20 into one or more ferromagnetic thin-film nanomagnetic structures 2 or from one or more ferromagnetic thin-film nanomagnetic structures 2 into the ionic material layer(s) 20. In this way, the magnetic anisotropy and magnetic state of said one or more of the ferromagnetic thin-film nanomagnetic structures 2 can be changed. A readout of the magnetic device may be used as an input to ionic material layer(s) 20, and in this way a 'self-training' system can be made. An output of the magnetic device may be used as an input to a voltage applied to gate electrode(s) 21 electrically connected to the ionic material layer(s) 20, and in this way a feedback loop can be created. Sixth magnetic device 1, le Referring to Figure 10, a sixth magnetic device 1, le is shown. The sixth magnetic device 1, le is similar to the first magnetic device 1, li in that it comprises the same layers arranged in the same order in the z direction, but is different to the first magnetic device 1, li, in that it has differently sized ferromagnetic thin-film nanomagnetic structures 2a, 2b, 2C. In particular, the sixth magnetic device 1, le has a first type of ferromagnetic thin-film nanomagnetic structures 2a having a first length and first width, a second type of ferromagnetic thin-film nanomagnetic structures 2b having a second length greater than the first length and a second width equal to the first width, and a third type of ferromagnetic thin-film nanomagnetic structures 2c having a third length equal to the first length and a third width less than the first width. There is no need, however, for different types of ferromagnetic thin-film nanomagnetic structures 2 (that is, different subsets of the ferromagnetic thin-film nanomagnetic structures 2) to have equal length or equal width. Rather, different types of ferromagnetic thin-film nanomagnetic structures 2, 2a, 2b, 2c may have different lengths, different widths, and different thicknesses, and may be formed from different materials. Seventh magnetic device 1, 1? Referring to Figure 11, a seventh magnetic device 1, I7 is shown. The seventh magnetic device 1, I7 is a specific example of the sixth magnetic device 1, 16 in which ferromagnetic thin-film nanomagnetic structures 2 are arranged in an alternating pattern of rows of ferromagnetic thin-film nanomagnetic structures 2a with a first width and rows of ferromagnetic thin-film nanomagnetic structures 2C with equal length but different width. The first subset of the ferromagnetic thin-film nanomagnetic structures 2a are sized to have a first resonant frequency, and the second subset of the ferromagnetic thin-film nanomagnetic structures 2c are sized to have a second resonant frequency which is different to the first resonant frequency. By information, such as time series information, being encoded in the power and frequency of an AC electrical current driven through the spin current generation layer 3, selective writing to the ferromagnetic thin-film nanomagnetic structures 2a, 2c can be performed. The information may be multidimensional information. Neuromorphic memcomputing In each of the first to seventh magnetic devices 1, li, I2, I3, I4, Is, le, I7 and modified versions thereof described herein, input data taking the form of field(s) or signal(s) provided to the magnetic device 1, li, I2, Is, I4, Is, le, I7 can correspond to an input layer of a physical neural network, the ferromagnetic thin-film nanomagnetic structures 2 can correspond to one or more hidden layers of the physical neural network, and output data taking the form of the measurable response(s) of the magnetic states can correspond to an output layer of a physical neural network. The magnetic device 1, li, I2, I3, I4, Is, 16, I7 can include intrinsic non-linearities and be suitable for use as, for example, a feed-forward neural network, or a recurrent neural network, or a convolutional neural network, or a trainable deep neural network, or a reservoir computing scheme. The magnetic device 1, li, 12, I3, I4, Is, le, I7 may be for use as a physical neural network. The magnetic device 1, li, I2, I3, I4, Is, le, I7 may comprise a plurality of artificial neurons, each artificial neuron comprising one or more of the ferromagnetic thin-film nanomagnetic structures 2. Pairs of artificial neurons may be configured to be coupled by dipolar coupling so as to define artificial synapses, each artificial synapse having a trainable synaptic weight that is responsive to switching of the magnetic states or modification of spin-wave dynamics of one or more of the magnetic states. Some or all of the artificial neurons may comprise only one ferromagnetic thin-film nanomagnetic structure 2, said ferromagnetic thin-film nanomagnetic structure 2 having a switchable magnetic state. Some or all of the artificial neurons may comprise two or more ferromagnetic thin-film nanomagnetic structures 2. In the case that an artificial neuron comprises two or more ferromagnetic thin-film nanomagnetic structures 2, either some of the magnetic states may be pinned (that is, not switchable) or all of the magnetic states may be switchable. The magnetic states are non-volatile. The magnetic states have history dependent dynamics. In other words, the magnetic states have hysteresis. The ferromagnetic thin-film nanomagnetic structures 2 can form a network having fading memory. Fading memory means that the response to a certain input depends on previous inputs. Fading memory is a property required for predictive tasks such as blood hormone prediction, robotic gesture movement and predicting the environment. Training a software network to have the same functionality can be highly challenging and power intensive, whereas this functionality is available for free via the physics of the magnetic device 1, li, I2, I3, I4, Is, le, I7. The magnetic device 1, li, I2, I3, I4, Is, le, 1? can have strong coupling and communication between adjacent ferromagnetic thin-film nanomagnetic structures, linear and non-linear responses to external data input, and inherent data storage capabilities. The magnetic device 1, li, I2, Is, I4, Is, le, I7 can thus be used to perform neuromorphic in-memory computing. The magnetic device 1, li, I2, 13, 14, Is, le, 17 can be reconfigured for various neuromorphic architectures from simple, non-trainable 'reservoir computers' for timeseries processing, to fully trainable deep-neural networks for inference. Eighth magnetic device 1. la Referring to Figure 12, an eighth magnetic device 1, la is shown. The eighth magnetic device 1, Is is similar to the first magnetic device 1, li in that it comprises the same layers arranged in the same order in the z direction, but is different to the first magnetic device 1, li, in that it is implemented in a crossbar array architecture. The crossbar array architecture includes a first wiring layer having first wires 22i, 22i+i,...,22m running along a first direction x and a second wiring layer having second wires 22j, 22j+i,...22n running along a second direction y which is different to the first direction x, optionally orthogonal to the first direction x. The first wires 22,, 22i+i,...,22m and the second wires 22j, 22j+i,...22n are formed by the spin current generation layer 3, and the ferromagnetic thin-film nanomagnetic structures 2 are disposed on the spin current generation layer 3 at vertices of the first and second wiring layers. In this arrangement, the ferromagnetic thin-film nanomagnetic structures 2 may be individually addressed by selecting which of the first and second wires of the first and second wiring layers to apply current to. The ferromagnetic thin-film nanomagnetic structures 2 may be spaced apart from each other so as to be magnetically separated from each other. In other words, separated from each other so as to not be coupled by dipolar coupling, or magnetically isolated from each other. By magnetically separating the ferromagnetic thin-film nanomagnetic structures 2 from each other, the magnetic stability of the device 1 can be improved. This can be useful when using the magnetic device 1, optionally the eighth magnetic device 1, Is, for data storage. Ninth magnetic device 1, 19 Referring to Figure 13, a ninth magnetic device 1, I9 is shown. The ninth magnetic device 1, 1g is similar to the eighth magnetic device 1, Is in that it comprises the same layers and is implemented in a crossbar array architecture, but is different from the eighth magnetic device 1, Is in that its ferromagnetic thin-film nanomagnetic structures 2 are interposed between the first and second wiring layers at vertices instead of disposed on both the first and second wiring layers. Reading a state of the magnetic device 1 A variety of approaches can be used to read out a state of the magnetic device 1, li, I2, Is, I4, Is, Is, I7, Is, I9. In particular, approaches based on performing one or more of spin-torque ferromagnetic resonance measurements, ferromagnetic resonance measurements, magneto-optical Kerr effect measurements, electrical measurements such as magnetoresistance measurements, and Hall effect measurements, can be used. Referring to Figure 14, a method of reading a state of the magnetic device 1, li, 12, I3, I4, Is, le, I7, Is, I9 (hereinafter referred to as the "first method") now be described. The first method includes performing spin torque ferromagnetic resonance measurements for the magnetic device 1, li, I2, I3, I4, Is, 16, I7, Is, I9, the spin torque magnetic resonance measurements indicating characteristic frequencies corresponding to spin wave modes in the magnetic device 1, li, I2, I3, I4, Is, le, I7, Is, I9. An AC electrical current is driven through the spin current generation layer 3 of the magnetic device 1, li, I2, I3, I4, Is, le, I7, Is, I9 (step 1.1). The magnetic device 1, li, I2, Is, I4, Is, le, I7, Is, I9 is exposed to a magnetic field and / or a DC electrical current is driven through the spin current generation layer 3 of the magnetic device 1, li, I2, I3, I4, Is, le, I7, Is, I9 (step 1.2). A rectified voltage across the spin current generation layer 3 is sensed (step 1.3). As a result of the magnetic device 1, li, I2, 13, 14, Is, le, 17, Is, 19 including the spin current generation layer 3 and all the ferromagnetic thin-film nanomagnetic structures 2 being disposed on the spin current generation layer 3, all-electrical data readout from all the ferromagnetic thin-film nanomagnetic structures 2 can be performed using the first method. The AC electrical current may have a constant amplitude and / or a frequency of at least 1 GHz. Driving the AC electrical current through the spin current generation layer 3 may include sweeping the frequency of the AC electrical current from a first frequency to a second frequency. Exposing the magnetic device 1, li, I2, I3, I4, Is, le, I7, Is, I9 to a magnetic field may include sweeping the magnetic field from a first magnetic field strength to a second magnetic field strength. Driving the DC electrical current through the spin current generation layer 3 may include sweeping the DC electrical current from a first magnitude to a second magnitude. Data for constructing spin-torque ferromagnetic resonance spectra can thus be obtained. Examples of spin-torque ferromagnetic resonance spectra are shown In Figures 22, 23B, 23C, which are described in more detail hereinafter. Spin torque ferromagnetic measurements may be performed using an AC electrical current having a first constant amplitude, then again using an AC electrical current having a second constant amplitude. Inputting data to the magnetic device 1 A variety of approaches can be used to input data to the magnetic device 1, li, 12, I3, I4, Is, le, I7, Is, I9. Data can be input to the magnetic device 1, li, I2, Is, I4, Is, le, I7, Is, I9 by switching one or more of the magnetic states of the ferromagnetic thin-film nanomagnetic structures 2 and / or exerting control over spin-wave dynamics of one or more of the magnetic states of the ferromagnetic thin-film nanomagnetic structures 2. Inputting data to the magnetic device 1, li, I2, I3, I4, Is, Is, I7, Is, I9 may include exposing the magnetic device to one or more of a magnetic field, an electric field, and illumination. In particular, the step of inputting data to the magnetic device may comprise generating one or more of: a time-varying magnetic field for magnetic switching; microwave excitation for microwave-assisted magnetic switching and / or microwave switching; DC electrical current for switching via spin-torque; optical illumination for all-optical magnetic switching; a locally applied magnetic field from a scanning probe for topological magnetic writing; heat for heat-assisted switching; a local field for application by a read / write head; and a current-induced field from a nanostructure proximate to the ferromagnetic thin-film nanomagnetic structures, such as a nanopatterned stripline. Referring to Figure 15, a method of inputting data to the magnetic device 1, li, I2, I3, 14, Is, 16, I?, Is, I9 (hereinafter referred to as the "second method") will now be described. An AC electrical current is driven through the spin current generation layer 3, the AC electrical current having a frequency at or near a resonant frequency of one or more of the ferromagnetic thin-film nanomagnetic structures 2 and an amplitude sufficient to induce ferromagnetic resonance in said one or more of the ferromagnetic thin-film nanomagnetic structures 2 (step S2.1). While the AC electrical current is being driven through the spin current generation layer 3, a DC electrical current is also driven through the spin current generation layer 3 to cause the magnetic states of only said one or more of the ferromagnetic thin-film nanomagnetic structures to switch 2; alternatively, or additionally, while the AC electrical current is being driven through the spin current generation layer 3, the magnetic device 1, li, I2, Is, 14, Is, le, I7, Is, I9 is exposed to a magnetic field to cause the magnetic states of only said one or more of the ferromagnetic thin-film nanomagnetic structures to switch (step S2.2). By virtue of the magnetic device 1, li, 12, I3, 14, Is, le, I7, Is, I9 including the spin current generation layer 3 and all the ferromagnetic thin-film nanomagnetic structures 2 being disposed on the spin current generation layer 3, all-electrical data input from all the ferromagnetic thin-film nanomagnetic structures 2 can be performed using the second method. The magnetic field may be a DC magnetic field or a magnetic field having a timevarying amplitude. Data (also referred to herein as "information") to be input to the ferromagnetic thin-film nanomagnetic structures 2 may be encoded into the AC electrical current. Thus, multidimensional data can be input to the magnetic device 1, li, I2, I3, I4, Is, le, I7, Is, I9 in a manner which does not involve a series of sets of data (for example, onedimensional time series data) being written sequentially. In other words, multidimensional data can be written to the magnetic device in a single shot. Data to be input to the ferromagnetic thin-film nanomagnetic structures 2 may be encoded into the frequency and amplitude of the AC electrical current. The AC electrical current may comprise a plurality of components each having a respective amplitude and frequency. Another method of inputting data to the magnetic device 1, li, I2, 13, 14, Is, le, I7, Is, I9 will now be described. A DC electrical current is driven through the spin current generation layer 3 to cause the magnetic states of one or more of the ferromagnetic thin-film nanomagnetic structures 2 to switch. Alternatively, or additionally, the magnetic device 1, li, I2, I3, I4, Is, 16, I?, Is, I9 is exposed to a DC magnetic field to cause the magnetic states of one or more of the ferromagnetic thin-film nanomagnetic structures 2 to switch. Since the coercive field of a ferromagnetic thin-film nanomagnetic structure 2 is dependent on its sizing, as well as the magnetic states of adjacent ferromagnetic thin-film nanomagnetic structures 2 in the case of the magnetic devices 1, li, I2, I3, 14, Is, 16, I7, and the relative orientation of the direction of current flow with respect to any given ferromagnetic thin-film nanomagnetic structure 2, some ferromagnetic thin-film nanomagnetic structures 2 can switch a lower applied DC electrical current I DC magnetic field than other ferromagnetic thin-film nanomagnetic structures 2. The magnetic states of one or more of the ferromagnetic thin-film nanomagnetic structures 2 can therefore be selectively switched without an AC electrical current using this alternative method. Training the magnetic device 1 Referring to Figures 16A and 16B, a method of operating the magnetic device 1, li, I2, Is, 14, Is, 16, 17, 18, I9 (hereinafter referred to as the "third method") will now be described. The third method includes inputting data to the magnetic device 1, li, I2, Is, I4, Is, le, I7, Is, I9 by switching one or more of the magnetic states of the ferromagnetic thin-film nanomagnetic structures 2 and / or exerting control over spin-wave dynamics of one or more of the magnetic states of the ferromagnetic thin-film nanomagnetic structures 2 (step S3.1), reading a state of the magnetic device 1, li, I2, I3, I4, Is, le, I7, Is, I9 (step S3.2), and training the magnetic device 1, li, I2, I3, I4, Is, le, I7, Is, I9 (step S3.3). The step of training the magnetic device 1, li, I2, I3, I4, Is, le, I7, Is, I9 (step S3.3) can include determining whether a performance metric and / or a loss function and / or an error value of the device is improved for a given task after one or more magnetic states are switched (step S3.3a). The step of training the magnetic device 1, li, I2, I3, I4, Is, le, I7, la, I9 (step S3.3) can also include: in response to a positive determination of improvement of the performance metric or the loss function or the error value, keeping said one or more magnetic states are switched; and / or in response to a negative determination of improvement of the performance metric or the loss function or the error value, switching back said one or more magnetic states (step S3.3b). The third method may be performed using an algorithm that is random or recursive, for example by using backpropagation or equilibrium propagation or by using a gradient descent technique or by using an evolutionary or genetic search algorithm and / or a regression algorithm including ridge or linear regression. Integrated circuit 25 Referring to Figure 17, an integrated circuit 25 is shown. The integrated circuit 25 includes the magnetic device magnetic device 1, li, I2, I3, 14, 15, 16, 17, 18, 19. The integrated circuit 25 may include an electrical current source 26. The electrical current source 26 may be configured to drive an AC electrical current and / or a DC electrical current through the spin current generation layer 3. The integrated circuit 25 may include a magnetic field source 27. The magnetic field source 27 may be integral to the Integrated circuit and configured to expose the magnetic device 27 to a magnetic field. The integrated circuit 25 may include a voltage sensor 28. The voltage sensor 28 may be configured to sense the rectified voltage across the spin current generation layer 3. The integrated circuit 25 may include a voltage source 29. The voltage source 29 may be configured to apply a voltage across the ionic material layer(s) 20. The integrated circuit 25 may include a controller 30. The controller 30 may be configured to control one or more of the electrical current source(s) 26, the magnetic field source 27, the voltage sensor 28, and the voltage source 29. The controller 30 may be configured to encode data to be input to the ferromagnetic thin-film nanomagnetic structures 2 into the AC electrical current. The controller 30 may be configured to perform the step of training the magnetic device 1, li, 12, I3, I4, Is, le, I7, Is, 19. The integrated circuit 25 may be operable to perform time-series processing and / or forecasting of time series, where functionality is achieved by writing the magnetisations of the ferromagnetic thin-film nanomagnetic structures 2. The integrated circuit 25 may be operable to perform recurrent neural network computation. The integrated circuit 25 may be for neuromorphic computing, and / or inference-based computing, and / or logic-based computing (for example, for use as a logic gate). Apparatus 31 Referring to Figure 18, an apparatus 31 is shown. The apparatus 31 includes the Integrated circuit 25. The apparatus may also include an external magnetic field source 32. The external magnetic field source 32 may be external to the integrated circuit 25 and configured to expose the magnetic device 1, li, I2, I3, 14, Is, 16, 17, Is, 19 to a magnetic field. The apparatus 31 may also include an external controller (not shown) configured to control the external magnetic field source 32. System 33 The integrated circuit 25 and / or the apparatus 31 can be suitable for deployment with a battery or with energy harvesting technology. In this way, and in view of the memory and processing of a physical neural network being provided by the physics of a device itself, the integrated circuit 25 and / or the apparatus 31 can be deployed In a remote location with a poor data connection. This can mean that the integrated circuit 25 and / or the apparatus 31 can be used for edge computing. In particular, the integrated circuit 25 and / or the apparatus 31 may be operable at a power of 10 W or less, 5 W or less, 2 W or less, 1 W or less, 500 mW or less, 200 mW or less, 100 mW or less, 50 mW or less, 20 mW or less, 10 mW or less, 5 mW or less, 2 mW or less, or 1 mW or less. Referring to Figure 19, a system 33 is shown. The system 33 includes at least one of the integrated circuit 25 and the apparatus 31. The system 33 also includes an off-grid power source 34 configured to power the at least one of the integrated circuit 25 and / or the apparatus 31. The off-grid power source 34 may be a battery, a solar panel, or the like. The off-grid power source 34 may not be connected to an established power distribution grid. The off-grid power source 34 may be portable. Fabricating the magnetic device 1 Referring to Figure 20A, a method of rotational deposition will now be described. A body 35 provided on its main surface 36 with a patterned layer of resist 37 having an undercut profile and which exposes portions of the main surface 36. Material 38 is deposited while the body 35 is rotated so as to cause the material 38 to enter the gaps in the patterned layer of resist 37 at a broad range of angles. In this way, the material 38 can cover most or all of the exposed portions of the main surface 36. Referring also to Figure 20B, a method of static deposition will now be described. A body 35 provided on its main surface 36 with a patterned layer of resist 37 having an undercut profile and which exposes portions of the main surface 36. Material 38 is deposited while the body 35 is static so as to cause the material 38 to enter the gaps in the patterned layer of resist 37 at a narrow range of angles. In this way, the material 38 cannot cover all the exposed portions of the main surface 36. Rather, the material 38 can only cover exposed portions of the main surface 36 which are not shadowed by the upper edges of the layer of resist 37. Thus, by providing a body 35 with a patterned layer of resist 37 having an undercut profile and which exposes portions of the main surface 36 of the body, depositing a first one or more layers then depositing a second one or more layers can result in the second one or more layers being wholly disposed on the first one or more layers and a need to perform separate lithography steps for the first one or more layers and the second one or more layers can be negated. Referring to Figure 21, a method of fabricating the magnetic device 1 (hereinafter referred to as the "fourth method") will now be described. First, a body 35 having a planar main surface 36 on which the magnetic device 1, li, 12, 13, 14, Is, Is, 17, 18, 19 is to be fabricated is provided (step S4.1). Next, a lithographically patterned layer of resist 37 is provided on the planar main surface 36, the lithographically patterned layer of resist 37 exposing areas of the planar main surface 36 on which material 38 is to be deposited, the lithographically patterned layer of resist 37 having an undercut profile (step S4.2). Next, while rotating the body 35, a first one or more layers including the spin current generation layer 3 (step S4.3) is deposited. Next, after depositing the spin current generation layer 3 and while not rotating the body 35, a second one or more layers including the ferromagnetic thln-film nanomagnetic structures 2 is deposited on the first one or more layers (step S4.4). The magnetic device 1, li, 12, I3, I4, Is, le, I7, Is, I9 can be fabricated in a single lithography step using the fourth method. The layer of resist 37 may be lithographically patterned by photolithography, electron beam lithography, proton beam writing (or "p-beam writing"), ion beam lithography, direct-write lithography, or soft lithography. The resist may be a positive resist, a negative resist, or another suitable type of resist. In the cases of photolithography, electron beam lithography, proton beam writing (or "p-beam writing"), ion beam lithography, direct-write lithography, the undercut profile may result from scattering (namely, of light, electrons, protons, ions, respectively etc.). In the case of soft lithography, the undercut profile may be achieved by performing stamping first along one direction then along another direction. After steps S4.1 to S4.4 are complete, the patterned layer of resist 37 and any layers deposited on the patterned layer of resist 37 may be removed. The first one or more layers may include the spacer layer 9. The ionic material layer 20 may be included in either the first one or more layers or the second one or more layers. The one or more gate electrodes 21 may be deposited on the magnetic device in a subsequent lithography step. The ninth magnetic device 1, 19 can be fabricated using the fourth method in a single lithography step by depositing a first portion of the spin current generation layer 3 as one of the first one or more layers and a remainder of the spin current generation layer as one of a third one or more layers, the third one or more layers deposited after the second one or more layers and preferably while rotating the body 35. Experimental data Referring to Figure 22, a spin-torque ferromagnetic resonance spectra of a magnetic device having a 20 pm x 20 pm array of ferromagnetic thin-film nanomagnetic structures is shown. The spin-torque ferromagnetic resonance spectra is obtained from measurements of the magnetic device 1, 14 shown in Figure 7 and is plotted as a function of AC electrical current frequency and applied magnetic field. The spectra shows resonances across a broad frequency range (2-10 GHz at 0 Oe). The richness of this spectra arises from the presence of multiple magnetic lattices (namely, from the ferromagnetic thin-film nanomagnetic structures 2, and one of the patterned spin current generation layer 3) and the strong coupling between those lattices. Typically, square artificial spin ices possess around one to around three main resonances, whereas the spectra shown in Figure 22 shows at least six. The broad spectral range can be an advantage for computing. By occupying a larger frequency range, the number of uncorrelated readout channels can increase, which can in turn increase the number of weights that can be applied to the readout during training. Referring to Figures 23A to 23D, a demonstration of microwave assisted switching in an array of ferromagnetic thin-film nanomagnetic structures 2 is shown. Here, AC electrical current is driven through 20 nm thick ferromagnetic thin-film nanomagnetic structures 2 arranged as zigzags. These are shown in light contrast In Figure 23A, which is a scanning electron micrograph of these ferromagnetic thin-film nanomagnetic structures 3. Figures 23B and 23C show spin torque ferromagnetic resonance spectra at AC electrical current powers of 20 dBm and 26 dBm. Here, a bias magnetic field is applied (x axis) whilst sweeping the frequency of the AC electrical current (y axis). At 20 dBm, two magnetisation switching events can be seen. These are evident from the discontinuities in the spectra as a function of field and occur at 38 mT and 42 mT respectively. At 26 dBm, the switching fields for these two events can be seen to reduce to 25 mT and 32 mT respectively. Figure 23D shows the switching field as a function of frequency for an AC electrical current power of 26 dBm. Here, the frequency of the AC electrical current was kept fixed whilst the magnetic field was swept. The switching field can be seen to be frequency dependent, from which It can be concluded that the reduction of the coercive field is due to magnetic resonance rather than heating effects. Switching fields were found to reduce by up to 57%. By material engineering, for example by reducing the magnetic material thickness, this can be further reduced. Modifications It will be appreciated that various modifications may be made to the embodiments hereinbefore described. Such modifications may involve equivalent and other features which are already known in the fields of nanoscale artificial spin systems and neuromorphlc hardware, and component parts thereof, and which may be used instead of, or in addition to, features already described herein. Features of one embodiment may be replaced or supplemented by features of another embodiment. Although claims have been formulated in this application to particular combinations of features, it should be understood that the scope of the disclosure of the present invention also includes any novel features or any novel combination of features disclosed herein either explicitly or implicitly or any generalization thereof, whether or not it relates to the same invention as presently claimed in any claim and whether or not it mitigates any or all of the same technical problems as does the present invention. The applicant hereby gives notice that new claims may be formulated to such features and / or combinations of such features during the prosecution of the present application or of any further application derived therefrom.

Claims

1. A magnetic device comprising:ferromagnetic thin-film nanomagnetic structures spaced apart from each other, each having a respective magnetic state; anda spin current generation layer having a surface on which the ferromagnetic thin-film nanomagnetic structures are disposed, the spin current generation layer configured to convert an electrical current driven through the spin current generation layer parallel to the surface into a spin current capable of controlling the magnetic state(s) of one or more of the ferromagnetic thin-film nanomagnetic structures.

2. The magnetic device of claim 1, wherein:the spin current is capable of controlling the magnetic state(s) of one or more of the ferromagnetic thin-film nanomagnetic structures by exerting a spin torque on the ferromagnetic thin-film nanomagnetic structures.

3. The magnetic device of claim 1 or 2, wherein:the spin current generation layer is ferromagnetic and magnetically coupled to the ferromagnetic thin-film nanomagnetic structures; andthe spin current is capable of controlling the magnetic state(s) of one or more of the ferromagnetic thin-film nanomagnetic structures by exerting a spin torque on the spin current generation layer.

4. The magnetic device of any one of claims 1 to 3, wherein:the spin current generation layer is patterned to Include voids in areas adjacent to the ferromagnetic thin-film nanomagnetic structures.

5. The magnetic device of any one of claims 1 to 4, wherein:the ferromagnetic thin-film nanomagnetic structures are wholly disposed on the spin current generation layer.

6. The magnetic device of any one of claims 1 to 5, wherein:a spacer layer is interposed between each ferromagnetic thin-film nanomagnetic structure and the spin current generation layer.

7. The magnetic device of claim 6, wherein:the spacer layer is patterned to include voids in areas adjacent to the ferromagnetic thin-film nanomagnetic structures.

8. The magnetic device of claim 6 or 7, wherein:each of the ferromagnetic thin-film nanomagnetic structures is wholly disposed on the spacer layer.

9. The magnetic device of any one of claims 6 to 8, wherein:the spin current generation layer is magnetic;the spacer layer is non-magnetic; andeach ferromagnetic thin-film nanomagnetic structure is magnetically coupled to the spin current generation layer.

10. The magnetic device of any one of claims 1 to 9, wherein:the spin current generation layer comprises:a heavy metal material; and / ora magnetic material.

11. The magnetic device of any one of claims 1 to 10, wherein:each ferromagnetic thin-film nanomagnetic structure has a respective main surface coated with an ionic material layer.

12. The magnetic device of claim 11, wherein:the magnetic device is a magnetic device according to claim 6 and the spacer layer is an ionic material layer; orthe magnetic device is a magnetic device according to claim 6 and an ionic material layer is interposed between each thin-film nanomagnetic structure and the spacer layer; oreach ionic material layer is wholly disposed on a respective ferromagnetic thin-film nanomagnetic structure.

13. The device of any one of claims 1 to 12, wherein:a first subset of the ferromagnetic thin-film nanomagnetic structures are sized to have a first resonant frequency; anda second subset of the ferromagnetic thin-film nanomagnetic structures are sized to have a second resonant frequency which is different to the first resonant frequency.

14. The magnetic device of any one of claims 1 to 13, wherein:the ferromagnetic thin-film nanomagnetic structures are arranged to form an artificial spin ice.

15. The magnetic device of any one of claims 1 to 14, wherein:the magnetic device is implemented in a crossbar array.

16. A method of reading a state of the magnetic device of any one of claims 1 to 15, the method comprising:performing spin torque ferromagnetic resonance measurements on the magnetic device, the spin torque magnetic resonance measurements indicating characteristic frequencies corresponding to spin wave modes in the magnetic device, the step of performing the spin torque ferromagnetic resonance measurements comprising:driving an AC electrical current through the spin current generation layer of the magnetic device;exposing the magnetic device to a magnetic field and / or driving a DC electrical current through the spin current generation layer of the magnetic device; andsensing a rectified voltage across the spin current generation layer.

17. A method of inputting data to the magnetic device of any one of claims 1 to 15, the method comprising:driving an AC electrical current through the spin current generation layer, the AC electrical current having a frequency at or near a resonant frequency of the one or more of the ferromagnetic thin-film nanomagnetic structures and an amplitude sufficient to induce ferromagnetic resonance in the one or more of the ferromagnetic thin-film nanomagnetic structures; anddriving a DC electrical current through the spin current generation layer to cause only the magnetic state(s) of the one or more of the ferromagnetic thin-film nanomagnetic structures to switch; or exposing the magnetic device to a magnetic field to cause only the magnetic state(s) of the one or more of the ferromagnetic thin-film nanomagnetic structures to switch.

18. The method of claim 17, wherein:data to be input to the ferromagnetic thin-film nanomagnetic structures is encoded into the AC electrical current.

19. A method of operating a magnetic device according to claims 12 and 14, the method comprising:applying a voltage across the ionic material layer(s) to cause ions to migrate from the ionic material layer(s) into one or more ferromagnetic thin-film nanomagnetic structures or to cause ions to migrate from one or more ferromagnetic thin-film nanomagnetic structures into the ionic material layer(s).

20. A method of operating the magnetic device of any one of claims 1 to 15, the method comprising:inputting data to the magnetic device by switching the magnetic state(s) of one or more of the ferromagnetic thin-film nanomagnetic structures and / or exerting control over spin-wave dynamics of the magnetic state(s) of one or more of the ferromagnetic thin-film nanomagnetic structures;reading a state of the magnetic device; andtraining the magnetic device,wherein:training the magnetic device comprises:determining whether a performance metric and / or a loss function and / or an error value of the magnetic device is improved for a given task after one or more magnetic states are switched; andtraining the device further comprises:in response to a positive determination of Improvement of the performance metric or the loss function or the error value, keeping said one or more magnetic states switched; and / orin response to a negative determination of improvement of the performance metric or the loss function or the error value, switching back said one or more magnetic states.

21. An integrated circuit comprising the magnetic device of any one of claims 1 to 15.

22. The integrated circuit of claim 21, further comprising one or more of:an electrical current source;a magnetic field source;a voltage sensor;a voltage source; anda controller.

23. An apparatus comprising:the Integrated circuit of claim 21 or 22; andan external magnetic field source.

24. A system comprising:the integrated circuit of claim 21 or 22 or the apparatus of claim 23; andan off-grid power source configured to power the integrated circuit or the apparatus.

25. A method of fabricating the magnetic device of any one of claims 1 to 15, the method comprising:providing a body having a planar main surface on which the magnetic device is to be fabricated;providing a lithographically patterned layer of resist on the planar main surface, the lithographically patterned layer of resist exposing areas of the planar main surface on which material is to be deposited, the lithographically patterned layer of resist having an undercut profile;while rotating the body, depositing a first one or more layers including the spin current generation layer; andafter depositing the spin current generation layer and while not rotating the body, depositing a second one or more layers including the ferromagnetic thin-film nanomagnetlc structures on the first one or more layers.41

Citation Information

Patent Citations

  • Circuits and devices based on spin hall effect to apply a spin transfer torque with a component perpendicular to the plane of magnetic layers

    US20160276006A1