Classification system and corresponding process
A magnetic component with two chambers and a junction element addresses the challenges of large surface area and high energy consumption in neuromorphic networks, enabling efficient and compact signal classification.
Patent Information
- Authority / Receiving Office
- FR · FR
- Patent Type
- Patents
- Current Assignee / Owner
- GOLANA COMPUTING
- Filing Date
- 2023-12-21
- Publication Date
- 2026-06-19
Smart Images
Figure 00000030_0000 
Figure 00000030_0001 
Figure 00000030_0002
Abstract
Description
Title of the invention: Classification system and corresponding method. Technical field of the invention
[0001] The present invention relates to a classification system for classifying an input signal into a signal class. This type of classification system is particularly well-suited for applications in the development of impulse neural networks, also known as "in-materio computing" according to established Anglo-Saxon terminology. This refers to the use of physical devices to artificially reproduce the behavior of a neural network, taking advantage of the properties of the physical device(s) in question, and in particular the material(s) they comprise.
[0002] The present invention also relates to a classification method for classifying an input signal into a signal class, this classification method being implemented by a classification system as previously described. State of the art
[0003] There are different ways of artificially reproducing a neural network, for example in the field of photonics, or in the field of spin electronics.
[0004] Existing solutions in the field of photonics have made it possible to replicate the behavior of individual elements and to demonstrate the functionalities of more complete devices. However, these implementations have complex architectures and generally involve significant energy consumption.
[0005] Existing spin electronics solutions are also interesting because they exhibit good endurance compared to photonic systems, and are compatible with CMOS technologies (for Complementary Metal Oxide Semiconductor according to the established Anglo-Saxon terminology).
[0006] However, the various solutions are currently limited to proof of concept of simple devices and are not suitable for the realization of a functional neuromorphic device.
[0007] Document EP4160484A1 discloses such a functional neuromorphic network capable, in particular, of performing speech recognition of spoken digits. This network consists of a set of magnetic devices, each reproducing the behavior of a neuron. In particular, these magnetic devices are configured to reproduce the properties of accumulation, loss or leakage, and firing, beyond a certain threshold.
[0008] To reproduce a complete neuromorphic network, chambers are arranged in several rows and separated by gates, the rows all being connected to a central nucleation chamber which serves as the starting point for the propagation of a domain wall. This propagation of the domain wall occurs when a physical excitation corresponding to the signal to be processed is generated. For this purpose, a magnetic domain of a given orientation is nucleated in the central nucleation chamber. This magnetic orientation is opposite to that of the rest of the magnetic component, which allows the domain wall to be formed. In this configuration, the physical excitation will excite and displace the magnetic domain wall located at the boundary between two domains of opposite magnetizations.Chambers will thus be filled and gates crossed until a network is obtained in which some chambers are in a first magnetic orientation and in which the other chambers are in a second magnetic orientation. The image obtained of the complete network is then characteristic of the information to be processed that was used to excite the system.
[0009] This solution has the advantage of allowing the processing of a signal, without having to resort to very large computing capacities in particular for pre-processing calculations and for the calculation of synaptic weights.
[0010] However, in order to obtain a neuromorphic network capable of discriminating a wide variety of signals, it is necessary to increase the number of chamber chains. Given that the size of a chamber is on the order of 0.1 pm to 10 pm on a side, and that the number of chains can exceed 10,000, the total surface area of the classification system can quickly become large, or even unsuitable for certain applications.
[0011] Furthermore, the energy consumption required for physical excitation by a magnetic field increases with the surface area of the device. In the case of using an electric current as the physical quantity of excitation, the energy consumption increases with the length of the wires.
[0012] There is therefore a need to find a compact and low energy consumption solution that allows for the reproduction of a complete neuromorphic network.
[0013] Object and invention
[0014] The present invention aims to provide a solution that addresses all or part of the aforementioned problems.
[0015] This goal can be achieved by implementing a conversion device to convert an initial signal into a data set, the conversion device comprising: - a magnetic component comprising a magnetic material configured to vary locally between a first magnetic state exhibiting a first magnetization, and a second magnetic state having a second magnetization different from the first magnetization, said magnetic component being subdivided into a plurality of zones, where each zone is either in the first magnetic state or in the second magnetic state; each zone in the first magnetic state being separated from a zone in the second magnetic state by a magnetic domain wall; - a junction element separating the magnetic component between an upstream chamber and a downstream chamber and ensuring magnetic communication between the upstream chamber and the downstream chamber, said junction element being characterized by a magnetic transmission value corresponding to a capacity of the junction element to allow a displacement of the magnetic domain wall in a direction of propagation, said direction of propagation being defined from the upstream chamber to the downstream chamber; - an initialization unit configured to place the magnetic component in an initial magnetic configuration, in which the magnetic domain wall is arranged at the level of the junction element; - an excitation unit configured to generate a physical excitation in order to excite the magnetic material, so as to move the magnetic domain wall in the direction of propagation, the physical excitation being generated as a function of the initial signal; - a detection unit disposed at the downstream chamber, said detection unit being configured to detect and record at least one trigger characteristic parameter when a physical parameter associated with a magnetic domain wall deformation state exceeds a predetermined threshold value, the dataset including said at least one trigger characteristic parameter.
[0016] The above-described arrangements allow for the proposal of a conversion device capable of converting an initial signal into a data set comprising one or more characteristic triggering parameters. This type of device is particularly well-suited for artificial neural network applications. Furthermore, the use of only two chambers results in a more compact conversion device requiring less electrical power.
[0017] By "magnetic transmission value" is meant the capacity of the junction element to allow transmission through a magnetic domain wall. In other words, this "magnetic transmission value" corresponds to the inverse of a resistance that the junction element can produce on the passage through a magnetic domain wall.
[0018] By “magnetic domain wall” we mean a boundary separating two areas of the same continuous magnetic medium which are not in the same magnetic state.
[0019] The conversion device may also have one or more of the following characteristics, taken alone or in combination.
[0020] According to one embodiment, the conversion device includes a nucleation member configured to place the magnetic component in a magnetic nucleation configuration, in which all areas of the magnetic component are placed in the first magnetic state, except for one nucleation area which is placed in the second magnetic state, said nucleation area being contained in the upstream chamber.
[0021] In general, the magnetic material of the magnetic component is inhomogeneous. In other words, in most configurations, the magnetic component comprises a plurality of zones, where at least one zone of the plurality of zones is in the first magnetization state, and where at least another zone of the plurality of zones is in the second magnetization state.
[0022] According to one embodiment, the initialization unit includes the nucleation element.
[0023] According to one embodiment, the physical parameter associated with a deformation state of the magnetic domain wall is a distance measured in the direction of propagation from the junction member, the characteristic triggering parameter being detected and recorded by the detection unit when this distance exceeds a predetermined threshold distance.
[0024] Alternatively, the physical parameter associated with a deformation state of the magnetic domain wall is a measurement of area or of an area ratio, the characteristic trigger parameter being detected and recorded by the detection unit at the time when this area or area ratio exceeds the predetermined threshold value.
[0025] According to one embodiment, the dataset is a number corresponding to the number of times a trigger characteristic parameter is recorded by the detection unit. In other words, the dataset is equal to the sum of the detected trigger characteristic parameters.
[0026] According to one embodiment, the excitation unit is configured to stop at least temporarily the physical excitation of the magnetic material when the following condition is met: at least one triggering characteristic parameter is detected by the detection unit.
[0027] According to one embodiment, the conversion device further comprises a closed-contour peripheral delimiter in which the magnetic component is fully contained.
[0028] Thus, it is possible to limit the expansion of the domain wall when it meets the peripheral boundary.
[0029] Generally, the peripheral boundary is a non-magnetic zone. For example, a groove etched into the material constituting the magnetic component, or a zone that has been irradiated by a laser beam or ionic irradiation to form said non-magnetic zone.
[0030] According to one embodiment, the peripheral delimitation is quadrilateral. Thus, the manufacture of the conversion device is simplified.
[0031] According to one embodiment, the initialization unit is configured to place the magnetic component in the initial magnetic configuration when the following condition is met: at least one triggering characteristic parameter is detected by the detection unit.
[0032] The ability to reset the magnetic component by placing it in the initial magnetic configuration after each detection of a triggering characteristic parameter makes it possible to accelerate the return of the magnetic component to the initial magnetic configuration, which makes it possible to detect a maximum of triggering characteristic parameters in time by the conversion device.
[0033] According to one embodiment, the initialization unit includes an expansion member configured to move the magnetic domain wall up to the junction member, so as to place the magnetic component in the initial magnetic configuration.
[0034] Thus, it is possible to place the conversion device in a configuration where the physical excitation generated by the excitation unit only contributes to moving the magnetic domain wall beyond the junction member into the downstream chamber.
[0035] According to one embodiment, the initialization unit includes the excitation unit, the excitation unit being configured to move the magnetic domain wall to the junction member, so as to place the magnetic component in the initial magnetic configuration.
[0036] According to one embodiment, the conversion device further includes a reconfiguration unit configured to modify the magnetic transmission value of the junction member.
[0037] Thus, it is possible to modify the magnetic transmission value of the junction element to adapt it to the initial signal that one wishes to convert.
[0038] According to one embodiment, the reconfiguration unit is configured to modify the magnetic transmission value of the junction member each time the detection unit detects a characteristic triggering parameter.
[0039] Thus, it is possible to modify the magnetic transmission value of the junction element when the magnetic component is placed in the initial magnetic configuration. This makes it possible to artificially reproduce a succession of chambers separated by junction elements having varying magnetic transmission values. In other words, the junction element is reconfigurable.
[0040] According to one embodiment, the excitation unit includes a current generator configured to generate an electric current.
[0041] Synergistically, the use of a current generator as an excitation unit in a conversion device comprising only two chambers makes it possible to reduce the energy consumption of the system while maintaining good conversion quality of the initial signal by the conversion device.
[0042] According to one embodiment, the physical excitation corresponds to a signal comprising pulses of electric current.
[0043] The object of the invention can also be achieved through the implementation of a classification system for classifying an input signal into a signal class, the classification system comprising: - at least one pre-processing device intended to transform the input signal into at least one initial signal; - at least one conversion device as described above and taking as input an initial signal from said at least one initial signal and converting said initial signal taken as input into a set of data; - an identification unit taking as input the dataset and configured to associate the input signal with a signal class according to the dataset, said signal class being chosen from a set of predetermined signal classes stored in a memory of the classification system.
[0044] The provisions described above make it possible to propose a classification system capable of classifying an input signal according to the response detected by the conversion device.
[0045] According to one embodiment, the classification system comprises at least two conversion devices.
[0046] According to one embodiment, the classification system comprises at least two conversion devices taking as input the same initial signal or each a distinct initial signal obtained by transforming the same input signal by means of a distinct pre-processing device.
[0047] For example, the classification system may take a single input signal and may include one or more pre-processing devices. The pre-processing device(s) may convert the input signal into an initial signal The signal can be identical for all conversion devices, or alternatively, a plurality of distinct or identical initial signals for each conversion device. In other words, the same input signal can be pre-processed differently by the pre-processing device for each conversion device.
[0048] In this way, it is possible to perform at least two conversions of the initial signal into a data set, to improve the signal conversion.
[0049] The object of the invention can also be achieved through the implementation of a method for converting an initial signal into a data set, the conversion method being implemented by a conversion device as described above, and comprising the following phases: - an initialization phase including an initial step in which the magnetic component is placed in the initial magnetic configuration; - an excitation phase in which the excitation unit generates physical excitation according to the initial signal, so as to move the magnetic domain wall; - a reconfiguration phase implemented if the detection unit detects that at least one physical parameter associated with the deformation state of the magnetic domain wall exceeds the predetermined threshold value, the reconfiguration phase then including a detection step in which the detection unit records at least one characteristic trigger parameter in the dataset.
[0050] The arrangements described above make it possible to propose a conversion method for converting an initial signal into a set of data by means of a conversion device which may use only two chambers of a magnetic material.
[0051] It is therefore well understood that the excitation phase is implemented continuously, and that several detection steps can be implemented, in particular each time that at least one physical parameter associated with the deformation state of the magnetic domain wall exceeds the predetermined threshold value.
[0052] The conversion process may also have one or more of the following characteristics, taken alone or in combination.
[0053] According to one embodiment, the initial signal and the physical excitation are time-dependent. In this case, it can be provided that the conversion process is stopped at the temporal end of the physical excitation.
[0054] According to one embodiment, the initialization unit is configured to place the magnetic component in the initial magnetic configuration, for example via the expansion member.
[0055] According to one embodiment, the initialization unit comprises a nucleation member, the initialization phase then comprising a nucleation step, implemented before the initial step, in which the nucleation member places the magnetic component in the magnetic nucleation configuration in which all areas of the magnetic component are placed in the first magnetic state, with the exception of a nucleation area which is placed in the second magnetic state, said nucleation area being contained in the upstream chamber.
[0056] Thus, it is possible to prepare the conversion device to perform a conversion, in particular when it has not been used for a long time.
[0057] According to one embodiment, the initial step is carried out simultaneously with the excitation phase. In other words, it is the physical excitation generated during the excitation phase that allows the domain wall to be moved towards the junction element so as to transition from the nucleation magnetic configuration to the initial magnetic configuration.
[0058] According to one embodiment the initial magnetic configuration corresponds to a configuration where the inhomogeneous magnetic material is in the second magnetic state in the upstream chamber, and in the first magnetic state in the downstream chamber, or vice versa.
[0059] According to one embodiment, the reconfiguration phase is implemented when the magnetic material present in the downstream chamber is entirely in the second magnetic state.
[0060] Thus, the detection of the exceeding of the threshold distance by the magnetic wall is simpler.
[0061] According to one embodiment, the reconfiguration phase includes a stop step, in which the excitation unit stops at least temporarily the implementation of the excitation phase; a new excitation phase being implemented at the end of the reconfiguration phase.
[0062] Stopping the excitation phase allows the magnetic component to return to the initial magnetic configuration before implementing the excitation phase again.
[0063] According to one embodiment, a new initialization phase is implemented at the end of the reconfiguration phase.
[0064] Thus, it is possible to place the magnetic component in its initial magnetic configuration, even when the downstream chamber is entirely in the second magnetic state. It is also possible to actively return the magnetic component to its initial magnetic configuration to save time.
[0065] According to one embodiment, the reconfiguration phase further includes a modification step implemented after the detection step, in which the reconfiguration unit modifies the magnetic transmission value of the junction member.
[0066] In this way, it is possible to artificially reproduce a succession of chambers separated by junction elements having a magnetic transmission value that varies.
[0067] The object of the invention can also be achieved through the implementation of a classification method for classifying an input signal into a signal class, the classification method being implemented by a classification system as described above and comprising the following steps: - a pre-processing step in which the pre-processing device transforms the input signal into at least one initial signal; - at least one conversion step in which a conversion process as described above is applied by the conversion device to an initial signal among said at least one initial signal, so as to obtain a data set corresponding to said initial signal to which the conversion process is applied; - an identification step in which the identification unit associates the input signal with a signal class based on the data set obtained during at least one conversion step
[0068] The provisions described above make it possible to propose a method of classifying an input signal into a signal class by means of a compact conversion device.
[0069] According to one embodiment, the pre-processing step may include transforming the input signal into a plurality of initial signals. Each initial signal in the plurality of initial signals may be identical to or different from another of the initial signals. Thus, during each conversion step, each conversion device applies the conversion process to the initial signal associated with it.
[0070] In this way, it is possible to pre-process the input signal differently during the pre-processing stage, depending on the characteristics that one wishes to extract from the input signal, or depending on the conversion device that will apply the conversion process to the initial signal thus pre-processed.
[0071] Brief description of the drawings
[0072] Other aspects, objectives, advantages and features of the invention will become clearer upon reading the following detailed description of preferred embodiments thereof, given by way of non-limiting example, and made with reference to the accompanying drawings in which:
[0073] [Fig.1] Fig.1 is a schematic view of a classification system according to a particular embodiment of the invention.
[0074] [Fig.2] Fig.2 is a schematic view showing different ways of putting in works a nucleation step, via a nucleation organ.
[0075] [Fig.3] Fig.3 is a schematic view of a conversion device according to a a particular embodiment of the invention implementing certain steps of the conversion process.
[0076] [Fig.4] Fig.4 is a schematic perspective view of a device conversion according to a particular embodiment of the invention, and more particularly presenting the detection unit.
[0077] [Fig. 5] Fig. 5 is a graph representing the evolution of the physical parameter associated with a state of deformation of the domain wall, as a function of time.
[0078] [Fig.6] The [Fig.6] is a schematic side and top view of a conversion device according to a particular embodiment of the invention comprising a fixed connecting member.
[0079] [Fig.7] Fig.7 is a schematic side and top view of a conversion device according to a particular embodiment of the invention comprising a reconfigurable junction element.
[0080] [Fig.8] Fig.8 is a schematic view showing different ways of putting in implements a modification step, via a reconfiguration unit.
[0081] [Fig.9] Fig.9 is a schematic view showing different ways of putting in implements a modification step, via a reconfiguration unit.
[0082] [Fig. 10] The [Fig. 10] is a schematic view showing some steps of the initialization phase.
[0083] [Fig. 11] The [Fig. 11] is a schematic view of a classification process according to a particular embodiment of the invention.
[0084] [Fig. 12] The [Fig.1] is a schematic view of a classification process according to a particular embodiment of the invention. Detailed description
[0085] In the figures and throughout the description, the same reference numerals represent identical or similar elements. Furthermore, the various elements are not drawn to scale in order to enhance the clarity of the figures. Moreover, the different embodiments and variants are not mutually exclusive and may be combined.
[0086] As illustrated in Figures 1 to 9, the invention relates to a conversion device 10 for converting an initial signal denoted "SI" into a data set denoted "SC". For example, the initial signal SI may correspond to an input signal denoted "SE" having been pre-processed by a pre-processing device 3, which will be described later. The input signal SE may correspond to information to be processed, which is pre-processed into an initial signal SI that can be converted by the conversion device 10. For example, and without limitation, it is possible that the input signal SE may be pre-processed in the form of a succession of pulses as described in European patent application No. 22315278.6 filed in November 2022. It is also possible that the initial signal SI may correspond to a variation in amplitude, frequency, time response, or any other type of signal capable of exciting a magnetic material.
[0087] The conversion device first comprises a magnetic component 20 which includes a magnetic material configured to vary locally between a first magnetic state, denoted "M1", exhibiting a first magnetization, and a second magnetic state, denoted "M2", exhibiting a second magnetization different from the first magnetization. As can be seen in [Fig. 1], the conversion device 10 may include a closed-contour peripheral boundary 11 in which the magnetic component 20 is entirely contained. Thus, and as will be shown later, it is possible to limit the expansion of a magnetic domain wall P when it encounters the peripheral boundary 11.Generally, the peripheral boundary 11 is a non-magnetic zone, for example, a groove etched into the material of the magnetic component 20, a zone irradiated by a laser beam, or by ion irradiation to form said non-magnetic zone. It is also possible, but not limited to, such a peripheral boundary 11 being quadrilateral. This simplifies the manufacture of the conversion device 10.
[0088] The magnetic component 20 is subdivided into a plurality of zones, where each zone is either in the first magnetic state M1 or in the second magnetic state M2. Each zone in the first magnetic state M1 is separated from a zone in the second magnetic state M2 by a magnetic domain wall P. By "magnetic domain wall P" is meant a boundary separating two zones of the same continuous magnetic medium that are not in the same magnetic state. Generally, the magnetic material of the magnetic component 20 is inhomogeneous. In other words, in most configurations, the magnetic component 20 comprises a plurality of zones, where at least one zone of the plurality of zones is in the first magnetic state M1, and where at least one other zone of the plurality of zones is in the second magnetic state M2.
[0089] The conversion device 10 also includes a junction member 30 separating the magnetic component 20 between an upstream chamber 21 and a downstream chamber 23. Figure 1 shows, in particular, an embodiment in which the chamber The upstream chamber 21 is entirely in the second magnetic state M2, and the downstream chamber 23 is entirely in the first magnetic state M1. The junction element ensures magnetic communication between the upstream chamber 21 and the downstream chamber 23. This junction element 30 is characterized by a magnetic transmission value denoted "Rtr," corresponding to the junction element 30's capacity to allow the magnetic domain wall P to move in a propagation direction denoted "X," said propagation direction X being defined from the upstream chamber 21 to the downstream chamber 23. The "magnetic transmission value Rtr" refers to the junction element 30's capacity to allow transmission through a magnetic domain wall P. In other words, this "magnetic transmission value Rtr" corresponds to the inverse of the resistance that the junction element 30 can produce when passing through a magnetic domain wall P.In general, the propagation of the magnetic domain wall P within one of the two chambers 21, 23 is easier than its propagation through the junction member 30. The junction member 30 therefore acts as a gate constituting a brake on the propagation of the magnetic domain wall P between the upstream chamber 21 and the downstream chamber 23.
[0090] The conversion device 10 also includes an initialization unit 40 configured to place the magnetic component 20 in an initial magnetic configuration denoted "Cl", in which the magnetic domain wall P is arranged at the level of the junction member 30. Such an initial magnetic configuration Cl is shown for example in [Fig. 1].
[0091] According to one embodiment, the conversion device 10 comprises a nucleation element 41 configured to place the magnetic component 20 in a magnetic nucleation configuration denoted C0, in which all areas of the magnetic component 20 are placed in the first magnetic state M1, with the exception of one nucleation area which is placed in the second magnetic state M2, said nucleation area being contained in the upstream chamber 21. It is also possible that the initialization unit 40 comprises the nucleation element 4L. [Fig. 2] presents different variants of nucleation elements 4L. The placement of the magnetic component 20 in the magnetic nucleation configuration C0 is particularly useful in the case where the entire conversion device 10 is in the first magnetic state ML. In this case, the nucleation element 41 makes it possible to form an area in the second magnetic state M2, and thus create a magnetic domain wall P.
[0092] Figure 2 A shows a nucleation element 41 configured to apply a local magnetic field to form the nucleation zone. Generally, a nucleation element 41 may include an electromagnet-type device or A permanent magnet allows the application of a local magnetic field. Figure 2B shows a nucleation element 41 configured to apply a local magnetic field, which is associated with the presence of an intrinsic defect 42 in the magnetic material, or one created intentionally. The creation of a defect 42 in the upstream chamber 21 thus serves as a preferred center to facilitate nucleation. This defect 42 can notably be created by a local modification of the properties of the magnetic layer by various means such as: a laser beam, a FIB beam (for Focused Ion Beam, according to established Anglo-Saxon terminology), or an optical or electron beam lithography step followed by an etching step. Figure 2C shows a nucleation element 41 configured to apply a local magnetic field when it is associated with the presence of a magnetic layer 44 applied to the magnetic component 20.A non-magnetic layer 46 may or may not be interposed between the magnetic component 20 and the magnetic layer 44. This magnetic layer 44, or this stack of layers 44, 46, may in particular exhibit planar magnetic anisotropy. Figure 2D shows a nucleation element 41 configured to apply a local or global magnetic field associated with local heating. The nucleation element 41 may then include a local heating device 47 configured to generate such local heating by means of an electric current, a localized laser beam, or any other means. Figure 2E shows a nucleation element 41 configured to apply a local magnetic field in the presence of a local electric field. In this case, the electric field can be used to lower the magnetic anisotropy barrier. Finally, Figure 2D shows a nucleation element 41 configured to apply a local magnetic field in the presence of a local electric field. In this case, the electric field can be used to lower the magnetic anisotropy barrier. Finally, Figure 2E shows a nucleation element 41 configured to apply a local magnetic field in the presence of a local electric field.[2] F represents a nucleation device 41 configured to inject a spin-polarized current, for example by defining on the upstream chamber 21 a pillar of a magnetic stack of the magnetic tunnel junction or spin valve type. The current is then injected as shown [Fig.2] F. It is well understood that these different embodiments of the nucleation device 41 are not limiting.
[0093] According to one embodiment, the initialization unit 40 includes an expansion member 43 configured to move the magnetic domain wall P to the junction member 30, so as to place the magnetic component 20 in the initial magnetic configuration CL. Thus, it is possible to place the conversion device 10 in a configuration in which a physical excitation generated by the excitation unit 50, described below, only contributes to moving the magnetic domain wall P beyond the junction member 30 into the downstream chamber 23. According to a first variant, the expansion member 43 is configured to move the magnetic domain wall P in the propagation direction X, for example, once the nucleation member 41 has formed a nucleation zone. According to another variant, it is possible for the expansion member 43 to be configured to move the magnetic domain wall P in the opposite direction to the propagation direction X, for example to return the magnetic domain wall P to the level of the junction member 30 when it has moved inside the downstream chamber 23. The arrangements described above therefore allow the expansion member 43 to replace the magnetic component 20 in the initial magnetic configuration Cl.
[0094] Different variants can be used to implement the expansion member 43. According to a first variant, the expansion member 43 includes a magnetic field generator, such as an electromagnet or permanent magnet device for applying a local magnetic field. According to a second variant, the expansion member 43 includes a spin-polarized current generator. It is also possible for the expansion member 43 to include the nucleation member 41.
[0095] The conversion device 10 further includes an excitation unit 50 configured to generate a physical excitation to excite the magnetic material, so as to move the magnetic domain wall P in the propagation direction X, the physical excitation being generated according to the initial signal SI. For example, the physical excitation corresponds to a signal comprising electric current pulses or magnetic field pulses. For this purpose, the excitation unit 50 may include a current generator 51 configured to generate an electric current, or a magnetic field generator.
[0096] According to one embodiment, the initialization unit 40 includes the excitation unit 50, the excitation unit 50 being configured to move the magnetic domain wall P to the junction member 30, so as to place the magnetic component 20 in the initial magnetic configuration CL. In other words, the excitation unit 50 includes the expansion member 43.
[0097] According to one embodiment, the excitation unit 50 can be configured to place the magnetic component 20 in the initial magnetic configuration Cl in the case where the magnetic domain wall P has passed the junction member 30 and is placed in the downstream chamber 23.
[0098] Synergistically, the use of a current generator 51 as an excitation unit 50 in a conversion device 10 comprising only two chambers 21, 23 reduces the system's energy consumption while maintaining good conversion quality of the initial signal SI by the conversion device 10. Furthermore, the current generator 51 can act both as an expansion element 43 and as an excitation unit 50. [Fig. 3] illustrates the conversion device 10 in different configurations. [Fig. 3] A shows the conversion device 10 when the magnetic component 20 is in the nucleation configuration C0. [Fig. 3] B shows the conversion device 10 when the magnetic component 20 is in the initial configuration Cl. The current generator 51 can be configured to: - inject an electric current to move the magnetic domain wall P in the direction of propagation X to move the magnetic component 20 from the CO nucleation magnetic configuration to the initial Cl magnetic configuration; - inject an electric current to move the magnetic domain wall P in the direction of propagation X beyond the junction element 30; - inject a reverse electric current, to move the magnetic domain wall P in a direction opposite to the direction of propagation X, to return the magnetic domain wall P to the level of the junction element 30, in the initial magnetic configuration Cl.
[0099] Thus, the three functions are fulfilled by a single element.
[0100] The conversion device 10 further includes a detection unit 60 located in the downstream chamber 23. The detection unit 60 is configured to detect and record at least one trigger characteristic parameter, denoted "Nbf," when a physical parameter associated with a deformation state of the magnetic domain wall P exceeds a predetermined threshold value, denoted "Vs." Different variants can be used for implementing the detection unit 60. According to a first variant, the detection unit 60 includes an optical detector such as a Kerr optical microscope, either wide-field or focused. Figure 4 illustrates a second variant in which the detection unit 60 operates by electrical detection.In this case, the detection unit 60 can electrically detect at least one triggering parameter by means of a magnetic tunnel junction (tunnel magnetoresistance signal) or a spin valve (giant magnetoresistance signal). More specifically, the detection unit 60 can include a pillar 61 disposed on the magnetic component 20 vertically above the downstream chamber 23. This pillar 61 does not need to be centered on the downstream chamber 23. Indeed, and advantageously, the position of the pillar 61 relative to the junction element 30 helps to set the predetermined threshold value VS. According to one embodiment, the pillar 61 can include a spin valve-type stack or a magnetic tunnel junction-type stack comprising: - a contact electrode 62 to read the electrical signal; - a layer 63 of an insulating material (for example MgO, AlOx, HfO2, ... in the context of a magnetic tunnel junction stack), or of a non-ferromagnetic electrically conductive material in the case of a spin valve type stack; - a layer of a magnetic stack 64 constituting the reference magnetic electrode.
[0101] According to a first embodiment, the contact electrode 62 is a magnetic layer magnetically coupled to the magnetic material constituting the magnetic component 20, the local magnetic state then being read at the spin valve stack or the magnetic tunnel junction stack. Alternatively, the magnetic layer 62 can be omitted. The layer 63 of an insulating material, in the case of a magnetic tunnel junction, or the layer 63 of a non-ferromagnetic electrically conductive material in the case of a spin valve stack, is deposited directly onto the magnetic material constituting the magnetic component 20. The local magnetic state is then read at the spin valve stack or the magnetic tunnel junction stack.
[0102] The detection unit 60 may also include an electrical device for reading the resistance of the pillar 61, as well as contacts which are not shown in [Fig.4].
[0103] According to one embodiment, the physical parameter associated with a deformation state of the magnetic domain wall P is a distance denoted "d" measured in the propagation direction X from the junction member 30, the characteristic triggering parameter Nbf being detected and recorded by the detection unit 60 when this distance d exceeds a predetermined threshold distance VS. Figures 3CF illustrate the conversion device 10 when the magnetic domain wall P moves in the downstream chamber 23. In one possibility, the predetermined threshold value VS is a fixed distance measured between the junction member 20 and a position in the downstream chamber 23, as illustrated in Figures 3C and D. In this case, a triggering parameter Nbf can be detected when the magnetic domain wall P exceeds this fixed distance VS, as illustrated in [Fig. 3]D. This fixed distance VS not being crossed in [Fig.3] C, a characteristic Nbf triggering parameter, is therefore not detected.
[0104] Alternatively, the physical parameter associated with a magnetic domain wall deformation state P can be a surface area measurement or a surface area ratio. The characteristic trigger parameter Nbf can then be detected and recorded by the detection unit 60 when this surface area or surface area ratio exceeds the predetermined threshold value VS. For example, the predetermined threshold value VS may correspond to a maximum area Amax of the downstream chamber 23, and the physical parameter may correspond to the area A of the downstream chamber 23 that is in the second magnetic state M2, as shown in [Fig. 3] D. In this case, a characteristic trigger parameter Nbf can then be detected when the entire downstream chamber 23 is in the second magnetic state M2, i.e. that is when A=Amax, or when the ratio A / Amax=l, as represented on [Fig.3] E.
[0105] The [Fig.5] is a graph which represents the variation of a physical parameter of the magnetic component 20, associated with a deformation state of the domain wall in the downstream chamber 23, as a function of the current pulses generated by the physical excitation of the excitation unit 50.
[0106] Regardless of the variant considered, the SC dataset includes said at least one Nbf trigger characteristic parameter. For example, the SC dataset is a number corresponding to the number of times a Nbf trigger characteristic parameter is recorded by the detection unit 60. In other words, the SC dataset is equal to the sum of the detected Nbf trigger characteristic parameters.
[0107] Generally, the detection unit 60 may include a memory configured to record the SC data set.
[0108] According to a non-limiting variant, the excitation unit 50 can be configured to stop at least temporarily the physical excitation of the magnetic material when the following condition is met: at least one Nbf triggering characteristic parameter is detected by the detection unit 60. However, it is generally expected that the physical excitation will only be stopped by the excitation unit 50 at the temporal end of the initial signal SI.
[0109] Advantageously, the initialization unit 40 can be configured to place the magnetic component 20 in the initial magnetic configuration Cl when the following condition is met: at least one trigger characteristic parameter Nbf is detected by the detection unit 60. The ability to reset the magnetic component 20 by placing it in the initial magnetic configuration Cl after each detection of a trigger characteristic parameter Nbf accelerates the return of the magnetic component 20 to the initial magnetic configuration Cl, thus enabling the conversion device 10 to detect a maximum number of Nbf trigger characteristic parameters in time. Figure 3 F illustrates the conversion device 10 in a configuration where the initialization unit 40 is in the process of returning the magnetic component 20 to the initial magnetic configuration Cl
[0110] Although the magnetic transmission value Rtr of the junction member 30 can be fixed (as shown in [Fig. 6]), it is also possible to modify this magnetic transmission value Rtr (as shown in [Fig. 7]). In this case, and as illustrated in [Fig. 7], the conversion device 10 can include a reconfiguration unit 70 configured to modify the magnetic transmission value Rtr of the junction member 30. This allows to adapt the junction element 30 to the initial signal SI to be converted. For example, the reconfiguration unit 70 can be configured to modify the magnetic transmission value Rtr of the junction element 30 each time the detection unit 60 detects a characteristic trigger parameter Nbf. Thus, it is possible to modify the magnetic transmission value Rtr of the junction element 30 at the moment the magnetic component 20 is placed in the initial magnetic configuration Cl. This makes it possible to artificially reproduce a succession of chambers separated by junction elements 30 having a varying magnetic transmission value Rtr. In other words, the junction element 30 is reconfigurable. There are different variations for reconfiguring junction elements 30.In general, the production or reconfiguration of junction elements 30 is implemented by controlling and modifying the local magnetic properties of the magnetic material constituting the magnetic component 20. The various variants proposed below are different embodiments enabling those skilled in the art to implement reconfiguration units, and are not limiting.
[0111] Figure AB [Fig. 8] represents a reconfiguration unit 70 capable of modifying the magnetic transmission value Rtr of the junction member 30 by applying a local magnetic field. Figure A [Fig. 8] more particularly represents a reconfiguration unit 70 configured to apply a local magnetic field by means of a magnetic element 71 whose magnetization is perpendicular to the plane. This magnetic element 71 can be separated from the junction member 30 by a non-magnetic element 73. Figure B [Fig. 8] represents a reconfiguration unit 70 configured to apply a local magnetic field by means of a magnetic element 75 whose magnetization is in the plane. This magnetic element 75 can be separated from the junction member 30 by a non-magnetic element 73. Other variants not shown, such as the use of a spin valve-type structure or a magnetic tunnel junction, can also be used to obtain a reconfiguration unit 70.
[0112] A person skilled in the art may, for example, refer to European patent application number 23315104.2 filed on April 25, 2023 to obtain a reconfiguration unit 70.
[0113] Fig. 9 A represents a reconfiguration unit 70 capable of modifying the value magnetic transmission Rtr of the junction element 30 by means of a local electric field produced, for example, by the application of an electric voltage V. More specifically, the local electric field allows modulation of the magnetic properties of a magnetic layer, and in particular its anisotropy and the Dzyaloshinskii Moryia interaction. Figure 9 A shows, in particular, a conducting electrode 72 to which a voltage V is applied by means of a voltage generator. 74. According to this embodiment, it is possible for a dielectric layer 76 to be interposed between the conducting electrode 72, and the junction member 30.
[0114] Figures 9B and C represent a reconfiguration unit 70 capable of modifying the magnetic transmission value Rtr of the junction member 30 by means of local heating. More specifically, [Fig. 9]B represents a conductive layer 77a into which an electric current, denoted "I", is injected by means of a current generator 77b in order to raise the temperature of the conductive layer 77a by Joule heating. The generated temperature is thus transmitted to the junction member 30 directly, or via an intermediate layer 78 disposed between the junction member and the conductive layer 77a. This intermediate layer 78 can, for example, be a layer made of a dielectric material, so as to electrically insulate the junction member 30 from the conductive layer 77a into which the current is injected.Local heating can also be achieved by means of a hot tip, for example an atomic force microscopy or AFM tip (for Atomic Force Microscopy according to the established Anglo-Saxon terminology), brought close to the junction organ, or by means of a laser 77c focused locally as shown in [Fig.9] C. For this, it is possible to arrange an absorption layer 77d configured to absorb the energy received from the laser 77c.
[0115] The set of arrangements described above makes it possible to propose a conversion device 10 capable of converting an initial signal SI into a data set SC comprising one or more characteristic trigger parameters Nbf. This type of device is particularly suitable for artificial neural network applications. Furthermore, the use of only two chambers 21, 23 makes it possible to obtain a more compact conversion device 10 requiring less electrical power.
[0116] As previously stated, the invention also relates to a classification system 1 for classifying an input signal SE into a signal class denoted "Ci". An example of such a classification system 1 is shown in [Fig. 1].
[0117] Classification system 1 comprises: - at least one pre-processing device 3 intended to transform the input signal SE into an initial signal SI; - at least one conversion device 10 as described above taking as input said initial signal SI and converting the initial signal SI into a data set SC; - an identification unit 5 taking as input the dataset SC and configured to associate the input signal SE with a signal class Ci according to the dataset SC, said signal class Ci being chosen among a set of predetermined signal classes stored in a memory 7 of the classification system 1.
[0118] The arrangements described above make it possible to propose a classification system 1 capable of classifying an input signal SE according to the response detected by the conversion device 10.
[0119] According to an unrepresented variant, the classification system 1 may comprise at least two conversion devices 10. These different conversion devices 10 may have the same initial signal SI as input. Alternatively, the conversion devices 10 may have a distinct and different input signal SI for each conversion device 10, said initial signal SI being obtained by transforming the same input signal SE by means of a separate pre-processing device 3 associated with each of the conversion devices 10. In this way, it is possible to improve the classification quality of the input signal SE into a dataset SC. It is also possible to perform at least two conversions of the initial signal SI into a dataset SC to improve the signal conversion.
[0120] According to a first possibility, the classification system 1 may include a pre-processing device 3 capable of converting the input signal SE into a plurality of distinct or identical initial signals SI. Each initial signal SI of the plurality of initial signals SI is then associated with one of the conversion devices 10. Thus, it is possible to obtain different sets of data SC from each of the conversion devices 10. According to a second possibility, the classification system 1 may include several pre-processing devices 3, each intended to convert the input signal SE into an initial signal SI. It is therefore understood that for each of the conversion devices 10, it is possible to obtain different initial signals SI from a single input signal SE.
[0121] The invention also relates to a method for converting an initial signal SI into a data set SC. Generally, the initial signal SI and the physical excitation are time-dependent. In this case, the conversion method can be stopped at the temporal end of the physical excitation. This conversion method is implemented by a conversion device 10 as described above. Figures 3 and 10 to 12 show different phases and steps of the conversion method.
[0122] The conversion process includes an PL initialization phase
[0123] According to one embodiment, the initialization unit 40 of the conversion device 10 may include a nucleation element 4L. In this case, it is possible that the initialization phase PI includes a nucleation step Eli, in which the nucleation element 41 places the magnetic component 20 in the magnetic nucleation configuration C0 in which all areas of the magnetic component 20 are placed in the first magnetic state M1, with the exception of a nucleation zone which is placed in the second magnetic state M2, said nucleation zone being contained in the upstream chamber 21. Thus, it is possible to prepare the conversion device 10 to perform a conversion, in particular when it has not been used for a long time.
[0124] The initialization phase PI also includes an initial step E13 in which the magnetic component 20 is placed in the initial magnetic configuration Cl. In the case where a nucleation step Eli is implemented, the initialization unit 40 can, for example, be configured to place the magnetic component 20 in the initial magnetic configuration Cl, for example via the expansion member 43. For example, the initial magnetic configuration Cl corresponds to a configuration where the inhomogeneous magnetic material is in the second magnetic state M2 in the upstream chamber 21, and in the first magnetic state M1 in the downstream chamber 23, or vice versa.
[0125] The conversion process also includes an excitation phase P2 in which the excitation unit 50 generates physical excitation according to the initial signal SI, so as to move the magnetic domain wall P. As illustrated in [Fig. 10], the initial step E13 can be implemented simultaneously with the excitation phase P2. In other words, it is the physical excitation generated during the excitation phase P2 that allows the magnetic domain wall P to be moved towards the junction element 30 so as to transition from the magnetic nucleation configuration C0 to the initial magnetic configuration CL
[0126] The conversion method also includes a reconfiguration phase P3 implemented if the detection unit 60 detects that at least one physical parameter associated with the deformation state of the magnetic domain wall P exceeds the predetermined threshold value Vs. The reconfiguration phase P3 then includes a detection step E31 in which the detection unit 60 records at least one characteristic trigger parameter Nbf in the SC dataset. It is therefore understood that the excitation phase P2 is implemented continuously, and that several detection steps E31 can be implemented, in particular each time at least one physical parameter associated with the deformation state of the magnetic domain wall P exceeds the predetermined threshold value Vs.According to a first variant, the detection step can be implemented when the magnetic material present in the downstream chamber 23 is entirely in the second magnetic state M2. Thus, detecting the exceedance of the predetermined threshold value Vs by the physical parameter associated with the deformation of the magnetic wall is simpler. However, such a variant is not limiting, and it is also... It is possible that the detection step E31 is implemented when the magnetic domain wall P crosses a threshold distance Vs as shown in [Fig.3] D.
[0127] According to an unshown embodiment, the reconfiguration phase P3 may include a stop step E32, in which the excitation unit 50 at least temporarily stops the implementation of the excitation phase P2; a new excitation phase P2 being implemented at the end of the reconfiguration phase P3. Stopping the excitation phase P2 allows the magnetic component 20 to return to the initial magnetic configuration Cl before implementing the excitation phase P2 again.
[0128] Each time a characteristic trigger parameter is detected during the detection step E31, it is advantageous to implement a new initialization step PI at the end of the reconfiguration phase P3. More specifically, once a detection step E31 is implemented, the initialization unit 40 can return the magnetic component 20 to the initial magnetic configuration Cl by implementing a new initial step El3. It is therefore understood that, according to this embodiment, a new initial step is implemented each time a detection step is implemented. As specified with reference to the conversion device, the initial step E13 can be implemented by the excitation unit 50 when the initialization unit 40 includes the excitation unit 50.In the specific case where the excitation unit 50 includes a current generator 51, it is advantageous for the initial steps E13, implemented after the detection steps E31, to be carried out by injecting an electric current from the current generator 51 in the opposite direction to the propagation direction X. This makes it possible to actively return the magnetic component 20 to the initial magnetic configuration Cl, thus saving time. It is also well understood that the implementation of a nucleation step E13 is independent of the implementation of an initial step Eli. For example, the conversion process may include a single nucleation step El1 and a plurality of initial steps E13 implemented (actively or passively) after each detection step E31.
[0129] According to a non-limiting embodiment in which the conversion device 10 includes a reconfiguration unit, the reconfiguration phase P3 may also include a modification step E33 implemented after the detection step E31, in which the reconfiguration unit 70 modifies the magnetic transmission value Rtr of the junction member 30. In this way, it is possible to artificially reproduce a succession of chambers separated by junction members having a varying magnetic transmission value Rtr. Such an embodiment is shown, for example, in [Fig. 12].
[0130] The arrangements described above make it possible to propose a conversion method for converting an initial signal SI into a data set SC by means of a conversion device 10 which may use only two chambers of a magnetic material.
[0131] Finally, and as illustrated in Figures 11 and 12, the invention relates to a classification method for classifying an input signal SE into a signal class Ci. The classification method is implemented by a classification system 1 as described above and comprises the following steps: - a pre-processing step E0 in which the pre-processing device 3 transforms the input signal SE into an initial signal SI; - at least one conversion step El in which a conversion process as described above is applied by the conversion device 10 to the initial signal SI, so as to obtain a data set SC corresponding to the initial signal SI; - an identification step E2 in which the identification unit 5 associates the input signal SE with a signal class Ci according to the data set SC obtained during at least one conversion step El
[0132] The arrangements described above make it possible to propose a method for classifying an input signal SE into a signal class by means of a compact conversion device 10. The classification method can therefore comprise one or more conversion steps El, depending in particular on the number of conversion devices 10.
[0133] According to an unrepresented variant, in which the classification system 1 comprises N (N being an integer greater than or equal to 2) conversion devices 10, it is possible that the preprocessing step E0 comprises the transformation of the input signal SE into N initial signals SI. For example, these initial signals SI may be different or identical. Each of these initial signals SI may be associated with one of the N conversion devices 10.
[0134] Thus, it is possible to adapt the pre-processing step E0 to obtain an initial signal either according to the conversion device 10 to which it is associated, or according to a particular characteristic that one wishes to extract, or both. For example, it is possible to implement the pre-processing step according to the magnetic transmission values Rtr of each of the junction elements 30 of the N conversion devices 10.
[0135] According to this variant, the classification process may comprise N conversion steps E1, implemented by the N conversion devices 10, and the conversion process will yield N data sets SC. The identification step E2 can then consist of associating to the input signal SE, a class of signal, according to the N data sets SC obtained during the N conversion steps El.
Claims
1. Demands Conversion device (10) for converting an initial signal (IS) into a data set (DS) for an artificial neural application, the conversion device (10) comprising: • a magnetic component (20) comprising a magnetic material configured to vary locally between a first magnetic state (M1) having a first magnetization, and a second magnetic state (M2) having a second magnetization different from the first magnetization, said magnetic component (20) being subdivided into a plurality of zones, where each zone is either in the first magnetic state (M1) or in the second magnetic state (M2); each zone in the first magnetic state (M1) being separated from a zone in the second magnetic state (M2) by a magnetic domain wall (P); • a junction member (30) separating the magnetic component (20) between an upstream chamber (21) and a downstream chamber (23) and ensuring magnetic communication between the upstream chamber (21) and the downstream chamber (23), said junction member (30) being characterized by a magnetic transmission value (Rtr) corresponding to a capacity of the junction member (30) to allow a displacement of the magnetic domain wall (P) in a direction of propagation (X), said direction of propagation (X) being defined from the upstream chamber (21) to the downstream chamber (23); • an initialization unit (40) configured to place the magnetic component (20) in an initial magnetic configuration (Cl), in which the magnetic domain wall (P) is arranged at the level of the junction member (30); • an excitation unit (50) configured to generate a physical excitation to excite the magnetic material, so as to move the magnetic domain wall (P) in the direction of propagation (X), the physical excitation being generated as a function of the initial signal (SI); • a detection unit (60) disposed at the downstream chamber (23), said detection unit (60) being configured to detect and record at least one trigger characteristic parameter (Nbf) when a physical parameter associated with a deformation state of the magnetic domain wall (P) exceeds a predetermined threshold value (Vs), the data set (SC) including said at least one trigger characteristic parameter (Nbf); conversion device (10) in which the initialization unit (40) is configured to place the magnetic component (20) in the initial magnetic configuration (Cl) when the following condition is met: at least one trigger characteristic parameter (Nbf) is detected by the detection unit (60).
2. Conversion device (10) according to claim 1, further comprising a closed contour peripheral boundary (11) in which the magnetic component (20) is fully contained.
3. Conversion device (10) according to any one of claims 1 or 2, wherein the initialization unit (40) includes an expansion member (43) configured to move the magnetic domain wall (P) up to the junction member (30), so as to place the magnetic component (20) in the initial magnetic configuration (Cl).
4. Conversion device (10) according to any one of claims 1 to 3, further comprising a reconfiguration unit (70) configured to modify the magnetic transmission value (Rtr) of the junction member (30).
5. Conversion device (10) according to any one of claims 1 to 4, wherein the excitation unit (50) comprises a current generator (51) configured to generate an electric current.
6. Classification system (1) for classifying an input signal (SE) into a signal class (Ci), the classification system (1) comprising: • at least one pre-processing device (3) for transforming the input signal (SE) into at least one initial signal (SI);
7.
8. • at least one conversion device (10) according to any one of claims 1 to 5 taking as input an initial signal (SI) from said at least one initial signal (SI) and converting said initial signal (SI) taken as input, into a data set (SC); • an identification unit (5) taking as input the data set (SC) and configured to associate the input signal (SE) with a signal class (Ci) according to the data set (SC), said signal class (Ci) being chosen from a set of predetermined signal classes stored in a memory (7) of the classification system (1). Method for converting an initial signal (IS) into a data set (DS) for an artificial neuron application, the conversion method being implemented by a conversion device (10) according to any one of claims 1 to 5, and comprising the following phases: • an initialization phase (PI) comprising an initial step (E13) in which the magnetic component (20) is placed in the initial magnetic configuration (Cl); • an excitation phase (P2) in which the excitation unit (50) generates the physical excitation as a function of the initial signal (SI), so as to move the magnetic domain wall (P); • a reconfiguration phase (P3) implemented if the detection unit (60) detects that at least one physical parameter associated with the deformation state of the magnetic domain wall (P) exceeds the predetermined threshold value (Vs), the reconfiguration phase (P3) then comprising a detection step (E31) in which the detection unit (60) records at least one trigger characteristic parameter (Nbf) in the data set (SC); A conversion method wherein a new initialization phase (IP) is implemented at the end of the reconfiguration phase (P3). A conversion method according to claim 7, wherein the initialization unit (40) comprises a nucleation element (41), the phase initialization (PI) then comprising a nucleation step (Eli), implemented before the initial step (E13), in which the nucleation member (41) places the magnetic component (20) in the magnetic nucleation configuration (CO) in which all areas of the magnetic component (20) are placed in the first magnetic state (M1), with the exception of a nucleation area which is placed in the second magnetic state (M2), said nucleation area being contained in the upstream chamber (21).
9. A conversion method according to any one of claims 7 or 8, wherein the reconfiguration phase (P3) includes a stop step (E32), in which the excitation unit (50) stops at least temporarily the implementation of the excitation phase (P2); a new excitation phase (P2) being implemented at the end of the reconfiguration phase (P3).
10. A conversion method according to any one of claims 7 to 9, implemented by a conversion device (10) according to claim 4, wherein the reconfiguration phase (P3) further comprises a modification step (E33) implemented after the detection step (E31), in which the reconfiguration unit (70) modifies the magnetic transmission value (Rtr) of the junction member (30).
11. A classification method for classifying an input signal (ES) into a signal class (Ci), the classification method being implemented by a classification system (1) according to claim 6 and comprising the following steps: • a pre-processing step (E0) wherein the pre-processing device (3) transforms the input signal (ES) into at least one initial signal (IS); • at least one conversion step (E1) wherein a conversion method according to any one of claims 7 to 10 is applied by the conversion device (10) to an initial signal (IS) from among said at least one initial signal (IS), so as to obtain a data set (DS) corresponding to said initial signal (IS) to which the conversion method is applied; • an identification step (E2) wherein the identification unit (5) associates the input signal (ES) with a signal class (Ci) as a function of the data set (SC) obtained during at least one conversion step (El)