Information processing device, information processing method, and program

Abstract

To provide an information processing device, an information processing method, and a program for realizing a reservoir computer with higher accuracy.SOLUTION: An information processing device 1 includes an input part 5 and a magnetic body 62. The input part generates an AC magnetic field Hin that changes according to an input signal S1 from the outside. The magnetic body is magnetized to have a spontaneous magnetization M along a predetermined magnetization direction, and induces a spin wave including two or more of a first mode, a second mode, and a third mode by applying an AC magnetic field from the input part. The first mode is a mode propagating along a magnetization direction x, the second mode is a mode propagating along a first intersecting direction y intersecting the magnetization direction, and the third mode is a mode propagating along a second intersecting direction Z intersecting both the magnetization direction and the first intersecting direction.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to an information processing device, an information processing method, and a program. [Background technology]

[0002] Patent Document 1 discloses a technology relating to a reservoir calculation device with high calculation accuracy that can obtain many different reservoir states. The reservoir calculation device according to this technology includes a light source, a reservoir calculation element having an active layer that interacts with light from the light source, and a light adjusting means for adjusting the light from the light source, and the reservoir calculation element has a light-transmitting portion that allows light from the light source to reach at least a partial region of the active layer. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Publication No. 2024-66709 Summary of the Invention [Problem to be solved by the invention]

[0004] However, in order to handle a wider variety of computations in reservoir computing, various response characteristics to inputs are required. [Means for solving the problem]

[0005] According to one aspect of the present invention, there is provided an information processing device comprising an input section and a magnetic body, wherein the input section is configured to generate an AC magnetic field that changes in response to an external input signal, and the magnetic body is magnetized to have spontaneous magnetization along a predetermined magnetization direction, and wherein application of the AC magnetic field from the input section induces spin waves including two or more modes from a first mode, a second mode, and a third mode, wherein the first mode is a mode that propagates along the magnetization direction, the second mode is a mode that propagates along a first intersecting direction that intersects the magnetization direction, and the third mode is a mode that propagates along a second intersecting direction that intersects both the magnetization direction and the first intersecting direction.

[0006] With this configuration, a reservoir computer with higher accuracy can be realized. [Brief explanation of the drawings]

[0007] [Figure 1] FIG. 2 is a diagram illustrating an example of a hardware configuration of the information processing device according to the first embodiment. [Figure 2] 2 is a plan view of the first embodiment of the information processing device shown in FIG. 1, seen from the Y direction. [Figure 3] 10 is a diagram for explaining a propagation mode of a spin wave SW using an input unit 5. FIG. [Figure 4] 1A and 1B are diagrams showing the propagation modes of (a) forward volume spin waves, (b) backward volume spin waves, and (c) surface spin waves as different modes of spin wave propagation. [Figure 5] 2 is a diagram showing an example of a propagation mode of a spin wave SW in response to an AC magnetic field in the magnetic material of the information processing device 1 shown in FIG. [Figure 6] FIG. 1 is a diagram showing an information processing device 1 used in this simulation. [Figure 7] FIG. 10 is a diagram showing the distribution of the Y component of the spin of the magnetic body 62 for each angle θ, obtained by a simulation related to the first embodiment. [Figure 8]FIG. 10 is a diagram showing the dependence of the frequency spectrum of the detection signal of the central detection terminal on the angle θ. [Figure 9] This figure shows the waveform of the output signal S5 decoded by the decoder based on the detection signal from the detection terminal in this simulation, and the sawtooth waveform corresponding to the training data, for angles θ = 30° and 90°. [Figure 10] FIG. 10 is a diagram showing simulation results for the dependence of the root mean square error (RMSE) of the predicted signal on the angle θ. [Figure 11] FIG. 10 is a diagram showing a simulation result regarding the dependency of the performance improvement index PI on the angle θ. [Figure 12] FIG. 2 is a diagram illustrating an example of a hardware configuration of an information processing device 1 according to a second embodiment. [Figure 13] FIG. 10 is a diagram showing the distribution of the Y component of the spin of the magnetic body 62 for each angle θ′, obtained by a simulation relating to the second embodiment. [Figure 14] 10 is a diagram showing a simulation result of the angle θ′ dependency of NMSE of the second embodiment of the information processing device 1. FIG. [Figure 15] 10A is a diagram showing the angle θ′ dependency of the total memory capacity of the information processing device 1 according to the first embodiment, and FIG. 10B is a diagram showing the angle θ′ dependency of the total memory capacity of the information processing device 1 according to the second embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0008] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described below with reference to the accompanying drawings. Various features shown in the following embodiments can be combined with each other.

[0009] Incidentally, the program for realizing the software appearing in one embodiment may be provided as a non-transitory computer-readable medium, or may be provided so that it can be downloaded from an external server, or may be provided so that the program is started on an external computer and its functions are realized on a client terminal (so-called cloud computing).

[0010] Furthermore, various information processing according to an embodiment may realize input and output corresponding to the input. Here, the form of information referenced in such information processing (hereinafter referred to as reference information) is not limited as long as an output is obtained as a result of the input. The reference information may be, for example, rule-based information such as a database, a lookup table, or a predetermined function (including a decision formula such as a regression formula constructed using a statistical method), a trained model that has previously trained the correlation between input and output, or a generative AI such as a large-scale language model or a visual language model that can output a desired result in response to a prompt input.

[0011] In one embodiment, a "unit" may include, for example, a combination of hardware resources implemented by a circuit in the broad sense and software information processing that can be specifically realized by these hardware resources. In one embodiment, various information is handled, and this information is represented, for example, by physical values ​​of signal values ​​representing voltage and current, high and low signal values ​​as a binary bit set consisting of 0 or 1, or quantum superposition (so-called quantum bits), and communication and calculations can be performed on a circuit in the broad sense.

[0012] Furthermore, a circuit in the broad sense is a circuit realized by at least an appropriate combination of a circuit, circuitry, processor, memory, etc. The processor may be a general-purpose processor or a dedicated circuit. That is, it includes an application specific integrated circuit (ASIC), a programmable logic device (e.g., a simple programmable logic device (SPLD), a complex programmable logic device (CPLD), and a field programmable gate array (FPGA)), etc.

[0013] 1. First embodiment In this chapter, a first embodiment of the hardware configuration of the information processing device will be described.

[0014] 1.1. Hardware configuration example 1 is a diagram illustrating an example of the hardware configuration of a first embodiment of an information processing device. Hereinafter, for convenience of explanation, it is assumed that an orthogonal coordinate system is defined in which the three directions used in the diagram, that is, the X direction, the Y direction, and the Z direction, are orthogonal to one another. As shown in FIG. 1, the information processing device 1 includes an input device 2, an encoder 3, a current source 4, an input unit 5, an external field application unit MAG, a physical reservoir 6, a detection unit 7, an extraction unit 8, a decoder 9, and an output terminal 10.

[0015] The input device 2 is configured to receive an external input signal S1. The input device 2 can be designed appropriately depending on the form of the input signal S1, and can be particularly configured to receive an input signal that changes over time. For example, the input device 2 can be a sound collector capable of detecting audio signals, a camera capable of detecting continuous images that change over time, a vibration detector capable of detecting device vibration, or a photodetector capable of detecting changes in light over time. The input device 2 outputs the external input signal S1 as an electrical signal.

[0016] The encoder 3 is configured to convert an input signal S1 from the input device 2 into a control signal S2 for a current source 4, which will be described later. The encoder 3 may also function as an amplifier that amplifies the input signal S1 input from the input device 2. The encoder 3 may also be incorporated into the input device 2 or the current source 4.

[0017] The current source 4 outputs an input current I as an electrical signal based on the control signal S2. in The input current I in is an alternating current that changes over time.

[0018] The input section 5 receives an input signal S1 (in this embodiment, an input current I in ) which varies with the alternating magnetic field H in In this embodiment, the input unit 5 is a conductor antenna that extends linearly along the input direction, and receives the input current I from the current source 4. in In other words, the input unit 5 transmits an electric signal along the input direction in response to the input signal S1, thereby generating an input current I as an electric signal. in The alternating magnetic field H with time change inIn this embodiment, the input direction (i.e., the direction in which the input unit 5 extends) is defined to intersect with at least two of the X, Y, and Z directions. In this embodiment, the input direction is perpendicular to the Z direction and extends in an XY plane defined by the X and Y directions so as to intersect with both the X and Y directions. In this chapter, for convenience of explanation, the angle of the input direction with respect to the X direction is denoted as θ. In this embodiment, the tilt angle indicating the angle of the magnetization direction with respect to the input direction is equal to the angle θ. The angle θ can be in the range of 0°<θ<90°. Note that if the angle θ=0°, the input direction coincides with the X direction, and if the angle θ=90°, the input direction coincides with the Y direction. However, in this embodiment, the input direction is defined so as not to coincide with either the X or Y direction in relation to the magnetization direction, which will be described later. Specifically, the angle θ is, for example, 5, 10, 15, 20, 25, 30, 35, 40, 45, 50, 55, 60, 65, 70, 75, 80, or 85°, and may be within a range between any two of the values ​​exemplified here. In particular, from the viewpoint of the performance of the physical reservoir 6, the angle θ is preferably 15 to 75°, and more preferably 15 to 45°.

[0019] The external field application unit MAG applies an AC magnetic field H along a direction that intersects with the input direction. in According to such a configuration, the magnetization direction of the magnetic body 62 can be controlled by applying the external magnetic field H0, thereby increasing the flexibility of the arrangement of the input unit 5. For example, the external magnetic field H0 can be set to a value different from the external magnetic field H in The external field application unit MAG may be a permanent magnet or an electromagnet. In this embodiment, the external field application unit MAG is configured to apply an external magnetic field H0 along the X direction.

[0020] The physical reservoir 6 is an alternating magnetic field H in The magnetic material 62 is configured to function as an intermediate layer in reservoir computing for the above-mentioned semiconductor device. For example, the magnetic material 62 includes a substrate 61 and a magnetic material.

[0021] The substrate 61 may be made of a non-magnetic material such as a silicon substrate, an aluminum oxide substrate, etc. The substrate 61 may also be a flexible substrate.

[0022] The magnetic body 62 is magnetized to have spontaneous magnetization M along a predetermined magnetization direction, and is irradiated with the AC magnetic field H from the input unit 5. in The magnetic material 62 is configured so that spin waves SW are induced by applying a magnetic field. The propagation mode of the spin waves SW will be described later. For example, the magnetic material 62 is a ferromagnetic layer stacked on the substrate 61. The magnetic material 62 has a thickness in the Z direction and extends planarly in the X and Y directions perpendicular to the Z direction. The ferromagnetic material may be, for example, a simple metal such as iron, cobalt, nickel, or gadolinium, or a compound such as ferrite, yttrium iron garnet (YIG), or chromium dioxide. From the viewpoint of uniformly aligning the magnetic domains, the magnetic material 62 is not limited to a ferromagnetic material, but may also be a so-called weak ferromagnetic material such as a ferrimagnetic material or a tilted antiferromagnetic material. Furthermore, the magnetic easy axis and the spin easy axis of the magnetic material 62 may be different. From the viewpoint of complex propagation of spin waves, a magnetic material with antiferromagnetic order, such as a ferrimagnetic material or a tilted antiferromagnetic material, is preferable. In this embodiment, the magnetic material 62 is a single-crystal thin film containing ferrite garnet as a main component.

[0023] The magnetic body 62 is magnetized to have spontaneous magnetization M along the X direction based on the external magnetic field H0 generated by the external field application unit MAG. As a result, the magnetization direction intersects with the input direction at an angle θ in the XY plane. In the first embodiment, the X direction corresponds to the magnetization direction, the Y direction intersecting (specifically, perpendicular to) the X direction corresponds to a first intersecting direction intersecting with the magnetization direction, and the Z direction corresponds to a second intersecting direction intersecting with both the X direction and the Y direction. Therefore, the input direction can intersect with at least two of the magnetization direction, the first intersecting direction, and the second intersecting direction. With this configuration, a stable AC magnetic field H can be generated through an electrical signal. in can be applied to the magnetic body 62.

[0024] In this embodiment, the magnetic body 62 includes a propagation region R1 as a first region, a damping region R2 as a second region, and a boundary portion B.

[0025] The propagation region R1 is a region where the AC magnetic field H in The propagation region R1 is a region where the spin waves SW excited by the magnetic field H propagate (preferentially compared to the damping region R2 described later). The propagation region R1 is, for example, a flat plane formed in the center of the magnetic body 62 and perpendicular to the Z direction, and in this embodiment, is defined as a square with a side length of L1. In this embodiment, the input unit 5 applies an AC magnetic field H in is placed so as to lie on the propagation region R1 so as to apply a voltage to the propagation region R1. For example, the input section 5 is placed so as to be in contact with the propagation region R1.

[0026] The propagation region R1 includes at least one detection region R11. The detection region R11 is a position where the spin wave SW is detected. In this embodiment, five point-like detection regions R11 are arranged in each of the X and Y directions, for a total of 25 detection regions R11. The shape of the detection region R11 is determined appropriately depending on the specific aspect of the detection unit 7 described below, and may be virtual within the propagation region R1. For example, the detection region R11 may be a point-like (zero-dimensional) region as in this embodiment, a linear (one-dimensional) region extending along the X or Y direction, or a two-dimensional region having a certain area in the XY plane.

[0027] The damping region R2 is defined to cover the outer periphery of the first region. For example, the damping region R2 is defined as a square with a side length of L2 that surrounds the entire outer periphery of the propagation region R1. The ratio L1 / L2 of the length L1 of the propagation region R1 to the length L2 of the damping region R2 is specifically, for example, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, or 0.9, and may be within a range between any two of the values ​​exemplified here.

[0028] The boundary portion B defines the boundary between the propagation region R1 and the damping region R2 and is configured to reflect or absorb spin waves propagating from the propagation region R1 toward the damping region R2. An example of the configuration of the boundary portion B will now be described. FIG. 2 is a plan view of the first embodiment of the information processing device shown in FIG. 1 , viewed from the Y direction. As shown in FIG. 2, the propagation region R1 protrudes in the Z direction relative to the damping region R2. In this case, the boundary portion B is defined as a step surface connecting the outer periphery of the propagation region R1 and the inner periphery of the damping region R2. A portion SW1 of the spin wave SW induced by the input portion 5 propagates toward the center of the magnetic body 62, while a portion SW2 of the spin wave SW propagates toward the outer periphery of the propagation region R1. The spin wave SW2 propagating toward the propagation region R1 enters the step surface, is absorbed or attenuated, is reflected, and then propagates toward the inner region of the propagation region R1, or is transmitted toward the damping region R2, where it is attenuated and disappears. In this embodiment, the propagation region R1 and the damping region R2 are integrally formed. For example, the propagation region R1 and the damping region R2 can be obtained by laminating the same material at different thicknesses. This configuration allows for adjustment of the reflection of spin waves at the boundary of the propagation region R1, thereby promoting the propagation of spin waves with higher nonlinearity. Any method can be used to create a difference in the damping constant between the propagation region R1 and the damping region R2. Examples include changing the laminate structure between the propagation region R1 and the damping region R2 (e.g., evaporating a different material onto the surface of only the damping region R2), changing the state of the material between the propagation region R1 and the damping region R2 (e.g., doping only the damping region R2 with a different material), or changing the film thickness between the propagation region R1 and the damping region R2.

[0029] Returning to FIG. 1, the detection unit 7 is configured to detect temporal changes in the internal magnetic field of the magnetic body 62 due to the propagation of spin waves. Here, the detection unit 7 detects temporal changes in the internal magnetic field in each of the multiple detection regions R11 so as to be synchronized in time. For example, the detection unit 7 may be a magnetic sensor (e.g., a pickup coil, a magnetoresistive element, a Hall element, etc.) that detects temporal changes in the magnetic distribution on the surface of the detection region R11 of the magnetic body 62, a shape change in the magnetic body 62 caused by spin waves (e.g., a piezoelectric element capable of detecting spin surface acoustic waves), or a photodetector that detects reflected light of light (e.g., a laser) irradiated onto the detection region R11. The detection unit 7 outputs the detection result of each detection region R11 as a detection signal S3.

[0030] The extraction unit 8 extracts the AC magnetic field H from the time change in the internal magnetic field of the magnetic body 62 detected by the detection unit 7. in The extraction unit 8 extracts the component corresponding to the second harmonic of the AC magnetic field H. This configuration makes it possible to provide a reservoir computing with higher performance. in Even if the filter is an analog circuit (for example, a high-pass filter or a band-pass filter) that preferentially passes a specific frequency component depending on the frequency, the filter may be configured to perform a Fourier transform on the detection result and process the specific frequency component preferentially. As a result, the extraction unit 8 outputs the detection signal S3 output from the detection unit 7 as a processed signal S4.

[0031] The decoder 9 outputs an output signal S5 in accordance with a pre-learned manner based on the detection signal S3 of the detection unit 7 (in this embodiment, the processed signal S4 extracted by the extraction unit 8). When the information processing device 1 is configured to function as a determiner, the output signal S5 outputs a determination result of the state of the input signal S1 represented by the processed signal S4. For example, when the information processing device 1 is used to detect a failure of a certain device, the output signal S5 outputs a determination result indicating whether or not the device has failed.

[0032] The output terminal 10 is configured to perform desired information processing in accordance with the output signal S5 output from the decoder 9. For example, the output terminal 10 is configured to visually present the determination result represented in the output signal S5 to a user via a display unit. Furthermore, for example, the output terminal 10 may be configured to control a device in accordance with the determination result represented in the output signal S5.

[0033] 1.2. Spin wave propagation Next, a description will be given of the propagation mode of the spin wave induced in the magnetic body 62. FIG.

[0034] As shown in FIG. 3, an input current I in When the current flows, an alternating magnetic field H in is induced. The induced AC magnetic field H in A part of the AC magnetic field H in The direction of the input current I in This alternating magnetic field H in As the time changes, a disturbance is induced in the spins in the magnetic material 62, and the change in the state of the spins due to the disturbance induces a change in the state of other spins through exchange interaction. The propagation of such a change in the state of the spin system is excited in the spin system of the magnetic material 62 as a spin wave SW characterized by a wave vector k. Here, the spin wave SW characterized by the wave vector k propagates in a single propagation mode with the same group velocity.

[0035] The spin waves propagating through the magnetic material 62 in this embodiment can be expressed as a composite wave of multiple modes with different group velocities. FIG. 4 shows the propagation modes of (a) forward volume spin waves, (b) backward volume spin waves, and (c) surface spin waves as different modes of spin wave propagation. In FIG. 4, H indicates the magnetization direction of the magnetic material 62 (in other words, the direction of the external magnetic field H0), and k indicates the propagation direction (i.e., wave vector) of spin waves propagating at the same group velocity. For convenience of explanation, H is referred to as the magnetization direction here. As shown in FIGS. 4(a) to 4(c), forward volume spin waves (FVSW) are a propagation mode of spin waves having a wave vector k perpendicular to the component of the spontaneous magnetization of the thin-film magnetic material 62 in the thickness direction of the thin film. Backward volume spin waves (BVSW) are a propagation mode of spin waves having a wave vector k parallel to the in-plane component of the spontaneous magnetization of the thin-film magnetic material 62. Surface spin waves (SSW) are a propagation mode of spin waves having a wave vector k perpendicular to the in-plane component of the spontaneous magnetization of the thin-film magnetic material 62. In other words, backward volume spin waves are a mode that propagates along the magnetization direction (X direction in the first embodiment). Surface spin waves are a mode that propagates along the first intersecting direction (Y direction in the first embodiment). Forward volume spin waves are a mode that propagates along the second intersecting direction (Z direction in the first embodiment).

[0036] 5 is a diagram showing an example of the propagation mode of the spin wave SW in the magnetic material of the information processing device 1 shown in FIG. 1 with respect to the AC magnetic field. As shown in FIG. 5, the input section 5 is defined so as to intersect in the XY plane with the magnetization direction aligned by the external magnetic field H0. Therefore, the AC magnetic field H in At this time, the alternating magnetic field H in From the region where the AC magnetic field H inA spin wave SW is excited according to the direction of the magnetic field, and the spin wave SW travels, for example, according to a wave vector k along the input direction of the input unit 5. The spin wave SW is described as a composite vector of the wave vectors of multiple modes, and the wave vectors of these modes propagate at different group velocities, causing reflection and absorption at the boundary B of the magnetic material 62.

[0037] At this time, the wave vector k intersects with the magnetization direction of the magnetic body 62 (in other words, the direction of the external magnetic field H0) at an angle θ, so that the spin waves SW propagating through the magnetic body 62 include two or more modes among backward volume spin waves, which are an example of a first mode, surface spin waves, which are an example of a second mode, and forward volume spin waves, which are an example of a third mode. in This can improve the nonlinearity of the response of the magnetic material 62 to the magnetic field while improving the damping memory performance of the physical reservoir 6. In this embodiment, the spin waves SW propagate in the XY plane in the magnetic material 62 having a magnetization direction in the XY plane, and therefore may contain, in particular, backward volume spin waves and surface spin waves as major components.

[0038] These modes propagate at different group velocities in the magnetic material 62. As a result, the spin waves SW propagating in the magnetic material 62 exhibit more complex time variations than simple plane waves propagating linearly from the input unit 5. As a result, the information processing device 1 can provide reservoir computing with higher nonlinearity and the like than conventional spin-based physical reservoirs that propagate spin waves containing only a single mode.

[0039] 1.3. Simulation of the First Embodiment of the Information Processing Device 1 Next, a simulation relating to the first embodiment of the information processing device 1 will be described.

[0040] Setup First, the inventors performed a magnetic simulation of a chip of magnetic material 62 using COMSOL Multiphysics (manufactured by COMSOL Inc.). Figure 6 is a diagram showing the information processing device 1 used in this simulation. As shown in Figure 6, the magnetic material 62 used in this simulation is a square chip of 3 μm × 3 μm. YIG was used as the material constituting the magnetic material 62. Furthermore, the angle θ was defined by the input direction in which the input section 5 extends, with one side of the propagation region R1 parallel to the direction of the external magnetic field H0 as the reference. The parameters used in this simulation are as shown in the table below.

[0041] [Table 1]

[0042] In the propagation region R1, a 5×5 spot of detection terminals is provided as a detection region R11. These detection terminals detect the Y component of each spin of the magnetic material 62, and can directly represent a single waveform.

[0043] 1.3.2. Changes in spin distribution Next, we will explain the distribution of the Y-component of spin as one of the results of a simulation based on the above setup. FIG. 7 is a diagram showing the distribution of the Y-component of spin of the magnetic body 62 for each angle θ, obtained by a simulation related to the first embodiment. The square region in the center of FIG. 7 corresponds to the propagation region R1, and the region covering the propagation region R1 corresponds to the damping region R2. The elongated square region extending across the propagation region R1 corresponds to the input section 5. Here, we performed a simulation of the time evolution of spin over a period of 15 ns. The angles θ set for (a) to (g) in FIG. 7 were (a) 0°, (b) 15°, (c) 30°, (d) 45°, (e) 60°, (f) 75°, and (g) 90°. That is, in the cases of (a) and (g), the input direction coincides with the X-direction or the Y-direction, and in the other cases, the input direction intersects both the X-direction and the Y-direction.

[0044] As shown in Figures 7(a) and 7(g), when the angle θ = 0° or 90°, a plane wave-like pattern with a periodic wavefront along the direction perpendicular to the input direction was observed in the propagation region R1. This suggests that the spin waves SW propagate with a nearly uniform group velocity throughout the entire propagation region R1, in other words, propagate in a single mode (SSW for θ = 0°, BVSW for θ = 90°).

[0045] On the other hand, as shown in Figure 7(b) to (f), when the angle θ ≠ 0°, 90°, a complex pattern with poor periodicity was observed in the direction perpendicular to the input direction compared to when the angle θ = 0°, 90°. This suggests that the spin waves SW propagating perpendicular to the input direction propagate with different group velocities in the X and Y directions.

[0046] 1.3.3. Dependence of the frequency spectrum of the signal detected by the detection terminal on the angle θ The dependence of the frequency spectrum of the signal detected by the central detection terminal on the angle θ will now be described. FIG. 8 is a diagram showing the dependence of the frequency spectrum of the signal detected by the central detection terminal on the angle θ. As shown in FIG. 8, the output signal is in frequency f in It was found that the frequency of the 5 GHz signal contains peaks at 3.93 GHz, 10 GHz (second harmonic), 15 GHz (third harmonic), and 20 GHz (fourth harmonic). By extracting the frequency components corresponding to these peaks using the extraction unit 8 and decoding them using the decoder 9, it is possible to obtain the desired output, such as a determination result. In particular, the peak at 10 GHz, which is the second harmonic, was observed at all angles θ in the simulation.

[0047] 1.3.4. Simulation results regarding the dependence of the evaluation index of the output signal S5 on the angle θ Next, the dependence of the performance index of the information processing device 1 on the angle θ will be described as a result of this simulation. First, for learning the decoder 9, the input current I in Decoder 9 was trained using a set of training data of 8 ns, which output a sawtooth wave that oscillates periodically in 20 steps (each step corresponds to a time of 0.01 ns) for the input signal. The training method used was ridge regression, with the normalization parameter set to 0.01. 200 steps, equivalent to 2 ns, were then used as test data to evaluate how well the output signal reproduced the sawtooth wave of the training data.

[0048] FIG. 9 shows the waveform of the output signal S5 decoded by the decoder based on the detection signal from the detection terminal in this simulation, and the waveform of the sawtooth wave corresponding to the teacher data, for angles θ=30° and 90°. In FIG. 9, the teacher signals TS1 and TS2 corresponding to the "Teacher signal" indicate the (correct) waveform corresponding to the teacher data, and the predicted signals PS1 and PS2 corresponding to the "Predicted signal" indicate the waveform of the output signal S5 output from the information processing device 1 (decoder 9) learned through this simulation. The smaller the difference between the predicted signals PS1 and PS2 and the teacher signals TS1 and TS2, the higher the accuracy of the output result of the information processing device 1. As shown in FIG. 9, the difference between the predicted signal PS1 and the teacher signal TS1 when θ=30° is smaller than the difference between the predicted signal PS2 and the teacher signal TS2 when θ=90°, which can be intuitively observed from the comparison of the waveforms.

[0049] Next, a quantitative evaluation was performed on the difference between the predicted signal and the teacher signal. FIG. 10 shows the simulation results on the dependency of the root mean square error (RMSE) of the predicted signal on the angle θ. FIG. 11 shows the simulation results on the dependency of the performance improvement index PI on the angle θ. In FIGS. 10 and 11, "Training RMSE" indicates the RMSE of the predicted signal for the teacher data, and "Testing RMSE" indicates the RMSE of the predicted signal for the test data. The dashed line in FIG. 10 indicates the value of the evaluation index when a physical reservoir is not used.

[0050] The PI (Performance Improvement) shown in FIG. 11 is defined as follows using the RMSE used in FIG.

number

[0051] In the above formula, "RMSE of proposal" is the RMSE of the information processing device 1 obtained in this simulation (that is, "Training RMSE" or "Testing RMSE" shown in FIG. 10).

[0052] As shown in Figure 10, the use of a physical reservoir reduced the RMSE in all angle θ regions. In particular, both RMSEs were reduced in the 0°<θ<90° region compared to when the angle θ was 0° and 90°. Among these, the RMSE was particularly reduced when the angle θ was 30°≦θ≦75°. Furthermore, as shown in Figure 11, both performance improvement indices PI were improved in the 0°<θ<90° region compared to when the angle θ was 0° and 90°. Among these, particularly when the angle θ was 30°≦θ≦75°, a particularly high performance improvement indices were obtained.

[0053] The above simulation results suggest that the first embodiment of the information processing device 1 has higher performance than the prior art.

[0054] 2. Second embodiment Next, a description will be given of a second embodiment of the information processing device 1. For the sake of convenience, the following description may omit the description of the configuration of the second embodiment of the information processing device 1 that is common to the first embodiment of the information processing device 1 by assigning the same reference numerals.

[0055] 2.1. Hardware Configuration Example FIG. 12 is a diagram illustrating an example of the hardware configuration of the information processing device 1 according to the second embodiment. The configuration of the information processing device 1 according to the second embodiment, excluding the external field application unit MAG, the input unit 5, and the physical reservoir 6, is the same as that of the information processing device 1 according to the first embodiment, and is therefore omitted from FIG. 12. In the second embodiment, the external field application unit MAG is configured to apply an external magnetic field H0 in a direction intersecting both the X and Y directions. As a result, the magnetization direction of the spontaneous magnetization M of the magnetic material 62 intersects both the X and Y directions, just like the external magnetic field H0. In this chapter, the magnetization direction, the first intersecting direction, and the second intersecting direction are parallel to the X, Y, and Z directions, respectively, in the first embodiment. In the second embodiment, the magnetization direction is parallel to the X direction, while the input unit 5 extends along the Y direction, so that the input direction extends along the Y direction. In the second embodiment, the magnetization direction is indicated by the angle θ′. The angle θ' is defined as the angle of the magnetization direction (direction of the external magnetic field H0) relative to the X direction, and has a relationship of θ' = 90° - θ with respect to the angle θ in the first embodiment (i.e., the tilt angle indicating the angle between the input direction and the magnetization direction). In addition, the external field application unit MAG may be configured to be able to control the direction of the external magnetic field H0 by, for example, rotating around the magnetic body 62. This configuration makes it easy to adjust the arrangement so that the performance of the physical reservoir 6 can be maximized.

[0056] 2.2. Simulation of the second embodiment of the information processing device 1 Next, a simulation of the information processing device 1 according to the second embodiment will be described. Note that the setup of the simulation differs in that the input direction is along the Y direction, the magnetization direction is rotated instead of the input direction, and θ' is used as the angle instead of θ, but the parameter values ​​and simulation procedure are the same, so a description thereof will be omitted.

[0057] 13 is a diagram showing the distribution of the Y-component of the spin of the magnetic body 62 for each angle θ′ obtained by a simulation related to the second embodiment. As shown in FIG. 13, at angles θ′=0° and 90°, a relatively simple pattern with periodicity in both the X and Y directions was observed, similar to the simulation results related to the first embodiment, similar to the spin wave SW propagating in a single mode. However, in the range of 0°<θ′<90°, it was observed that the width of the plane wave in the Y direction became smaller as the distance from the input section 5 in the X direction increased, and the periodicity along the Y direction in particular was significantly reduced. This suggests that the intersection of the input direction and the magnetization direction can induce a spin wave SW including multiple modes, regardless of the shape of the propagation region R1.

[0058] Fig. 14 is a diagram showing simulation results of the angle θ' dependency of NMSE in the second embodiment of the information processing device 1. As shown in Fig. 14, in the second embodiment as well, when 0°<θ'<90°, it was observed that the NMSE of the information processing device 1 was reduced compared to when the angle θ'=0°, 90°, and an excellent NMSE reduction effect was observed particularly in the range of 45°≦θ'≦75° (i.e., 15°≦θ≦45°).

[0059] 15 is a diagram showing the angle θ' dependence of (a) the total memory capacity of the first embodiment of the information processing device 1 and (b) the total memory capacity of the second embodiment of the information processing device 1. As shown in FIG. 15, in both the first and second embodiments, it was shown that the total memory capacity of the physical reservoir 6 can be increased by using spin waves SW including multiple modes.

[0060] From the above, it has been shown that the information processing devices 1 of the first and second embodiments can both function as physical reservoir computers superior to the prior art.

[0061] [others] The above embodiment may be modified as follows.

[0062] In the above embodiments, the boundary B is defined as a step between the propagation region R1 and the damping region R2, but this is not limiting. For example, the boundary B may be defined as a groove or a protrusion between the propagation region R1 and the damping region R2. The boundary B may also be defined by constructing the propagation region R1 and the damping region R2 from different materials. In this case, the damping region R2 may be constructed from a material that does not have long-range magnetic order.

[0063] In each of the above embodiments, the electrical signal transmitted to the input unit 5 is an input current I in Not limited to, AC magnetic field H in Any suitable electrical signal may be used as long as it can generate an electric field. For example, the electrical signal transmitted to the input unit 5 may be an AC voltage signal. Furthermore, the input unit 5 is not limited to being provided in a closed circuit through which a current flows, but may be incorporated in an open circuit that induces a time-varying electric field in space due to the application of an AC voltage.

[0064] In each of the above embodiments, the input unit 5 is not limited to a conductor antenna that extends linearly along the input direction. That is, the input unit 5 is configured to apply an AC magnetic field H in The direction of application intersects at least two of the magnetization direction, the first intersecting direction, and the second intersecting direction. With this configuration, the input AC magnetic field H in This can improve the nonlinearity of the response of the magnetic material 62 to the magnetic field while improving the attenuation memory performance of the magnetic material 62 as a physical reservoir.

[0065] In each of the above embodiments, the information processing device 1 was used in a state in which the magnetization direction of the magnetic body 62 was aligned by the external field application unit MAG, but as long as the magnetic domains of the magnetic body 62 can be maintained in a state in which they are aligned to a certain extent, the information processing device 1 does not need to be equipped with the external field application unit MAG.

[0066] In the above embodiments, the magnetization direction, the first intersecting direction, and the second intersecting direction are defined to be perpendicular to each other, but these directions may be set appropriately depending on the reference direction of the propagation mode of the spin wave. For example, the magnetization direction, the first intersecting direction, and the second intersecting direction may be defined to be perpendicular to each other at an acute angle. Such directions may be set appropriately depending on, for example, the crystal structure or spin structure of the magnetic body 62.

[0067] The above embodiment is not limited to the information processing device 1, and may be an information processing method or a program. That is, the information processing method includes the following steps: In the magnetic field generation step, an AC magnetic field H in In the inducing step, an AC magnetic field H is applied to the magnetic material 62 having spontaneous magnetization along a predetermined magnetization direction. in By applying a magnetic field H to the magnetic material 62, spin waves including two or more modes selected from a first mode, a second mode, and a third mode are induced in the magnetic material 62. The first mode is a mode that propagates along the magnetization direction. The second mode is a mode that propagates along a first intersecting direction that intersects with the magnetization direction. The third mode is a mode that propagates along a second intersecting direction that is orthogonal to both the magnetization direction and the first intersecting direction. With this configuration, the input AC magnetic field H in It is possible to provide reservoir computing that achieves both the nonlinearity of the response of the magnetic material 62 to the magnetic field and the attenuation memory performance of the magnetic material 62 as a physical reservoir. The program causes a computer to execute each step of the information processing method.

[0068] The information processing device 1 and the like may be provided in the following aspects.

[0069] (1) An information processing device comprising an input unit and a magnetic body, wherein the input unit is configured to generate an AC magnetic field that changes in response to an external input signal, the magnetic body is magnetized to have spontaneous magnetization along a predetermined magnetization direction, and is configured such that application of the AC magnetic field from the input unit induces spin waves including two or more modes from among a first mode, a second mode, and a third mode, wherein the first mode is a mode that propagates along the magnetization direction, the second mode is a mode that propagates along a first intersecting direction that intersects with the magnetization direction, and the third mode is a mode that propagates along a second intersecting direction that intersects both the magnetization direction and the first intersecting direction.

[0070] With this configuration, a reservoir computer with higher accuracy can be realized.

[0071] (2) In the information processing device described in (1) above, the input unit is configured to apply the AC magnetic field to at least a portion of the magnetic body along an application direction, and the application direction intersects at least two of the magnetization direction, the first intersecting direction, and the second intersecting direction.

[0072] With this configuration, a reservoir computer with higher accuracy can be realized.

[0073] (3) In the information processing device described in (1) or (2) above, the input unit is configured to generate the AC magnetic field in accordance with the time change of the electrical signal by transmitting an electrical signal along an input direction in response to the input signal, and the input direction intersects at least two of the magnetization direction, the first intersecting direction, and the second intersecting direction.

[0074] With this configuration, a stable AC magnetic field can be applied to the magnetic body through an electric signal.

[0075] (4) In the information processing device described in (3) above, the magnetic body comprises a first region through which the spin waves propagate, a second region covering the outer edge of the first region, and a boundary portion that reflects or absorbs the spin waves propagating from the first region toward the second region.

[0076] According to this configuration, the reflection of the spin waves at the boundary of the first region can be adjusted, and therefore, propagation of the spin waves with higher nonlinearity can be promoted.

[0077] (5) An information processing device according to (4) above, further comprising an external field application unit, wherein the external field application unit is configured to apply an external magnetic field different from the AC magnetic field along a direction intersecting the input direction.

[0078] According to this configuration, the magnetization direction of the magnetic material can be controlled by applying an external magnetic field, thereby increasing the flexibility in the arrangement of the input section.

[0079] (6) An information processing device according to any one of (1) to (5) above, further comprising a detection unit and an extraction unit, wherein the detection unit is configured to detect the time change in the internal magnetic field of the magnetic body due to the propagation of the spin waves, and the extraction unit extracts a component corresponding to the second harmonic of the AC magnetic field from the detected time change in the internal magnetic field of the magnetic body.

[0080] Such a configuration can provide higher performance reservoir computing.

[0081] (7) An information processing method, comprising the following steps: in the magnetic field generating step, an AC magnetic field is generated in response to an external input signal; and in the inducing step, spin waves including two or more modes selected from a first mode, a second mode, and a third mode are induced in the magnetic body by applying the AC magnetic field to a magnetic body having spontaneous magnetization along a predetermined magnetization direction, wherein the first mode is a mode that propagates along the magnetization direction, the second mode is a mode that propagates along a first intersecting direction that intersects with the magnetization direction, and the third mode is a mode that propagates along a second intersecting direction that is perpendicular to both the magnetization direction and the first intersecting direction.

[0082] With this configuration, a reservoir computer with higher accuracy can be realized.

[0083] (8) A program that causes a computer to execute each step of the information processing method described in (7) above. Of course, this is not the case.

[0084] Finally, while various embodiments of the present disclosure have been described, they are presented as examples and are not intended to limit the scope of the invention. The novel embodiments may be embodied in various other forms, and various omissions, substitutions, and modifications may be made without departing from the spirit of the invention. Such embodiments and modifications are intended to be included within the scope and spirit of the invention, as well as within the scope of the inventions and their equivalents as defined in the claims. [Explanation of symbols]

[0085] 1: Information processing device 2: Input device 3: Encoder 4: Current source 5: Input section 6: Physical reservoir 61: Circuit board 62:Magnetic material 7: Detection unit 8:Extraction part 9: Decoder 10: Output terminal B: Boundary H0: External magnetic field H in :AC magnetic field I in : Input current M: spontaneous magnetization MAG: External field application section PS1: Prediction signal PS2: Predictive signal R1: Propagation region R11: Detection area R2: Damping region S1: Input signal S2: Control signal S3: Detection signal S4: Processed signal S5: Output signal SW: spin wave TS1: Teacher signal TS2: Teacher signal k: wave vector θ,θ': angles

Claims

1. An information processing device, An input unit and a magnetic body, the input unit is configured to generate an AC magnetic field that changes in response to an external input signal; The magnetic body is magnetized to have a spontaneous magnetization along a predetermined magnetization direction; a spin wave including two or more modes selected from a first mode, a second mode, and a third mode is induced by applying the AC magnetic field from the input unit; the first mode is a mode that propagates along the magnetization direction, the second mode is a mode propagating along a first intersecting direction that intersects the magnetization direction, An information processing device, wherein the third mode is a mode that propagates along a second intersecting direction that intersects both the magnetization direction and the first intersecting direction.

2. 2. The information processing device according to claim 1, An information processing device, wherein the input unit is configured to apply the AC magnetic field to at least a portion of the magnetic body along an application direction, the application direction intersecting at least two of the magnetization direction, the first intersecting direction, and the second intersecting direction.

3. 2. The information processing device according to claim 1, the input unit is configured to generate the AC magnetic field in accordance with a time change of the electrical signal by transmitting an electrical signal along an input direction in response to the input signal; An information processing device, wherein the input direction intersects at least two of the magnetization direction, the first intersecting direction, and the second intersecting direction.

4. 4. The information processing device according to claim 3, The magnetic body includes a first region through which the spin waves propagate, a second region that covers an outer periphery of the first region, and a boundary portion that reflects or absorbs the spin waves propagating from the first region toward the second region. Information processing device.

5. 5. The information processing device according to claim 4, Further, an external field applying unit is provided, The external field application unit is configured to apply an external magnetic field different from the AC magnetic field along a direction intersecting the input direction.

6. 2. The information processing device according to claim 1, Further, a detection unit and an extraction unit are provided, the detection unit is configured to detect a time change in an internal magnetic field of the magnetic body due to propagation of the spin wave, The extraction unit extracts a component corresponding to a second harmonic of the AC magnetic field from the detected change over time in the internal magnetic field of the magnetic body.

7. An information processing method, comprising: It includes the following steps: In the magnetic field generating step, an AC magnetic field is generated in response to an external input signal; In the inducing step, the AC magnetic field is applied to a magnetic body having spontaneous magnetization along a predetermined magnetization direction, thereby inducing spin waves including two or more modes selected from a first mode, a second mode, and a third mode in the magnetic body; the first mode is a mode that propagates along the magnetization direction, the second mode is a mode propagating along a first intersecting direction that intersects the magnetization direction, The information processing method, wherein the third mode is a mode that propagates along a second intersecting direction that is orthogonal to both the magnetization direction and the first intersecting direction.

8. A program, A program causing a computer to execute each step of the information processing method according to claim 7.

Images