Spin electronic device, storage pool device, and storage pool neural network architecture

Abstract

Provided in the present disclosure are a spin electronic device, a storage pool device, and a storage pool neural network architecture. The spin electronic device comprises a magnetic domain device and at least two electrodes. The magnetic domain device comprises: a substrate; a spin-orbit coupling layer disposed on the top surface of the substrate; a magnetic layer disposed on the top surface of the spin-orbit coupling layer; and a barrier layer disposed on the top surface of the magnetic layer. The electrodes are inserted into the substrate from the top surface of the magnetic domain device, and the electrodes are configured to form an observation port of the spin electronic device; and the combined action of the antisymmetric interaction and dipole interaction of the spin-orbit coupling layer and the magnetic layer allows for the formation of a complex, stable, and periodically alternating magnetic domain structure. In the storage pool device, a volatile nonlinear response is obtained via application of an external magnetic field or electric pulse excitation, thus enabling a recognition function of the storage pool network.

Description

Spintronic devices, memory pool devices, and memory pool neural network architectures Technical Field

[0001] The present disclosure relates to the field of magnetic domain wall technology, and more specifically, to a spintronic device, a storage pool device, and a storage pool neural network architecture. Background Art

[0002] The development of artificial intelligence has promoted the coordinated development of neural networks and their hardware, with applications in areas such as pattern recognition and autonomous driving. The development of hardware-friendly, high-speed, and low-power neural networks and their corresponding hardware has become a key area of ​​scientific research. The memory pool layer in a memory pool neural network, due to its inherent connectivity and the lack of additional training, significantly improves the speed and energy efficiency of pattern recognition tasks.

[0003] In the process of realizing the concept of the present disclosure, the inventors discovered that there are at least the following problems in the related art: the overall connection complexity of existing spintronic devices is insufficient, making it difficult to handle complex recognition tasks, and the storage pool computing efficiency is low.

[0004] Summary of the Invention

[0005] In view of this, embodiments of the present disclosure provide a spintronic device, a storage pool device, and a storage pool neural network architecture.

[0006] One aspect of an embodiment of the present disclosure provides a spintronic device, including:

[0007] Magnetic domain device, the magnetic domain device comprising:

[0008] substrate;

[0009] A spin-orbit coupling layer is provided on the top surface of the substrate;

[0010] a magnetic layer, disposed on the top surface of the spin-orbit coupling layer;

[0011] a barrier layer, disposed on top of the magnetic layer;

[0012] At least two electrodes, the electrodes being arranged to be inserted from the top surface of the magnetic domain device into the substrate, and the electrodes being configured to serve as observation ports of the spintronic device;

[0013] Wherein, under the joint action of the antisymmetric interaction and the dipole interaction between the spin-orbit coupling layer and the magnetic layer, a complex, stable, and periodically alternating magnetic domain structure is formed.

[0014] According to an embodiment of the present disclosure, the magnetic domain device further includes:

[0015] The protective layer is arranged on the top surface of the barrier layer.

[0016] According to an embodiment of the present disclosure, the material of the magnetic layer includes at least one of CoFeB, CoFe, NiFe, IrMn, GdFeCo, Co, Fe, two-dimensional CrI3 and Fe3GeTe2;

[0017] The material of the spin-orbit coupling layer includes at least one of Pt, W, Ta, Ru, Au, Ir and Pd.

[0018] Another aspect of an embodiment of the present disclosure provides a storage pool device, including:

[0019] A magnetic domain device, wherein one or more spintronic devices are arrayed on the magnetic domain device to form a storage pool device;

[0020] In which, when an external magnetic field excitation is applied to the above-mentioned storage pool device, the above-mentioned magnetic domain device forms different magnetic textures. When the spin Hall magnetoresistance and anisotropic magnetoresistance of different magnetic textures change, the above-mentioned storage pool device outputs a nonlinear voltage response.

[0021] According to an embodiment of the present disclosure, a spatial multiplexing technology is used to divide the magnetic domain structure formed by the spintronic device to obtain a plurality of first magnetic domain regions;

[0022] A physical node is added to at least one magnetic domain area among the plurality of the first magnetic domain areas to expand the network scale of the storage pool device, wherein the network scale of the storage pool device is related to the short-term memory capacity of the storage pool device.

[0023] According to an embodiment of the present disclosure, when an external magnetic field is applied to the spintronic device, the magnetization direction of the spintronic device is parallel to the direction of the input current, and the resistance of the spintronic device reaches the upper limit of the resistance range;

[0024] When no external magnetic field is applied to the spintronic device, the spintronic device relaxes to a complex magnetic domain state, and the generated magnetic domain walls reduce the resistance of the spintronic device.

[0025] Another aspect of the disclosed embodiments provides a storage pool neural network architecture, including:

[0026] Input layer, used to input pulse excitation signal;

[0027] An intermediate storage layer, comprising a spintronic device or a storage tank device, wherein the intermediate storage layer is used to process the pulse excitation signal and generate a response signal;

[0028] The output and recognition layer is used to collect, train or infer the above response signals.

[0029] According to an embodiment of the present disclosure, when a pulse excitation signal with the same magnetic field but different pulses is input into the above-mentioned input layer, the above-mentioned intermediate storage layer generates a response signal corresponding to the pulse according to the pulse excitation signal with different pulses, wherein the above-mentioned pulse excitation signal includes an excitation signal converted from an image pixel or a voice signal and a magnetic signal.

[0030] According to an embodiment of the present disclosure, the magnetic domain structure formed in the intermediate storage layer is divided by using a spatial multiplexing technology to obtain a plurality of second magnetic domain regions, wherein each of the second magnetic domain regions includes a pulse input terminal;

[0031] When the pulse excitation signal is input to a plurality of the pulse input terminals, the output is a nonlinear response signal that is time-multiplexed with the identification layer output.

[0032] According to an embodiment of the present disclosure, the time-multiplexed nonlinear response signal constitutes a plurality of virtual nodes;

[0033] Wherein, when the output and recognition layer is used for training, the intermediate storage layer outputs linearly separable response voltage signals corresponding to different virtual nodes;

[0034] Using the output and recognition layer to perform linear regression training on the response voltage signal to obtain recognition information corresponding to the response voltage signal;

[0035] Among them, the above-mentioned storage pool neural network architecture is used for image or speech recognition.

[0036] According to the embodiments of the present disclosure, the spintronic device, storage pool device, and storage pool neural network architecture of the present disclosure have the following effects:

[0037] (1) The magnetization reversal can reach the nanosecond level, enabling high-speed storage pool computing;

[0038] (2) The input of the storage pool neural network architecture can be stimulated by magnetic signals, which means that it can be extended to the integrated sensing, storage and computing system;

[0039] (3) Under large in-plane magnetic field resetting, a new magnetic domain structure can be reset to form a new complex connection situation, thereby realizing a reconfigurable storage pool device.

[0040] (4) By using spatial multiplexing technology to increase the number of physical nodes, the number of virtual nodes is reduced while keeping the memory capacity of the storage pool device unchanged, which can improve the accuracy of the virtual node response signal. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The above and other objects, features and advantages of the present disclosure will become more apparent through the following description of the embodiments of the present disclosure with reference to the accompanying drawings, in which:

[0042] FIG1 schematically shows a structural diagram of a spintronic device according to an embodiment of the present disclosure;

[0043] FIG2 schematically shows a schematic diagram A of the nonlinear resistance response of a spintronic device under an in-plane magnetic field sweep according to an embodiment of the present disclosure;

[0044] FIG3 schematically shows a schematic diagram B of the nonlinear resistance response of a spintronic device under an in-plane magnetic field sweep according to an embodiment of the present disclosure;

[0045] FIG4 schematically shows a structural diagram of a storage pool device according to an embodiment of the present disclosure;

[0046] FIG5 schematically shows a schematic diagram of the resistance response of a storage cell device under an external magnetic field pulse according to an embodiment of the present disclosure;

[0047] FIG6 schematically shows a schematic diagram of the reconfigurable characteristics of a storage pool device according to an embodiment of the present disclosure; and

[0048] FIG7 schematically shows a schematic diagram of the reconfigurable characteristics of a storage pool device according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0049] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present disclosure. In the detailed description below, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0050] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise," "include," etc. used herein indicate the presence of the features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0051] In the context of this disclosure, when a layer / element is referred to as being "on top of" another layer / element, the layer / element can be directly on the other layer / element, or there can be an intervening layer / element therebetween. Additionally, if a layer / element is "on top of" another layer / element in one orientation, it can be "under" the other layer / element when the orientation is reversed.

[0052] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0053] When expressions such as "at least one of A, B and C, etc." are used, they should generally be interpreted in accordance with the meaning of the expression commonly understood by those skilled in the art (for example, "a system having at least one of A, B and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).

[0054] FIG1 schematically shows a structural diagram of a spintronic device according to an embodiment of the present disclosure.

[0055] According to an embodiment of the present disclosure, as shown in FIG1 , a spintronic device includes:

[0056] Magnetic domain devices, magnetic domain devices include:

[0057] substrate;

[0058] A spin-orbit coupling layer is provided on the top surface of the substrate;

[0059] a magnetic layer disposed on a top surface of the spin-orbit coupling layer;

[0060] a barrier layer disposed on a top surface of the magnetic layer;

[0061] At least two electrodes, the electrodes are arranged to be inserted into the substrate from the top surface of the magnetic domain device, and the electrodes are configured to be observation ports of the spintronic device;

[0062] Among them, under the joint action of the antisymmetric interaction (Dzyaloshinskii-Moriya interaction, DMI) and dipole interaction between the spin-orbit coupling layer and the magnetic layer, a complex, stable, and periodically alternating magnetic domain structure is formed.

[0063] According to an embodiment of the present disclosure, the magnetic domain structure includes a plurality of randomly positioned magnetic domain regions, and the boundary between two adjacent magnetic domain regions is a magnetic domain wall.

[0064] According to an embodiment of the present disclosure, the number of electrodes can be adjusted according to actual conditions, for example, it can be any integer such as 2, 3, 4, etc.

[0065] According to an embodiment of the present disclosure, the magnetic domain device further includes:

[0066] The protective layer is arranged on the top surface of the barrier layer.

[0067] According to an embodiment of the present disclosure, the material of the magnetic layer includes at least one of CoFeB, CoFe, NiFe, IrMn, GdFeCo, Co, Fe, two-dimensional CrI3 and Fe3GeTe2;

[0068] Materials of the protection layer and the spin-orbit coupling layer include at least one of Pt, W, Ta, Ru, Au, Ir, and Pd.

[0069] According to an embodiment of the present disclosure, the electrode is composed of Ti / Au or Ni / Au, which plays the role of conductivity and lead-out port; the barrier layer includes at least one of MgO, AlOx and hBN, and the substrate is a thermally oxidized silicon substrate.

[0070] According to an embodiment of the present disclosure, the spintronic device is specifically used as follows:

[0071] First, the specific magnetic layer and spin-orbit coupling layer structure can form a complex domain structure that is stable at room temperature. The magnetic domain structures with magnetization upward and magnetization downward are periodically and randomly arranged, and magnetic domain walls exist to form complex connections.

[0072] Secondly, under the excitation of an external in-plane magnetic field, due to the existence of magnetic anisotropic resistance and spin Hall resistance, the magnetization of the spin electronic device produces a nonlinear volatile resistance change. After the read current is applied, it manifests as a nonlinear volatile voltage response of the MOKE signal in the time domain, thereby realizing the functions required by the storage pool neural network system.

[0073] Finally, under large in-plane magnetic field reset, a completely new magnetic domain structure can be reset to form a new complex connection situation, thereby realizing a reconfigurable storage pool device.

[0074] According to embodiments of the present disclosure, by regulating the antisymmetric exchange interaction (DMI) strength, complex magnetic domain walls are formed that are stable at room temperature. When an external magnetic field pulse is applied, the resistance changes accordingly, enabling a volatile nonlinear response signal that is dependent on the pulse application history. Furthermore, after being driven by magnetic fields of varying magnitudes, the initial resistance signal can be reset, forming a reconfigurable storage pool device. This spintronic device's basic structural unit operates in a method and has corresponding image recognition capabilities for the storage pool device, and the spintronic device can be expanded to array applications.

[0075] Figure 2 schematically shows a schematic diagram A of the nonlinear resistance response of a spintronic device under an in-plane magnetic field sweep according to an embodiment of the present disclosure. Figure 3 schematically shows a schematic diagram B of the nonlinear resistance response of a spintronic device under an in-plane magnetic field sweep according to an embodiment of the present disclosure.

[0076] According to an embodiment of the present disclosure, FIG2 shows the nonlinear resistance response of a spintronic device under an in-plane magnetic field sweep. FIG2(a) shows the radial resistance R of the spintronic device. xx The resistance changes obtained by sweeping the field in the x and y directions. When the magnetic field is swept along the x direction. As the magnetic field increases, the resistance increases and tends to saturation at around 5000Oe. When the magnetic field decreases, the resistance value returns to its initial value, reflecting the effect of anisotropic magnetoresistance (AMR) on the resistance. When the magnetic field is swept along the y direction. As the magnetic field increases, the resistance decreases and tends to saturation at around 5000Oe. When the magnetic field decreases, the resistance value returns to its initial value, reflecting the effect of spin-Hall magnetoresistance (SMR) on the resistance. Figure 2(b) shows the Hall resistance R xy As Hx changes, it also shows a volatile nonlinear response with a small generalization rank. xx and R xy The resistance that can be used as a detection response. Its change comes from the response of the complex magnetic domain wall structure to the excitation of the external magnetic field, Figure 3(a), the in-plane field H x = 3000Oe, when the device is saturated, the magnetization direction is completely parallel to the current and the resistance is the highest; Figure 3(b), the in-plane field H x = 0 Oe. In the absence of an external magnetic field, the device relaxes to a complex magnetic domain state, generating magnetic domain walls that reduce the resistance of the spintronic device. The interaction between adjacent magnetizations in a spintronic device contributes to the complexity of the storage pool device, resulting in a high kernel rank.

[0077] FIG4 schematically shows a structural diagram of a storage pool device according to an embodiment of the present disclosure.

[0078] According to an embodiment of the present disclosure, as shown in FIG4 , the storage pool device includes:

[0079] A magnetic domain device, wherein one or more spintronic devices are arrayed on the magnetic domain device to form a storage pool device;

[0080] Among them, when an external magnetic field excitation is applied to the magnetic domain device, the magnetic domain device forms different magnetic textures. When the spin Hall magnetoresistance and anisotropic magnetoresistance of different magnetic textures change, the storage pool device outputs a nonlinear voltage response.

[0081] According to the embodiments of the present disclosure, when used as a storage pool device, the interaction between the spin magnetization of the spintronic device is used as a source of complexity to increase the core level of the storage pool device and enhance the ability of the storage pool device to solve complex problems.

[0082] According to an embodiment of the present disclosure, a magnetic domain structure formed by a spintronic device is divided and processed using a spatial multiplexing technology to obtain a plurality of first magnetic domain regions;

[0083] A physical node is added to at least one magnetic domain area among the plurality of first magnetic domain areas to expand the network scale of the storage pool device, wherein the network scale of the storage pool device is related to the short-term memory capacity of the storage pool device.

[0084] According to the embodiments of the present disclosure, when used as a storage pool device, different magnetic textures are formed under the application of external magnetic field excitation. Since the spin Hall magnetoresistance and anisotropic magnetoresistance (SMR / AMR) of different magnetic textures vary, the generalization level is low and a nonlinear voltage response can be obtained.

[0085] According to an embodiment of the present disclosure, when used as a storage pool device, spatial multiplexing technology is used to divide the magnetic domain structure into multiple first magnetic domain areas, increase the number of physical nodes, and increase the network scale of the storage pool device to increase the memory capacity of the storage pool device and improve the degree of recalling historical input stimuli from the current storage pool state.

[0086] According to an embodiment of the present disclosure, when an external magnetic field is applied to the storage pool device, the magnetization direction of the storage pool device is parallel to the direction of the input current, and the resistance value of the storage pool device reaches the upper limit of the resistance range;

[0087] When no external magnetic field is applied to the storage cell device, the storage cell device relaxes to a complex magnetic domain state, and the generated magnetic domain walls reduce the resistance of the storage cell device.

[0088] FIG5 schematically shows a resistance response diagram of a storage cell device under an external magnetic field pulse according to an embodiment of the present disclosure.

[0089] According to an embodiment of the present disclosure, FIG5(a) shows a pulse with a pulse width of 0.1s and a magnetic field of 900Oe, and FIG5(b) shows the resistance R of the storage cell device. xx The response process. It can be seen that when the pulse is applied, the resistance changes suddenly. When the 0.2s pulse is removed, the resistance shows a nonlinear time-varying response. The time-varying signal can be sampled at uniform intervals, and the sampling points serve as virtual nodes to record the changes in the pulse history.

[0090] FIG6 schematically shows a schematic diagram of the reconfigurable characteristics of a storage pool device according to an embodiment of the present disclosure.

[0091] According to an embodiment of the present disclosure, Figure 6 illustrates the reconfigurable nature of the storage pool device. Under in-plane magnetic field pulses of varying magnitudes, magnetic domains produce periodic arrangements with varying densities, generating new, complex connections between the spins of the magnetic material, thereby forming a new storage pool device. This storage pool device, however, remains unchanged under a fixed magnetic field, thus achieving a reconfigurable storage pool. For example, using magnetic field pulses of varying magnitudes, a complex magnetic texture can be reconstructed under zero field conditions. This complex magnetic texture achieves the purpose of resetting different complex connections, forming a new storage pool device.

[0092] FIG7 schematically shows a schematic diagram of the reconfigurable characteristics of a storage pool device according to an embodiment of the present disclosure.

[0093] According to an embodiment of the present disclosure, as shown in FIG7 , the storage pool neural network architecture includes:

[0094] Input layer, used to input pulse excitation signal;

[0095] An intermediate storage layer, including a spintronic device or a storage tank device, is used to process the pulse excitation signal and generate a response signal;

[0096] The output and recognition layer is used to collect, train, or infer response signals.

[0097] According to an embodiment of the present disclosure, when a pulse excitation signal with the same magnetic field but different pulses is input into the input layer, the intermediate storage layer generates a response signal corresponding to the pulse according to the pulse excitation signal with different pulses, wherein the pulse excitation signal includes an electrical excitation signal and a magnetic excitation signal converted from an image pixel or a voice signal.

[0098] According to an embodiment of the present disclosure, a magnetic domain structure formed in the middle storage layer is divided and processed using a spatial multiplexing technology to obtain a plurality of second magnetic domain regions, wherein each second magnetic domain region includes a pulse input terminal;

[0099] When a pulse excitation signal is input to a plurality of pulse input terminals, a nonlinear response signal is output that is time-multiplexed with the output of the identification layer.

[0100] According to an embodiment of the present disclosure, by dividing the magnetic domain structure, multiple second magnetic domain areas are obtained, thereby generating M pulse input terminals, and then forming M physical nodes. After the intermediate storage layer receives the input pulse excitation signal, it generates a time-multiplexed nonlinear response signal, forming N virtual nodes (M, N are integers, and the specific size depends on the reading accuracy), thereby forming a basic intermediate storage layer.

[0101] According to an embodiment of the present disclosure, the time-multiplexed nonlinear response signal constitutes a plurality of virtual nodes;

[0102] Wherein, when the output and recognition layers are used for training, the intermediate storage layer outputs linearly separable response voltage signals corresponding to different virtual nodes;

[0103] The output and recognition layers are used to perform linear regression training on the response voltage signal to obtain recognition information corresponding to the response voltage signal;

[0104] Among them, the storage pool neural network architecture is used for image or speech recognition.

[0105] According to an embodiment of the present disclosure, during the training process, the response currents of different virtual nodes are read out, and the corresponding patterns of the current responses of the storage pool devices are obtained through training, thereby realizing image / speech recognition and intrapersonal recognition in applications of inference recognition.

[0106] According to an embodiment of the present disclosure, FIG7 illustrates a storage pool neural network architecture, which includes a pulsed magnetic field input layer, an intermediate storage layer, and an output and recognition layer. The output and recognition layer is composed of a pulsed magnetic field sequence. When the storage pool neural network performs a recognition task, image or speech information is converted into a 0 / 1 pulse sequence in the input layer. Furthermore, because spintronic devices or storage pool devices can produce a direct nonlinear response under stimulation, the input signal (i.e., the pulsed excitation signal) can also be an artificial or natural magnetic signal, which is input into the spintronic device or storage pool device, serving as the basis for integrated sensing and computing.

[0107] According to an embodiment of the present disclosure, a spintronic device having a complex magnetic domain structure or a storage pool device of an n*n array thereof constitutes an intermediate storage layer. After receiving an input pulse excitation signal of a pulse sequence, the intermediate storage layer generates a nonlinear volatile response. During the volatile response cycle, N virtual nodes (where N is an integer, the size of which is related to the recognition accuracy and actual task requirements) can be set during detection to detect and output the response voltage (i.e., the response signal).

[0108] According to the embodiments of the present disclosure, the collection of response voltages and the training and inference circuits can be used as the output and recognition layer. When the training set information of a suitable recognition task is input into the intermediate storage layer, the spintronic device or storage pool device will obtain a linearly separable voltage output. At this time, it is only necessary to perform linear regression training on the output signal to make the output voltage signal correspond to the required recognition information (such as 0-9 in digital and speech recognition). After training is completed, using the same input method and storage pool calculation principle, the storage pool neural network system can recognize the task set.

[0109] The embodiments of the present disclosure are described above. However, these embodiments are for illustrative purposes only and are not intended to limit the scope of the present disclosure. Although each embodiment has been described separately above, this does not mean that the measures in each embodiment cannot be used in combination to advantage. The scope of the present disclosure is defined by the appended claims and their equivalents. Without departing from the scope of the present disclosure, those skilled in the art may make various substitutions and modifications, which should all fall within the scope of the present disclosure.

Claims

1. A spin electronic device, comprising: A magnetic domain device, comprising: substrate; A spin-orbit coupling layer is disposed on the top surface of the substrate; A magnetic layer, disposed on the top surface of the spin-orbit coupling layer; A barrier layer, disposed on the top surface of the magnetic layer; at least two electrodes, the electrodes are arranged to be inserted into the substrate from the top surface of the magnetic domain device, and the electrodes are configured as observation ports of the spintronic device; Wherein, under the joint action of the antisymmetric interaction and the dipole interaction between the spin-orbit coupling layer and the magnetic layer, a complex, stable, and periodically alternating magnetic domain structure is formed.

2. The spintronic device according to claim 1, wherein: The magnetic domain device further comprises: The protection layer is arranged on the top surface of the barrier layer.

3. The spintronic device according to claim 1, wherein: The material of the magnetic layer includes at least one of CoFeB, CoFe, NiFe, IrMn, GdFeCo, Co, Fe, two-dimensional CrI3 and Fe3GeTe2; The material of the spin-orbit coupling layer includes at least one of Pt, W, Ta, Ru, Au, Ir and Pd.

4. A storage pool device, comprising: A magnetic domain device, wherein one or more spin electronic devices according to any one of claims 1 to 3 are arranged in an array on the magnetic domain device to form a storage pool device; Wherein, when an external magnetic field is applied to the storage pool device, the magnetic domain device forms different magnetic textures, and when the spin Hall magnetoresistance and anisotropic magnetoresistance of different magnetic textures change, the storage pool device outputs a nonlinear voltage response.

5. The storage pool device according to claim 4, wherein: Using a spatial multiplexing technology to divide the magnetic domain structure formed by the spintronic device to obtain a plurality of first magnetic domain regions; A physical node is added to at least one magnetic domain area among the plurality of the first magnetic domain areas to expand the network scale of the storage pool device, wherein the network scale of the storage pool device is related to the short-term memory capacity of the storage pool device.

6. The storage pool device according to claim 4, wherein: When an external magnetic field is applied to the storage pool device, the magnetization direction of the storage pool device is parallel to the direction of the input current, and the resistance value of the storage pool device reaches the upper limit of the resistance value range; When no external magnetic field is applied to the storage pool device, the storage pool device relaxes to a complex magnetic domain state, generating The magnetic domain walls reduce the resistance of the storage pool device.

7. A storage pool neural network architecture, comprising: Input layer, used to input pulse excitation signal; An intermediate storage layer, comprising the spin electronic device according to any one of claims 1 to 3 or the storage pool device according to any one of claims 4 to 6, wherein the intermediate storage layer is used to process the pulse excitation signal to generate a response signal; The output and recognition layer is used to collect, train or infer the response signal.

8. The storage pool neural network architecture of claim 7, wherein: When a pulse excitation signal with the same magnetic field but different pulses is input into the input layer, the intermediate storage layer generates a response signal corresponding to the pulse according to the pulse excitation signal with different pulses, wherein the pulse excitation signal includes an electrical excitation signal and a magnetic excitation signal converted from image pixels or voice signals.

9. The storage pool neural network architecture of claim 7, wherein: Using a spatial multiplexing technology to divide the magnetic domain structure formed by the intermediate storage layer to obtain a plurality of second magnetic domain regions, wherein each of the second magnetic domain regions includes a pulse input terminal; When the pulse excitation signal is input to a plurality of the pulse input terminals, the output and the identification layer output a time-multiplexed nonlinear response signal.

10. The storage pool neural network architecture of claim 9, wherein: The time-multiplexed nonlinear response signal constitutes a plurality of virtual nodes; Wherein, when the output and recognition layer is used for training, the intermediate storage layer outputs linearly separable response voltage signals corresponding to different virtual nodes; Using the output and recognition layer to perform linear regression training on the response voltage signal to obtain recognition information corresponding to the response voltage signal; Among them, the storage pool neural network architecture is used for image or speech recognition.

Images