Magnetic pulse nerve component and application thereof

Abstract

The invention discloses a magnetic pulse nerve component and application thereof, and the device comprises a magnetic tunnel junction which is used for responding to a driving electric signal to emit a spike pulse electric signal, the magnetic tunnel junction comprises a cap layer, a magnetic free layer, a barrier layer, a pinning layer and an antiferromagnetic layer which are sequentially stacked, and the magnetization direction of the magnetic free layer is in-plane orientation; a first electrode layer; a second electrode layer; a large thermal resistance interface is formed between the cap layer and the magnetic free layer, the large thermal resistance interface is used for generating a thermally-driven self-rotation moving torque and thermal noise under the action of a driving electric signal, and the magnetization direction of the magnetic free layer is continuously overturned under the action of the thermally-driven self-rotation moving torque and an electrically-driven self-rotation moving torque generated by the driving electric signal. The magnetic pulse nerve component can generate a peak pulse signal capable of being modulated under excitation of external direct current voltage or pulse voltage, the area is remarkably saved, and power consumption is reduced.

Description

Technical Field

[0001] The present application belongs to the technical field of pulse neuron devices, and specifically relates to a magnetic pulse neuron device and its application. Background Art

[0002] With the continuous improvement of artificial intelligence theory and the continuous development of technology, the computing power and energy efficiency of existing computing devices are gradually unable to meet the needs of the sustainable development of the field of artificial intelligence. At present, mainstream chips are based on the von Neumann architecture with storage and computing separation, which naturally has bottlenecks in improving computing power and energy efficiency. The human brain has a brain neural network composed of hundreds of billions of neurons and hundreds of trillions of synapses to complete various complex tasks, with a power consumption of only 24 watts. To this end, people began to simulate the working mechanism of the brain and explore high-computing power, high-energy efficiency, and brain-like neuromorphic computing. Brain-like neuromorphic computing needs to rely on the neural network architecture to be realized. The most discussed and studied are artificial neural networks (ANN) and spiking neural networks (SNN). Compared with ANN processing continuous value information, SNN communicates with each other through discrete pulse sequences in the time dimension, has lower energy consumption, and is conducive to processing complex, sparse and noisy spatiotemporal information data.

[0003] The hardware systems that the neuromorphic computing performed by the neural network relies on mainly include artificial neurons and artificial synapses, among which artificial neurons are used to process input signals. Existing spin neuromorphic devices based on magnetic tunnel junctions can use their spin diode effect, oscillator characteristics, and random flipping characteristics to realize the functions of nonlinear activation functions such as ReLU and Sigmoid of neurons in ANN from the hardware bottom layer, but they cannot be applied to directly output pulse signals in SNN. In order to realize the functions of neurons in SNN, the existing technology combines neurons, capacitors, inductors, adders, comparators, and corresponding customized circuits built using CMOS technology to realize the output of pulse signals. This technology requires the use of multiple devices to realize the function of a neuron, which has a large amount of calculation, power consumption, and area, and cannot meet the requirements of the field of artificial intelligence for reducing chip size and power consumption. Summary of the invention

[0004] The purpose of the present application is to provide a magnetic pulse neuron device and its application, so as to solve the technical problem that the spin neuromorphic device based on magnetic tunnel junction in the prior art cannot be applied to SNN, multiple devices are needed to realize the function of a pulse neuron, and the requirements of the field of artificial intelligence for reducing chip size and power consumption are not met.

[0005] In order to achieve the above-mentioned object, the first aspect of the present application provides a magnetic pulse neuron device, comprising:

[0006] A magnetic tunnel junction, used for emitting a spike pulse electrical signal in response to a driving electrical signal, wherein the magnetic tunnel junction comprises a cap layer, a magnetic free layer, a barrier layer, a pinning layer and an antiferromagnetic layer stacked in sequence, and the magnetization direction of the magnetic free layer is in-plane oriented;

[0007] A first electrode layer, arranged on one side of the magnetic tunnel junction;

[0008] A second electrode layer, arranged at the other side of the magnetic tunnel junction, the second electrode layer being used to cooperate with the first electrode layer to input the driving electrical signal to the magnetic tunnel junction;

[0009] Among them, a large thermal resistance interface is formed between the cap layer and the magnetic free layer, and the large thermal resistance interface is used to generate heat accumulation under the action of the driving electrical signal, so as to generate a thermally driven spin transfer torque and thermal noise through a thermal effect. The magnetization direction of the magnetic free layer is continuously flipped under the action of the thermally driven spin transfer torque and the electrically driven spin transfer torque generated by the driving electrical signal to generate a spike pulse electrical signal, and the spike pulse electrical signal has a certain randomness under the action of the thermal noise.

[0010] In one or more embodiments, the driving electrical signal is a DC voltage signal, and the voltage intensity of the driving electrical signal is less than or equal to 5V.

[0011] In one or more embodiments, the driving electrical signal is a pulse voltage signal, and the voltage intensity of the driving electrical signal is less than or equal to 5V, and the signal frequency is less than or equal to 1 GHz.

[0012] In one or more embodiments, the material of the cap layer is one or more combinations of MgO, Pt, IrMn, MgO / W / MgO, and the thickness of the cap layer is 0.01-10 nm.

[0013] In one or more embodiments, the material of the magnetic free layer is Co 20 Fe 60 B 20 , one or more combinations of NiFe, FeB, Py, Co, CoFe, FePt, and Co / Ni, and the thickness of the magnetic free layer is 0.1 to 20 nm.

[0014] In one or more embodiments, the barrier layer is made of one or more combinations of MgO, Al2O3, MgAl2O4, and TiO2, and the thickness of the barrier layer is 0.5 to 10 nm.

[0015] In one or more embodiments, the material of the pinning layer is Co 40 Fe 40 B20 / Ru / Co 70 Fe 30 , one or more combinations of FeB, Py, Co, CoFe, FePt, and Co / Ni, and the thickness of the pinning layer is 0.1 to 20 nm.

[0016] In one or more embodiments, the material of the first electrode layer and the second electrode layer is Ta, Ru, Mo, Pt, Al, W, Cu, CuN, Au, Cr, Ti, Ni, Pb, TMDCs, SiC, MoO2, ITO, Ti3C2T x , graphene or one or more combinations thereof, the thickness of the first electrode layer and the second electrode layer is 5 to 200 nm.

[0017] In one or more embodiments, a magnetic field generating device is further included, wherein the magnetic field generating device is used to generate an external magnetic field based on an input control electrical signal to control the magnetization direction reversal of the magnetic free layer and assist in achieving the output of a spike pulse electrical signal.

[0018] In order to achieve the above-mentioned purpose, the second aspect of the present application provides an application of the magnetic pulse neuron device described in any of the above-mentioned embodiments as a pulse neuron in a pulse neural network.

[0019] Different from the prior art, the beneficial effects of this application are:

[0020] The magnetic pulse neuron device of the present application can generate a modulated spike pulse signal under the stimulation of an external DC voltage or a pulse voltage, and can independently realize the function of neurons in a pulse neural network without being combined with other CMOS devices, thus significantly saving area and reducing power consumption.

[0021] The magnetic pulse neuron device of the present application has the characteristic of being able to respond to pulse voltage, and can be used to construct a scalable spike neural network; it can also be integrated into a CMOS system to reduce the production cost of a neuromorphic chip constructed using this device. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0023] Figure 1 This is a schematic structural diagram of an embodiment of a magnetic pulse neuron device of the present application;

[0024] Figure 2 It is a structural schematic diagram of another embodiment of the magnetic pulse neuron device of the present application;

[0025] Figure 3 is an output signal diagram of the magnetic pulse neuron device of Example 1 of the present application under a linear voltage signal;

[0026] Figure 4 is an output signal diagram of the magnetic pulse neuron device of Example 2 of the present application under a pulse voltage signal;

[0027] Figure 5 It is the pulse time distribution of the peak pulse electrical signal output by the magnetic pulse neuron device of Example 3 of the present application under a DC voltage signal. DETAILED DESCRIPTION

[0028] In order to enable those skilled in the art to better understand the technical solutions in the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work should fall within the scope of protection of the present application.

[0029] Compared with artificial neural networks (ANNs), spiking neural networks (SNNs) communicate with each other through discrete pulse sequences in the time dimension, so they have lower energy consumption and are conducive to processing complex, sparse and noisy temporal information data.

[0030] Existing spin neuromorphic devices based on magnetic tunnel junctions can use their spin diode effect, oscillator characteristics, and random flipping characteristics to implement the functions of nonlinear activation functions such as ReLU and Sigmoid of neurons in ANN from the hardware bottom layer. The rich characteristics of magnetic tunnel junction devices make them promising for large-scale application as the basic unit of neuromorphic chips.

[0031] However, the spin neuromorphic device based on the magnetic tunnel junction cannot be used alone as a pulse neuron of the SNN. The main reason is that its magnetic free layer cannot realize the automatic continuous random flipping of the magnetization direction under a single voltage, resulting in the inability to output a spike pulse electrical signal. Specifically, when using existing neurons to generate pulse signals, it is necessary to input a forward voltage signal to flip the magnetization direction of the magnetic free layer to one direction, and then input a reverse voltage signal to reset the magnetization direction to complete a pulse process. Therefore, in order to realize the magnetization direction flipping and resetting process, the prior art needs to use neurons, capacitors, inductors, adders, comparators and corresponding customized circuits constructed using CMOS technology to achieve the output of pulse signals. This technology requires the use of multiple devices to realize the function of a neuron, and the calculation amount, power consumption and area are large, which cannot meet the requirements of the field of artificial intelligence for reducing chip size and power consumption. In particular, the pulse signals output by controlling the random flipping of the magnetization direction of the magnetic free layer in the prior art are all square wave pulse signals, which represent the state in which the free layer stays between two resistance states, and the output of spike pulse signals cannot be achieved.

[0032] In order to solve the above problems, the applicant has developed a magnetic pulse neuron device, which does not need to be combined with other CMOS devices and can independently realize the functions of neurons in a pulse neural network, significantly saving area and reducing power consumption.

[0033] Specifically, see Figure 1 , Figure 1 It is a schematic structural diagram of an embodiment of a magnetic pulse neuron device of the present application.

[0034] like Figure 1 As shown, the magnetic pulse neuron device includes a magnetic tunnel junction 10 , a first electrode layer 20 and a second electrode layer 30 .

[0035] The magnetic tunnel junction 10 is used to emit a spike pulse electrical signal in response to a driving electrical signal. The magnetic tunnel junction 10 includes a cap layer 101, a magnetic free layer 102, a barrier layer 103, a pinned layer 104 and an antiferromagnetic layer 105 which are stacked in sequence. The magnetization direction of the magnetic free layer 102 is in-plane oriented.

[0036] The first electrode layer 20 is arranged on one side of the magnetic tunnel junction 10 ; the second electrode layer 30 is arranged on the other side of the magnetic tunnel junction 10 , and the second electrode layer 30 is used to cooperate with the first electrode layer 20 to input a driving electrical signal to the magnetic tunnel junction 10 .

[0037] There is a large thermal resistance at the interface between the cap layer 101 and the magnetic free layer 102, that is, a large thermal resistance interface a is formed. After applying a driving electrical signal to the magnetic tunnel junction 10, the large thermal resistance interface a will generate heat accumulation, which will lead to a thermal effect, and under the action of the thermal effect, a thermally driven spin transfer torque will be generated on the magnetized free layer. Synchronously, the driving electrical signal will also generate an electrically driven spin transfer torque on the magnetized free layer.

[0038] Under the action of thermally driven spin transfer torque and electrically driven spin transfer torque, the magnetization direction of the magnetized free layer will be continuously flipped, thereby generating a spike pulse electrical signal.

[0039] The working process of the magnetic pulse neuron device of the present application is described in detail below:

[0040] When a driving electrical signal is applied to the magnetic pulse neuron device in the initial state, the magnetic tunnel junction 10 is in a low-resistance state, and the current tunnels under the tunneling magnetoresistance effect. The large thermal resistance interface a gradually accumulates heat, and the thermally driven spin transfer torque generated by the thermal effect gradually increases until the magnetization direction of the magnetized free layer is reversed under the action of the thermally driven spin transfer torque and the electrically driven spin transfer torque. The magnetic tunnel junction 10 switches to a high-resistance state, the tunneling current decreases rapidly, the heat is rapidly dissipated, the effect of the thermally driven spin transfer torque is weakened, and the magnetization direction of the magnetized free layer is rapidly reversed, so that the magnetic tunnel junction 10 returns to a low-resistance state, and this is repeated in sequence. The magnetization direction of the magnetized free layer is continuously reversed, thereby generating a spike pulse electrical signal.

[0041] Specifically, the magnetization dynamic behavior of the magnetization free layer of the present application can be expressed by the Landau-Lifshitz-Gilbert equation (LLG equation):

[0042]

[0043] Where m is the unit magnetization vector of the magnetization free layer, m = m / Ms, Ms is the saturation magnetization, m is the magnetization of the magnetization free layer, P is the direction of the spin polarization current, and h eff is the effective field, which includes the external magnetic field, the anisotropic field in the free layer and the demagnetization field, h ther is the thermal field, τ STT is the spin torque, γ is the gyromagnetic ratio, and α is the Gilbert damping.

[0044] As can be seen from the above formula, the frequency and time interval of the peak pulse electrical signal generated by the magnetic tunnel junction 10 can be controlled by adjusting the intensity of the driving voltage.

[0045] In addition, the thermal effect not only generates the thermally driven spin transfer torque, but also provides thermal noise. The amplitude of the thermal noise can be expressed as follows:

[0046]

[0047] Where Ms is the saturation magnetization, γ is the gyromagnetic ratio, k B is the Boltzmann constant, T is the temperature of the magnetic tunnel junction 10, V FL is the volume of the magnetic free layer 102, and Δt is the time length.

[0048] Under the influence of thermal noise, the spike pulse electrical signal output by the magnetic pulse neuron device has a certain randomness, which can be expressed by a typical exponential function:

[0049] N counts =Aexp(-t / τ gap time )

[0050] Where N counts Indicates the number of spike pulses that occur within a time length t, where t is the test duration, τ gaptime It represents the expected value of the spike pulse time interval, and A is a constant.

[0051] Therefore, the appearance of spike pulse electrical signals follows Poisson distribution. The number of pulses of the spike pulse electrical signals generated by the magnetic pulse neuron device within a certain period of time can be predicted, but the intervals between peaks are random, which meets the randomness requirements of the spike neural network and helps to build a spike neural network.

[0052] It is contemplated that, in one embodiment, Figure 1 As shown, the driving electrical signal may be a DC voltage signal 50 , and the magnetic pulse neuron device may directly output a spike pulse electrical signal under the action of the DC voltage signal 50 .

[0053] Specifically, in one embodiment, the voltage strength of the DC voltage signal 50 may be less than or equal to 5 V. In other embodiments, the voltage strength of the DC voltage signal 50 may also be adjusted based on actual working conditions.

[0054] In another embodiment, the driving electrical signal may also be a pulse voltage signal 60, which may also realize the output of a peak pulse electrical signal, and both may achieve the effect of this embodiment.

[0055] Specifically, in one embodiment, the voltage strength of the pulse voltage signal 60 may be less than or equal to 5V, and the signal frequency may be less than or equal to 1 GHz. In other embodiments, the voltage strength and signal frequency of the pulse voltage signal 60 may also be adjusted based on actual working conditions.

[0056] The materials of each layer of the magnetic pulse neuron device of the present application are described in detail below.

[0057] In one embodiment, the material of the cap layer 101 may be one or more combinations of MgO, Pt, IrMn, MgO / W / MgO, and the thickness of the cap layer 101 may be 0.01-10 nm.

[0058] In one embodiment, the material of the magnetic free layer 102 can be Co 20 Fe 60 B 20 , NiFe, FeB, Py, Co, CoFe, FePt, Co / Ni or one or more combinations thereof, the thickness of the magnetic free layer 102 can be 0.1 to 20 nm.

[0059] In one embodiment, the material of the barrier layer 103 may be one or more combinations of MgO, Al 2 O 3 , MgAl 2 O 4 , and TiO 2 , and the thickness of the barrier layer 103 may be 0.5 to 10 nm.

[0060] In one embodiment, the material of the pinning layer 104 can be Co 40 Fe 40 B 20 / Ru / Co 70 Fe 30 , FeB, Py, Co, CoFe, FePt, Co / Ni or one or more combinations thereof, the thickness of the pinning layer 104 may be 0.1 to 20 nm.

[0061] In one embodiment, the material of the first electrode layer 20 and the second electrode layer 30 can be Ta, Ru, Mo, Pt, Al, W, Cu, CuN, Au, Cr, Ti, Ni, Pb, TMDCs, SiC, MoO2, ITO, Ti3C2T x , graphene or one or more combinations thereof, the thickness of the first electrode layer 20 and the second electrode layer 30 may be 5 to 200 nm.

[0062] In one embodiment, the antiferromagnetic layer 105 may be made of IrMn, PtMn, FeMn, NiO, NiCoO or a combination thereof, and may have a thickness of 0.1 nm to 20 nm.

[0063] In another embodiment, to assist in achieving the output of a spike pulse electrical signal, refer to Figure 2 , Figure 2 It is a schematic structural diagram of another embodiment of the magnetic pulse neuron device of the present application.

[0064] like Figure 2As shown, the magnetic pulse neuron device also includes a magnetic field generating device 40, which is used to generate an external magnetic field based on an input control electrical signal to control the magnetization direction reversal of the magnetic free layer 102 and assist in achieving the output of a spike pulse electrical signal.

[0065] It can be understood that the state of the magnetic tunnel junction 10 itself is related to its structure. The magnetic tunnel junction 10 needs to be adjusted to an initial state of low resistance in order to quickly achieve the output of a spike pulse electrical signal. For this reason, a magnetic field generating device 40 is also provided in this embodiment. The magnetic field generating device 40 quickly adjusts the magnetic tunnel junction 10 to the initial state by applying an external magnetic field, thereby accelerating the output of the spike pulse.

[0066] The magnetic pulse neuron device of each of the above embodiments can generate a modulated spike pulse signal under the stimulation of an external DC voltage or a pulse voltage, and can independently realize the function of a neuron in a pulse neural network without being combined with other CMOS devices, thereby significantly saving area and reducing power consumption.

[0067] In addition, due to the characteristic of being able to respond to pulse voltage, the magnetic pulse neuron device of the above-mentioned embodiments can also be used to build a physical neural network, perform on-chip learning and calculation, and thus construct a scalable spike neural network; it can also be integrated into a CMOS system to reduce the production cost of a neuromorphic chip constructed using this device.

[0068] The effects of the technical solution of the present application are further described in detail below in conjunction with specific embodiments.

[0069] Embodiment 1:

[0070] A magnetic pulse neuron device comprises a first electrode layer, a magnetic tunnel junction and a second electrode layer which are stacked, and a magnetic field generating device.

[0071] Among them, the first electrode layer is a Ta / Ru / Cu / Ta multilayer electrode with a thickness of 10nm, and the second electrode layer is a Ta / Ru / Ta / CuN / Ta / CuN / Ta multilayer electrode with a thickness of 10nm.

[0072] The magnetic tunnel junction includes a 5nm thick MgO cap layer, a 10nm thick Co 20 Fe 60 B 20 Magnetic free layer, 5nm thick MgO barrier layer, 5nm thick Co 40 Fe 40 B 20 / Ru / Co 70 Fe 30 Pinning layer, 10nm thick PtMn antiferromagnetic layer.

[0073] A -70 mV DC voltage signal is applied to the magnetic pulse neuron device of Example 1, and the external magnetic field strength Hext generated by the magnetic field generating device is controlled to be -260 Oe. The signal diagram output by the device is recorded to obtain Figure 3 , Figure 3 This is an output signal diagram of the magnetic pulse neuron device of Example 1 of the present application under a linear voltage signal.

[0074] like Figure 3 As shown, the magnetic pulse neuron device of Example 1 directly outputs a spike pulse electrical signal under the drive of a DC voltage signal, and the intervals between the peaks are random in duration, meeting the requirements of the pulse neural network.

[0075] Embodiment 2:

[0076] A magnetic pulse neuron device comprises a first electrode layer, a magnetic tunnel junction and a second electrode layer which are stacked, and a magnetic field generating device.

[0077] Among them, the first electrode layer is a Ta / Ru / Cu / Ta multilayer electrode with a thickness of 10nm, and the second electrode layer is a Ta / Ru / Ta / CuN / Ta / CuN / Ta multilayer electrode with a thickness of 10nm.

[0078] The magnetic tunnel junction includes a 10nm thick MgO cap layer, a 20nm thick Co 20 Fe 60 B 20 Magnetic free layer, 10nm thick MgO barrier layer, 10nm thick Co 40 Fe 40 B 20 / Ru / Co 70 Fe 30 Pinning layer, 10nm thick IrMn antiferromagnetic layer.

[0079] A pulse voltage signal of -100 mV and 200 kHz was applied to the magnetic pulse neuron device of Example 2, and the external magnetic field strength Hext generated by the magnetic field generating device was controlled to be -300 Oe. The signal graph output by the device was recorded to obtain Figure 4 , Figure 4 This is an output signal diagram of the magnetic pulse neuron device of Example 2 of the present application under a pulse voltage signal.

[0080] like Figure 4 As shown, the magnetic pulse neuron device of Example 2 directly outputs a spike pulse electrical signal under the drive of a pulse voltage signal, and the intervals between the peaks are random in duration, meeting the requirements of the pulse neural network.

[0081] Embodiment 3:

[0082] A magnetic pulse neuron device comprises a first electrode layer, a magnetic tunnel junction and a second electrode layer which are stacked, and a magnetic field generating device.

[0083] Among them, the first electrode layer is a Ta / Ru / Cu / Ta multilayer electrode with a thickness of 10nm, and the second electrode layer is a Ta / Ru / Ta / CuN / Ta / CuN / Ta multilayer electrode with a thickness of 10nm.

[0084] The magnetic tunnel junction includes a 2nm thick MgO cap layer, a 2nm thick Co 20 Fe 60 B 20 Magnetic free layer, 5nm thick MgO barrier layer, 10nm thick Co 40 Fe 40 B 20 / Ru / Co 70 Fe 30 Pinning layer, 5nm thick IrMn antiferromagnetic layer.

[0085] A DC voltage signal of -130 mV is applied to the magnetic pulse neuron device of Example 3, and the external magnetic field strength Hext generated by the magnetic field generating device is controlled to be -270 Oe, τ gaptime =30.2ns; record the number of pulse peaks of the magnetic pulse neuron device in a fixed period of time, and obtain Figure 5 , Figure 5 It is the pulse time distribution of the peak pulse electrical signal output by the magnetic pulse neuron device of Example 3 of the present application under a DC voltage signal.

[0086] like Figure 5 As shown, there is a regularity in the distribution of pulse intervals of spike pulse electrical signals.

[0087] It is obvious to those skilled in the art that the present application is not limited to the details of the exemplary embodiments described above, and that the present application can be implemented in other specific forms without departing from the spirit or essential features of the present application. Therefore, the embodiments should be regarded as exemplary and non-limiting in all respects, and the scope of the present application is defined by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims be included in the present application. Any reference numeral in a claim should not be regarded as limiting the claim to which it relates.

[0088] In addition, it should be understood that although the present specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

Claims

1. A magnetic pulse neuron device, characterized in that: include: A magnetic tunnel junction, used for emitting a spike pulse electrical signal in response to a driving electrical signal, wherein the magnetic tunnel junction comprises a cap layer, a magnetic free layer, a barrier layer, a pinning layer and an antiferromagnetic layer stacked in sequence, and the magnetization direction of the magnetic free layer is in-plane oriented; A first electrode layer, arranged on one side of the magnetic tunnel junction; A second electrode layer, arranged at the other side of the magnetic tunnel junction, the second electrode layer being used to cooperate with the first electrode layer to input the driving electrical signal to the magnetic tunnel junction; Among them, a large thermal resistance interface is formed between the cap layer and the magnetic free layer, and the large thermal resistance interface is used to generate heat accumulation under the action of the driving electrical signal, so as to generate a thermally driven spin transfer torque and thermal noise through a thermal effect. The magnetization direction of the magnetic free layer is continuously flipped under the action of the thermally driven spin transfer torque and the electrically driven spin transfer torque generated by the driving electrical signal to generate a spike pulse electrical signal, and the spike pulse electrical signal has a certain randomness under the action of the thermal noise.

2. The magnetic pulse neuron device according to claim 1, characterized in that: The driving electrical signal is a DC voltage signal, and the voltage intensity of the driving electrical signal is less than or equal to 5V.

3. The magnetic pulse neuron device according to claim 1, characterized in that: The driving electrical signal is a pulse voltage signal, the voltage intensity of the driving electrical signal is less than or equal to 5V, and the signal frequency is less than or equal to 1GHz.

4. The magnetic pulse neuron device according to claim 1, characterized in that: The material of the capping layer is one or more combinations of MgO, Pt, IrMn, MgO / W / MgO, and the thickness of the capping layer is 0.01-10 nm.

5. The magnetic pulse neuron device according to claim 1, characterized in that: The material of the magnetic free layer is Co 20 Fe 60 B 20 , one or more combinations of NiFe, FeB, Py, Co, CoFe, FePt, and Co / Ni, and the thickness of the magnetic free layer is 0.1 to 20 nm.

6. The magnetic pulse neuron device according to claim 1, characterized in that: The material of the barrier layer is one or more combinations of MgO, Al2O3, MgAl2O4, and TiO2, and the thickness of the barrier layer is 0.5-10 nm.

7. The magnetic pulse neuron device according to claim 1, characterized in that: The material of the pinning layer is Co 40 Fe 40 B 20 / Ru / Co 70 Fe 30 , one or more combinations of FeB, Py, Co, CoFe, FePt, and Co / Ni, and the thickness of the pinning layer is 0.1 to 20 nm.

8. The magnetic pulse neuron device according to claim 1, characterized in that: The materials of the first electrode layer and the second electrode layer are Ta, Ru, Mo, Pt, Al, W, Cu, CuN, Au, Cr, Ti, Ni, Pb, TMDCs, SiC, MoO2, ITO, Ti3C2T x , graphene or one or more combinations thereof, the thickness of the first electrode layer and the second electrode layer is 5 to 200 nm.

9. The magnetic pulse neuron device according to claim 1, characterized in that: It also includes a magnetic field generating device, which is used to generate an external magnetic field based on an input control electrical signal to control the magnetization direction of the magnetic free layer to flip and assist in outputting a peak pulse electrical signal.

10. Use of the magnetic pulse neuron device according to any one of claims 1 to 9 as a pulse neuron in a pulse neural network.

Images