阅读说明：本技术 全电控自旋纳米振荡器神经元器件 (Full electric control spinning nano oscillator nerve component ) 是由 邢国忠 王迪 林淮 刘龙 牛洁斌 刘宇 王艳 许晓欣 刘明 于 2021-07-05 设计创作，主要内容包括：本发明公开了全电控自旋纳米振荡器神经元器件,该器件包括：左电极、右电极、顶电极、重金属层、非磁性金属层、左反铁磁钉扎层、右反铁磁钉扎层和MTJ；MTJ包括由下至上的铁磁自由层、势垒隧穿层和铁磁参考层；在全电场调控下,直流电流产生的自旋转移矩与形状各向异性、DM反对称交换作用相互竞争,驱动磁畴壁往复运动实现磁化分量周期性振荡；该器件具有两种振荡模式,可产生稳定、均匀微波信号以实现神经元的振荡特性。可见,该器件可在全电场的情况下实现周期性振荡,无需依靠外部磁场。(The invention discloses a fully electrically controlled spin nanooscillator neural component, which comprises: the MTJ structure comprises a left electrode, a right electrode, a top electrode, a heavy metal layer, a nonmagnetic metal layer, a left antiferromagnetic pinning layer, a right antiferromagnetic pinning layer and an MTJ; the MTJ comprises a ferromagnetic free layer, a barrier tunneling layer and a ferromagnetic reference layer from bottom to top; under the regulation and control of a full electric field, a spin transfer torque generated by direct current competes with shape anisotropy and DM antisymmetric exchange action, and a magnetic domain wall is driven to reciprocate to realize the periodic oscillation of a magnetization component; the device has two oscillation modes, and can generate stable and uniform microwave signals to realize the oscillation characteristics of neurons. Therefore, the device can realize periodic oscillation under the condition of full electric field without depending on external magnetic field.)

1. A fully electronically controlled spin nanooscillator neural component, the neural component comprising: the MTJ structure comprises a left electrode, a right electrode, a top electrode, a heavy metal layer, a nonmagnetic metal layer, a left antiferromagnetic pinning layer, a right antiferromagnetic pinning layer and an MTJ; the MTJ comprises a ferromagnetic free layer, a barrier tunneling layer and a ferromagnetic reference layer from bottom to top;

wherein the ferromagnetic free layer has a shape anisotropic structure; the left antiferromagnetic pinning layer and the right antiferromagnetic pinning layer are respectively arranged at two ends of the upper surface of the ferromagnetic free layer; the left antiferromagnetic pinning layer and the right antiferromagnetic pinning layer have opposite magnetization directions, so that a magnetic domain wall moves in a range between two boundary pinning regions without annihilation;

depositing the non-magnetic metal layer around the ferromagnetic free layer, wherein the left electrode and the right electrode are respectively attached to two ends of the non-magnetic metal layer; the non-magnetic metal layer is used for enabling the current density injected into the ferromagnetic free layer to be uniform;

the heavy metal layer is attached to the lower surface of the ferromagnetic free layer and used for generating a DM (direct memory) antisymmetric exchange effect with the ferromagnetic free layer at an interface;

the top electrode is attached to the ferromagnetic reference layer;

under the regulation and control of a full electric field, a spin transfer torque generated by a direct current competes with shape anisotropy and DM antisymmetric exchange action, a magnetic domain wall is driven to reciprocate to realize the periodic oscillation of a magnetization component, and the magnetic domain wall oscillation shows two oscillation modes according to different densities of the introduced direct current; when the current is removed, the oscillation curve shows nonlinear attenuation and has short-term memory characteristics.

2. The fully electronically controlled spin nanooscillator neural component of claim 1,

the material of the nonmagnetic metal layer comprises one of the following nonmagnetic metals: cu, Ag, Au;

the materials of the ferromagnetic reference layer and the ferromagnetic free layer include one or more of the following ferromagnetic materials with perpendicular magnetic anisotropy: CoFeB, Co/Pt, CoFeAl;

the material of the barrier tunneling layer comprises one of the following materials: MgO and Al₂O₃；

The material of the left antiferromagnetic pinning layer and the right antiferromagnetic pinning layer comprises one or more of the following antiferromagnetic materials: PtMn, IrMn;

the material of the heavy metal layer comprises one of the following materials: pt, Ta, W, Pt alloy, Ta alloy, W alloy;

the left electrode, the right electrode, and the top electrode comprise one of the following materials: Ti/Au, Ti/Pt, Cr/Au, Ta/CuN.

3. A method for preparing a fully electronically controlled spin nanooscillator neural component, the method being used for preparing a fully electronically controlled spin nanooscillator neural component as claimed in any one of claims 1 to 2; the method comprises the following steps:

preparing MTJ;

the injection and pinning of a magnetic domain wall are realized by respectively depositing a left antiferromagnetic pinning layer and a right antiferromagnetic pinning layer with opposite magnetization directions at two ends of the upper surface of the ferromagnetic free layer, so that the magnetic domain wall moves in the range between two boundary pinning regions without annihilation;

depositing a heavy metal layer below the ferromagnetic free layer to introduce a DM (direct magnetic) antisymmetric effect to generate a non-uniform magnetization texture and avoid the magnetic domain wall from being static in a balanced state;

depositing a non-magnetic metal layer around the ferromagnetic free layer, and modulating the shape of the non-magnetic metal layer to modulate the current density flowing through the ferromagnetic free layer to be uniform;

and a top electrode is attached to a ferromagnetic reference layer in the MTJ, and a left electrode and a right electrode are respectively attached to two ends of the nonmagnetic metal layer.

4. A spin nanooscillator comprising the fully electronically controlled spin nanooscillator neural component of claims 1-2 above; the oscillation characteristic and the threshold current of the spin nano-oscillator can be adjusted by modulating one or more parameters of saturation magnetization, perpendicular magnetic anisotropy, exchange strength, damping coefficient, DM (direct mode) antisymmetric action strength and device central angle of a ferromagnetic free layer in the nerve element.

5. A circuit, comprising:

the fully electronically controlled spin nanooscillator neural component and the peripheral circuit as claimed in claims 1-2 above;

in the peripheral circuit, a microwave amplifier is connected in series with a capacitor, and a direct current power supply is connected in series with an inductor; the capacitor and the inductor are connected to a top electrode in the nerve component together, and the direct-current power supply and the microwave amplifier are connected to a right electrode in the nerve component together.

6. A two-dimensional oscillation network based on a fully electronically-controlled spin nanooscillator neural component is characterized by comprising the following components: the device comprises a spinning nano oscillator neuron two-dimensional array, a microwave strip and a write-in circuit;

wherein the two-dimensional array of spin nanooscillator neurons comprises: a plurality of fully electronically controlled spin nanooscillator neural components as claimed in claims 1-2 arranged in two dimensions;

the writing circuit is used for injecting current for the nerve element;

the microwave strip is provided with a microwave signal obtained by encoding a signal to be processed;

the microwave band is applied to the spin nano-oscillator neuron two-dimensional array, the current injected by each neuron is adjusted, the injection-locked phase diagram of the current is matched with the frequency of the microwave signal, and the identification of the signal to be processed is realized.

7. A three-dimensional oscillation network based on a fully-electronically-controlled spin nanooscillator neural component is characterized by comprising the following components in parts by weight: the device comprises a spinning nanooscillator neuron three-dimensional array, a microwave strip and a write-in circuit;

wherein the three-dimensional array of spin nanooscillator neurons comprises: a plurality of fully electronically controlled spin nanooscillator neural components as claimed in claims 1-2 arranged in three dimensions; in the three-dimensional array, all nerve components are coupled with each other;

the writing circuit is used for injecting current for the nerve element;

the microwave strip is provided with a microwave signal obtained by encoding a signal to be processed;

the microwave band is applied to the spin nano-oscillator neuron three-dimensional array, the current injected by each neuron is adjusted, the injection-locked phase diagram of the current is matched with the frequency of the microwave signal, and the identification of the signal to be processed is realized.

8. A two-dimensional reserve pool computational structure based on full electrically controlled spin nanooscillator neural components, comprising: the device comprises an input layer, a reserve pool formed by a spinning nanooscillator neuron two-dimensional array and an output layer;

wherein the two-dimensional array of spin nanooscillator neurons comprises: a plurality of fully electronically controlled spin nanooscillator neural components as claimed in claims 1-2 arranged in two dimensions;

and injecting a pulse sequence formed by encoding a signal to be processed into a reserve pool formed by a spin nano-oscillator neuron two-dimensional array to generate an output sequence with time dynamic characteristics.

9. The two-dimensional pool computing architecture of claim 8,

and aiming at each full-electric-control spinning nano-oscillator nerve component, the nerve component is connected with a detection circuit.

10. A three-dimensional reserve pool computational structure based on full electronic control spin nanooscillator neural components, characterized by, includes: the device comprises an input layer, a reserve pool formed by a spinning nanooscillator neuron three-dimensional array and an output layer;

wherein the three-dimensional array of spin nanooscillator neurons comprises: a plurality of fully electronically controlled spin nanooscillator neural components as claimed in claims 1-2 arranged in three dimensions; in the three-dimensional array, all nerve components are coupled with each other;

and injecting a pulse sequence formed by encoding a signal to be processed into a reserve pool formed by a spin nano oscillator neuron three-dimensional array to generate an output sequence with time dynamic characteristics.

11. The three-dimensional pool computing structure of claim 10,

and aiming at each full-electric-control spinning nano-oscillator nerve component, the nerve component is connected with a detection circuit.

Technical Field

The application relates to the technical field of information safety, in particular to a fully-electronically-controlled spinning nanooscillator neural component.

Background

The conventional microelectronic device only utilizes the intrinsic property of the charge of the electron, and the characteristics and advantages of the electron spin property gradually appear along with the shrinking of the size of the silicon device. Since the discovery of Giant Magnetoresistance (GMR), spintronics to study the spin properties of electrons has been emerging. At present, a Magnetic Random Access Memory (MRAM) driven by Spin Transfer Torque (STT) and Spin Orbit Torque (SOT) has been developed greatly, however, research on a Spin oscillator, which is another research direction of spintronics, is relatively less.

Aiming at the research of the spin nano oscillator, the device structure is mainly based on a nano column structure and a point contact structure, and the magnetic equilibrium state comprises in-plane magnetization, in-plane/out-of-plane magnetization, perpendicular magnetization, a structure of a multi-free layer or the circumferential motion of a Sgmon, and the like. Under the drive of Spin transfer torque generated by vertical injection current or Spin polarization current induced by Spin Orbit Coupling (SOC) effect of a heavy metal layer, the magnetization direction of a free layer is driven to precess at high speed, and GHz or even THz high-frequency microwave signals are excited.

However, current spin-based nanooscillators require external magnetic field assistance.

Disclosure of Invention

The invention provides a fully electrically controlled spin nanooscillator neural component, which aims to solve or partially solve the technical problem that the prior spin nanooscillator needs an external magnetic field for assistance, and can realize periodic oscillation of a domain wall under the condition of a full electric field so as to lead the magnetization direction of a free layer to precess at a high speed in a plane.

In order to solve the above technical problem, the present invention provides a fully electronically controlled spin nanooscillator neural device, including: the MTJ structure comprises a left electrode, a right electrode, a top electrode, a heavy metal layer, a nonmagnetic metal layer, a left antiferromagnetic pinning layer, a right antiferromagnetic pinning layer and an MTJ; the MTJ comprises a ferromagnetic free layer, a barrier tunneling layer and a ferromagnetic reference layer from bottom to top;

wherein the ferromagnetic free layer has a shape anisotropic structure; the left antiferromagnetic pinning layer and the right antiferromagnetic pinning layer are respectively arranged at two ends of the upper surface of the ferromagnetic free layer; the left antiferromagnetic pinning layer and the right antiferromagnetic pinning layer have opposite magnetization directions, so that a magnetic domain wall moves in a range between two boundary pinning regions without annihilation;

depositing the non-magnetic metal layer around the ferromagnetic free layer, wherein the left electrode and the right electrode are respectively attached to two ends of the non-magnetic metal layer; the non-magnetic metal layer is used for enabling the current density injected into the ferromagnetic free layer to be uniform;

the heavy metal layer is attached to the lower surface of the ferromagnetic free layer and used for generating a DM (direct memory) antisymmetric exchange effect with the ferromagnetic free layer at an interface;

the top electrode is attached to the ferromagnetic reference layer;

under the regulation and control of a full electric field, a spin transfer torque generated by a direct current competes with shape anisotropy and DM antisymmetric exchange action, a magnetic domain wall is driven to reciprocate to realize the periodic oscillation of a magnetization component, and the magnetic domain wall oscillation shows two oscillation modes according to different densities of the introduced direct current; when the current is removed, the oscillation curve shows nonlinear attenuation and has short-term memory characteristics.

Preferably, the material of the nonmagnetic metal layer includes one of the following nonmagnetic metals: cu, Ag, Au;

the materials of the ferromagnetic reference layer and the ferromagnetic free layer include one or more of the following ferromagnetic materials with perpendicular magnetic anisotropy: CoFeB, Co/Pt, CoFeAl;

the material of the barrier tunneling layer comprises one of the following materials: MgO and Al₂O₃；

The material of the left antiferromagnetic pinning layer and the right antiferromagnetic pinning layer comprises one or more of the following antiferromagnetic materials: PtMn, IrMn;

the material of the heavy metal layer comprises one of the following materials: pt, Ta, W, Pt alloy, Ta alloy, W alloy;

the left electrode, the right electrode, and the top electrode comprise one of the following materials: Ti/Au, Ti/Pt, Cr/Au, Ta/CuN.

The invention also discloses a preparation method of the fully electrically controlled spin nanooscillator neural component, which is used for preparing the fully electrically controlled spin nanooscillator neural component; the method comprises the following steps:

preparing MTJ;

the injection and pinning of a magnetic domain wall are realized by respectively depositing a left antiferromagnetic pinning layer and a right antiferromagnetic pinning layer with opposite magnetization directions at two ends of the upper surface of the ferromagnetic free layer, so that the magnetic domain wall moves in the range between two boundary pinning regions without annihilation;

depositing a heavy metal layer below the ferromagnetic free layer to introduce a DM (direct magnetic) antisymmetric effect to generate a non-uniform magnetization texture and avoid the magnetic domain wall from being static in a balanced state;

depositing a non-magnetic metal layer around the ferromagnetic free layer, and modulating the shape of the non-magnetic metal layer to modulate the current density flowing through the ferromagnetic free layer to be uniform;

and a top electrode is attached to a ferromagnetic reference layer in the MTJ, and a left electrode and a right electrode are respectively attached to two ends of the nonmagnetic metal layer.

The invention also discloses a spinning nanooscillator, which comprises the fully-electrically-controlled spinning nanooscillator nerve component; the oscillation characteristic and the threshold current of the spin nano-oscillator can be adjusted by modulating one or more parameters of saturation magnetization, perpendicular magnetic anisotropy, exchange strength, damping coefficient, DM (direct mode) antisymmetric action strength and device central angle of a ferromagnetic free layer in the nerve element.

The invention also discloses a circuit comprising:

the fully electric control spinning nano oscillator comprises the nerve component and a peripheral circuit;

in the peripheral circuit, a microwave amplifier is connected in series with a capacitor, and a direct current power supply is connected in series with an inductor; the capacitor and the inductor are connected to a top electrode in the nerve component together, and the direct-current power supply and the microwave amplifier are connected to a right electrode in the nerve component together.

The invention also discloses a two-dimensional oscillation network based on the fully electrically controlled spinning nano-oscillator neural component, which comprises the following components: the device comprises a spinning nano oscillator neuron two-dimensional array, a microwave strip and a write-in circuit;

wherein the two-dimensional array of spin nanooscillator neurons comprises: a plurality of fully electrically controlled spinning nanooscillator nerve components which are arranged in a two-dimensional way;

the writing circuit is used for injecting current for the nerve element;

the microwave strip is provided with a microwave signal obtained by encoding a signal to be processed;

the microwave band is applied to the spin nano-oscillator neuron two-dimensional array, the current injected by each neuron is adjusted, the injection-locked phase diagram of the current is matched with the frequency of the microwave signal, and the identification of the signal to be processed is realized.

The invention also discloses a three-dimensional oscillation network based on the fully-electronically-controlled spinning nano-oscillator neural component, which comprises the following components in percentage by weight: the device comprises a spinning nanooscillator neuron three-dimensional array, a microwave strip and a write-in circuit;

wherein the three-dimensional array of spin nanooscillator neurons comprises: a plurality of fully electrically controlled spinning nanooscillator nerve components which are arranged in a three-dimensional way; in the three-dimensional array, all nerve components are coupled with each other;

the writing circuit is used for injecting current for the nerve element;

the microwave strip is provided with a microwave signal obtained by encoding a signal to be processed;

the microwave band is applied to the spin nano-oscillator neuron three-dimensional array, the current injected by each neuron is adjusted, the injection-locked phase diagram of the current is matched with the frequency of the microwave signal, and the identification of the signal to be processed is realized.

The invention also discloses a two-dimensional reserve pool computing structure based on the fully electrically controlled spinning nanooscillator neural component, which comprises the following steps: the device comprises an input layer, a reserve pool formed by a spinning nanooscillator neuron two-dimensional array and an output layer;

wherein the two-dimensional array of spin nanooscillator neurons comprises: a plurality of fully electrically controlled spinning nanooscillator nerve components which are arranged in a two-dimensional way;

and injecting a pulse sequence formed by encoding a signal to be processed into a reserve pool formed by a spin nano-oscillator neuron two-dimensional array to generate an output sequence with time dynamic characteristics.

Preferably, for each fully electrically controlled spin nanooscillator neural component, the neural component is connected with a detection circuit.

The invention also discloses a three-dimensional reserve pool computing structure based on the full-electronic spin nanooscillator neural component, which comprises the following steps: the device comprises an input layer, a reserve pool formed by a spinning nanooscillator neuron three-dimensional array and an output layer;

wherein the three-dimensional array of spin nanooscillator neurons comprises: a plurality of fully electrically controlled spinning nanooscillator nerve components which are arranged in a three-dimensional way; in the three-dimensional array, all nerve components are coupled with each other;

and injecting a pulse sequence formed by encoding a signal to be processed into a reserve pool formed by a spin nano oscillator neuron three-dimensional array to generate an output sequence with time dynamic characteristics.

Preferably, for each fully electrically controlled spin nanooscillator neural component, the neural component is connected with a detection circuit.

Through one or more technical schemes of the invention, the invention has the following beneficial effects or advantages:

(1) the invention discloses a fully electrically controlled spin nanooscillator neural component, which comprises: the MTJ structure comprises a left electrode, a right electrode, a top electrode, a heavy metal layer, a nonmagnetic metal layer, a left antiferromagnetic pinning layer, a right antiferromagnetic pinning layer and an MTJ; the MTJ comprises a ferromagnetic free layer, a barrier tunneling layer and a ferromagnetic reference layer from bottom to top; under the regulation and control of a full electric field, a spin transfer torque generated by direct current competes with shape anisotropy and DM antisymmetric exchange action, and a magnetic domain wall is driven to reciprocate to realize the periodic oscillation of a magnetization component; the device has two oscillation modes, and can generate stable and uniform microwave signals to realize the oscillation characteristics of neurons. Therefore, the device can realize periodic oscillation under the condition of full electric field without depending on external magnetic field.

(2) The spin nanooscillator neural component can realize the periodic oscillation of a domain wall under the condition of a full electric field, so that the magnetization direction of the free layer precesses at a high speed in the plane, and a high-frequency microwave signal can be generated by combining the MTJ and an external circuit to simulate the oscillation characteristic of human brain neurons;

(3) the oscillation characteristic of the spin nanooscillator neural component can be used for constructing an oscillation network to complete tasks such as voice and image recognition;

(4) the nonlinear characteristic of the spin nanooscillator neural component and the characteristic of mutual coupling among the components can be used for constructing tasks such as time-sharing multiplexing reserve pool calculation or physical reserve pool calculation.

(5) The spin nanooscillator neural component can be used in the fields of auxiliary writing of a hard disk magnetic head, high-frequency microwave signal generators, communication and the like.

The foregoing description is only an overview of the technical solutions of the present invention, and the embodiments of the present invention are described below in order to make the technical means of the present invention more clearly understood and to make the above and other objects, features, and advantages of the present invention more clearly understandable.

Drawings

Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The drawings are only for purposes of illustrating the preferred embodiments and are not to be construed as limiting the invention. Also, like reference numerals are used to refer to like parts throughout the drawings. In the drawings:

FIG. 1 shows a schematic structural diagram of a fully electronically controlled spin nanooscillator neural component and an output circuit according to an embodiment of the present invention;

FIG. 2 shows free layer oscillation characteristics and Fourier frequency domain analysis of a fully electronically controlled spin nanooscillator neural component in accordance with an embodiment of the invention;

FIG. 3 shows the variation curves of the oscillation frequency and the saturation magnetization, the perpendicular magnetic anisotropy, the damping coefficient, the exchange strength and the central angle of the fully electronically controlled spin nanooscillator neural component according to one embodiment of the invention;

FIG. 4 is a graph showing the threshold current versus DM antisymmetric effect of the fully electronically controlled spin nanooscillator neural component in accordance with an embodiment of the present invention;

FIG. 5 is a schematic diagram showing the structural principle of an oscillation network based on a fully electrically controlled spin nanooscillator neural component according to an embodiment of the present invention;

fig. 6 is a schematic diagram showing the structural principle of reservoir calculation based on a fully electrically controlled spin nanooscillator neural component according to an embodiment of the present invention.

Detailed Description

In order to make the present application more clearly understood by those skilled in the art to which the present application pertains, the following detailed description of the present application is made with reference to the accompanying drawings by way of specific embodiments.

In view of the above problems, the present invention provides a fully electrically controlled spin nanooscillator neural device, which can be used for hard disk magnetic head assisted writing, high frequency microwave signal generator, communication, pulse generator, wireless communication, transceiver, magnetic field detection, disk drive or neural morphology calculation, etc. Specifically, under the regulation and control of a full electric field, the spin transfer torque generated by direct current and the shape anisotropy and DM antisymmetric exchange interaction compete with each other, and the magnetic domain wall is driven to reciprocate to realize the periodic oscillation of the magnetization component; the device has two oscillation modes, and can generate stable and uniform microwave signals to realize the oscillation characteristics of neurons. Based on the method and the device, a scheme of a two-dimensional and three-dimensional oscillation network is provided, and tasks such as voice and image recognition are facilitated to be completed. Meanwhile, based on the nonlinear short-time memory characteristic of the nerve component, the scheme of two-dimensional reserve pool calculation and three-dimensional physical reserve pool is further provided, and the method is suitable for dynamic recognition tasks. The device is compatible with the existing CMOS (complementary metal oxide semiconductor) integrated process, is beneficial to large-scale preparation, and is beneficial to realizing the integrated application of high-density neural morphology calculation.

Fig. 1 shows a fully electronically controlled spin nanooscillator neural component and an output circuit according to the technical solution. Fig. 1(a) is a schematic structural diagram of a fully electronically controlled spin nanooscillator neural component according to the technical solution, where the neural component includes: a left electrode, a right electrode, a top electrode, a heavy metal layer, a nonmagnetic metal layer, a left antiferromagnetic pinning layer, a right antiferromagnetic pinning layer, and an MTJ.

Further, the MTJ includes, from bottom to top, a ferromagnetic free layer, a barrier tunneling layer, and a ferromagnetic reference layer.

Wherein the materials of the ferromagnetic reference layer and the ferromagnetic free layer include one or more of the following ferromagnetic materials having perpendicular magnetic anisotropy: CoFeB, Co/Pt, CoFeAl. Wherein Co/Pt represents: co and Pt multilayer materials, namely: a layer of Co and a layer of Pt. The material of the barrier tunneling layer comprises one of the following materials: MgO and Al₂O₃. The ferromagnetic free layer has a shape anisotropic structure; the shape anisotropic structures are various, such as a sector, a triangle, a trapezoid, etc., and may be substituted for each other. The ferromagnetic reference layer and the barrier tunneling layer may be circular, elliptical, rectangular, etc., and may be interchanged.

And a left antiferromagnetic pinning layer and a right antiferromagnetic pinning layer are respectively arranged at two ends of the upper surface of the ferromagnetic free layer. Specifically, the left antiferromagnetic pinning layer and the right antiferromagnetic pinning layer may be attached to both ends of the upper surface of the ferromagnetic free layer. In addition, the local thickness of the two ends of the upper surface of the ferromagnetic free layer can be increased. The left antiferromagnetic pinning layer and the right antiferromagnetic pinning layer have opposite magnetization directions, so that a magnetic domain wall moves in a range between two boundary pinning regions without annihilation. The left antiferromagnetic pinning layer and the right antiferromagnetic pinning layer have the shape of a ring and a sector respectively. Both materials include one or more of the following antiferromagnetic materials: PtMn and IrMn.

And depositing a nonmagnetic metal layer around the ferromagnetic free layer, wherein the material of the nonmagnetic metal layer comprises one of the following nonmagnetic metals: cu, Ag, Au, etc. The non-magnetic metal layer is used for enabling the current density injected into the ferromagnetic free layer to be uniform and equal; the shape of the non-magnetic metal layer is adjusted according to the self resistivity, the shape of the ferromagnetic free layer and the shape resistivity of the ferromagnetic free layer so as to ensure that the current density injected into the ferromagnetic free layer is uniform and equal.

The two ends of the non-magnetic metal layer are respectively attached with a left electrode and a right electrode for injecting current, and specifically, the current flows into the right electrode from the left electrode and flows out from the right electrode. The top electrode is attached to the ferromagnetic reference layer. The left electrode, the right electrode and the top electrode comprise one of the following materials: Ti/Au, Ti/Pt, Cr/Au, Ta/CuN. Wherein, Ti/Au represents a multilayer material of Ti and Au, namely a layer of Ti and a layer of Au. Other materials containing "/" signs also represent multiple layers of material and are not described in detail herein.

The lower surface of the ferromagnetic free layer is pasted with a heavy metal layer for generating DM antisymmetric exchange effect with the ferromagnetic free layer at the interface. The material of the heavy metal layer comprises one of the following materials: pt, Ta, W, Pt alloy, Ta alloy, W alloy.

Under the regulation and control of the full electric field, direct current is injected, the spin transfer torque generated by the direct current competes with the shape anisotropy and DM antisymmetric exchange action, and the magnetic domain wall is driven to reciprocate to realize the periodic oscillation of the magnetization component. In addition, the oscillation position of the magnetic domain wall is different according to different densities of the introduced direct current, and the magnetic domain wall oscillation shows two oscillation modes: when the direct current density is low, the ultrahigh frequency oscillation frequency of linear negative correlation is obtained; and at high direct current density, the extremely high frequency oscillation frequency is linearly and positively correlated. The oscillation shows good nonlinear characteristic, when the current is removed, the oscillation curve shows nonlinear attenuation, and the short-time memory characteristic is achieved.

In a specific implementation process, the magnetization direction of the left antiferromagnetic pinning layer is along the-z direction, and the magnetization direction of the right antiferromagnetic pinning layer is along the + z direction, so that a magnetic domain wall can move in a range between two boundary pinning regions without annihilation; the ferromagnetic reference layer magnetization direction is along the y-direction. In the initial state, the domain wall in the ferromagnetic free layer remains near the right antiferromagnetic pinning layer boundary due to shape anisotropy. When direct current in the + x direction is introduced, the magnetic domain wall moves towards the-x direction under the drive of the STT, under the assistance of non-uniform magnetic textures induced by the DM anti-symmetric effect, the magnetic domain wall is always in an unbalanced state, namely the DM anti-symmetric exchange effect induces an equivalent magnetic field, under the competition of the STT with the shape anisotropy effect and the DM anti-symmetric exchange effect, the magnetic domain wall reciprocates left and right, the periodical oscillation precession of the magnetization direction is realized while moving, and corresponding microwave signals can be output by utilizing the tunneling magneto-resistance effect.

Based on the same inventive concept, the embodiment of the invention discloses a preparation method of a fully electronically controlled spin nanooscillator neural component, which is used for preparing the fully electronically controlled spin nanooscillator neural component described in the embodiment. It is noted that the sequence of the steps in the preparation method can be adjusted according to the actual situation, and the method comprises the following steps:

preparing MTJ;

the injection and pinning of a magnetic domain wall are realized by respectively depositing a left antiferromagnetic pinning layer and a right antiferromagnetic pinning layer with opposite magnetization directions at two ends of the upper surface of the ferromagnetic free layer, so that the magnetic domain wall moves in the range between two boundary pinning regions without annihilation;

depositing a heavy metal layer below the ferromagnetic free layer, introducing a DM (direct magnetic) antisymmetric effect to generate a non-uniform magnetization texture, and preventing a magnetic domain wall from being static in a balanced state;

depositing a non-magnetic metal layer around the ferromagnetic free layer, and modulating the shape of the non-magnetic metal layer to modulate the current density flowing through the ferromagnetic free layer to be uniform;

a top electrode is attached to a ferromagnetic reference layer in the MTJ, and a left electrode and a right electrode are attached to two ends of the nonmagnetic metal layer, respectively.

After the nerve cell component is prepared, direct current is applied to the right electrode, and the oscillation phenomenon that magnetic domain walls compete with each other under the actions of spin transfer torque, shape anisotropy, DM antisymmetric exchange and the like is realized. When different DC current densities are applied, different high-frequency microwave signals can be output by combining the MTJ and an external circuit.

It is noted that the above definitions of devices and methods are not limited to the specific structures, shapes or patterns mentioned in the examples, and that the dimensions of the devices and layers therein may be scaled according to the process.

Fig. 2 shows an oscillation characteristic curve and fourier frequency domain analysis of the fully electronically controlled spin nanooscillator neural component according to the technical solution. FIG. 2(a) is an oscillation curve of the y-direction normalized magnetization component after DC filtering, in this embodiment, the DC current density is 5 × 10⁷A/cm²In the case of (2), uniform and stable oscillation characteristics are exhibited. FIG. 2(b) is a Fourier frequency domain analysis of the oscillation curve of FIG. 2(a) with an applied 5X 10⁷A/cm²The oscillation signal of 3.04GHz can be obtained by the direct current density of (2).

Fig. 4 shows a variation relationship between the threshold current of the fully electronically controlled spin nanooscillator neural component and the DM antisymmetric effect, and under the driving of STT, the DM antisymmetric effect introduces an uneven magnetization texture, which results in a blocked angular momentum transfer and an increase in the threshold current density, and it can also be seen from fig. 5 that the DM antisymmetric effect strength is increased, and the threshold current density is increased accordingly.

The spin nanooscillator neural component can realize the periodic oscillation of a domain wall under the condition of a full electric field, so that the magnetization direction of the free layer precesses at a high speed in the plane, and a high-frequency microwave signal can be generated by combining the MTJ and an external circuit, thereby simulating the oscillation characteristic of human brain neurons. Therefore, based on the same inventive concept, the embodiment of the invention discloses a circuit for outputting a stable microwave signal, and fig. 1(b) shows a readout circuit of a fully electronically controlled spin nanooscillator neural component of the technical scheme. The method comprises the following steps: the fully electronically controlled spin nanooscillator neural components and peripheral circuits described in the above embodiments. In the peripheral circuit, a microwave amplifier is connected in series with a capacitor, and a direct current power supply is connected in series with an inductor; the capacitance and the inductance are jointly connected to a top electrode in the nerve element, the direct-current power supply and the microwave amplifier are jointly connected to a right electrode in the nerve element, and the nerve element is combined with the tee joint (the capacitance and the inductance) and the microwave amplifier to output stable microwave signals. Further, the microwave amplifier is connected with an oscilloscope.

Fig. 3 shows an oscillation frequency variation modulation curve of the fully electronically controlled spin nanooscillator neural component of the technical scheme. FIG. 3(a) shows the relationship between the oscillation frequency and the saturation magnetization of the spin nanooscillator, and the oscillation frequency decreases with the increase of the saturation magnetization; FIG. 3(b) is a graph showing the relationship between the oscillation frequency of the spin nanooscillator and the perpendicular magnetic anisotropy, in which the oscillation frequency increases with the increase of the perpendicular magnetic anisotropy; FIG. 3(c) is a graph showing the relationship between the oscillation frequency of the spin nanooscillator and the exchange intensity, wherein the oscillation frequency increases with the increase of the exchange intensity; FIG. 3(d) shows the relationship between the oscillation frequency and the damping coefficient of the spin nanooscillator, and the oscillation frequency increases with the increase of the damping coefficient; FIG. 3(e) shows the relationship between the oscillation frequency of the spin nanooscillator and the central angle, and the oscillation frequency decreases as the central angle increases. Appropriate material and device parameters can be modulated to achieve the appropriate oscillation frequency based on the above relationship.

Therefore, the spin nano-oscillator with different oscillation characteristics and threshold currents can be realized by modulating parameters such as saturation magnetization, perpendicular magnetic anisotropy, exchange strength, damping coefficient, DM (direct magnetic resonance) antisymmetric action strength, device central angle and the like of the free layer.

Based on the same inventive concept, the embodiment of the invention discloses a spin nanooscillator, which comprises a fully electrically controlled spin nanooscillator nerve component described in the embodiment; the oscillation characteristic and the threshold current of the spin nano-oscillator can be adjusted by modulating one or more parameters of saturation magnetization, perpendicular magnetic anisotropy, exchange intensity, damping coefficient, DM (direct magnetic) antisymmetric action intensity and device central angle of a ferromagnetic free layer in a nerve element.

The oscillation characteristic of the spin nanooscillator neural component can be used for constructing an oscillation network to complete tasks such as voice and image recognition. Therefore, based on the same inventive concept, the embodiment discloses a two-dimensional oscillation network based on a fully electronically controlled spin nanooscillator neural component, which is characterized by comprising the following steps: the device comprises a spinning nano oscillator neuron two-dimensional array, a microwave strip and a write-in circuit;

wherein the two-dimensional array of spin nanooscillator neurons comprises: a plurality of fully electronically controlled spin nanooscillator nerve elements described in the above embodiments arranged in two dimensions;

the writing circuit is used for injecting current for the nerve element;

the microwave strip is provided with a microwave signal obtained by encoding a signal to be processed;

the microwave strip is applied to the spin nanooscillator neuron two-dimensional array, the current injected by each neuron is adjusted, the injection locked phase diagram of the current is matched with the frequency of a microwave signal, and the identification of a signal to be processed is realized.

Fig. 5 is a schematic diagram and a schematic diagram of an oscillation network structure of a fully electronically controlled spin nanooscillator neural component according to the technical scheme. FIG. 5(a) is a schematic diagram of a two-dimensional oscillation network structure based on an electrically controlled spin nanooscillator neural component, wherein f_A+f_BFor encoding signals (signals to be processed) of images, speech, etc. into f_AAnd f_BMicrowave signals with two frequencies are close to the oscillation frequency of the spin nano-oscillator nerve component and are applied to the two-dimensional spin nano-oscillator neuron array through the strips. According to the principle of injection locking, when the applied external microwave signal and the oscillation frequency of the spin nano-oscillator are close enough, the coupling effect of the two is very strong, so that the oscillation frequency of the spin nano-oscillator is limited andthe external microwave signals have the same frequency. When different direct currents are respectively introduced into the spinning nano-oscillators in the array, different oscillation frequencies can be generated, injection locking is formed between the spinning nano-oscillators and different external microwave signals, and injection locking phase diagrams under different conditions can be generated. Fixing the microwave signal encoded by the image or the voice signal, training and adjusting the direct current introduced by different spinning nano-oscillator neurons, and enabling the injection-locked phase diagram to deviate towards the designated signal. Finally, each injection locking area corresponds to a specific image signal and a specific voice signal respectively, and recognition of the image signal, the voice signal and other signals is achieved, namely the function of the oscillation network is achieved.

Based on the same invention concept, the embodiment of the invention discloses a three-dimensional oscillation network based on a fully-electronically-controlled spin nanooscillator neural component, which is characterized by comprising the following steps: the device comprises a spinning nanooscillator neuron three-dimensional array, a microwave strip and a write-in circuit;

wherein the three-dimensional array of spin nanooscillator neurons comprises: a plurality of fully electronically controlled spin nanooscillator nerve components described in the above embodiments arranged in three dimensions; in the three-dimensional array, all nerve components are coupled with each other;

the writing circuit is used for injecting current for the nerve element;

the microwave strip is provided with a microwave signal obtained by encoding a signal to be processed;

the microwave strip is applied to the spin nanooscillator neuron three-dimensional array, the current injected by each neuron is adjusted, the injection locked phase diagram of the current is matched with the frequency of a microwave signal, and the identification of a signal to be processed is realized.

Fig. 5(b) is a schematic diagram of a three-dimensional oscillation network structure based on an electrically controlled spin nanooscillator neural component, and the principle is similar to that of a two-dimensional oscillation network, so that details are not repeated here, and along with more complex functions, more classified numbers and larger oscillation networks, the significance of constructing a three-dimensional spin nanooscillator neuron array is more significant, so that not only can the area overhead be reduced, but also good spatial coupling can be formed between neurons, and superior performance and accuracy can be expected to be exhibited. It is noted that the image, voice wait processing signal can be encoded into microwave signals of three or even more frequencies to construct more complex and diverse oscillation networks.

Due to the nonlinear characteristic of the spin nanooscillator neural component and the characteristic of mutual coupling among the components, the method can be used for constructing tasks such as time-sharing multiplexing reserve pool calculation or physical reserve pool calculation. Therefore, based on the same invention concept, the embodiment of the invention discloses a two-dimensional reserve pool computing structure based on a fully-electronically-controlled spin nanooscillator neural component, which is characterized by comprising the following steps of: the device comprises an input layer, a reserve pool formed by a spinning nanooscillator neuron two-dimensional array and an output layer;

wherein the two-dimensional array of spin nanooscillator neurons comprises: a plurality of fully electronically controlled spin nanooscillator nerve elements described in the above embodiments arranged in two dimensions;

and injecting a pulse sequence formed by encoding a signal to be processed into a reserve pool formed by a spin nano-oscillator neuron two-dimensional array to generate an output sequence with time dynamic characteristics.

Optionally, for each fully electronically controlled spin nanooscillator neural component, the neural component is connected to the detection circuit. The detection circuit comprises a capacitor, an inductor, a direct current power supply and a diode.

One end of the direct current power supply is connected with the right electrode, the other end of the direct current power supply is connected with the inductor, a connecting circuit led out by the top electrode and the inductor are connected with one end of the capacitor together, and the other end of the capacitor is connected with the diode.

Fig. 6 is a schematic structural diagram and a schematic diagram of reservoir calculation of the fully electronically controlled spin nanooscillator neural component according to the technical scheme. Fig. 6(a) is a structural schematic diagram of a two-dimensional reserve pool calculation based on a spin nano-oscillator neural component, which samples and codes signals such as voice, image signals, or video signals into pulse sequences with a certain amplitude, each pulse sequence and the spin nano-oscillator neural component make corresponding responses, and outputs microwave voltage signals through a tee (capacitor, inductor) and a diode detection circuit, and then the spin nano-oscillator neural component enters a short-time memory stage, and when the next pulse arrives, the spin nano-oscillator neural component contains certain characteristics of a pre-pulse signal, and generates an output sequence with time dynamic characteristics. In the training stage, linear weights of each neuron in the reserve pool to an output layer are continuously adjusted according to expected output until the neuron can generate accurate output, so that the neuron can generate correct output aiming at a specific dynamic nonlinear prediction task. The good predicted value can be obtained only by simply training the weight of linear mapping from the spin nanooscillator neurons to the output sequence. For some dynamic problems, such as voice processing tasks, the weight from the reserve pool to the output layer is trained, so that the whole reserve pool can finally directly output the processed signals. The input signals do not need to be decomposed, calculated in a complex mode and the like, and therefore the problems of time delay, power consumption and the like caused by operation can be reduced.

Based on the same invention concept, the embodiment of the invention discloses a three-dimensional reserve pool computing structure based on a fully-electronically-controlled spin nanooscillator neural component, which is characterized by comprising the following steps: the device comprises an input layer, a reserve pool formed by a spinning nanooscillator neuron three-dimensional array and an output layer;

wherein the three-dimensional array of spin nanooscillator neurons comprises: a plurality of fully electronically controlled spin nanooscillator nerve components described in the above embodiments arranged in three dimensions; in the three-dimensional array, all nerve components are coupled with each other;

and injecting a pulse sequence formed by encoding a signal to be processed into a reserve pool formed by a spin nano oscillator neuron three-dimensional array to generate an output sequence with time dynamic characteristics.

Optionally, for each fully electronically controlled spin nanooscillator neural component, the neural component is connected to the detection circuit. The detection circuit comprises a capacitor, an inductor, a direct current power supply and a diode.

One end of the direct current power supply is connected with the right electrode, the other end of the direct current power supply is connected with the inductor, a connecting circuit led out by the top electrode and the inductor are connected with one end of the capacitor together, and the other end of the capacitor is connected with the diode.

Fig. 6(b) is a schematic diagram of a three-dimensional physical reserve pool computing structure based on a spin nano-oscillator neural component, and the basic principle is similar to that of fig. 6(a), and thus, the description is omitted. On the basis, the spinning nano-oscillator nerve elements are constructed into a three-dimensional array, and the structure, the position and various material parameters of the device can be relatively randomly modulated, so that the spinning nano-oscillator nerve elements are coupled on the physical layer through stray fields, the complexity of the reserve pool is improved, and more superior performance and accuracy are expected to be shown. Furthermore, process fluctuations are to some extent favorable for the three-dimensional physical reservoir computational structure, which therefore reduces the difficulty of the process.

Through one or more embodiments of the present invention, the present invention has the following advantageous effects or advantages:

(1) the invention discloses a fully electrically controlled spin nanooscillator neural component, which comprises: the MTJ structure comprises a left electrode, a right electrode, a top electrode, a heavy metal layer, a nonmagnetic metal layer, a left antiferromagnetic pinning layer, a right antiferromagnetic pinning layer and an MTJ; the MTJ comprises a ferromagnetic free layer, a barrier tunneling layer and a ferromagnetic reference layer from bottom to top; under the regulation and control of a full electric field, a spin transfer torque generated by direct current competes with shape anisotropy and DM antisymmetric exchange action, and a magnetic domain wall is driven to reciprocate to realize the periodic oscillation of a magnetization component; the device has two oscillation modes, and can generate stable and uniform microwave signals to realize the oscillation characteristics of neurons. Therefore, the device can realize periodic oscillation under the condition of full electric field without depending on external magnetic field.

(2) The spin nanooscillator neural component can realize the periodic oscillation of a domain wall under the condition of a full electric field, so that the magnetization direction of the free layer precesses at a high speed in the plane, and a high-frequency microwave signal can be generated by combining the MTJ and an external circuit to simulate the oscillation characteristic of human brain neurons;

(3) the oscillation characteristic of the spin nanooscillator neural component can be used for constructing an oscillation network to complete tasks such as voice and image recognition;

(4) the nonlinear characteristic of the spin nanooscillator neural component and the characteristic of mutual coupling among the components can be used for constructing tasks such as time-sharing multiplexing reserve pool calculation or physical reserve pool calculation.

(5) The spin nanooscillator neural component can be used in the fields of auxiliary writing of a hard disk magnetic head, high-frequency microwave signal generators, communication and the like.

While the preferred embodiments of the present application have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. Therefore, it is intended that the appended claims be interpreted as including preferred embodiments and all alterations and modifications as fall within the scope of the application.

It will be apparent to those skilled in the art that various changes and modifications may be made in the present application without departing from the spirit and scope of the application. Thus, if such modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is intended to include such modifications and variations as well.