A brain-like computational neuron based on tilt anisotropy

Abstract

This invention discloses a brain-like computing neuron based on tilted anisotropy. The invention utilizes a CoPt multilayer film grown on a slanted substrate, microfabricated into a nanowire structure, to serve as the brain-like computing neuron, thus realizing a magnetic thin film system with tilted anisotropy. Based on the different domain wall motion velocities under positive and negative currents in a tilted anisotropic magnetic thin film material system, this invention designs a brain-like computing neuron that can be used to realize the conversion of analog signals to digital signals before and after a threshold voltage in brain-like devices.

Description

Technical Field

[0001] This invention belongs to the field of advanced chips, specifically relating to a brain-like computing neuron based on tilt anisotropy. Technical Background

[0002] Neuromorphic computing is a next-generation computing technology system modeled after the brain. It retains the existing advantages of computers while incorporating the many advantages of the human brain in information processing, potentially breaking free from the constraints of the von Neumann architecture and achieving integrated storage and processing, ultra-low energy consumption, and massively parallel information processing. Its core is inspired by biological nervous systems, aiming to simulate brain-like behavior through artificial neural networks. In living organisms, neurons transmit signals via neurotransmitters. The neuronal membrane potential undergoes a leakage-integration process. When a neuron receives stimulation from other neurons, it accumulates a membrane potential of charged ions. As stimulation continues, the membrane potential reaches a certain threshold, at which point the neuron releases all the ions, returning to its initial state. This is the Leaky-Integrate-Fire (LIF) behavior of biological neurons. We can consider this process as a conversion between analog and digital signals. Before reaching the threshold, the stimulation intensity and neuronal voltage are analog signals, with the neuron's charge gradually increasing with the stimulation intensity. However, after reaching the threshold voltage, the signal changes abruptly, allowing no intermediate states; this is equivalent to a digital signal with only two states: 0 and 1.

[0003] To achieve neuromorphic computing, a device that functions similarly to a neuron is first needed. In recent years, neuromorphic devices have been continuously developing. The polymorphic storage behavior of phase-change memories, resistive random access memories, and magnetic memories has been used to mimic the analog signal behavior of a neuron before the stimulus reaches a threshold voltage. However, while current devices can simulate the analog signal behavior before reaching the threshold voltage, they cannot simultaneously achieve the digital behavior after reaching the threshold voltage. To return the device to its initial state, a much larger stimulus must be applied. How to enable neuromorphic devices to return to their initial state more quickly after reaching the threshold voltage, achieving the switching from analog to digital signals, without increasing energy consumption, remains a key research challenge. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention proposes a brain-like computing neuron based on tilted anisotropy. Based on the different domain wall movement velocities under positive and negative currents in a tilted anisotropic magnetic thin film material system, this invention designs a brain-like computing neuron that can be used to realize the conversion of analog signals to digital signals in brain-like devices before and after a threshold voltage.

[0005] The present invention provides a brain-like computing neuron based on tilted anisotropy, specifically by growing a CoPt multilayer film on a slanted substrate as a brain-like computing neuron.

[0006] Preferably, the CoPt multilayer film grown on the obliquely cut substrate is microfabricated into a nanowire structure and then used as a brain-like computing neuron.

[0007] Preferably, the obliquely cut substrate of the multilayer film is an alumina obliquely cut substrate with a substrate angle of 7 degrees.

[0008] Preferably, the CoPt thin film layer structure is as follows from top to bottom: Co, Pt, Co, Pt, Co, Pt, Pt. The thickness of Pt in the upper layer is 0.4-0.6 nm, the thickness of Co in the upper layer is 0.8-1 nm, the thickness of Pt in the middle layer is also 0.5-0.6 nm, the thickness of Co in the middle layer is 0.3-0.5 nm, the thickness of Pt in the bottom layer is 0.8-1 nm, and the thickness of Co in the bottom layer is 0.3-0.5 nm.

[0009] Preferably, a protective layer is grown on the CoPt multilayer film to prevent oxidation. The protective layer material is Ta and the thickness is 1.5 nm.

[0010] Preferably, the CoPt multilayer film is microfabricated into a nanowire structure, wherein the nanowire size is 2000nm x 100nm x 10nm.

[0011] Preferably, the conductive layer of the CoPt multilayer film is connected to a control current and a measurement voltage, so that its changing resistance can be obtained by continuously applying pulse current; the applied magnetic field is a uniform magnetic field with Hx = -25mt applied along the length of the nanowire.

[0012] Compared with the prior art, the beneficial effects of the present invention are:

[0013] 1. This invention proposes a brain-like computational neuron based on tilt anisotropy, which, compared to other artificial neuron architectures, can not only simulate signals before a threshold voltage, but also realize digital signals after the threshold voltage is reached.

[0014] 2. The artificial neuron device prepared by this invention exhibits good linearity in the leakage-integration process.

[0015] 3. Compared to Skyrmion-based artificial neurons, the artificial neuron device fabricated in this invention can be quickly reset after reaching the threshold voltage by applying only a single anti-phase current pulse with the same amplitude and pulse width as the driving current. For Skyrmion, which lacks a ratchet effect, the reset process requires a large current, and the reset speed is relatively slow with higher power consumption. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the easy axis tilt direction of a Co / Pt thin film;

[0017] Figure 2 This is a schematic diagram of the Co / Pt thin film structure;

[0018] Figure 3 This is a schematic diagram of the internal magnetic domain structure of the nanowire and the directions of the applied current and magnetic field.

[0019] Figure 4 This is a LIF behavior map based on tilted anisotropic artificial neurons. Detailed Implementation

[0020] The purpose of this invention is to propose a brain-like computational neuron based on tilt anisotropy.

[0021] The following is in conjunction with the appendix Figure 1 , 2 3. Further explanation of the present invention:

[0022] A brain-like computational neuron based on tilt anisotropy, specifically:

[0023] CoPt multilayer films grown on obliquely shaped substrates will be microfabricated into nanowire structures with dimensions of 2000 nm x 100 nm x 10 nm, which will then be used as brain-like computing neurons. The obliquely shaped substrate of the multilayer films is an alumina obliquely shaped substrate, such as... Figure 1 As shown, the easy axis tilt direction of the Co / Pt thin film makes an angle of θ = 7° with the Z-axis, which is the angle of the oblique cut substrate; Figure 2 As shown, the CoPt thin film layer structure, from top to bottom, is Co, Pt, Co, Pt, Co, Pt. The thickness of Pt in the upper layer is 0.4-0.6 nm, and the thickness of Co in the upper layer is 0.8-1 nm. The thickness of Pt in the middle layer is also 0.5-0.6 nm, and the thickness of Co in the middle layer is 0.3-0.5 nm. The thickness of Pt in the bottom layer is 0.8-1 nm, and the thickness of Co in the bottom layer is 0.3-0.5 nm. This structure achieves vertical anisotropy by creating a vertical compositional gradient, breaking the vertical symmetry, and tilting the easy axis of the film by obliquely cutting the substrate, thus achieving tilted anisotropy. A protective layer is grown on the CoPt multilayer film to prevent oxidation. The protective layer material is Ta, and the thickness is 1.5 nm. Figure 3 As shown, the nanowire structure is fabricated from Co / Pt thin film micromachining. The applied electric and magnetic fields are both along the nanowire direction. The interior of the nanowire is a magnetic domain structure, and the domain walls move along the nanowire direction.

[0024] The objective of this invention is achieved through the following technical solution:

[0025] 1) In a system with tilted anisotropy, a nanowire with dimensions of 2000nm x 100nm x 10nm (length x width x thickness) was constructed, and a domain wall was created on the nanowire with an initial position of 400nm. Seven negative driving current pulses and one positive reset current pulse were set, and the threshold distance was controlled to be around 900nm.

[0026] 2) Apply an input peak potential (negative pulse current sequence) in the x direction to drive the domain walls to move slowly along the nanowire.

[0027] 3) When the domain wall moves to the threshold distance, the domain wall is rapidly reversed to the initial position of the nanowire by the output peak potential (positive pulse current), thus successfully simulating the LIF behavior of artificial neurons.

[0028] To further illustrate the effectiveness of the invention, it was verified in micromagnetic simulation software.

[0029] like Figure 4 As shown, the tilt anisotropy is set to (0.15, 0, 1); the saturation magnetization Ms is 580 kA / m; the applied Ku is 0.5 MJ / m³; the applied magnetic field is a uniform magnetic field applied in the x-direction with Hx = -25 mt; the nanowire size is 2000 nm x 100 nm x 10 nm; the thin film size is 5 nm x 5 nm x 10 nm; the current is a current applied in the x-direction with an amplitude of 1.3e¹² A / m. 2 A uniform current sequence with a frequency of 0.4 GHz and a pulse duration of 2 ns was used. The initial position of the domain wall was 400 nm, and the threshold distance was 900 nm. By applying a negative current pulse sequence in the x-direction, the domain wall was driven to move along the nanowire. The movement distance of the domain wall represented the membrane potential of the biological neuron. When the domain wall moved to the threshold distance, a positive pulse current was applied to quickly drive the domain wall back to the initial position of the nanowire, thus successfully simulating the LIF behavior of the artificial neuron.

Claims

1. A brain-inspired computing neuron based on tilted anisotropy, characterized in that: The CoPt multilayer film grown on a bevelled substrate is made into a nanowire structure by microprocessing and used as a brain-like computing neuron; The bevelled substrate is an alumina bevelled substrate with a substrate angle of 7 degrees; The conductive layer of the CoPt multilayer film is connected to a current regulating and voltage measuring device, so that the resistance change of the CoPt multilayer film is obtained by continuously applying a pulse current; the applied magnetic field is a uniform magnetic field with Hx=-25mt applied along the length direction of the nanowire. The CoPt thin film layer structure from top to bottom is Co, Pt, Co, Pt, Co, Pt, the thickness of Pt in the upper layer is 0.4-0.6nm, the thickness of Co in the upper layer is 0.8-1nm, the thickness of Pt in the middle layer is also 0.5-0.6nm, the thickness of Co in the middle layer is 0.3-0.5nm; the thickness of Pt in the lowermost layer is 0.8-1nm, the thickness of Co in the lowermost layer is 0.3-0.5nm, and the vertical component gradient of this structure obtains a vertical anisotropy gradient.

2. The brain-inspired computing neuron based on tilted anisotropy according to claim 1, wherein, A protective layer is grown on the CoPt multilayer film to prevent oxidation, and the protective layer material is Ta with a thickness of 1.5nm.

3. The brain-inspired computing neuron based on tilted anisotropy according to claim 1, wherein, The CoPt multilayer film is made into a nanowire structure by microprocessing, and the nanowire size is 2000nm x 100nm x 10nm.

Images