Silicon-based integrated oxide heterostructure and spin-orbit torque device
By using a water-soluble sacrificial layer on a silicon-based substrate to prepare a self-supporting oxide single crystal thin film and grow an oxide spin source layer and a magnetic layer, the lattice mismatch problem of heterogeneous integration of transition metal oxide and silicon-based substrate is solved, and high-quality heterogeneous integration and the preparation of high-performance spin orbit torque devices are achieved.
Patent Information
- Application Number
- CN202411662729.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-20
- Publication Date
- 2025-04-18
AI Technical Summary
The heterogeneous integration of transition metal oxides and silicon-based substrates faces lattice mismatch, differences in chemical properties and thermal expansion coefficient, resulting in difficulties in growth of high-quality single crystal thin films and hindered spin transport.
A self-supporting oxide single crystal thin film was prepared using a water-soluble sacrificial layer and transferred to a silicon substrate as a buffer layer. By epitaxially growing the oxide spin source layer and the magnetic layer, a silicon-based integrated oxide heterostructure with perpendicular magnetic anisotropy was formed.
High-quality heterogeneous integration of transition metal oxide films on silicon-based substrates is achieved, which improves the preparation yield and stability of the device, enhances the current density withstands, and maintains material performance, which is suitable for the preparation of spin-orbit torque devices.
Smart Images

Figure CN119855476A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of semiconductor technology, and more specifically, to a silicon-based integrated oxide heterostructure and a spin-orbit torque device. Background Art
[0002] As an emerging technology, spin electronics provides new ideas for building high-performance, low-power information storage and processing devices by utilizing the spin freedom of electrons. Among them, devices based on the spin-orbit torque (SOT) effect have attracted much attention due to their unique advantages. SOT uses spin current to manipulate the magnetization state of magnetic materials, has the characteristics of sub-nanosecond fast response and read-write separation, and can achieve high-speed and stable data storage and processing. At the same time, SOT devices have the advantages of non-volatility, low power consumption, high durability and scalability, and are compatible with CMOS processes, and are expected to achieve large-scale integration. These characteristics make SOT devices show broad application prospects in the field of new memory. In addition to applications in the field of storage, SOT devices show unique advantages in neuromorphic computing. Unlike traditional computers, the information storage and processing of the human brain are highly parallel and low-power, and its energy efficiency is several orders of magnitude higher than that of traditional computers. With its ultra-fast dynamic controllability and non-volatility, SOT devices are expected to simulate biological synaptic functions and achieve efficient neuromorphic computing. Therefore, by building a brain-like computing system based on SOT devices, it is expected to overcome the bottleneck of the traditional von Neumann architecture, achieve efficient big data processing and complex cognitive tasks, and provide new possibilities for the innovative development of future information processing technology.
[0003] Transition metal oxides are the preferred materials for building efficient SOT devices due to their strong spin-orbit coupling (SOC) and unique electronic band structure, which can efficiently convert charge flow into spin flow. Integrating transition metal oxide-based SOT devices onto silicon-based substrates and combining them with CMOS technology is crucial for their practical application in SOT neuromorphic computing. However, the heterogeneous integration of transition metal oxides with silicon-based substrates faces many challenges. First, the SOT performance of transition metal oxides is heavily dependent on crystal quality and structure, and high-quality single-crystalline films are usually required to be grown on lattice-matched single-crystalline substrates. However, due to the significant lattice mismatch between oxides and silicon-based substrates, as well as differences in chemical properties and thermal expansion coefficients, it is extremely challenging to epitaxially grow high-quality transition metal oxide single-crystalline films directly on silicon-based substrates. In addition, defects between the oxide film and the silicon-based substrate can also hinder effective spin transport.
[0004] Therefore, there is an urgent need for a heterogeneous integrated structure of transition metal oxides and silicon-based substrates. Summary of the invention
[0005] In view of the above problems, an object of the present invention is to provide a silicon-based integrated oxide heterostructure and a spin-orbit torque device to solve at least one of the above technical problems.
[0006] According to the first aspect of the present invention, a silicon-based integrated oxide heterostructure is provided, which includes, from bottom to top, a silicon substrate, an oxide buffer layer, an oxide spin source layer, and a magnetic layer, wherein the magnetic layer has perpendicular magnetic anisotropy. The oxide buffer layer is tightly combined with the silicon substrate without a gap, which not only ensures high-quality epitaxial growth of the oxide film on the silicon substrate, but also improves the preparation yield and stability of the device and increases the withstand current density of the device.
[0007] Furthermore, a preferred structure is that the oxide buffer layer is a self-supporting oxide single crystal thin film prepared from a water-soluble sacrificial layer.
[0008] Furthermore, a preferred structure is that the oxide spin source layer is SrRuO3, SrIrO3 or IrO2.
[0009] Furthermore, a preferred structure is that the silicon substrate is amorphous SiO2 / Si.
[0010] Furthermore, a preferred structure is that the oxide buffer layer is one of SrTiO3, CaTiO3 and LaAlO3.
[0011] Further, the preferred structure is that the magnetic layer is a Co / Pt multilayer film, a Co / Pd multilayer film, a CoFeB / MgO multilayer film, a Co 1-x Pt x Multilayer film, Co 1-x Pd x A type of multilayer film.
[0012] According to a second aspect of the present invention, a spin-orbit torque device is provided, which adopts the above-mentioned silicon-based integrated oxide heterostructure. When current flows through the oxide spin source layer, due to the strong spin-orbit coupling therein, the charge flow is converted into a spin current. The spin current generates spin accumulation at the interface of the oxide spin source layer / magnetic layer, and after being transferred to the magnetic layer, a spin-orbit torque is generated through angular momentum exchange, which drives the magnetic torque to deviate from the equilibrium position and realizes the reversal of the magnetization state.
[0013] Furthermore, the preferred structure is that the critical reversal current density of the spin-orbit torque device is 1.3×10 10 A.m -2 The spin-orbit torque device has excellent charge-spin conversion capability. SH reachable The efficiency of the spin-orbit torque device prepared by the present invention is comparable to that of the device prepared by the functional oxide film epitaxially grown on a single crystal substrate; this means that while achieving silicon-based heterogeneous integration, the performance of the material and the device is not lost, fully demonstrating the technical advantages of the present invention.
[0014] According to a third aspect of the present invention, there is provided a magnetic random access memory using the above-mentioned spin-orbit torque device.
[0015] According to a fourth aspect of the present invention, there is provided a magnetic tunnel junction using the above-mentioned spin-orbit torque device.
[0016] According to a fifth aspect of the present invention, there is provided a neuromorphic device, which employs the above-mentioned spin-orbit torque device.
[0017] According to a sixth aspect of the present invention, a method for preparing a silicon-based integrated oxide heterostructure is provided, the method comprising: preparing a self-supporting single crystal film by a water-soluble sacrificial layer method, transferring the self-supporting single crystal film to a silicon substrate as a buffer layer; growing an oxide spin source layer and a magnetic layer on the surface of the buffer layer; the magnetic layer having perpendicular magnetic anisotropy.
[0018] Furthermore, in a preferred method, the growth temperature of the sacrificial layer, the self-supporting single crystal film and the oxide spin source layer is 600-750° C., and the oxygen pressure is 0.002-0.2 mbar.
[0019] Furthermore, a preferred method is that the annealing temperature after transferring the self-supporting single crystal film onto a silicon substrate as a buffer layer is 600-800° C. and the oxygen pressure is 0.01-0.2 mbar.
[0020] According to a seventh aspect of the present invention, a method for preparing a spin-orbit torque device is provided, the method comprising: preparing a silicon-based integrated oxide heterostructure using the above-mentioned method for preparing a silicon-based integrated oxide heterostructure, and then preparing the required spin-orbit torque device using micro-nano processing technology.
[0021] Further, the preferred method is that the method for preparing the required spin-orbit torque device using micro-nano processing technology includes: ultrasonically treating the silicon-based integrated oxide heterostructure for 30 seconds, removing surface impurities, and drying; preparing a Hall bar pattern on the silicon-based integrated oxide heterostructure using micro-nano processing technology; wherein the photoresist is AZ5214, and the exposure energy density is 1mW / cm 2; The obtained pattern is subjected to argon ion etching to etch away the oxide spin source layer and the magnetic layer; wherein the etching ion energy is 350eV, the beam current is 60mA, and the etching time is 270s; UV exposure overlay and development are performed to prepare an electrode pattern; Ti and Au metal layers are sequentially deposited and grown in the electrode pattern region by electron beam evaporation to form the electrode of the spin-orbit torque device; wherein the thickness of Ti is 10nm and the thickness of Au is 100nm. In the micro-nano processing process, the presence of the oxide buffer layer prevents the silicon substrate from being directly etched, effectively reduces the unstable factors generated during device measurement, and improves the accuracy and reliability of device performance evaluation.
[0022] Furthermore, a preferred method is to achieve continuous regulation of the magnetization and resistance state of the spin-orbit torque device by adjusting the magnitude of the applied current and the number of pulses.
[0023] According to the eighth aspect of the present invention, a brain-like computing method is provided, the method comprising: preparing a spin-orbit torque device using the above-mentioned method for preparing the spin-orbit torque device; using the spin-orbit torque device as a synapse or neuron in an artificial neural network; adjusting the current magnitude and the number of pulses applied to the spin-orbit torque device, and regulating the magnetization state of the spin-orbit torque device by adjusting the current magnitude and the number of pulses, thereby simulating some functions of biological synapses and / or neurons. By changing the amplitude and repetition number of the write current pulse, the multi-state magnetization state of the device can be further controlled to simulate the plasticity of the synapse and the integrated discharge behavior of the neuron. The artificial neural network constructed based on the SOT synapse and neuron can complete image recognition tasks and obtain an accuracy rate of up to 88%, showing broad application prospects in the fields of brain-like computing and artificial intelligence.
[0024] The silicon-based integrated oxide heterostructure and spin-orbit torque device of the present invention have the following beneficial effects:
[0025] 1) The present invention selects transition metal oxides as the basic research object, dissolves the sacrificial layer in water, prepares the self-supporting buffer layer single crystal film, transfers it to the silicon substrate as the buffer layer, grows various oxides on it by high-temperature epitaxial growth, and further deposits the magnetic layer; the present invention not only realizes the heterogeneous integration of transition metal oxide films on silicon-based substrates, but also expands the application platform of oxide materials, and also provides the possibility for large-scale preparation and integration based on functional oxide films. 2) In the silicon-based integrated oxide heterostructure obtained by the present invention, the self-supporting oxide single crystal film is tightly combined with the silicon substrate, and there is no gap at the interface. The high-quality heterogeneous interface in the silicon-based integrated oxide heterostructure of the present invention has good structural stability and impact resistance. 3) The silicon-based integrated oxide heterostructure obtained by the present invention uses transition metal oxide as a buffer layer to obtain a surface with orderly lattice and smooth surface, which is conducive to the subsequent high-quality epitaxial growth of functional single crystal films, which lays a material foundation for the realization of high-performance devices. 4) After obtaining the silicon-based integrated oxide heterostructure by the preparation method of the silicon-based integrated oxide heterostructure, the present invention further obtains a low critical reversal current density as low as 1.3×10 10 A.m -2 Spin-orbit torque devices, magnetic tunnel junctions, and magnetic random access memories; by building a brain-like computing system based on SOT devices, it is expected to overcome the bottleneck of the traditional von Neumann architecture, achieve efficient big data processing and complex cognitive tasks, and provide new possibilities for the innovative development of future information processing technology.
[0026] In order to achieve the above and related purposes, one or more aspects of the present invention include features that will be described in detail later and are particularly pointed out in the claims. The following description and the accompanying drawings describe some exemplary aspects of the present invention in detail. However, these aspects indicate only some of the various ways in which the principles of the present invention can be used. In addition, the present invention is intended to include all of these aspects and their equivalents. BRIEF DESCRIPTION OF THE DRAWINGS
[0027] By referring to the following description and claims in conjunction with the accompanying drawings, and with a more comprehensive understanding of the present invention, other objects and results of the present invention will become more apparent and easier to understand. In the accompanying drawings:
[0028] Figure 1 A schematic diagram of a process for preparing a silicon-based integrated oxide heterostructure according to an embodiment of the present invention;
[0029] Figure 2 is a schematic diagram of a preparation process of a silicon-based integrated oxide heterostructure according to an embodiment of the present invention;
[0030] Figure 3 Schematic diagram of a silicon-based integrated oxide heterostructure according to an embodiment of the present invention;
[0031] Figure 4 The results of atomic force microscopy measurements of the integrated film surface morphology show that the surface has a relatively low roughness (0.3nm);
[0032] Figure 5 The X-ray diffraction characterization results of silicon-based integrated oxide films;
[0033] Figure 6 This is a physical picture of silicon-based integrated oxide film;
[0034] Figure 7 These are the results of high-resolution transmission electron microscopy measurements of the side section of a silicon-based integrated thin film structure.
[0035] Figure 8 (a) is a schematic diagram of the structure and measurement of the oxide spin-orbit torque and neuromorphic device in an embodiment of the present invention, (b) is a measurement curve of the anomalous Hall resistance of the device, (c) is a first-order harmonic Hall signal of the spin-orbit torque device under an external magnetic field of 400Oe, (d) is a second-order harmonic Hall signal of the spin-orbit torque device under an external magnetic field of 400Oe, (e) is the relationship between the second-order harmonic Hall signal and the magnetic field within an external magnetic field of 6000Oe, and (f) is a curve showing the relationship between the effective field generated by the spin-orbit torque and the current density under different current densities;
[0036] Fig. 9 (a) is a schematic diagram of the measurement principle of the spin-orbit torque device in an embodiment of the present invention; (b) is a write and read timing diagram of the spin-orbit torque device; (c) is a magnetization reversal curve under different magnetic fields; (d) is a magnetization reversal curve under different current densities.
[0037] Fig.10 (a) is a schematic diagram of the working mechanism of the artificial neuron based on spin-orbit torque in an embodiment of the present invention; (b) is a curve of the anomalous Hall resistance of the device changing with the current density; (c) is the relationship between the anomalous Hall resistance state of the device and the number of pulse sequences, simulating the synaptic long-term potentiation plasticity and long-term inhibition plasticity functions based on spin-orbit torque; (d) is a schematic diagram of the working mechanism of the constructed artificial neural network; (e) is the confusion matrix when training the 30,000th batch; (f) is a comparison of the recognition accuracy of the artificial neural network constructed based on artificial synapses and the ideal numerical model. DETAILED DESCRIPTION
[0038] In the following description, for the purpose of illustration, in order to provide a comprehensive understanding of one or more embodiments, many specific details are set forth. However, it is apparent that these embodiments may also be implemented without these specific details. In other examples, for ease of describing one or more embodiments, known structures and devices are shown in the form of block diagrams.
[0039] Example 1
[0040] A silicon-based integrated oxide heterostructure is provided, which includes, from bottom to top, a silicon substrate, an oxide buffer layer, an oxide spin source layer, and a magnetic layer, wherein the magnetic layer has perpendicular magnetic anisotropy. The oxide buffer layer is a self-supporting oxide single crystal film prepared by a water-soluble sacrificial layer. The self-supporting oxide film of the present invention opens up a new way for the heterogeneous integration of oxide materials and the research and development of new devices. Through the water-soluble sacrificial layer method, the separation of the oxide film from the growth substrate is successfully achieved, freeing the constraints of the substrate. This self-supporting film not only retains the excellent magnetoelectric properties and multi-field coupling characteristics of the oxide material itself, but also has the advantages of easy integration and lightweight. More importantly, the self-supporting oxide film can be transferred to various substrates including silicon substrates, realizing the integration of high-quality epitaxial oxide films and nanostructures with traditional CMOS technology. This integration strategy is expected to produce new devices that meet future development needs, which is of great significance to promoting basic scientific research and the development of applied industries.
[0041] The sacrificial layer, buffer layer and spin source layer are oxides with similar crystal structures and similar lattice constants. The oxide spin source layer is an oxide with strong spin-orbit coupling, such as SrRuO3, SrIrO3 or IrO2. The silicon substrate is amorphous SiO2 / Si. The oxide buffer layer is an insulating dielectric material such as SrTiO3, CaTiO3, LaAlO3, etc. The magnetic layer is a Co / Pt multilayer film, a Co / Pd multilayer film, a Ti / CoFeB / MgO multilayer film, a Co 1-x Pt x Multilayer film, Co 1-x Pd x Multilayer films, etc.
[0042] Figure 1 and Figure 2 The process of the silicon-based integrated oxide heterostructure of the embodiment of the present invention is generally described; wherein, Figure 1 The present invention shows a general process of a method for preparing a silicon-based integrated oxide heterostructure. Figure 2 Schematic diagram of the preparation process of the silicon-based integrated oxide heterostructure described in this embodiment.
[0043] like Figure 1 and Figure 2 As shown, the method includes the following steps S110-S130.
[0044] Preparation Example 1
[0045] S110. Prepare a self-supporting single crystal film by a water-soluble sacrificial layer method.
[0046] In the specific implementation process, S110 includes steps S111 and S112. S111, a Sr3Al2O6 (SAO) sacrificial layer and a SrTiO3 (STO) oxide buffer layer are sequentially deposited on the SrTiO3 substrate layer by a pulsed laser deposition method. When preparing a self-supporting single crystal film, the sacrificial layer is etched by soaking in deionized water. The obtained self-supporting film has no wrinkles or cracks, and there is no gap between the film and the silicon substrate after transfer. Among them, the deposition conditions of the Sr3Al2O6 sacrificial layer are: temperature 750°C, oxygen pressure 0.002mbar; the deposition conditions of the SrTiO3 buffer layer are: temperature 750°C, oxygen pressure 0.15mbar. S112, AR-30080 and PMMA are sequentially spin-coated on the surface of the structure obtained in step S111 as supporting layers, and the sample is gently scraped around with a blade and purged with a nitrogen gun. Then the sample is fixed with a PDMS elastic film and placed in a container filled with deionized water for wet etching. After about 24 h, the Sr3Al2O6 sacrificial layer was completely dissolved to obtain a self-supporting SrTiO3 film.
[0047] S120, transferring the self-supporting single crystal film to a silicon substrate as a buffer layer. Specifically, the self-supporting SrTiO3 film obtained in step S112 is transferred to a silicon substrate, placed in a vacuum drying oven, and baked at 105°C for 1 hour. After being taken out, it is placed in an acetone solution for about 10 minutes to obtain a silicon substrate / SrTiO3 heterostructure.
[0048] S130, grow an oxide spin source layer and a magnetic layer on the surface of the buffer layer; the magnetic layer has perpendicular magnetic anisotropy. In the specific implementation process, S130 includes steps S131 and S132. S131, heat the silicon substrate / SrTiO3 heterostructure obtained in step S120 at 800°C and 0.15mbar oxygen pressure for 10 minutes, and then grow a 20nm thick SrRuO3 (SRO) film on its surface by pulsed laser deposition to obtain a silicon substrate / SrTiO3 / SrRuO3 structure; wherein the growth conditions are: temperature 700°C, oxygen pressure 0.15mbr. S132, deposit a CoPt multilayer film (magnetic film) with a total thickness of 4nm on the silicon substrate / SrTiO3 / SrRuO3 structure obtained in step S131 by magnetron sputtering, and its structure is Pt(1nm) / Co(0.5nm) / Pt(1nm) / Co(0.5nm) / Pt(1nm). The final heterostructure is Pt(1nm) / Co(0.5nm) / Pt(1nm) / Co(0.5nm) / Pt(1nm) / SrRuO3(20nm) / SrTiO3(25nm) / SiO2 / Si, e.g. Figure 7In the silicon-based integrated oxide heterostructure obtained by the present invention, the self-supporting oxide single crystal film is tightly combined with the silicon substrate, and there is no gap at the interface. The high-quality heterointerface in the silicon-based integrated oxide heterostructure of the present invention has good structural stability and impact resistance.
[0049] The sputtering conditions of Pt and Co are: argon flow rate of 15 SCCM, power supply power of 50 W, and argon flow rate of 20 SCCM, power supply power of 50 W. The obtained CoPt multilayer film should have perpendicular magnetic anisotropy.
[0050] For the silicon-based integrated oxide heterostructure obtained in Preparation Example 1, Figure 3 Schematic diagram of a silicon-based integrated oxide heterostructure according to an embodiment of the present invention; Figure 2 and Figure 3 As shown, Sr3Al2O6 (SAO) is used as a sacrificial layer, SrTiO3 (STO) is used as a buffer layer, and SrRuO3 (SRO) is used as a spin source layer. The self-supporting STO single crystal film is prepared by dissolving SAO in water, transferred to a silicon substrate as a buffer layer, and various oxides such as SRO are grown epitaxially on it at high temperature. The heterogeneous integration of transition metal oxide films on silicon-based substrates is achieved. Magnetic layers such as Co / Pt are further deposited.
[0051] In addition, the silicon-based integrated oxide heterostructure obtained in Preparation Example 1 was measured by atomic force microscopy, characterized by X-ray diffraction, and the side section of the silicon-based integrated oxide heterostructure was measured by high-resolution transmission electron microscopy; the test results are as follows: Figure 4 , Figure 5 and Figure 7 shown. Figure 6 The actual image of the silicon-based integrated oxide heterostructure obtained by the above method in this embodiment is shown. Specifically, Figure 4 The results of atomic force microscopy measurements of the integrated film surface morphology show that the surface has a low roughness (0.3nm); Figure 5 The X-ray diffraction characterization results of silicon-based integrated oxide films show that both SrTiO3 and SrRuO3 are single crystal epitaxial; Figure 6 The high-resolution transmission electron microscopy measurement results of the side section of the silicon-based integrated oxide heterostructure show that each interface is clear and there is no element diffusion, especially there is no gap at the SrTiO3 / SiO2 interface, indicating that the integrated oxide film is tightly bonded to the silicon substrate with strong bonding force, which helps to improve the thermal and mechanical stability of the device. Figure 5 and 7As shown, the silicon-based integrated oxide heterostructure obtained by the present invention uses transition metal oxide as a buffer layer to obtain a surface with an orderly lattice and a flat surface, which is conducive to the subsequent high-quality epitaxial growth of functional single crystal thin films, thus laying a material foundation for the realization of high-performance devices. This method expands the application platform of oxide materials and also provides the possibility for large-scale preparation and integration based on functional oxide thin films.
[0052] Preparation Example 2
[0053] S110. Prepare a self-supporting single crystal film by a water-soluble sacrificial layer method.
[0054] In the specific implementation process, S110 includes steps S111 and S112. S111, a Sr3Al2O6 (SAO) sacrificial layer and a CaTiO3 oxide buffer layer are sequentially deposited on the SrTiO3 substrate layer by a pulsed laser deposition method. When preparing a self-supporting single crystal film, the sacrificial layer is etched by soaking in deionized water. The obtained self-supporting film has no wrinkles or cracks, and there is no gap between the film and the silicon substrate after transfer. Among them, the deposition conditions of the Sr3Al2O6 sacrificial layer are: temperature 600°C, oxygen pressure 0.2mbar; the deposition conditions of the CaTiO3 buffer layer are: temperature 750°C, oxygen pressure 0.15mbar. S112, AR-30080 and PMMA are spin-coated as support layers on the surface of the structure obtained in step S111, and the sample is gently scraped around with a blade and purged with a nitrogen gun. Then the sample is fixed with a PDMS elastic film and placed in a container filled with deionized water for wet etching. After about 24 h, the Sr3Al2O6 sacrificial layer was completely dissolved to obtain a self-supporting CaTiO3 film.
[0055] S120, transferring the self-supporting single crystal film to a silicon substrate as a buffer layer. Specifically, the self-supporting CaTiO3 film obtained in step S112 is transferred to a silicon substrate, placed in a vacuum drying oven, and baked at 105°C for 1 hour. After being taken out, it is placed in an acetone solution for about 10 minutes to obtain a silicon substrate / CaTiO3 heterostructure.
[0056] S130, grow an oxide spin source layer and a magnetic layer on the surface of the buffer layer; the magnetic layer has perpendicular magnetic anisotropy. In the specific implementation process, S130 includes steps S131 and S132. S131, heat the silicon substrate / CaTiO3 heterostructure obtained in step S120 at 600°C and 0.01mbar oxygen pressure for 10 minutes, and then grow a 20nm thick SrIrO3 film on its surface by pulsed laser deposition to obtain a silicon substrate / CaTiO3 / SrIrO3 structure; wherein the growth conditions are: temperature 600°C, oxygen pressure 0.12mbar. S132, deposit a CoPd multilayer film (magnetic film) with a total thickness of 4nm on the silicon substrate / CaTiO3 / SrIrO3 structure obtained in step S131 by magnetron sputtering, and its structure is Pd(1nm) / Co(0.5nm) / Pd(1nm) / Co(0.5nm) / Pd(1nm). The final heterostructure obtained is Pd (1nm) / Co (0.5nm) / Pd (1nm) / Co (0.5nm) / Pd (1nm) / SrIrO3 (20nm) / CaTiO3 (25nm) / SiO2 / Si. In the silicon-based integrated oxide heterostructure obtained by the present invention, the self-supporting oxide single crystal film is tightly combined with the silicon substrate, and there is no gap at the interface. The high-quality heterointerface in the silicon-based integrated oxide heterostructure of the present invention has good structural stability and impact resistance.
[0057] The sputtering conditions of Pd and Co are respectively: argon flow rate of 20 SCCM, power supply power of 50 W, and argon flow rate of 20 SCCM, power supply power of 50 W. The obtained CoPd multilayer film should have perpendicular magnetic anisotropy.
[0058] Preparation Example 3
[0059] S110. Prepare a self-supporting single crystal film by a water-soluble sacrificial layer method.
[0060] In the specific implementation process, S110 includes steps S111 and S112. S111, a pulsed laser deposition method is used to sequentially deposit a Sr3Al2O6 (SAO) sacrificial layer and a LaAlO3 oxide buffer layer on the substrate layer of SrTiO3. When preparing a self-supporting single crystal film, deionized water is used to soak and etch the sacrificial layer. The resulting self-supporting film has no wrinkles or cracks, and there is no gap between the film and the silicon substrate after transfer. Among them, the deposition conditions of the Sr3Al2O6 sacrificial layer are: temperature 700°C, oxygen pressure 0.1mbar; the deposition conditions of the LaAlO3 buffer layer are: temperature 750°C, oxygen pressure 0.15mbar. S112, AR-30080 and PMMA are sequentially spin-coated as support layers on the surface of the structure obtained in step S111, and the sample is gently scraped around with a blade and purged with a nitrogen gun. Then the sample is fixed with a PDMS elastic film and placed in a container filled with deionized water for wet etching. After about 24 h, the Sr3Al2O6 sacrificial layer was completely dissolved to obtain a self-supporting LaAlO3 film.
[0061] S120, transferring the self-supporting single crystal film to a silicon substrate as a buffer layer. Specifically, the self-supporting LaAlO3 film obtained in step S112 is transferred to a silicon substrate, placed in a vacuum drying oven, and baked at 105°C for 1 hour. After being taken out, it is placed in an acetone solution for about 10 minutes to obtain a silicon substrate / LaAlO3 heterostructure.
[0062] S130, grow an oxide spin source layer and a magnetic layer on the surface of the buffer layer; the magnetic layer has perpendicular magnetic anisotropy. In the specific implementation process, S130 includes steps S131 and S132. S131, heat the silicon substrate / LaAlO3 heterostructure obtained in step S120 at 700°C and 0.2mbar oxygen pressure for 10 minutes, and then use pulsed laser deposition to grow a 20nm thick IrO2 film on its surface to obtain a silicon substrate / / IrO2 structure; wherein the growth conditions are: temperature 750°C, oxygen pressure 0.08mbr. S132, use a magnetron sputtering method to deposit a total thickness of 2nm Co on the silicon substrate / LaAlO3 / IrO2 structure obtained in step S131 50 Pt 50 Magnetic film (i.e. Co 1-x Pt x , x = 0.5). The final heterostructure is (2nm)Co 50 Pt 50 / IrO2(20nm) / LaAlO3(25nm) / SiO2 / Si. In the silicon-based integrated oxide heterostructure obtained by the present invention, the self-supporting oxide single crystal film is tightly combined with the silicon substrate, and there is no gap at the interface. The high-quality heterogeneous interface in the silicon-based integrated oxide heterostructure of the present invention has good structural stability and impact resistance.
[0063] Co 50 Pt 50 The sputtering conditions are: Co and Pt co-sputtering, argon gas flow rate of 15 SCCM, power supply power of 50W and 30W respectively. 50 Pt 50 The film should have perpendicular magnetic anisotropy.
[0064] The silicon-based integrated oxide heterostructures obtained in Preparation Examples 2 and 3 have the above-mentioned characteristics of Preparation Example 1. In the specific implementation process, the CoFeB / MgO multilayer film or Co 1-x Pd x (e.g. x=0.5) multilayer film as the magnetic layer of silicon-based integrated oxide heterostructure; and silicon-based integrated oxide heterostructure based on CoFeB / MgO multilayer film, and silicon-based integrated oxide heterostructure based on Co 1-x Pd x The multi-layered silicon-based integrated oxide heterostructure was subjected to similar tests as the silicon-based integrated oxide heterostructure of Preparation Example 1, and the test results obtained all met the standards of the silicon-based integrated oxide heterostructure of the present invention.
[0065] Example 2
[0066] According to a second aspect of the present invention, a spin-orbit torque device (STO) is provided, which adopts the above silicon-based integrated oxide heterostructure. The spin-orbit torque device realizes the characteristics of low power consumption magnetization switching and high reliability operation.
[0067] When current flows through the oxide spin source layer (SrRuO3 in this embodiment), the charge flow is converted into spin current due to the strong spin-orbit coupling therein. The spin current generates spin accumulation at the interface of the oxide spin source layer / magnetic layer (SrRuO3 / CoPt in this embodiment). After being transferred to the magnetic layer, the spin-orbit torque is generated through angular momentum exchange, driving the magnetic torque to deviate from the equilibrium position and realizing the magnetization state reversal. Furthermore, the preferred structure is that the critical reversal current density of the spin-orbit torque device is relatively low, which is 1.3×10 10 A.m -2 The spin-orbit torque device has excellent charge-spin conversion capability. SH reachable
[0068] A method for preparing a spin-orbit torque device is provided, the method comprising: S140, preparing a silicon-based integrated oxide heterostructure by using the above-mentioned method for preparing a silicon-based integrated oxide heterostructure, and then preparing the required spin-orbit torque device by using micro-nano processing technology.
[0069] Further, the preferred method is that the method for preparing the required spin-orbit torque device using micro-nano processing technology includes: S141, ultrasonically treating the silicon-based integrated oxide heterostructure for 30 seconds, removing surface impurities, and drying; S142, preparing a Hall bar pattern on the silicon-based integrated oxide heterostructure using micro-nano processing technology; wherein the photoresist is AZ5214, and the exposure energy density is 1mW / cm 2 .
[0070] A Hall bar pattern for magneto-electric transport testing is prepared on the surface of the silicon-based integrated oxide heterostructure obtained in step S130 using micro-nano processing technology, and 10 nm Ti and 100 nm Au films are sequentially deposited on the electrode pattern, wherein Ti serves as an adhesion layer for the Au electrode to enhance the stability of the electrode.
[0071] S143, argon ion etching is performed on the obtained pattern to etch away the oxide spin source layer and the magnetic layer; wherein the etching ion energy is 350eV, the beam current is 60mA, and the etching time is 270s; S144, ultraviolet exposure overlay and development are performed to prepare an electrode pattern; S145, Ti and Au metal layers are deposited and grown in the electrode pattern area in sequence by electron beam evaporation to form electrodes of the spin-orbit torque device; wherein the thickness of Ti is 10nm and the thickness of Au is 100nm. Furthermore, the spin-orbit torque device is continuously regulated by adjusting the magnitude of the applied current and the number of pulses.
[0072] The following tests were performed on the spin-orbit torque device obtained in this embodiment.
[0073] First, the device structure and measurement of the spin-orbit torque device are carried out; Figure 8 (a) is a schematic diagram of the structure and measurement of the oxide spin-orbit torque and neuromorphic device in an embodiment of the present invention. When current is injected into the spin source layer SrRuO3 along the longitudinal direction (x-axis), a torque is induced in the magnetic layer. A magnetic field is further applied along the longitudinal and transverse (y) directions to measure the damping-like torque and field-like torque, respectively.
[0074] Secondly, the anomalous Hall resistance curve of the spin-orbit torque device was measured; Figure 8 (b) is the anomalous Hall resistance measurement curve of the device, from which it can be seen that the magnetic layer CoPt has good perpendicular magnetism and the anomalous Hall resistance is 0.72Ω. The magnetization state of the CoPt magnetic layer can also be effectively controlled by changing the current density.
[0075] Furthermore, in order to characterize the spin-orbit torque performance of the device, an alternating current was passed through the spin-orbit torque device prepared in this embodiment, and the first and second harmonic Hall voltage signals were tested by a lock-in amplifier. Figure 8 (c) The external magnetic field is 400 Oe and the current density is 1.21×10 9 Am -2 The first-order harmonic Hall signal corresponding to the device is Figure 8 (d) is the corresponding second-order harmonic Hall signal; by fitting the measured data (shown by the black line), the key parameters reflecting the properties of the device’s spin-orbit torque can be obtained. Figure 8 (f) shows the relationship between the effective field generated by the spin-orbit torque and the current density at different current densities. By linear fitting, it can be obtained that the effective field generated by the unit current density is 24.69 Oe / (10 11 A×m -2 ), and finally the charge-spin conversion efficiency is 0.155, and the corresponding spin conductivity is By measuring the relationship between the first-order and second-order Hall voltage signals of the spin-orbit torque device and the current through the harmonic Hall voltage test method, the relationship between the effective field and the current density can be obtained, thereby characterizing the spin-orbit torque properties of the device.
[0076] Finally, the magnetization reversal curves of the spin-orbit torque device prepared in this embodiment are measured under different magnetic fields and different current densities. Fig. 9 (a) and 9(b) are the measurement principle and read-write timing diagram of the spin-orbit torque device in the embodiment of the present invention. In the measurement, a pulse current is first injected along the x-axis, and an external magnetic field is applied as an auxiliary magnetic field to flip the CoPt magnetic moment; then, a continuous small current is applied to read the Hall resistance of the SrRuO3 / CoPt heterostructure device. Fig. 9 As shown in (b), after obtaining the silicon-based integrated oxide heterostructure through the preparation method of the silicon-based integrated oxide heterostructure, the present invention further obtains a low critical switching current density as low as 1.3×10 10 A.m -2 Spin-orbit torque device.
[0077] Fig. 9 (c) shows the magnetic moment reversal curves under different magnetic fields. Fig. 9 (c) It can be found that: first, when the external magnetic field is zero, the magnetic moment cannot be reversed; when the magnetic field is not zero, the magnetic moment can be reversed, but the chirality of the reversal curve is related to the magnetic field polarity. Specifically, when it is +100Oe, the reversal is in the counterclockwise direction; when it is -100Oe, it is clockwise. This is typical evidence of the spin-orbit torque driving the magnetic moment reversal, indicating that the main driving force for the magnetic moment reversal of the device comes from the current-induced spin-orbit torque. It is also found that, Figure 8 As shown in (b), the magnetization state of the CoPt magnetic layer can be effectively controlled by changing the current density. 10A m -2 Increased to 1.94×10 10 A m -2 ), the flip ratio gradually increases, that is, the device can achieve stable multi-resistance states, which provides the possibility for multi-state memory devices based on the spin-orbit torque effect.
[0078] Example 3
[0079] The spin-orbit torque device obtained in Example 2 is further used to obtain a magnetic tunnel junction based on the spin-orbit torque device. Among them, the magnetic tunnel junction (MTJ for short) is composed of a fixed ferromagnetic layer, an insulating tunneling barrier and a free ferromagnetic layer. The magnetization direction of the fixed ferromagnetic layer is fixed and unchanged, and it is a reference layer. The magnetization direction of the free ferromagnetic layer can be parallel or anti-parallel to the reference layer. When the magnetization directions of the fixed layer and the free layer are consistent, it is called a "parallel state", and the tunnel magnetoresistance (TMR) of the MTJ is low; when the magnetization directions are inconsistent, it is called an "anti-parallel state", and the TMR is high. The spin-orbit torque device obtained in Example 2 can be used in the free ferromagnetic layer. In MRAM, the spin-orbit torque writing method is adopted. The current passes through the underlying spin source layer, generates a spin current and injects it into the free layer. The spin-orbit torque is used to make the magnetization direction of the free layer precess, and a variety of methods are combined to make the magnetization direction produce a deterministic flip. This approach has the advantages of symmetrical read and write capabilities, separate and optimizable read and write paths, fast sub-nanosecond operation speeds, and low write power consumption.
[0080] Example 4
[0081] A magnetic random access memory is prepared based on the spin-orbit torque device obtained in Example 2. The magnetic random access memory (English: Magnetic Random Access Memory, MRAM for short) includes multiple magnetic tunnel junctions based on the spin-orbit torque device. The magnetic random access memory obtained in this embodiment uses SOT to increase the switching speed of the magnetization direction of the ferromagnetic layer in the magnetic tunnel junction, reduce the write power consumption, and improve the reliability of the magnetic random access memory.
[0082] Furthermore, based on the magnetic random access memory prepared by the spin-orbit torque device of Example 2, a chip is further prepared using multiple magnetic random access memories, multiple leads and multiple transistors.
[0083] Example 5
[0084] The spin-orbit torque device obtained based on Example 2 can be used to prepare neuromorphic devices. Specifically, the spin-orbit torque device can be used as a synapse or neuron of a neuromorphic device. One of the cores of neuromorphic devices is the synaptic function, which is usually achieved by a device with a memristive property. The memristive device can simulate the plasticity of biological synapses, that is, adjust its conductance or resistance state according to the intensity and time characteristics of the input signal to achieve information storage and processing. For example, by changing the intensity and duration of the voltage or current, the increase (long-term potentiation, LTP) or decrease (long-term inhibition, LTD) of the synaptic weight can be simulated. Another core of the neuromorphic device is the neuron function, which can be achieved by simulating the integral-discharge characteristics of biological neurons. For example, a neuromorphic device based on two-dimensional materials can simulate the discharge behavior of neurons, wherein the conductance change of the device can simulate the membrane potential change of the neuron, and when a certain threshold is reached, the device will send a pulse signal, similar to the action potential of a biological neuron. In addition, the manufacture of neuromorphic devices based on spin-orbit torque devices can also be compatible with existing CMOS integrated circuit technology.
[0085] Specifically, the spin-orbit torque device obtained in Example 2 is used as a synapse in an artificial neural network structure. Artificial neurons are a computing model that imitates the functions of biological neurons and are often used to construct artificial neural networks. They can process and respond to input signals and adjust their response methods through learning algorithms. The basic structure of an artificial neuron includes: Input: receiving multiple input signals, which can be different features or data points. Weight: Each input signal has a weight associated with it, which determines the degree of influence of the input signal on the output of the neuron. Activation function: a nonlinear function used to convert the weighted sum of input signals into the output of the neuron. Common activation functions include Sigmoid, Tanh, ReLU, etc. Output: The final output of the neuron, usually the output result of the activation function. That is, artificial neurons can process complex input signals and present nonlinear changes by using nonlinear activation functions. In order to realize artificial neurons based on spin-orbit torque, R is required AHE It can show nonlinear changes with the input stimulus signal. RAHE (Random Averaging Heuristic Ensemble) is an ensemble learning method in machine learning, which is usually used to improve the prediction performance of the model.
[0086] That is, in this embodiment, the front neuron and the back neuron are connected through the SrRuO3 / CoPt based spin-orbit torque synapse. AHEDefined as the synaptic weight, its size can be gradually changed as the magnetization state of the CoPt layer (i.e., the magnetic layer) is flipped.
[0087] Fig.10 (a) is a schematic diagram of the working mechanism of an artificial neuron based on spin-orbit torque in an embodiment of the present invention, in which the front and rear neurons are connected by a synapse based on a spin-orbit torque device, and the abnormal Hall resistance change of the device indicates the weight update. Fig.10 (b) is the curve of the anomalous Hall resistance of the device changing with current density. It can be found that as the current density (pulse width is 200μs) gradually increases (from 1.1×10 10 A m -2 Increased to 1.7×10 10 Am -2 ), the abnormal Hall resistance change value is first kept at 0Ω, and when it is greater than 1.2×10 10 A m -2 When the nonlinear increase begins, it reaches 1.6×10 10 A m -2 When , it reaches the saturation value of 0.4. This nonlinear change can be fitted by the modified Sigmoidal function, as shown in the black curve, and then used as the activation function of the neuron. Fig.10 (c) shows the relationship between the abnormal Hall resistance of the device and the number of constant current pulses (pulse width is 200 μs). In the present invention, positive and negative current pulses of 1.6×10 10 Am -2 , gradually regulating the resistance value, simulating the synaptic long-term potentiation plasticity (LTP) and long-term depression plasticity (LTD) functions based on spin-orbit torque. Fig.10 It can be seen that when a positive current is applied, R AHE The resistance increases monotonically; at this time, when the negative current pulse continues to be applied, the resistance state decreases monotonically and gradually returns to the initial value, that is, R AHE =0Ω. The above changes correspond to the long-term potentiation and long-term inhibition of biological synapses. Fig.10 The characteristic data in (c) is mapped to the resistance state of the device using the back-propagation gradient descent algorithm, thereby updating the synaptic weights. First, a two-layer artificial neural network consisting of 784 input neurons and 10 output neurons is constructed, as shown in Fig.10 (d) In the process of handwritten digit recognition, the image is first divided into a 28 by 28 pixel matrix, which corresponds to 784 input neurons. The 10 output neurons correspond to the numbers 0 to 9. Based on the constructed artificial neural network, up to 30,000 batches of training were carried out, and the training results are shown in Fig.10(e) and 10(f). From the confusion matrix after training, we can see that the predicted value corresponds well to the expected value, indicating that the recognition ability of all 10 numbers is good. Fig.10 (f) Comparison of the recognition accuracy between the neural network constructed based on artificial synapses and the ideal numerical model is shown. It can be found that the recognition accuracy rate reaches 88%, which is close to the simulation results based on the ideal numerical model, proving the effectiveness of the device of the present invention in the application of artificial neural networks. In summary, the brain-like computing method of this embodiment can be applied to artificial intelligence tasks such as image recognition, speech recognition, and natural language processing, and has higher computing efficiency and lower power consumption than traditional CMOS devices.
[0088] The present invention selects transition metal oxides as the basic research object, dissolves the sacrificial layer by water, prepares a self-supporting buffer layer single crystal film, transfers it to a silicon substrate as a buffer layer, grows various oxides epitaxially at high temperature on it, and further deposits a magnetic layer; the present invention not only realizes the heterogeneous integration of transition metal oxide films on silicon-based substrates, but also expands the application platform of oxide materials, and also provides the possibility for large-scale preparation and integration based on functional oxide films. In the silicon-based integrated oxide heterostructure obtained by the present invention, the self-supporting oxide single crystal film is tightly combined with the silicon substrate, and there is no gap at the interface. The high-quality heterogeneous interface in the silicon-based integrated oxide heterostructure of the present invention has good structural stability and impact resistance. The silicon-based integrated oxide heterostructure obtained by the present invention uses transition metal oxide as a buffer layer to obtain a surface with an orderly lattice and a smooth surface, which is conducive to the subsequent high-quality epitaxial growth of functional single crystal films, which lays a material foundation for the realization of high-performance devices. After the silicon-based integrated oxide heterostructure is obtained by the preparation method of the silicon-based integrated oxide heterostructure, the present invention further obtains a low critical reversal current density as low as 1.3×10 10 A.m -2 Spin-orbit torque devices, magnetic tunnel junctions, and magnetic random access memories; by building a brain-like computing system based on SOT devices, it is expected to overcome the bottleneck of the traditional von Neumann architecture, achieve efficient big data processing and complex cognitive tasks, and provide new possibilities for the innovative development of future information processing technology.
[0089] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.
Claims
1. A silicon-based integrated oxide heterostructure, characterized in that: From bottom to top, they include: Silicon substrate, Oxide buffer layer, Oxide spin source layer, and a magnetic layer having perpendicular magnetic anisotropy.
2. The silicon-based integrated oxide heterostructure according to claim 1, characterized in that: The oxide buffer layer is a self-supporting oxide single crystal film prepared from a water-soluble sacrificial layer.
3. The silicon-based integrated oxide heterostructure according to claim 1, characterized in that: The oxide spin source layer is SrRuO3, SrIrO3 or IrO2.
4. The silicon-based integrated oxide heterostructure according to claim 1, characterized in that: The silicon substrate is amorphous SiO2 / Si.
5. The silicon-based integrated oxide heterostructure according to claim 1, characterized in that: The oxide buffer layer is one of SrTiO3, CaTiO3 and LaAlO3.
6. The silicon-based integrated oxide heterostructure according to claim 1, characterized in that: The magnetic layer is a Co / Pt multilayer film, a Co / Pd multilayer film, a CoFeB / MgO multilayer film, a Co 1-x Pt x Multilayer film, Co 1-x Pd x A type of multilayer film.
7. A spin-orbit torque device, characterized in that: The silicon-based integrated oxide heterostructure as claimed in claim 1 is adopted.
8. The spin-orbit torque device according to claim 7, characterized in that: The critical reversal current density of the magnetic moment of the spin-orbit torque device is 1.3×10 10 A.m -2 .
9. A magnetic random access memory, characterized in that: A spin-orbit torque device as claimed in claim 7 is used.
10. A magnetic tunnel junction, characterized in that: A spin-orbit torque device as claimed in claim 7 is used.
11. A neuromorphic device, characterized in that A spin-orbit torque device as claimed in claim 7 is used.
12. A method for preparing a silicon-based integrated oxide heterostructure, characterized in that: Methods include: Preparation of free-standing single-crystalline thin films by dissolving a water-soluble sacrificial layer; transferring the self-supporting single crystal film onto a silicon substrate as a buffer layer; An oxide spin source layer and a magnetic layer are grown on the surface of the buffer layer; the magnetic layer has perpendicular magnetic anisotropy.
13. The method for preparing a silicon-based integrated oxide heterostructure according to claim 12, characterized in that: The growth temperature of the sacrificial layer, the self-supporting single crystal film and the oxide spin source layer is 600-750° C., and the oxygen pressure is 0.002-0.2 mbar.
14. The method for preparing a silicon-based integrated oxide heterostructure according to claim 12, characterized in that: The annealing temperature after the self-supporting single crystal film is transferred to a silicon substrate as a buffer layer is 600-800° C. and the oxygen pressure is 0.01-0.2 mbar.
15. A method for preparing a spin-orbit torque device, characterized in that: Methods include: After the silicon-based integrated oxide heterostructure is prepared by the method for preparing the silicon-based integrated oxide heterostructure as claimed in claim 12, the required spin-orbit torque device is prepared by micro-nano processing technology.
16. The method for preparing a spin-orbit torque device according to claim 15, characterized in that: The method of preparing the required spin-orbit torque device using micro-nano processing technology includes: The silicon-based integrated oxide heterostructure is subjected to ultrasonic treatment for 30 seconds to remove surface impurities and then blow-dried; The Hall bar pattern is prepared on the silicon-based integrated oxide heterostructure by using micro-nano processing technology; wherein the photoresist is AZ5214 and the exposure energy density is 1mW / cm 2 ; Performing argon ion etching on the obtained pattern to etch away the oxide spin source layer and the magnetic layer; wherein the etching ion energy is 350 eV, the beam current is 60 mA, and the etching time is 270 s; Performing ultraviolet exposure overlay and development to prepare electrode patterns; Ti and Au metal layers are sequentially deposited and grown in the electrode pattern region by electron beam evaporation method to form electrodes of the spin-orbit torque device; wherein the thickness of Ti is 10 nm, and the thickness of Au is 100 nm.
17. The method for preparing a spin-orbit torque device according to claim 15, characterized in that: By changing the magnitude of the applied current and the number of pulses, the magnetization state and the resistance state of the spin-orbit torque device can be continuously regulated.
18. A brain-like computing method, characterized in that: Methods include: A spin-orbit torque device is prepared by using the method for preparing a spin-orbit torque device as claimed in claim 15; Using the spin-orbit torque device as a synapse or neuron in an artificial neural network; By adjusting the current size and the number of pulses applied to the spin-orbit torque device, the weight update and the regulation of the neuron switch state are achieved, thereby achieving the training and learning of the neural network.