Spin synaptic device and in-memory computing apparatus
By designing a substrate, bottom electrode layer, complex oxide system layer, magnetic storage layer, and top electrode layer in a spin synapse device, and utilizing spin current to change the two-dimensional electron gas density, the problem of low on/off ratio in traditional spintronic devices is solved, thereby improving the magnetoresistive ratio of high and low resistance states and enhancing readout reliability.
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- INST OF MICROELECTRONICS CHINESE ACAD OF SCI LTD
- Filing Date
- 2024-11-04
- Publication Date
- 2026-05-07
AI Technical Summary
Traditional spintronic devices have a small switching ratio, which leads to complex external circuit design and is not conducive to improving read reliability.
Design a spin synapse device comprising a substrate, a bottom electrode layer, a complex oxide system layer, a magnetic storage layer, and a top electrode layer. Utilize the magnetic storage layer to generate a spin current under the read/write voltage of the peripheral circuit, thereby changing the two-dimensional electron gas density and realizing data reading or writing.
This improves the high and low resistance magnetoresistance ratio of spintronic devices, enhancing the readout reliability of the devices.
Smart Images

Figure CN2024129617_07052026_PF_FP_ABST
Abstract
Description
Spin synaptic devices and in-memory computing devices Technical Field
[0001] This application relates to the field of synapse technology, and more particularly to a spin synapse device and an in-memory computing device. Background Technology
[0002] The development of artificial intelligence has led to a rapid increase in the demand for computing power, while the development of traditional von Neumann architecture has struggled to keep pace with this rapid growth. Spintronic devices have become an emerging research direction in recent years due to their high speed, high energy efficiency, and long-term retention and durability. However, spintronic devices are limited by their relatively small switching ratio, resulting in complex external circuit designs and hindering improvements in read reliability.
[0003] Summary of the Invention
[0004] This application provides a spin synaptic device and an in-memory computing device to improve the high and low resistance magnetoresistance ratio of the spintronic device and enhance the reliability of device readout.
[0005] In a first aspect, embodiments of this application provide a spin synapse device, comprising:
[0006] A substrate; and a bottom electrode layer, a complex oxide system layer, a magnetic storage layer and a top electrode layer formed from bottom to top on the substrate, wherein the magnetic storage layer is connected to a conductive structure layer and the magnetic storage layer is connected to an external circuit through the top electrode layer and the conductive structure layer.
[0007] In this process, there is a two-dimensional electron gas region at the interface of the complex oxide system layer. When the magnetic storage layer receives the read / write voltage from the peripheral circuit, the magnetic storage layer generates a spin current. The spin current is injected into the complex oxide system layer to change the two-dimensional electron gas density of the two-dimensional electron gas region, thereby enabling the magnetic storage layer to complete the reading or writing of data.
[0008] In one possible implementation, the magnetic storage layer comprises a heavy metal layer and a magnetic layer formed from bottom to top on the complex oxide system layer, the heavy metal layer extending in opposite first and second directions beyond the boundaries of the magnetic layer in the first and second directions.
[0009] In one possible implementation, the conductive structure layer includes a first conductive layer and a second conductive layer, wherein the first conductive layer is connected to the portion of the heavy metal layer that extends beyond the complex oxide system layer in a first direction, and the second conductive layer is connected to the portion of the heavy metal layer that extends beyond the complex oxide system layer in the first direction.
[0010] In one possible implementation, the magnetic storage layer is used to generate spin currents with different polarization directions to enable data writing; the generated spin currents are then injected into the magnetic layer, which in turn injects them into a complex oxide system layer, altering the two-dimensional electron gas density to enable data reading.
[0011] In one possible implementation, during the read or write process of the spin synapse device, a write voltage is applied to the bottom electrode, a top auxiliary voltage is applied to the top electrode layer, the first conductive layer and the second conductive layer are grounded, and the charge flow is injected into the heavy metal layer, which is converted into a spin current through the spin-orbit coupling effect. The spin current is injected into the complex oxide system while entering the magnetic layer, changing the two-dimensional electron gas density.
[0012] In one possible implementation, the magnetic layer is a magnetic tunnel junction, wherein during the writing process of the magnetic storage layer, the top auxiliary voltage achieves the external field-free spin orbital flipping of the ferromagnetic material in the magnetic layer through spin-transfer torque, and injects spin current into the two-dimensional electron gas region, thereby changing the conductivity between the bottom electrode and the first conductive layer.
[0013] In one possible implementation, during the reading of the magnetic storage layer, the bottom electrode is grounded, a top auxiliary voltage is applied to the top electrode layer, a spin current is injected into the two-dimensional electron gas region to change the two-dimensional electron gas density, and the resistance between the bottom electrode and the first conductive layer of the bottom electrode is read, the resistance being used to indicate the read information.
[0014] In one possible implementation, the complex oxide system layer comprises one or more complex oxides.
[0015] Secondly, embodiments of this application provide an in-memory computing device, wherein the aforementioned spin synaptic device serves as a synaptic device in binary or multi-valued in-memory computing.
[0016] In one possible implementation,
[0017] During the in-memory computing weight modulation process, the spin orbital moment current of the magnetic storage layer drives the magnetic domains to move, and the proportion of the spin direction of the magnetic storage layer changes, thus linearly setting different weights.
[0018] During the in-memory computing process, the proportion of different spin directions is between the maximum and minimum charge, exhibiting different conductance values. When a reading voltage is applied, in-memory multiplication and addition operations are realized.
[0019] The spin synapse device and in-memory computing device provided in this application embodiment utilize a substrate; and, formed from bottom to top on the substrate, a bottom electrode layer, a complex oxide system layer, a magnetic storage layer, and a top electrode layer. The magnetic storage layer is connected to a conductive structure layer and is connected to peripheral circuits through the top electrode layer and the conductive structure layer. A two-dimensional electron gas region exists at the interface of the complex oxide system layer. When the magnetic storage layer receives read / write voltage from the peripheral circuit, it generates a spin current. This spin current is injected into the complex oxide system layer, changing the two-dimensional electron gas density in the two-dimensional electron gas region, enabling the magnetic storage layer to complete data reading or writing. Therefore, in this application, the spin synapse device generates a two-dimensional electron gas at the interface through the complex oxide layer, changing the two-dimensional electron gas density under spin-charge conversion. During information writing, different spin polarization directions lead to different changes in the two-dimensional electron gas density. During the reading phase, the two-dimensional electron gas density changes when reading spin polarization, enhancing the high-low resistance ratio of the device. The change in the density of the two-dimensional electron gas generated by the direction of regional magnetization leads to the change in the high and low resistance states of the device. This can be used as a synaptic device in binary or multi-valued in-memory computing to improve the magnetoresistance ratio of the high and low resistance states of spintronic devices and improve the reliability of device readout. Attached Figure Description
[0020] The accompanying drawings, which are incorporated herein by reference and constitute a part of the production cost specification, illustrate embodiments consistent with this application and, together with the specification, serve to explain the principles of this application.
[0021] Figure 1 is a schematic diagram of the structure of the spin synapse device provided in this application;
[0022] Figure 2 illustrates the read / write mechanism of the spin synapse device provided in this application;
[0023] Figure 3 is a flowchart of the fabrication process of the spin synapse device provided in this application;
[0024] Figure 4 shows the in-memory computation operation of the spin synapse device provided in this application.
[0025] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation
[0026] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.
[0027] Figure 1 is a schematic diagram of the spin synapse device provided in this application. As shown in Figure 1, the device includes:
[0028] Substrate; and, on the substrate, a bottom electrode layer, a complex oxide system layer, a magnetic storage layer and a top electrode layer are formed from bottom to top, wherein the magnetic storage layer is connected to a conductive structure layer and the magnetic storage layer is connected to an external circuit through the top electrode layer and the conductive structure layer.
[0029] In this complex oxide system layer, there is a two-dimensional electron gas region at the interface. When the magnetic storage layer receives the read / write voltage from the external circuit, the magnetic storage layer is used to generate a spin current. The spin current is injected into the complex oxide system layer to change the two-dimensional electron gas density in the two-dimensional electron gas region, enabling the magnetic storage layer to complete the reading or writing of data.
[0030] In this embodiment, a two-dimensional electron gas (2DEG) is an electronic system formed in a semiconductor heterojunction, in which electrons are confined to a two-dimensional space. This electron gas typically forms at the interface of the heterojunction, where, due to band discontinuities and quantum confinement, electrons are restricted to a two-dimensional region perpendicular to the interface. Within this region, electrons can move freely along the interface plane, exhibiting quantized electronic properties. The two-dimensional electron gas density (electron concentration of the two-dimensional electron gas) refers to the number of electrons per unit area in the two-dimensional electron gas system. A two-dimensional electron gas is a quantum state of electrons, in which the movement of electrons is restricted in one direction, while they can move freely in the other two directions.
[0031] In this embodiment, the spin synaptic device generates a two-dimensional electron gas at the interface through a complex oxide system. The upper heavy metal / ferromagnetic layer is injected with a spin flow, altering the density of the two-dimensional electron gas under spin-charge conversion. During information writing, different spin polarization directions lead to different changes in the two-dimensional electron gas density. During the read-out phase, the change in two-dimensional electron gas density during spin polarization enhances the high-low resistivity ratio of the device. This invention increases the magnetoresistivity ratio by introducing a two-dimensional electron gas and the spin-charge conversion effect, thereby improving the high-low resistivity ratio of the spintronic device and enhancing the reliability of the device readout.
[0032] In one example, the magnetic storage layer includes a heavy metal layer and a magnetic layer formed from bottom to top on a complex oxide system layer, the heavy metal layer extending in opposite first and second directions beyond the boundaries of the magnetic layer in the first and second directions.
[0033] Understandably, the combination of a heavy metal layer and a magnetic layer can form an information storage medium with unique physical properties. In this structure, the heavy metal layer typically provides the spin-orbit moment effect, while the magnetic layer is responsible for storing information. This combination can be used to realize novel storage technologies such as magnetic random access memory.
[0034] The spin-orbit moment effect is a novel magnetic effect that utilizes electric current in non-magnetic metals to modulate domain flipping, domain wall movement, and spin dynamics in ferromagnetic layers. In heavy metal / ferromagnetic metal heterostructures, current-driven magnetic moment manipulation can be achieved through interfacial spin-orbit coupling. Magnetic layers typically exhibit perpendicular magnetic anisotropy, meaning their magnetic moments tend to align along a specific direction. This characteristic contributes to improving the stability and interference immunity of storage media.
[0035] In one example, the conductive structure layer includes a first conductive layer and a second conductive layer. The first conductive layer is connected to the portion of the heavy metal layer that extends beyond the complex oxide system layer in a first direction, and the second conductive layer is connected to the portion of the heavy metal layer that extends beyond the complex oxide system layer in the first direction.
[0036] As shown in Figure 1, the overall structure of the device, from bottom to top, includes a bottom electrode, a first complex oxide layer, a second complex oxide layer, a two-dimensional electron gas region formed at the interface, a heavy metal layer, a magnetic layer, and a top electrode. The top and bottom electrodes are conductive materials connected to the external circuit, such as one or more of Ti / Au, Ni / Au, TiN, SrRuO3, W, Ru, Ta, and Pt. The heavy metal layer extends laterally, serving as the bottom electrode and the second conductive layer connected to the external circuit.
[0037] In one example, the complex oxide system consists of one or more complex oxides, such as SrTiO3, LaAlO3 / SrTiO3, BiFeO3 / SrTiO3, or a metal oxide / SrTiO3 system, forming a two-dimensional electron gas at the interface. The complex oxide layer mainly forms the two-dimensional electron gas at the interface through interfacial effects. The main materials include SrTiO3, LaMnO3 / SrTiO3, BiFeO3 / SrTiO3, LaAlO3 / SrTiO3, or other metal oxide / SrTiO3 systems, where the metal oxides include one or more of AlOx, RuO, ZnO, CuO, WOx, and TaOx. The magnetic layer, as a functional layer for storing information, can be a single-layer magnetic material, such as one or more of Co, CoFe, and CoFeB, or a heterostructure for storing spin information, such as one or more of Ta / CoFeB / MgO, W / CoFeB / MgO, Pt / CoFeB / MgO, Co / Pt, and IrMn / CoFeB / MgO multilayer structures, or a magnetic tunnel junction based on the basic structure of heavy metal layer / free layer / MgO / reference layer such as Ta / CoFeB / MgO / CoFeB, W / CoFeB / MgO / CoFeB, Pt / CoFeB / MgO / CoFeB, and IrMn / CoFeB / MgO / CoFeB.
[0038] In one example, the magnetic storage layer is used to generate spin currents with different polarization directions to enable data writing; the generated spin currents from the magnetic storage layer enter the magnetic layer, are injected into the complex oxide system layer, and change the two-dimensional electron gas density to enable data reading.
[0039] A heavy metal layer and a magnetic layer combine to form an information storage medium, adjacent to a complex oxide layer. Spin magnetization in the magnetic layer represents binary information. Charge is injected into the heavy metal layer, transforming into a spin current through spin-orbit coupling. Due to the spin current gradient, this current is injected into the interface of the adjacent magnetic layer and the complex oxide layer. Different spin directions generate different densities of two-dimensional electron gas, resulting in a resistive response ΔR. Consequently, the magnitude of the response voltage depends on the polarization direction of the spin injection and is non-volatile. During readout operations, the spin-charge conversion effect of the spin current in the ferromagnetic and complex oxide layers is utilized to improve the magnetoresistance ratio between high and low resistivity states.
[0040] In one example, during the read or write process of the spin synaptic device, a write voltage is applied to the bottom electrode, a top auxiliary voltage is applied to the top electrode, the first and second conductive layers are grounded, and the charge flow is injected into the heavy metal layer, which is converted into a spin flow through the spin-orbit coupling effect. The spin flow enters the magnetic layer and is also injected into the complex oxide system, changing the two-dimensional electron gas density.
[0041] During information writing, different spin polarization directions lead to different changes in the two-dimensional electron gas density. During the reading phase, the change in two-dimensional electron gas density during spin polarization reading enhances the high-low resistance ratio of the device. The change in two-dimensional electron gas density caused by the regional magnetization direction leads to changes in the high-low resistance state of the device, which can be used as a synaptic device in binary or multi-valued in-memory computing.
[0042] In one example, the magnetic layer is a magnetic tunnel junction, where, during the writing process of the magnetic storage layer, the top auxiliary voltage achieves the external field-free spin orbital flipping of the ferromagnetic material in the magnetic layer through spin-transfer torque, and injects spin current into the two-dimensional electron gas region, changing the conductivity between the bottom electrode and the first conductive layer.
[0043] The device read / write process is shown in Figure 2. During the write (0) process, current is injected into the heavy metal layer from the bottom electrode 1. Due to spin-orbit coupling, the heavy metal layer generates a longitudinal spin current. At this time, the magnetic layer can be magnetized and driven to a unified spin direction, such as spin-up, by the top auxiliary voltage of the top electrode. At this time, the device represents the storage state "0". During the write (1) process, the current polarity is changed, and the heavy metal layer generates a spin current with the opposite spin direction. Under the action of the top auxiliary voltage of the top electrode, it is driven to the opposite spin direction. At this time, the device represents the storage state "1".
[0044] In one example, during the magnetic storage layer readout process, the bottom electrode is grounded, a top auxiliary voltage is applied to the top electrode, a spin current is injected into the two-dimensional electron gas region to change the two-dimensional electron gas density, and the resistance between the bottom electrode layer and the first conductive layer is read. The resistance is used to indicate the readout information.
[0045] The device read / write process is shown in Figure 2. During the read process, a read voltage is applied to the top electrode. The spin current tunnels through the magnetic layer into the two-dimensional electron gas region. Under the action of spin-charge conversion, spin polarization is converted into electric charge, forming an electric charge flow under the action of a fixed electric field. This causes a change in the density of the two-dimensional electron gas, resulting in a difference in resistance. Therefore, during the read process, the magnetization information "0" and magnetization information "1" cause the resistance of the two-dimensional electron gas region to decrease or increase. This achieves the goal of using the resistance value of the two-dimensional electron gas region to represent ferromagnetic information, expanding the read margin of the device and improving device reliability while ensuring write speed and energy efficiency.
[0046] In one example, the fabrication process of the spin synapse device is shown in Figure 3:
[0047] First, a bottom electrode 3 and a complex oxide layer are grown on a silicon, silicon oxide, or SrTiO3 substrate using laser pulse deposition to form a basic two-dimensional electron gas structure. Then, a junction region is etched out using ion beam etching technology, and the conductive layer and the upper two-dimensional electron gas are isolated by SiO2.
[0048] Furthermore, a heavy metal layer, a magnetic layer, and a top electrode are grown on the material using magnetron sputtering. The heavy metal layer and the top electrode patterned areas are then patterned and etched using photolithography and ion beam etching techniques to fabricate the complete device.
[0049] Figure 4 shows the in-memory computation operation of the spin synaptic device provided in this application, where the spin synaptic device is used as a synaptic device in binary or multi-valued in-memory computation.
[0050] The spin synapse device can be found in the above method embodiments, and its implementation principle and technical effect are similar, so it will not be described again here.
[0051] In-Memory Computing (IMC) is a computing paradigm that tightly integrates data processing and storage, aiming to solve the "memory wall" problem in traditional von Neumann architectures. In traditional architectures, the processor and memory are separate, and data needs to be transferred back and forth between them, leading to significant performance bottlenecks. In-Memory Computing reduces the number of data moves by performing computations directly in the storage device, thereby reducing energy consumption and latency, and improving system efficiency and responsiveness. In-Memory Computing technologies typically utilize specific storage devices that can not only store data but also perform simple computational operations such as addition, multiplication, or comparison.
[0052] In one example, the in-memory computation method of a spin synaptic device involves the local switching of the magnetization direction of the spin orbital moment current region. The change in the two-dimensional electron gas density generated by spin-charge conversion leads to changes in the high and low resistance states of the device, enabling it to function as a synaptic device in binary or multi-valued in-memory computation. During in-memory computation weight modulation, the spin orbital moment current drives magnetic domain movement, altering the proportion of spin direction in the magnetic layer, and linearly setting different weights. During the readout process, different spin direction proportions lie between the maximum and minimum charge, exhibiting different conductance values. When a readout voltage is applied, in-memory multiplication and addition operations are achieved. In terms of in-memory computation in devices, current artificial intelligence training relies heavily on neural network models.
[0053] The neural network model shown in Figure 4a requires numerous matrix multiplication and addition operations of Y = Σ·w, where Y is the neuron output value, X is the input value, and w is the synaptic weight. In-memory computation utilizes the device's own conductance and Kirchhoff calculations. It greatly simplifies the time and complexity of multiplication and addition calculations.
[0054] In this embodiment, the introduction of a two-dimensional electron gas further enhances the reliability of the device during multi-value in-memory computation. During weight modulation in the in-memory computation training process, the spin orbital moment current drives the magnetic domains to move, changing the proportion of spin orientation in the magnetic layer. As shown in the spin orientation in the magnetic layer of Figures 4b-e, the proportion of spin orientation represents different multi-valued information, and different weights can be linearly written through the movement of magnetic domains. During in-memory computation, the proportion of spin also affects the electron density of the two-dimensional electron gas, causing a change in the device conductivity, which has a linear mapping relationship with the weight values. The consistent spin polarization in Figures 4b and 4e leads to a maximum increase or decrease in the two-dimensional electron gas, generating minimum and maximum read resistance values, representing the binarized "0" and "1" signals, respectively. These characteristics can be used as synapses in a binary neural network for binary information storage, weight modulation, and in-memory computation based on Kirchhoff's laws.
[0055] Furthermore, different spin directions, with proportions between the maximum and minimum charge, exhibit different conductance values, enabling multi-valued in-memory multiplication and addition operations when a read voltage is applied. As shown in Figures 4c-d, which illustrate the multi-valued implementation, the spin direction of the magnetization layer and the charge density of the two-dimensional electron region are illustrated. In scenarios requiring multi-valued weighting, the magnetization direction of the magnetic layer can be locally switched using spin orbital moment currents, generating spin-up and spin-down domains with different proportions within the magnetization layer. During the read process, the injection of spin currents with different polarization directions will cause the two-dimensional electron gas density to exhibit an intermediate state of resistance between its maximum (Figure 4b) and minimum (Figure 4e) values, as shown in Figures 4c and 4d. In these four types of spin electron distributions, the device conductance can be represented as σ1, σ2, σ3, and σ4, respectively, thereby achieving weighted adjustment of binary or multi-valued synapses. It is worth noting that the number of multivalued resistive states is not limited to the four types shown above, but can be achieved through the design of multivalued regions, where N≥4.
[0056] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.
[0057] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.
[0058] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.
[0059] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.
[0060] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described method.
[0061] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.
[0062] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.
[0063] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.
[0064] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.
[0065] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.
[0066] If the functionality is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0067] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.
[0068] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.
Claims
1. A spin synaptic device, characterized in that, include: Substrate; And, a bottom electrode layer, a complex oxide system layer, a magnetic storage layer and a top electrode layer are formed sequentially from bottom to top on the substrate, wherein the magnetic storage layer is connected to a conductive structure layer, and the magnetic storage layer is connected to an external circuit through the top electrode layer and the conductive structure layer; In this process, there is a two-dimensional electron gas region at the interface of the complex oxide system layer. When the magnetic storage layer receives the read / write voltage from the peripheral circuit, the magnetic storage layer generates a spin current. The spin current is injected into the complex oxide system layer to change the two-dimensional electron gas density of the two-dimensional electron gas region, thereby enabling the magnetic storage layer to complete the reading or writing of data.
2. The spin synapse device as described in claim 1, characterized in that, The magnetic storage layer comprises a heavy metal layer and a magnetic layer formed sequentially from bottom to top on the complex oxide system layer. The heavy metal layer extends in opposite first and second directions, beyond the boundaries of the magnetic layer in the first and second directions.
3. The spin synapse device as described in claim 2, characterized in that, The conductive structure layer includes a first conductive layer and a second conductive layer. The first conductive layer is connected to the portion of the heavy metal layer that extends beyond the complex oxide system layer in a first direction, and the second conductive layer is connected to the portion of the heavy metal layer that extends beyond the complex oxide system layer in the first direction.
4. The spin synapse device as described in claim 3, characterized in that, The magnetic storage layer is used to generate spin currents with different polarization directions to enable data writing; the spin currents generated by the magnetic storage layer enter the magnetic layer and are injected into the complex oxide system layer to change the two-dimensional electron gas density so as to enable data reading.
5. The spin synapse device as described in claim 4, characterized in that, During the read or write process of the spin synapse device, a write voltage is applied to the bottom electrode layer, a top auxiliary voltage is applied to the top electrode layer, the first conductive layer and the second conductive layer are grounded, and the charge flow is injected into the heavy metal layer. Through the spin-orbit coupling effect, it is converted into the spin current. The spin current enters the magnetic layer and is also injected into the complex oxide system layer, changing the two-dimensional electron gas density.
6. The spin synaptic device as described in claim 5, characterized in that, The magnetic layer is a magnetic tunnel junction. During the writing process of the magnetic storage layer, the top auxiliary voltage realizes the external field-free spin orbital flip of the ferromagnetic material in the magnetic layer through the spin-transfer torque of the spin current, and injects the spin current into the two-dimensional electron gas region, thereby changing the conductivity between the bottom electrode layer and the first conductive layer.
7. The spin synapse device as described in claim 5, characterized in that, During the reading process of the magnetic storage layer, the bottom electrode layer is grounded, the top auxiliary voltage is applied, the spin current is injected into the two-dimensional electron gas region, the density of the two-dimensional electron gas is changed, and the resistance between the bottom electrode layer and the first conductive layer is read. The resistance is used to indicate the read information.
8. The spin synaptic device as described in any one of claims 1 to 7, characterized in that, The complex oxidation system layer includes one or more complex oxides.
9. An in-memory computing device, characterized in that, The spin synaptic device described in any one of claims 1-8 is used for binary or multi-valued in-memory computation.
10. The in-memory computing device according to claim 9, characterized in that, During the in-memory computing weight modulation process, the spin orbital moment current of the magnetic storage layer drives the magnetic domains to move, and the proportion of the spin direction of the magnetic storage layer changes, thus linearly setting different weights. During the in-memory computing process, the proportion of different spin directions is between the maximum and minimum charge, exhibiting different conductance values. When a reading voltage is applied, in-memory multiplication and addition operations are realized.
Citation Information
Patent Citations
Spin electronic device and in-memory logic calculation device
CN110459254A
Full-voltage regulation and control logic device based on spin-orbit coupling effect
CN112466359A
Magnetoresistive memory unit, preparation method, array circuit and binary neural network chip
CN116403623A