Neuromorphic devices

Abstract

The neuromorphic device includes a first element group and a second element group, wherein the first element group and the second element group respectively include a plurality of magnetic domain wall movement elements, wherein the plurality of magnetic domain wall movement elements respectively include a magnetic domain wall movement layer, a ferromagnetic layer, and a non-magnetic layer sandwiched between the magnetic domain wall movement layer and the ferromagnetic layer. The length of the magnetic domain wall movement layer in the long side direction of each of the magnetic domain wall movement elements belonging to the first element group is shorter than the length of the magnetic domain wall movement layer in the long side direction of each of the magnetic domain wall movement elements belonging to the second element group. In terms of the resistance change rate when a specified pulse is input, each of the magnetic domain wall movement elements belonging to the first element group is larger than each of the magnetic domain wall movement elements belonging to the second element group.

Description

Technical Field

[0001] The present invention relates to a neuromorphic device.

[0002] This application claims priority based on PCT / JP2020 / 039957 filed on October 23, 2020, and cites its contents herein. Background Art

[0003] A magnetoresistive element is known that utilizes changes in resistance (magnetoresistance) caused by changes in the relative angle of magnetization between two ferromagnetic layers. In a magnetoresistive element, the current path used when writing data sometimes differs from the current path used when reading data. To control the currents in these different current paths, such a magnetoresistive element is connected to three switching elements. A magnetoresistive element controlled by three switching elements is called a three-terminal magnetoresistive element.

[0004] For example, the domain wall migration type magnetoresistive effect element described in Patent Document 1 is an example of a three-terminal magnetoresistive effect element.

[0005] In recent years, research on neuromorphic devices that artificially mimic the relationship between neurons and synapses in the human brain has been advancing.

[0006] For example, the magnetoresistance element described in Patent Document 2 is an example of a neuromorphic device that utilizes a domain wall migration type magnetoresistive effect element.

[0007] Prior art literature

[0008] Patent Literature

[0009] Patent Document 1: Patent No. 5441005

[0010] Patent Document 2: Patent No. 6617829 Summary of the Invention

[0011] Problems to be solved by the invention

[0012] Variable resistance elements, such as magnetoresistive elements, are often used in integrated form. To increase storage capacity, devices incorporating variable resistance elements need to be more highly integrated. As a method for increasing the integration of variable resistance elements, methods for reducing the size of variable resistance elements are being explored.

[0013] However, simply reducing the size of the resistance variable element reduces the recognition rate of the neuromorphic device.

[0014] The present invention has been made in view of the above-mentioned problems, and an object of the present invention is to provide a neuromorphic device with high integration of resistance variable elements and improved recognition rate.

[0015] Technical solutions to solve problems

[0016] (1) A first embodiment provides a neuromorphic device comprising a first element group and a second element group, wherein the first element group and the second element group each include a plurality of magnetic domain wall movement elements, wherein the plurality of magnetic domain wall movement elements each include a magnetic domain wall movement layer, a ferromagnetic layer, and a non-magnetic layer sandwiched between the magnetic domain wall movement layer and the ferromagnetic layer, wherein the length of the magnetic domain wall movement layer in the long side direction of each of the magnetic domain wall movement elements belonging to the first element group is shorter than the length of the magnetic domain wall movement layer in the long side direction of each of the magnetic domain wall movement elements belonging to the second element group, and wherein the resistance change rate when a pulse of a predetermined size is input is larger than that of each of the magnetic domain wall movement elements belonging to the first element group.

[0017] (2) In the neuromorphic device of the above aspect, the first element group and the second element group may be in a stacked structure, and the stacked structure may be stacked on a substrate.

[0018] (3) In the stacked structure of the neuromorphic device according to the above aspect, the second element group may be located farther from the substrate than the first element group.

[0019] (4) In the neuromorphic device of the above aspect, the number of magnetic domain wall moving elements belonging to the first element group may be greater than the number of magnetic domain wall moving elements belonging to the second element group.

[0020] (5) In the neuromorphic device of the above-mentioned scheme, the critical current density required for moving the magnetic domain walls of the magnetic domain wall moving elements may be smaller for each of the magnetic domain wall moving elements belonging to the second element group than for each of the magnetic domain wall moving elements belonging to the first element group.

[0021] (6) When the neuromorphic device of the above embodiment is viewed from above in the stacking direction of the magnetic domain wall moving elements, any one of the magnetic domain wall moving elements in the first element group and any one of the magnetic domain wall moving elements in the second element group at least partially overlap.

[0022] (7) In the longitudinal direction of any one of the magnetic domain wall moving elements in the second element group of the neuromorphic device according to the above-mentioned scheme, both ends of any one of the magnetic domain wall moving elements in the first element group may be located on the inner side of both ends of any one of the magnetic domain wall moving elements in the second element group.

[0023] (8) When the neuromorphic device of the above aspect is viewed from above in the stacking direction of the magnetic domain wall moving elements, the magnetic domain wall moving elements of the first element group and the magnetic domain wall moving elements of the second element group may not overlap.

[0024] (9) In the neuromorphic device of the above embodiment, the surface roughness of the lower surface of the magnetic domain wall moving element belonging to the second element group may be greater than the surface roughness of the lower surface of the magnetic domain wall moving element belonging to the first element group.

[0025] (10) The neuromorphic device according to the above aspect may further include a connection line connecting any one of the magnetic domain wall moving elements in the first element group and any one of the magnetic domain wall moving elements in the second element group.

[0026] (11) In the neuromorphic device of the above scheme, the first element group may perform a first product-sum operation, the second element group may perform a second product-sum operation, and the sum of the outputs from the plurality of magnetic domain wall movement elements belonging to the first element group may be input to the magnetic domain wall movement element belonging to the second element group.

[0027] (12) In the neuromorphic device of the above embodiment, the pulse length of the write pulse input to the magnetic domain wall moving element belonging to the second element group may be different from the pulse length of the write pulse input to the magnetic domain wall moving element belonging to the first element group.

[0028] (13) In the neuromorphic device of the above embodiment, the pulse amplitude of the write pulse input to the magnetic domain wall moving element belonging to the second element group may be different from the pulse amplitude of the write pulse input to the magnetic domain wall moving element belonging to the first element group.

[0029] (14) In the neuromorphic device of the above scheme, the first component group and the second component group may be respectively responsible for operations between different layers in the neural network.

[0030] (15) In the neuromorphic device of the above scheme, the first element group may be responsible for operations on the input layer side of the neural network compared to the second element group.

[0031] Effects of the Invention

[0032] The neuromorphic device of the above scheme has high integration of resistance variable elements and excellent recognition rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 This is an image diagram of the integrated device of the first embodiment.

[0034] Figure 2 This is a schematic diagram of a neural network.

[0035] Figure 3 is a circuit diagram of an integrated device for a neuromorphic device according to a first embodiment.

[0036] Figure 4 is a cross-sectional view of a characteristic portion of an integrated device for a neuromorphic device according to the first embodiment.

[0037] Figure 5 FIG. 1 is a top view of a characteristic portion of an integrated device for a neuromorphic device according to the first embodiment.

[0038] Figure 6 This is an enlarged top view of the vicinity of two magnetic domain wall moving elements used in the integrated device of the neuromorphic device according to the first embodiment.

[0039] Figure 7 is a perspective view of a characteristic portion of an integrated device for a neuromorphic device according to the first embodiment.

[0040] Figure 8 4 is a cross-sectional view of a magnetic domain wall moving element belonging to a first element group and a magnetic domain wall moving element belonging to a second element group used in the integrated device of the neuromorphic device according to the first embodiment.

[0041] Figure 9 This is a map showing how the resistance value changes with the number of pulses applied to the magnetic domain wall moving element.

[0042] Figure 10 FIG. 1 is a top view of a characteristic portion of an integrated device for a neuromorphic device according to a second embodiment.

[0043] Figure 11 This is an enlarged top view of the vicinity of two magnetic domain wall moving elements used in the integrated device of the neuromorphic device according to the second embodiment.

[0044] Figure 12 4 is a cross-sectional view of a magnetic domain wall moving element belonging to a first element group and a magnetic domain wall moving element belonging to a second element group used in an integrated device of a neuromorphic device according to a second embodiment.

[0045] Figure 13 FIG. 1 is a top view of a characteristic portion of an integrated device for a neuromorphic device according to a third embodiment.

[0046] Figure 14 This is an enlarged top view of the vicinity of two magnetic domain wall moving elements used in the integrated device of the neuromorphic device according to the third embodiment.

[0047] Figure 15is a cross-sectional view of a characteristic portion of an integrated device for a neuromorphic device according to a fourth embodiment.

[0048] Figure 16 is a cross-sectional view of a characteristic portion of an integrated device for a neuromorphic device according to a fifth embodiment.

[0049] Figure 17 This is a cross-sectional view of a characteristic portion of another example of an integrated device used in the neuromorphic device of the fifth embodiment.

[0050] Figure 18 This is a graph showing how the resistance value changes with the number of pulses applied to a variable resistance element included in an integrated device used in a neuromorphic device according to the fifth embodiment.

[0051] Figure 19 4 is a cross-sectional view of a magnetic domain wall moving element belonging to the first element group and a magnetic domain wall moving element belonging to the second element group used in the integrated device of the neuromorphic device according to Modification 1.

[0052] Figure 20 4 is a cross-sectional view of a magnetic domain wall moving element belonging to the first element group and a magnetic domain wall moving element belonging to the second element group, used in an integrated device of a neuromorphic device according to Modification 2.

[0053] Figure 21 4 is a cross-sectional view of a magnetic domain wall moving element belonging to the first element group and a magnetic domain wall moving element belonging to the second element group, used in an integrated device of a neuromorphic device according to Modification 3.

[0054] Figure 22 4 is a cross-sectional view of a magnetic domain wall moving element belonging to the first element group and a magnetic domain wall moving element belonging to the second element group used in an integrated device of a neuromorphic device according to Modification 4.

[0055] Figure 23 4 is a cross-sectional view of a magnetic domain wall moving element belonging to the first element group and a magnetic domain wall moving element belonging to the second element group, used in an integrated device of a neuromorphic device according to Modification 5. DETAILED DESCRIPTION

[0056] The present embodiment will be described in detail below with reference to the accompanying drawings as appropriate. To facilitate understanding of the features of the present invention, the accompanying drawings used in the following description may sometimes be partially enlarged, and the dimensional ratios of the various components may differ from the actual ones. The materials, dimensions, and other aspects illustrated in the following description are merely examples, and the present invention is not limited thereto. Appropriate modifications may be made to achieve the effects of the present invention.

[0057] First, define the direction. Figure 4) is defined as the x-direction, and the direction perpendicular to the x-direction is defined as the y-direction. The x-direction is, for example, the direction in which the magnetic domain wall movement layer of the magnetic domain wall movement element extends. The x-direction is an example of a longitudinal direction. The z-direction is a direction perpendicular to the x-direction and the y-direction. The z-direction is an example of a stacking direction. Hereinafter, the +z-direction may be expressed as "up" and the -z-direction as "down". The +z-direction is the direction away from the substrate Sb. Up and down do not necessarily coincide with the direction in which gravity is applied.

[0058] In this specification, "extending in the x direction" means that the length in the x direction is longer than the lengths in other directions. In this specification, "connected" is not limited to direct connection, and includes connection via an intermediate layer.

[0059] First embodiment

[0060] Figure 1 This is a circuit diagram of an integrated device ID according to the first embodiment. The integrated device ID includes, for example, a first circuit C1 and a second circuit C2. The first circuit C1 and the second circuit C2 each perform a different sum-of-products operation. The first circuit C1 and the second circuit C2 are interconnected. For example, the output of the sum-of-products operation by the first circuit C1 is input to the second circuit C2. The second circuit C2, for example, uses the result of the sum-of-products operation by the first circuit C1 as input to perform a further sum-of-products operation.

[0061] The integrated device ID of the first embodiment functions as, for example, a neuromorphic device. A neuromorphic device is a device that performs neural network calculations. A neuromorphic device artificially mimics the relationship between neurons and synapses in the human brain.

[0062] Figure 2 This is a schematic diagram of a neural network NN. The neural network NN has an input layer Lin, an intermediate layer L m And the output layer Lout. Figure 2 In the figure, the middle layer L is shown. m For a three-layer example, the middle layer L m The number of input layer Lin, intermediate layer L m The input layer Lin, the intermediate layer Lout and the output layer Lout each have multiple chips C, each chip C corresponds to a neuron in the brain. m The neural network NN learns through the transmission units (synapses) to improve its correct answer rate. Learning involves discovering behaviors that can be used in the future from information. The neural network NN learns by changing the weights applied to the transmission units. The transmission units perform weighted product operations on the input signals and sum operations on the sum of the product results. In other words, the transmission units perform product-sum operations.

[0063] In the neural network NN, the resolution may vary at each layer. For example, the weights applied to the transmission units may be varied more finely at layers closer to the output layer Lout. This structure improves the recognition rate of the neural network NN. In other words, in the neural network NN, layers closer to the output layer Lout require higher resolution.

[0064] Figure 1 The first circuit C1 shown is responsible for example for the first intermediate layer L m1 To the second intermediate layer L m2 The second circuit C2 is responsible for the product and sum operation from the second intermediate layer L m2 To the third intermediate layer L m3 The product and sum operation.

[0065] Figure 3 This is a circuit diagram of a portion of the integrated device ID according to the first embodiment. Figure 3 (a) is a circuit diagram of the first circuit C1, Figure 3 (b) is a circuit diagram of the second circuit C2. The first circuit C1 and the second circuit C2 have, for example, the same circuit structure.

[0066] The first circuit C1 and the second circuit C2 each include a plurality of variable resistance elements. The variable resistance element included in the first circuit C1 is, for example, a magnetic domain wall moving element 100. Furthermore, the variable resistance element included in the second circuit C2 is, for example, a magnetic domain wall moving element 110. The magnetic domain wall moving elements 100 and 110 are domain wall moving type magnetoresistive effect elements. In other words, the magnetic domain wall moving elements 100 and 110 are three-terminal elements. The number of magnetic domain wall moving elements 100 in the first circuit C1 is, for example, the same as or greater than the number of magnetic domain wall moving elements 110 in the second circuit C2. Input to one magnetic domain wall moving element 110 is, for example, the sum of the outputs from a plurality of magnetic domain wall moving elements 100 connected to a readout line RL connected to the magnetic domain wall moving element 110.

[0067] The first circuit C1 and the second circuit C2 each include a plurality of first switching elements SW1 , a plurality of second switching elements SW2 , a plurality of third switching elements SW3 , a plurality of write lines WL, a plurality of read lines RL, and a plurality of common lines CL.

[0068] In the first circuit C1, the magnetic domain wall moving elements 100 are arranged in a matrix, for example. In the second circuit C2, the magnetic domain wall moving elements 110 are arranged in a matrix, for example. Each magnetic domain wall moving element 100, 110 is connected to a first switching element SW1, a second switching element SW2, and a third switching element SW3, respectively. Any one of the first switching element SW1, the second switching element SW2, and the third switching element SW3 can be connected to multiple magnetic domain wall moving elements 100, 110.

[0069] When the first switching element SW1 and the second switching element SW2 connected to a specific magnetic domain wall moving element 100 or 110 are turned on, data is written to the specific magnetic domain wall moving element 100 or 110. The data is recorded as the resistance value in the stacking direction of the magnetic domain wall moving element 100 or 110. Furthermore, when the second switching element SW2 and the third switching element SW3 connected to the specific magnetic domain wall moving element 100 or 110 are turned on, the data written to the specific magnetic domain wall moving element 100 or 110 is read.

[0070] When reading data from the domain wall moving elements 100 and 110, current is passed from the read line RL to the common line CL. The current (output value) output from the common line CL varies depending on the resistance value of the domain wall moving elements 100 and 110 or the reciprocal of the resistance value, i.e., the conductance (weight). That is, the application of current from the read line RL to the common line CL corresponds to a multiplication operation in the neural network NN. Furthermore, the common line CL is connected to multiple domain wall moving elements 100 and 110 belonging to the same column, and the current detected at the end of the common line CL is the sum of the results of the multiplication operations performed by each domain wall moving element 100 and 110. Therefore, the integrated device ID functions as a product-sum operator of the neural network NN.

[0071] The currents applied to the readout lines RL of the integrated device ID are inputs to the product-sum operator, and the currents output from each of the common lines CL of the integrated device ID are outputs from the product-sum operator. The input signals to the product-sum operator can be controlled by pulse length, pulse amplitude, or pulse frequency.

[0072] The first switching element SW1, the second switching element SW2, and the third switching element SW3 are, for example, field-effect transistors. They may also be, for example, elements utilizing commutation in a crystal layer, such as an Ovonic Threshold Switch (OTS), elements utilizing changes in energy band structure, such as a Metal-Insulator Transition (MIT) switch, elements utilizing breakdown voltage, such as a Zener diode or an avalanche diode, or elements whose conductivity changes with changes in atomic position.

[0073] The first switching element SW1 is connected to the write line WL. The second switching element SW2 is connected to the common line CL. The third switching element SW3 is connected to the read line RL. The read line RL is a wiring line that allows current to flow when reading data. The write line WL is a wiring line that allows current to flow when writing data. The common line CL is a wiring line that allows current to flow both when writing data and when reading data.

[0074] Figure 4 It is a cross-sectional view of a characteristic portion of the integrated device ID of the first embodiment. Figure 5 It is a plan view of the characteristic portion of the integrated device ID according to the first embodiment. Figure 5 The read lines RL, write lines WL, and common lines CL are omitted from the illustration. Figure 6 1 is an enlarged plan view of the vicinity of two magnetic domain wall moving elements 100 and 110 of the integrated device ID according to the first embodiment. Figure 4 It is along Figure 5 and Figure 6 The xz section cut by line A-A in the figure. Figure 7 It is a perspective view of the characteristic part of the integrated device ID. Figure 7 The insulator In is removed for illustration.

[0075] The integrated device ID includes a substrate Sb and a stacked structure LS. The stacked structure LS is located on the substrate Sb.

[0076] The substrate Sb is, for example, a semiconductor substrate. The substrate Sb has a plurality of switching elements. The plurality of switching elements are insulated from each other by an inter-element insulator Ei. The plurality of switching elements controls each of the magnetic domain wall moving elements 100 and 110.

[0077] The plurality of switching elements may be, for example, a first switching element SW1 and a second switching element SW2. The third switching element SW3 may be located at a different position in the y-direction. The third switching element SW3 may be located in a peripheral region outside the integrated region where the magnetic domain wall movement elements 100 and 110 are integrated. The following example uses the case where the first switching elements SW1 and the second switching elements SW2 are arranged in a matrix within the integrated region.

[0078] The first switching element SW1 and the second switching element SW2 are each, for example, a field effect transistor Tr. Hereinafter, the first switching element SW1 and the second switching element SW2 may not be distinguished and are simply referred to as transistors Tr.

[0079] The transistors Tr are arranged, for example, in a matrix. Each transistor Tr includes, for example, a gate G, a gate insulating film GI, a source S, and a drain D. When viewed in the z direction, the gate G is located between the source S and the drain D. The gate G controls the flow of charge between the source S and the drain D. The terms "source S" and "drain D" are given according to the direction of current flow, and their positions change depending on the direction of current flow. The positional relationship of the source S and drain D shown in the figure is an example; the positional relationship of the source S and drain D of each transistor Tr may also be reversed.

[0080] The stacked structure LS includes a first element group, a second element group, wiring, and an insulator In. The first element group includes a plurality of magnetic domain wall moving elements 100. The second element group includes a plurality of magnetic domain wall moving elements 110. The first element group and the second element group are located at different levels. The second element group is located further away from the substrate Sb than the first element group. The first element group, for example, forms a first circuit C1 and performs a first product-sum operation. The second element group, for example, forms a second circuit C2 and performs a second product-sum operation. That is, the first element group and the second element group, for example, are responsible for Figure 2 The operations between the different successive layers of the neural network NN are shown. For example, the first component group is responsible for the operations of the layer closer to the input layer Lin than the second component group.

[0081] A layer is a functionally defined layer. The stacked structure LS is produced by repeating the stacking and processing steps. The unit of stacking in each stacking step is generally referred to as a layer. For example, in the stacked structure LS, wiring layers containing in-plane wiring and element layers containing domain wall moving elements are alternately stacked. The element layers can be any number of layers, two or more.

[0082] The multiple magnetic domain wall moving elements 100, 110 and the wiring are located within the insulator In. The insulator In is formed at each level. For example, the insulator In is divided into insulators In1, In2, In3, and In4 according to the level. The insulator In insulates the wiring or elements of the multilayer wiring. The insulator In is, for example, silicon oxide (SiO x ), silicon nitride (SiN x ), silicon carbide (SiC), chromium nitride, silicon carbide nitride (SiCN), silicon oxynitride (SiON), aluminum oxide (Al2O3), zirconium oxide (ZrO x )wait.

[0083] The wiring is conductive. The wiring includes, for example, any one selected from the group consisting of Ag, Cu, Co, Al, Au, and Ru. The wiring includes in-plane wiring and through-hole wiring VL. In-plane wiring is wiring that extends in any direction within the xy plane. Through-hole wiring VL is wiring that extends in the z direction. Through-hole wiring VL, for example, connects elements at different levels. Through-hole wiring VL can also be through-hole wiring that penetrates adjacent levels and reaches the level or substrate sandwiching the adjacent levels. Through-hole wiring, for example, connects each of the magnetic domain wall moving elements 100 and 110 to the transistor Tr of the substrate Sb, and penetrates a portion of the insulator In along the z direction. The through-hole wiring is, for example, continuous along the z direction.

[0084] For example, the read line RL, write line WL, common line CL, the wiring connecting these to the through-hole wiring VL, and the wiring connecting the through-hole wiring VL to each other are in-plane wiring. In-plane wiring is located, for example, at the level between the substrate Sb and the first element group, and at the level between the first element group and the second element group. The read line RL extends, for example, in the x-direction. For example, the read line RL includes read line RL1 connected to the magnetic domain wall moving element 100 and read line RL2 connected to the magnetic domain wall moving element 110. Read lines RL1 and RL2 are connected to the ferromagnetic layers 20 and 60 of the magnetic domain wall moving elements 100 and 110, for example, via electrodes E. The write line WL extends, for example, in the x-direction. The write line WL is connected to the through-hole wiring VL that reaches the transistor Tr via wiring extending in the y-direction. The common line CL extends, for example, in the y-direction.

[0085] The magnetic domain wall moving element 100 and the magnetic domain wall moving element 110 are located at different levels of the stacked structure LS. The magnetic domain wall moving element 100 is located at the first level, and the magnetic domain wall moving element 110 is located at the second level. The magnetic domain wall moving elements 100 and 110 are each connected to one of the transistors Tr on, for example, the substrate Sb. For example, transistors Tr adjacent to each other in the x-direction are connected to magnetic domain wall moving elements 100 and 110 at different levels. For example, transistors Tr adjacent to each other in the x-direction are connected to magnetic domain wall moving elements 100 and 110 at different levels. For example, the second and third columns of transistors Tr arranged along the x-direction control the magnetic domain wall moving element 100, while the first and fourth columns of transistors Tr control the magnetic domain wall moving element 110.

[0086] Figure 8 1 is a cross-sectional view of the magnetic domain wall moving elements 100 and 110 included in the integrated device ID according to the first embodiment. Figure 8 The cross sections of the magnetic domain wall moving elements 100 and 110 are respectively cut along the xz plane passing through the center of the width of the magnetic domain wall moving layers 10 and 50 in the y direction.

[0087] The magnetic domain wall moving elements 100 and 110 are, for example, Figures 4 to 7As shown, the domain wall moving element 100 is integrated into the integrated device ID. The x-direction length of the domain wall moving element 100 is shorter than that of the domain wall moving element 110. Therefore, the x-direction ends of the domain wall moving element 100 are positioned, for example, to lie inward of the x-direction ends of the domain wall moving element 110. When viewed in the z-direction, the domain wall moving element 110 at least partially overlaps with the domain wall moving element 100.

[0088] The domain wall moving element 100 includes a domain wall moving layer 10, a nonmagnetic layer 30, and a ferromagnetic layer 20. The domain wall moving layer 10 is, for example, located closer to the substrate Sb than the ferromagnetic layer 20. The domain wall moving element 110 includes a domain wall moving layer 50, a nonmagnetic layer 70, and a ferromagnetic layer 60. The domain wall moving layer 50 is, for example, located closer to the substrate Sb than the ferromagnetic layer 60. The domain wall moving elements 100 and 110 are three-terminal magnetoresistive elements, with their length in the x-direction being longer than their length in the y-direction. The domain wall moving element 100 and the domain wall moving element 110 have different lengths in the x-direction. The domain wall moving element 110 is longer in the x-direction than the domain wall moving element 100. The domain wall moving element 100 and the domain wall moving element 110 have substantially the same other structures and shapes.

[0089] The domain wall migration layers 10 and 50 extend in the x-direction. When viewed from above in the z-direction, the domain wall migration layers 10 and 50 are rectangular, for example, with the x-direction being the major axis and the y-direction being the minor axis. The domain wall migration layers 10 and 50 face the ferromagnetic layers 20 and 60, sandwiching the nonmagnetic layers 30 and 70. The domain wall migration layers 10 and 50 have a first end connected to the first switching element SW1 and a second end connected to the second switching element SW2.

[0090] The magnetic domain wall migration layer 10, 50 is made of a ferromagnetic material. The magnetic domain wall migration layer 10, 50 is a layer that can magnetically record information by changing the internal magnetic state. The magnetic domain wall migration layer 10, 50 can have a first magnetic region A1 and a second magnetic region A2 with different magnetic states. The magnetization M of the first magnetic region A1 is A1 and the magnetization M of the second magnetic region A2 A2 For example, the magnetization M of the first magnetic region A1 is oriented in the opposite direction. A1 Oriented in the +z direction, the magnetization M of the second magnetic region A2 A2 Oriented in the -z direction. The boundary between the first magnetic region A1 and the second magnetic region A2 is a magnetic domain wall DW. The domain wall migration layer 10, 50 can have a magnetic domain wall DW within it. If a current exceeding the critical current density of the domain wall migration layer 10, 50 flows in the longitudinal direction of the domain wall migration layer 10, 50, the domain wall DW migrates.

[0091] If the magnetic domain wall DW moves, the ratio of the first magnetic region A1 to the second magnetic region A2 in the magnetic domain wall moving layer 10, 50 changes. The magnetic domain wall DW moves by passing a write current along the x-direction of the magnetic domain wall moving layer 10, 50. If the ratio of the first magnetic region A1 to the second magnetic region A2 in the magnetic domain wall moving layer 10, 50 changes, the resistance value of the magnetic domain wall moving element 100, 110 changes. The resistance value of the magnetic domain wall moving element 100, 110 changes according to the relative angle of the magnetization of the ferromagnetic layers sandwiching the non-magnetic layers 30, 70. The resistance value of the magnetic domain wall moving element 100, 110 changes according to the magnetization M of the magnetic domain wall moving layer 10, 50. A1 、M A2 and the magnetization M of the ferromagnetic layers 20 and 60 20 、M 60 The relative angle of .

[0092] That is, the resistance value of the magnetic domain wall moving element 100, 110 depends on the ratio of the first magnetic region A1 to the second magnetic region A2 in the magnetic domain wall moving layer 10, 50, and the magnetization M of the magnetic domain wall moving layer 10, 50. A1 、M A2 and the magnetization M of the ferromagnetic layers 20 and 60 20 、M 60 The relative angle of .

[0093] If the ratio of the first magnetic region A1 becomes higher, the resistance value of the magnetic domain wall moving element 100, 110 decreases. When the ratio of the first magnetic region A1 is maximum, the magnetization of the magnetic domain wall moving layer 10, 50 and the magnetization M of the ferromagnetic layer 20, 60 are equal. 20 、M 60 When the ratio of the second magnetic region A2 is the largest, the magnetization of the magnetic domain wall moving layer 10, 50 and the magnetization of the ferromagnetic layer 20, 60 are equal. 20 、M 60 The two magnetic domain wall moving elements 100 and 110 are in an antiparallel relationship, and the resistance value of the magnetic domain wall moving elements 100 and 110 reaches a maximum resistance value.

[0094] The domain wall moving elements 100 and 110 change the resistance value to analog by changing the position of the domain wall DW. The resistance value or the reciprocal of the resistance value, i.e., the conductance, of the domain wall moving elements 100 and 110 corresponds to the weight of the transfer unit in the neural network NN.

[0095] When a predetermined pulse is applied to the domain wall moving elements 100 and 110, the resistance change rate of the domain wall moving element 100 is greater than the resistance change rate of the domain wall moving element 110. When a predetermined pulse is applied to a domain wall moving element having a resistance value R, the resistance change rate P is expressed by the following equation (1). Here, the maximum resistance value and the minimum resistance value of the domain wall moving element are respectively R max 、R min .

[0096] P=|R′-R| / (R max -R min )···(1)

[0097] Figure 9 : is a graph showing the relationship between the resistance value of the magnetic domain wall moving elements 100 and 110 and the number of pulses applied to the magnetic domain wall moving elements 100 and 110. Figure 9 , as an example, a map showing an increase in resistance due to pulse input is shown. However, in this embodiment, a magnetic domain wall moving element in which the resistance is reduced due to pulse input may also be used. Figure 9 (a) is a diagram of the magnetic domain wall moving element 100. Figure 9 (b) is a graph of the domain wall moving element 110. In the domain wall moving elements 100 and 110, the conductance changes linearly with respect to the number of applied pulses, but the dynamic range is narrow. Figure 9 As shown, the resistance value changes linearly. Thus, as the magnetic domain wall moving element, a magnetic domain wall moving element whose resistance value changes nonlinearly with respect to the number of applied pulses can be used, or a magnetic domain wall moving element whose resistance value changes linearly with respect to the number of applied pulses can be used. Figure 9 In FIG. 5 , the maximum resistance value Rmax and the minimum resistance value Rmin of the magnetic domain wall moving element 100 are the same as the maximum resistance value Rmax and the minimum resistance value Rmin of the magnetic domain wall moving element 110 , but they may be different.

[0098] The domain wall motion layer 50 of the domain wall motion element 110 is longer in the longitudinal direction than the domain wall motion layer 10 of the domain wall motion element 100. When a predetermined pulse is applied to the domain wall motion elements 100 and 110, the domain wall DW moves, and the ratio between the first magnetic region A1 and the second magnetic region A2 changes. Because the lengths of the domain wall motion layers 10 and 50 differ, even when the amount of domain wall DW movement is the same, the ratio of the change in the ratio between the first magnetic region A1 and the second magnetic region A2 caused by the application of the predetermined pulse differs between the domain wall motion element 100 and the domain wall motion element 110. In other words, the rate of change in resistance of the domain wall motion element 100 after application of the predetermined pulse is greater than that of the domain wall motion element 110. The predetermined pulse is any pulse greater than the critical current density of the domain wall motion layers 10 and 50. The critical current density is the current density required to move the domain wall DW.

[0099] Because the resistance change rate of the domain wall moving element 110 is smaller than that of the domain wall moving element 100, the domain wall moving element 110 can refine the resistance value. In other words, the domain wall moving element 110 has a higher resolution than the domain wall moving element 100. By placing the domain wall moving element 110 on the output layer Lout side of the neural network NN, it is possible to assign various weights to the transmitted information.

[0100] The resolution can also be adjusted by the pinning point. At the pinning point, the magnetic domain wall DW is not easy to move, and the magnetization M of the first magnetic region is A1 and the magnetization M of the second magnetic region A2 The pinning points are, for example, the unevenness of the magnetic domain wall motion layer 10 or 50. The magnetic domain wall motion layer 10 or 50 may also have multiple pinning points. From the perspective of improving resolution, the magnetic domain wall motion layer 50 may have more pinning points than the magnetic domain wall motion layer 10. For example, the surface roughness of the lower surface 501 of the magnetic domain wall motion layer 50 may be rougher than the surface roughness of the lower surface 11 of the magnetic domain wall motion layer 10.

[0101] The magnetic domain wall migration layers 10 and 50 are made of a magnetic material. They can also be ferromagnetic, ferrimagnetic, or a combination of these and an antiferromagnetic material whose magnetic state can be changed by an electric current. The magnetic domain wall migration layers 10 and 50 preferably contain at least one element selected from the group consisting of Co, Ni, Fe, Pt, Pd, Gd, Tb, Mn, Ge, and Ga. Examples of materials for the magnetic domain wall migration layers 10 and 50 include stacked films of Co and Ni, stacked films of Co and Pt, stacked films of Co and Pd, MnGa-based materials, GdCo-based materials, and TbCo-based materials. Ferrimagnetic materials such as MnGa-based materials, GdCo-based materials, and TbCo-based materials have low saturation magnetization, which reduces the threshold current required for magnetic domain wall DW migration. Furthermore, stacked films of Co and Ni, Co and Pt, and Co and Pd have high coercivity, which slows the migration speed of the magnetic domain wall DW. Examples of antiferromagnetic materials include Mn3X (where X is Sn, Ge, Ga, Pt, Ir, etc.), CuMnAs, and Mn2Au. The domain wall migration layers 10 and 50 may also be made of the same material as the ferromagnetic layers 20 and 60 described later. Furthermore, the domain wall migration layers 10 and 50 may also be made of the same material as the ferromagnetic layers 20 and 60 described later and have a laminated structure. For example, the domain wall migration layers 10 and 50 may be a laminated film of Co, Pd, and CoFeB.

[0102] The nonmagnetic layer 30 is stacked on the magnetic domain wall migration layer 10 , for example. The nonmagnetic layer 70 is stacked on the magnetic domain wall migration layer 50 , for example. The nonmagnetic layers 30 and 70 are located between the magnetic domain wall migration layers 10 and 50 and the ferromagnetic layers 20 and 60 .

[0103] The non-magnetic layers 30 and 70 are composed of, for example, non-magnetic insulators, semiconductors, or metals. Non-magnetic insulators include, for example, Al2O3, SiO2, MgO, MgAl2O4, and materials in which some of the Al, Si, or Mg is substituted with Zn, Be, or the like. These materials have large band gaps and excellent insulating properties. When the non-magnetic layers 30 and 70 are composed of non-magnetic insulators, they serve as tunnel barriers. Non-magnetic metals include, for example, Cu, Au, and Ag. Non-magnetic semiconductors include, for example, Si, Ge, CuInSe2, CuGaSe2, and Cu(In,Ga)Se2.

[0104] The thickness of the non-magnetic layers 30 and 70 is preferably More preferably If the thickness of the non-magnetic layer 30, 70 is thick, the resistance area product (RA) of the domain wall moving element 100, 110 becomes larger. The resistance area product (RA) of the domain wall moving element 100, 110 is preferably 1×10 4 Ωμm 2 More than 1×105 Ωμm 2 The resistance area product (RA) of the domain wall moving element 100, 110 is represented by the product of the element resistance of one domain wall moving element 100, 110 and the element cross-sectional area of ​​the domain wall moving element 100, 110 (the area of ​​the cross section of the non-magnetic layer 30, 70 cut along the xy plane).

[0105] The ferromagnetic layer 20 is located on the non-magnetic layer 30. The ferromagnetic layer 60 is located on the non-magnetic layer 70. The ferromagnetic layer 20 has a magnetization M oriented in one direction. 20 The ferromagnetic layer 60 has a magnetization M oriented in one direction. 60 The magnetization M of the ferromagnetic layers 20 and 60 is 20 、M 60 When a predetermined external force is applied, the magnetization M of the first magnetic area A1 and the second magnetic area A2 A1 、M A2 The predetermined external force is, for example, an external force applied to the magnetization by an external magnetic field or an external force applied to the magnetization by a spin-polarized current. The ferromagnetic layers 20 and 60 are sometimes referred to as magnetization pinned layers or magnetization reference layers.

[0106] The ferromagnetic layers 20 and 60 are made of a ferromagnetic material. For example, the ferromagnetic layers 20 and 60 are made of a material that facilitates coherent tunneling between the domain wall migration layers 10 and 50. The ferromagnetic layers 20 and 60 are made of, for example, a metal selected from the group consisting of Cr, Mn, Co, Fe, and Ni, an alloy containing one or more of these metals, or an alloy containing these metals and at least one of B, C, and N. Examples of the ferromagnetic layers 20 and 60 are Co-Fe, Co-Fe-B, or Ni-Fe.

[0107] The ferromagnetic layers 20 and 60 may also be, for example, Whistler alloys. Whistler alloys are semimetals with high spin polarization. Whistler alloys are intermetallic compounds with a chemical composition of XYZ or X2YZ, where X is a transition metal element or a noble metal element of the Co, Fe, Ni or Cu group on the periodic table, Y is a transition metal element of the Mn, V, Cr or Ti group or an element of X, and Z is a typical element of Group III to Group V. Examples of Whistler alloys include Co2FeSi, Co2FeGe, Co2FeGa, Co2MnSi, Co2Mn 1－a Fe a AlbSi 1－b 、Co2FeGe 1－c Ga c wait.

[0108] A magnetic layer may also be provided on the surface of the ferromagnetic layer 20, 60 opposite the non-magnetic layer 30, 70, with an isolation layer interposed therebetween. The ferromagnetic layer 20, 60, the isolation layer, and the magnetic layer form a synthetic antiferromagnetic structure (SAF structure). The synthetic antiferromagnetic structure is composed of two magnetic layers sandwiching a non-magnetic layer. Antiferromagnetic coupling between the ferromagnetic layer 20, 60 and the magnetic layer increases the coercive force of the ferromagnetic layer 20, 60 compared to a case without a magnetic layer. The magnetic layer may comprise, for example, a ferromagnetic material or an antiferromagnetic material such as IrMn or PtMn. The isolation layer may comprise, for example, at least one selected from the group consisting of Ru, Ir, and Rh.

[0109] The magnetization direction of each layer of the domain wall moving elements 100 and 110 can be confirmed by, for example, measuring a magnetization curve. This can be measured, for example, using the MOKE (Magneto Optical Kerr Effect). MOKE measurement involves applying linearly polarized light to an object and utilizing the magneto-optical effect (Kerr effect) that causes rotation of the polarization direction.

[0110] Next, the manufacturing method of the integrated device ID is described. The integrated device ID is formed by a process of stacking each layer and processing a portion of each layer into a predetermined shape. The stacking of each layer can be performed using sputtering, chemical vapor deposition (CVD), electron beam evaporation (EB evaporation), atomic laser deposition, etc. The processing of each layer can be performed using photolithography, etc.

[0111] First, impurities are doped into predetermined locations on a substrate Sb to form a source S and a drain D. Next, a gate insulating film GI and a gate G are formed between the source S and the drain D. The source S, the drain D, the gate insulating film GI, and the gate G form a transistor Tr. A commercially available semiconductor substrate having transistors Tr periodically arranged thereon can also be used as the substrate Sb.

[0112] Next, a wiring layer up to the first level is formed. The wiring layer can be formed using photolithography.

[0113] Next, the first element group of the first level is fabricated. First, the ferromagnetic layer, nonmagnetic layer, and ferromagnetic layer are stacked in sequence and processed into the desired shape. The ferromagnetic layer, nonmagnetic layer, and ferromagnetic layer become the domain wall migration layer 10, nonmagnetic layer 30, and ferromagnetic layer 20, respectively. The first element group can also be fabricated using photolithography.

[0114] Next, the integrated device ID is obtained by fabricating the wiring layer between the first and second layers and the second element group of the second layer using the same steps. The second element group can be fabricated using the same steps as the first element group. By sequentially stacking the ferromagnetic layer, the nonmagnetic layer, and the ferromagnetic layer and processing them into the desired shape, the ferromagnetic layer, the nonmagnetic layer, and the ferromagnetic layer become the magnetic domain wall migration layer 50, the nonmagnetic layer 70, and the ferromagnetic layer 60, respectively.

[0115] The neuromorphic device with the above structure highly integrates the magnetic domain wall moving elements 100 belonging to the first element group while improving the resolution of the magnetic domain wall moving elements 110 belonging to the second element group. In other words, the neuromorphic device of this embodiment has high integration and can improve the recognition rate. Furthermore, the neuromorphic device of this embodiment, which varies the length in the x-direction between the magnetic domain wall moving elements 100 belonging to the first element group and the magnetic domain wall moving elements 110 belonging to the second element group, thus introducing structural variations at each level, runs counter to conventional technical common sense, which reduces variations at each level by using identical elements within an array.

[0116] Furthermore, the neuromorphic device of this embodiment can also be configured to position the x-direction ends of the domain wall moving element 100 inward of the x-direction ends of the domain wall moving element 110 by varying the lengths of the domain wall moving element 100 and the domain wall moving element 110. This structure facilitates the formation of the wiring necessary to operate the neuromorphic device. Furthermore, the number of domain wall moving elements 100 and 110 that can be accommodated within a predetermined area is increased. This improves the integration of the variable resistance element within the integrated device ID.

[0117] Furthermore, the aforementioned relationship between the domain wall moving elements 100 and 110 can apply to all or any of the domain wall moving elements 100 and 110. For example, in the longitudinal direction of any domain wall moving element 110 belonging to the second element group, both ends of any domain wall moving element 100 belonging to the first element group can be located inward of both ends of any domain wall moving element 110 belonging to the second element group. Furthermore, any domain wall moving element 100 belonging to the first element group and any domain wall moving element 110 belonging to the second element group can at least partially overlap. Furthermore, in the aforementioned example, the figure illustrates a configuration in which the longitudinal directions of all the domain wall moving elements 100 and 110 belonging to the first and second element groups are in the x-direction. However, the longitudinal directions of any domain wall moving element belonging to the first and second element groups may also be in a direction other than the x-direction.

[0118] In addition, in the above-mentioned embodiment, the figure shows a case where the number of magnetic domain wall movement elements 100 belonging to the first element group and the number of magnetic domain wall movement elements 110 belonging to the second element group are different, but the number of magnetic domain wall movement elements 100 belonging to the first element group and the number of magnetic domain wall movement elements 110 belonging to the second element group can also be the same.

[0119] Furthermore, during operation of the integrated device ID, the pulse length of the write pulse input to the magnetic domain wall moving element 110 belonging to the second element group may be different from the pulse length of the write pulse input to the magnetic domain wall moving element 100 belonging to the first element group. For example, the pulse length of the write pulse input to the magnetic domain wall moving element 110 may be longer than the pulse length of the write pulse input to the magnetic domain wall moving element 100 belonging to the first element group.

[0120] Alternatively, the pulse amplitude of the write pulse input to the magnetic domain wall moving element 110 belonging to the second element group may be different from the pulse amplitude of the write pulse input to the magnetic domain wall moving element 100 belonging to the first element group. For example, the pulse amplitude of the write pulse input to the magnetic domain wall moving element 110 may be larger than the pulse amplitude of the write pulse input to the magnetic domain wall moving element 100.

[0121] By making the magnitude (pulse length, pulse amplitude) of the write pulse applied to the domain wall migration layer 50 different from that of the write pulse applied to the domain wall migration layer 10 , the resolution of each layer in the neuromorphic device can be precisely adjusted.

[0122] For example, increasing the magnitude of the write pulse applied to the domain wall moving layer 50 can prevent the magnetic domain walls DW from being trapped in the well. This improves the operational reliability of the domain wall moving element 110. The pulse magnitude is controlled, for example, by a write circuit connected to the integrated device ID.

[0123] (Second embodiment)

[0124] Figure 10 FIG. 1 is a top view of an integrated device ID1 used in a neuromorphic device according to the second embodiment. Figure 10 The read lines RL, write lines WL, and common lines CL are omitted from the illustration. Figure 11 This is an enlarged cross-sectional view of the vicinity of the magnetic domain wall moving element 100 belonging to the first element group and the magnetic domain wall moving element 111 belonging to the second element group of the integrated device ID1 in the second embodiment. Figure 12The cross-section of the domain wall moving elements 100 and 111 is taken along a yz plane passing through the x-center of the domain wall moving layers 10 and 51, respectively. The neuromorphic device of the second embodiment differs from the first embodiment in that the width w111 of the domain wall moving element 111 is wider than the width w100 of the domain wall moving element 100. In the second embodiment, components identical to those of the first embodiment are denoted by the same reference numerals, and their descriptions are omitted.

[0125] The domain wall moving element 111 includes a domain wall moving layer 51, a nonmagnetic layer 71, and a ferromagnetic layer 61. The domain wall moving element 111 differs from the domain wall moving element 110 in length in the y direction. The domain wall moving layer 51, nonmagnetic layer 71, and ferromagnetic layer 61 of the domain wall moving element 111, which belongs to the second element group, have longer y-direction lengths than the domain wall moving layer 10, nonmagnetic layer 30, and ferromagnetic layer 20 of the domain wall moving element 100, which belongs to the first element group. In other words, the width w111 of the domain wall moving element 111 is greater than the width w100 of the domain wall moving element 100. Here, the width of the domain wall moving element refers to the average width of the upper surface of the domain wall moving layer and the lower surface of the ferromagnetic layer.

[0126] When the integrated device ID1 is viewed from the z direction, the magnetic domain wall moving element 100 is covered by the magnetic domain wall moving element 111 .

[0127] The second element group is arranged closer to the output layer Lout of the neural network NN than the first element group. As described above, the number of magnetic domain wall moving elements 111 belonging to the second element group may be less than the number of magnetic domain wall moving elements 100 belonging to the first element group. When the number of magnetic domain wall moving elements 111 belonging to the second element group is less than the number of magnetic domain wall moving elements 100 belonging to the first element group, the number of magnetic domain wall moving elements 111 connected to one readout current RL is less than the number of magnetic domain wall moving elements 100 connected to one readout wiring RL. Therefore, the second element group is smaller than the first element group in terms of the total output of one readout wiring RL. When the total output of one readout wiring RL is so small that it cannot be distinguished, it becomes a cause of learning errors in the neuromorphic device.

[0128] In the neuromorphic device of the second embodiment, because the domain wall moving layer 51, nonmagnetic layer 71, and ferromagnetic layer 61 of the domain wall moving element 111 are long in the y-direction, the resistance in the stacking direction can be reduced. In other words, the current flowing through the domain wall moving element 111 during readout can be increased. Consequently, the output from the domain wall moving element 111 during readout can be increased. Consequently, learning errors in the neuromorphic device can be suppressed. Furthermore, the neuromorphic device of the second embodiment can achieve the same effects as the neuromorphic device of the first embodiment.

[0129] In addition, Figures 10-12 , the state in which the magnetic domain wall moving element 100 is covered by the magnetic domain wall moving element 111 when viewed from the z direction is shown. By setting such a configuration, the integration of the magnetic domain wall moving elements 100 and 111 in the integrated device ID1 is improved. However, the configuration of the magnetic domain wall moving element 100 and the magnetic domain wall moving element 111 of this embodiment is not limited to this example, and a structure in which the magnetic domain wall moving element 100 and the magnetic domain wall moving element 111 partially overlap as in the first embodiment may also be used. In addition, in Figure 12 , an example is shown in which the yx cross-sectional shape of the magnetic domain wall moving elements 100 and 111 is tilted, but a non-tilted structure is also possible.

[0130] (Third embodiment)

[0131] Figure 13 is a top view of an integrated device ID2 used in a neuromorphic device according to a third embodiment. Figure 13 The read lines RL, write lines WL, and common lines CL are omitted from the illustration. Figure 14 This is an enlarged top view of the vicinity of two magnetic domain wall moving elements 100 and 110 of the neuromorphic device of the third embodiment. The configuration of the magnetic domain wall moving elements 100 and 110 of the neuromorphic device of the third embodiment differs from that of the neuromorphic device of the first embodiment. In the third embodiment, components identical to those of the first embodiment are denoted by the same reference numerals, and their descriptions are omitted.

[0132] In this embodiment, the magnetic domain wall moving element 110 belonging to the second element group does not overlap with the magnetic domain wall moving element 100 belonging to the first element group when viewed from the z direction.

[0133] The further the presence of various structures below the domain wall moving element 110, the lower the flatness of the stacking surface during fabrication of the domain wall moving element 110. This is because the number of processing steps required to reach the stacking surface increases. By arranging the domain wall moving element 110 at a position that does not overlap with the domain wall moving element 100 as viewed in the z-direction, the flatness of the stacking surface during fabrication of the domain wall moving element 110 can be improved.

[0134] If the flatness of the stacked surface is high when the magnetic domain wall movement element 110 is manufactured, the flatness of the lower surface of the magnetic domain wall movement layer 50 is improved.

[0135] The neuromorphic device of the third embodiment achieves the same effects as the neuromorphic device of the first embodiment. In addition, the neuromorphic device of the third embodiment has excellent reliability in writing operation by reducing the difference in roughness between the domain wall migration layer 10 and the domain wall migration layer 50 .

[0136] (Fourth embodiment)

[0137] Figure 15 is a cross-sectional view of an integrated device ID3 for a neuromorphic device according to a fourth embodiment. Figure 15 In the embodiment, the magnetic domain wall moving element 100 and a part of the magnetic domain wall moving element 110 are connected via a connection wire CW. In the fourth embodiment, the same components as those in the first embodiment are denoted by the same reference numerals, and their description is omitted.

[0138] For example, the domain wall moving element 100 is electrically connected to the closest domain wall moving element 110, not via the substrate Sb, but via the connection wire CW. Not all domain wall moving elements 100 need to be connected to the domain wall moving element 110; any one of them may be connected. For example, the ferromagnetic layer 20 of the domain wall moving element 100 is connected to the domain wall moving layer 50 of the domain wall moving element 110.

[0139] The connection wiring CW may also include a vertical switching element VSW. The vertical switching element VSW is a switching element composed of a stacked film stacked in the z-direction. For example, elements that utilize commutation in crystal layers, such as the Ovonic Threshold Switch (OTS), elements that utilize changes in energy band structure, such as the Metal-Insulator Transition (MIT) switch, elements that utilize breakdown voltage, such as Zener diodes and avalanche diodes, and elements whose conductivity changes with changes in atomic position are examples of vertical switching elements VSW.

[0140] If the domain wall moving element 100 and the domain wall moving element 110 are connected by a connecting wiring CW, a current path is formed from the readout line RL2 through the domain wall moving elements 100 and 110 to the common line CL. That is, the combined resistance of the resistance value of the domain wall moving element 100 and the resistance value of the domain wall moving element 110 can be read. In a neuromorphic device, the resistance value of the domain wall moving element 100 and the reciprocal of the resistance value, that is, the conductance, corresponds to a weight. The above-mentioned current path can express a new weight that is a combination of the weights of the two domain wall moving elements 100 and 110. Therefore, the neuromorphic device using the integrated device ID2 of the fourth embodiment can express three weights using the two domain wall moving elements 100 and 110, enabling more complex calculations and improved expressiveness.

[0141] (Fifth embodiment)

[0142] In the above embodiment, the case where the first element group and the second element group include magnetic domain wall moving elements is illustrated, but the present invention is not limited to this example. The first element group and the second element group may also include other resistance change elements. The neuromorphic device of the fifth embodiment differs from the neuromorphic device of the first embodiment in that resistance change elements other than the magnetic domain wall moving elements 100 and 110 are used as resistance change elements. In addition, the neuromorphic device of the fifth embodiment differs from the neuromorphic device of the first embodiment in that the length of the resistance change elements belonging to the first element group in the long side direction may be longer than the length of the resistance change elements belonging to the second element group in the long side direction. In addition, in the neuromorphic device of the fifth embodiment, with respect to the volume of the portion that contributes to the resistance change of the resistance change element, it is preferred that the resistance change elements belonging to the second element group are larger than the resistance change elements belonging to the first element group. The other structures are the same as those of the neuromorphic device of the first embodiment, and detailed description is omitted.

[0143] A neuromorphic device according to a fifth embodiment includes a first element group and a second element group including a plurality of variable resistance elements. In the neuromorphic device according to the fifth embodiment, the resistance change rate of each variable resistance element in the first element group, when a predetermined pulse is input, is greater than that of each variable resistance element in the second element group. For ease of description, the variable resistance elements in the first element group are referred to as first variable resistance elements, and the variable resistance elements in the second element group are referred to as second variable resistance elements.

[0144] Any element in which there is a correlation between the number of applied pulses and the resistance value is used as the first resistance change element and the second resistance change element. As the first resistance change element and the second resistance change element, for example, an element using phase change memory (PCM), a resistance change type memory (ReRAM), a carbon nanotube memory (NRAM), an element using CeRAM (Correlated Electron Random Access Memory) using a strongly correlated electron system, etc. are used. PCM gradually controls the phase switching between crystalline and amorphous phases. ReRAM forms conductive filaments based on metal precipitation in a medium such as TaO2 and utilizes resistance changes. The types of resistance change elements used as the first resistance change element and the second resistance change element may be the same or different.

[0145] Figure 16This is an example of a cross-sectional view of an integrated device ID4 used in a neuromorphic device according to a fifth embodiment. The neuromorphic device according to the fifth embodiment differs from the neuromorphic device according to the first embodiment in that it includes variable resistance elements 120 and 130 formed of phase change memory. The variable resistance element 120 is, for example, a first variable resistance element in a first layer, and the variable resistance element 130 is, for example, a second variable resistance element in a second layer.

[0146] Unlike the magnetic domain wall displacement elements 100 and 110 of the first embodiment, the variable resistance elements 120 and 130 are two-terminal elements. The variable resistance elements 120 and 130 each include, for example, lower electrodes E1 and E3, functional layers 15 and 75, and upper electrodes E2 and E4, which overlap in the stacking direction. The lower electrodes E1 and E3 are connected to the through-hole wiring VL. The upper electrodes E2 and E4 of the variable resistance elements 120 and 130 are connected to the wiring L. The lower electrodes E1 and E3 and the through-hole wiring VL may be integral, and the upper electrodes E2 and E4 and the wiring L may be integral.

[0147] The variable resistance element 130 is provided at a position farther from the substrate Sb than, for example, the variable resistance element 120. For example, at least a portion of the variable resistance element 120 overlaps with the variable resistance element 130. When viewed from above in the z-direction, all ends of the variable resistance element 120 may be located inside the entirety of the variable resistance element 130. In such a structure, both ends of the variable resistance element 120 in the longitudinal direction are located inside the ends of the variable resistance element 130 in the longitudinal direction. For example, the number of variable resistance elements 120 may be greater than that of the variable resistance element 130.

[0148] The functional layers 15 and 75 contain chalcogenides such as Ge-Sb-Te, and are composed of chalcogenides, for example. If a high voltage is applied to the functional layers 15 and 75 and a large current flows, the functional layers 15 and 75 are Joule-heated, and if the melting point is exceeded, the area exceeding the melting point melts. Thereafter, if the voltage is suddenly reduced, the functional layers 15 and 75 are suddenly cooled, and the area melted by Joule heating changes from a crystalline state to an amorphous state, and the resistance value increases (reset action). Here, the area that changes from a crystalline state to an amorphous state depends on the voltage applied to the functional layers 15 and 75 and the area of ​​the lower electrodes E1 and E3 perpendicular to the stacking direction. The greater the voltage applied to the functional layers 15 and 75, the larger the area that changes from a crystalline state to an amorphous state. In addition, in the functional layers 15 and 75, the larger the area of ​​the functional layers 15 and 75 perpendicular to the stacking direction, the larger the area that can change from a crystalline state to an amorphous state. When a voltage lower than the voltage required for the reset operation but sufficiently higher than the voltage required for the reset operation is applied to the functional layer 15 or 75, which has changed from a crystalline state to an amorphous state, the functional layer 15 or 75 changes from the amorphous state to the crystalline state, and the resistance value decreases (set operation). Thus, the functional layer 15 or 75 is composed of a material whose resistance value changes due to Joule heating. The operation of increasing the resistance value is called the reset operation, and the operation of decreasing the resistance value is called the set operation.

[0149] The area of ​​the functional layer 15 when viewed from above in the z-direction may be smaller than that of the functional layer 75, being less than 2 / 3 times or less than 1 / 2 times the area of ​​the functional layer 75. Here, the length of the functional layer 15 in the x-direction and the length of the functional layer 15 in the y-direction may also be shorter than the length of the functional layer 75 in the x-direction and the length of the functional layer 75 in the y-direction. By configuring the areas of the functional layers 15 and 75 of the variable resistance elements 120 and 130 as described above, the region in the variable resistance element 130 where the crystalline state can be changed by Joule heating can be expanded, and the resistance value can be varied to a greater extent. Furthermore, due to the high integration of the variable resistance element 120, a larger number of variable resistance elements can be provided than the variable resistance element 130.

[0150] The thickness T75 of the functional layer 75 may be, for example, greater than the thickness T15 of the functional layer 15, and may be at least 3 / 2 times or at least 2 times the thickness T15. By configuring the thicknesses of the functional layers 15 and 75 of the variable resistance elements 120 and 130 as described above, the region in the variable resistance element 130 where the crystalline state can change due to Joule heating can be increased, allowing the resistance value to change over a wider range of values.

[0151] The area of ​​the lower electrode E1 perpendicular to the stacking direction can be smaller than the area of ​​the lower electrode E3 perpendicular to the stacking direction, for example, by being 2 / 3 or 1 / 2 times the area of ​​the lower electrode E3 perpendicular to the stacking direction. Reducing the area of ​​the lower electrodes E1 and E3 perpendicular to the stacking direction can reduce the reset current required for the reset operation. By adopting the above-described structure for the relationship between the areas of the lower electrodes E1 and E3 perpendicular to the stacking direction in the variable resistance elements 120 and 130, the reset current of the variable resistance element 120, which is larger in number than the variable resistance element 130, can be reduced, thereby reducing the required power.

[0152] Of the main surfaces of the functional layers 15 and 75 , the surface roughness of the surface on the substrate Sb side may be rougher in the functional layer 75 than in the functional layer 15 .

[0153] The integrated device ID4 is formed, for example, by laminating each layer and processing each layer into a predetermined shape. The layers can be laminated using the same method as the integrated device ID. Specifically, similar to the integrated device ID, after transistors Tr are provided on the substrate Sb, wiring layers up to the first level are formed.

[0154] After stacking to the first level, the wiring layer can be produced using photolithography. Next, the first element group of the first level is produced. First, the lower electrode, functional layer, and upper electrode are stacked in sequence and processed into the specified shape. The lower electrode, functional layer, and upper electrode become the lower electrode E1, functional layer 15, and upper electrode E2, respectively. The first element group can also be formed using photolithography. Next, by forming the wiring layer between the first and second levels and the second element group of the second level using the same steps as above, the integrated device ID4 is obtained.

[0155] In addition, Figure 16 In the illustrated example, the variable resistance element 120 and the variable resistance element 130 overlap when viewed from above in the z direction. However, the variable resistance element 120 and the variable resistance element 130 may not overlap when viewed from above in the z direction.

[0156] The change in resistance during a reset operation depends on the ratio (change ratio) of the area that changes from a crystalline state to an amorphous state relative to the overall size of the functional layers 15 and 75. The size of the area that changes from a crystalline state to an amorphous state is positively correlated with the magnitude of the pulse applied to the functional layers 15 and 75. In the functional layer 75, which is larger than the functional layer 15, by reducing the pulse magnitude, the change ratio can be finer than that of the functional layer 15.

[0157] Figure 17This is an example of a cross-sectional view of an integrated device ID5 used in a neuromorphic device according to a fifth embodiment. The materials of the functional layers 16 and 76 of the variable resistance elements 121 and 131 of the integrated device ID5 differ from those of the functional layers 15 and 75 of the variable resistance elements 120 and 130 of the integrated device ID4. Furthermore, the sizes of the lower electrodes E1 and E3 of the integrated device ID5 differ from those of the integrated device ID4. In the integrated device ID5, identical components to those of the integrated device ID4 are denoted by the same reference numerals, and their descriptions are omitted.

[0158] Like the variable resistance elements 120 and 130 , the variable resistance elements 121 and 131 are two-terminal elements. The variable resistance elements 121 and 131 are, for example, ReRAMs. Unipolar and bipolar ReRAMs, as well as filament and interface ReRAMs, can also be used. The variable resistance element 121 is, for example, a first variable resistance element in the first layer, and the variable resistance element 131 is, for example, a second variable resistance element in the second layer.

[0159] The variable resistance elements 121 and 131 each include lower electrodes E1 and E3, functional layers 16 and 76, and upper electrodes E2 and E4, stacked in the stacking direction. The lower electrode plugs E1 and E3 are connected to a via wiring VL. The upper electrodes E2 and E4 of the variable resistance elements 121 and 131 are connected to the wiring L.

[0160] The variable resistance element 131 is provided, for example, at a position farther from the substrate Sb than the variable resistance element 121. For example, at least a portion of the variable resistance element 121 overlaps with the variable resistance element 131. When viewed from above in the z-direction, all ends of the variable resistance element 121 may be located inside the entirety of the variable resistance element 131. In such a structure, both ends of the variable resistance element 121 in the longitudinal direction are located inside the ends of the variable resistance element 131 in the longitudinal direction. For example, the number of variable resistance elements 121 may be greater than that of the variable resistance element 131.

[0161] The functional layers 16 and 76 include a multi-element metal oxide containing multiple metal elements such as a perovskite-type metal oxide, or a binary metal oxide consisting of a metal element and oxygen, for example, a multi-element metal oxide or a binary metal oxide. Specifically, TiO is used as a material for the functional layers 16 and 76. x 、TaO x , HfO x , SrRuO3, graphene oxide, etc.

[0162] When a metal oxide with high insulation properties is used as the functional layers 16 and 76 , the variable resistance elements 121 and 131 exhibit, for example, a conductive filament-type operating mechanism.

[0163] In the conductive filament type action mechanism, a conductive filament is formed as a conductive path in a metal oxide. Then, a part of the conductive filament is opened and closed by the oxidation-reduction reaction of the metal oxide caused by Joule heat or electrochemical effect due to voltage application, thereby changing the resistance value of the resistance change element 121 or 131.

[0164] The specific operating principle is described below.

[0165] When a sufficiently high voltage is applied to the functional layers 16 and 76, a conductive path is formed (molded) from the upper electrodes E2 and E4 to the lower electrodes E1 and E3. Hereinafter, in this embodiment, this conductive path is referred to as a conductive filament.

[0166] Next, when a voltage is applied between the upper electrodes E2 and E4 and the lower electrodes E1 and E3, and a current is applied to the conductive filaments, a portion of the conductive filaments changes, and the resulting conductive filaments become disconnected. This clears the information in the variable resistance elements 121 and 131 (a reset operation). During the reset operation, the resistance of the functional layers 16 and 76 of the variable resistance elements 121 and 131 increases.

[0167] Then, when current flows between the upper electrodes E2 and E4 and the lower electrodes E1 and E3, conductive filaments connecting the upper electrodes E2 and E4 to the lower electrodes E1 and E3 are re-formed. In other words, data can be written to the variable resistance elements 121 and 131 (a set operation). During the set operation, the resistance value of the functional layers 16 and 76 of the variable resistance elements 121 and 131 decreases.

[0168] The number of conductive filaments formed in the functional layers 16 and 76 of the variable resistance elements 121 and 131, which demonstrate the principle of conductive filament-type operation, is not limited to one and may be multiple. The area of ​​the functional layer 76 perpendicular to the z-direction is larger than the area of ​​the functional layer 16 perpendicular to the z-direction. Therefore, the change in resistance value of the variable resistance element 131 when a single conductive filament is formed is smaller than that of the variable resistance element 121. Therefore, the resolution of the variable resistance element 131 is higher than that of the variable resistance element 121. Furthermore, the variable resistance element 121 is smaller than the variable resistance element 131, enabling higher integration.

[0169] When a semiconductor metal oxide is used as the functional layers 16 and 76 , the variable resistance elements 121 and 131 exhibit, for example, an interface-type operating mechanism.

[0170] In the interface-type operating mechanism, when a voltage is applied between the lower electrodes E2 and E4 and the upper electrodes E1 and E3, oxygen deficiency occurs near the bonding interface due to the electric field between the two electrodes, and the resistance value of the functional layers 16 and 76 changes.

[0171] The larger the element area of ​​the variable resistance elements 121 and 131, the more difficult it is to change the resistance value in the interface-type operating mechanism. Therefore, the variable resistance element 131 with a large element area consumes more power. To reduce power consumption, the number of variable resistance elements 131 can be smaller than the number of variable resistance elements 121.

[0172] The area of ​​the functional layer 16 when viewed from above in the z direction may be smaller than the area of ​​the functional layer 76, being less than 2 / 3 times or less than 1 / 2 times the area of ​​the functional layer 76. Here, the length of the functional layer 16 in the x direction and the length in the y direction may also be shorter than the length of the functional layer 76 in the x direction and the length in the y direction. By setting the area of ​​the functional layers 16 and 76 of the resistance change elements 121 and 131 to the structure as described above, it is possible to expand the area in the resistance change element 131 where resistance can be changed by conductive filament formation or interface oxidation-reduction, and the resistance value can be changed to a greater number of values. In addition, the resistance change element 121 has high integration and can be provided in a greater number than the resistance change element 131.

[0173] To improve the resolution of the variable resistance element 131, the thickness T76 of the functional layer 76 may be, for example, thicker than the thickness T16 of the functional layer 16, and may be at least 3 / 2 times or at least 2 times the thickness T16. On the other hand, if the number of variable resistance elements 131 is smaller than the number of variable resistance elements 121, the thickness T76 of the functional layer 76 of the variable resistance element 131 may be, for example, thinner than the thickness T16 of the functional layer 16 of the variable resistance element 121 to prevent the total current of the element group consisting of the variable resistance elements 131 from becoming undetectable.

[0174] The area of ​​the lower electrode E1 perpendicular to the stacking direction may be smaller than the area of ​​the lower electrode E3 perpendicular to the stacking direction, for example, 2 / 3 or 1 / 2 times the area of ​​the lower electrode E3 perpendicular to the stacking direction.

[0175] Of the main surfaces of the functional layers 16 and 76 , the surface roughness of the surface on the substrate Sb side may be rougher in the functional layer 76 than in the functional layer 16 .

[0176] The integrated device ID5 is manufactured by the same method as the integrated device ID4. Even the neuromorphic device including the integrated device ID5 can achieve the same effects as the neuromorphic device of the first embodiment.

[0177] Figure 18 This is a map diagram showing the relationship between the resistance values ​​of the first and second variable resistance elements and the number of pulses applied to the first and second variable resistance elements. Figure 18 (a) is a graph related to the first variable resistance element, Figure 18(b) is a graph about the second variable resistance element. If the number of pulses applied to the first variable resistance element and the second variable resistance element increases, the resistance values ​​of the first variable resistance element and the second magnetoresistive effect element increase. Figure 18 In the embodiment, as an example, an image diagram showing an increase in resistance value due to pulse input is shown, but in this embodiment, a resistance change element whose resistance value decreases due to pulse input can also be used. In addition, the resistance values ​​of the first resistance change element and the second resistance change element sometimes change nonlinearly with respect to the number of applied pulses. Even in such a case, the first resistance change element is larger than the second resistance change element in terms of the resistance change rate when a prescribed pulse is input. Here, even in the case of using the first resistance change element and the second resistance change element whose resistance value changes nonlinearly with respect to the number of applied pulses, the resistance change rate P is the resistance change rate shown by the same formula (1) as in the first embodiment.

[0178] P=|R′-R| / (R max -R min )···(1)

[0179] exist Figure 18 In, R max is the maximum resistance value of the first variable resistance element and the second variable resistance element, R min is the minimum value of the resistance value of the first variable resistance element and the second variable resistance element. In addition, in formula (1), |R′-R| is the maximum value of the change in the resistance value of the first variable resistance element and the second variable resistance element when a pulse is applied to the first variable resistance element and the second variable resistance element. In the first variable resistance element, |R′-R| = R2-R1, and in the second variable resistance element, |R′-R| = r2-r1. In addition, the maximum and minimum resistance values ​​of the first variable resistance element and the second variable resistance element may be the same or different.

[0180] The neuromorphic device of this embodiment also achieves the same effects as those of the neuromorphic device of the first embodiment.

[0181] (Variation)

[0182] In the above-described embodiment, the critical current density Jc2 for moving the magnetic domain walls DW of the domain wall moving element 110 may be smaller than the critical current density Jc1 for moving the magnetic domain walls DW of the domain wall moving element 100. That is, the domain wall moving elements 100 and 110 may also satisfy the relationship Jc2 < Jc1. By satisfying Jc2 < Jc1 for the domain wall moving elements 100 and 110, the power consumption required to move the magnetic domain walls DW of the domain wall moving element 110 can be reduced. The domain wall moving element 110 may require higher resolution than the domain wall moving element 100, and the number of pulses required to obtain an appropriate resistance value may be greater. The critical current densities Jc1 and Jc2 of the domain wall moving elements 100 and 110 can be varied, for example, based on the structure, shape, and material of the domain wall moving layers 10 and 50. For example, the domain wall moving elements 100 and 110 can satisfy the relationship Jc2 < Jc1 by the following method.

[0183] (Variation 1)

[0184] Figure 19 1 and 2 are cross-sectional views of the magnetic domain wall moving elements 100 and 112 taken along an xz plane passing through the y-direction centers of the magnetic domain wall moving layers 10 and 53. FIG. Figure 19 Figure 1 shows the magnetic domain wall moving element 100 belonging to the first element group and the magnetic domain wall moving element 112 belonging to the second element group. The magnetic domain wall moving element 112 includes a magnetic domain wall moving layer 52. The magnetic domain wall moving element 112 differs from the magnetic domain wall moving element 110 in that the thickness h52 of the magnetic domain wall moving layer 52 in the z direction is thinner. The remaining structures are identical to those of the magnetic domain wall moving element 110 and are designated by the same reference numerals, with description omitted. The thickness h52 of the magnetic domain wall moving layer 52 is thinner than the thickness h10 of the magnetic domain wall moving layer 10. Because the thickness h52 of the magnetic domain wall moving layer 52 is thinner than the thickness h10 of the magnetic domain wall moving layer 10, the critical current density Jc1 of the magnetic domain wall moving element 100 and the critical current density Jc2 of the magnetic domain wall moving element 112 satisfy the relationship Jc2 < Jc1.

[0185] (Variation 2)

[0186] Figure 20 The cross section of the magnetic domain wall moving element 100 and 113 is taken along the yz plane passing through the x-center of the magnetic domain wall moving layer 10 and 53. The magnetic domain wall moving element 113 belongs to the second element group. The magnetic domain wall moving element 113 includes the magnetic domain wall moving layer 53, the ferromagnetic layer 63, and the non-magnetic layer 73. The magnetic domain wall moving element 113 has a short length in the y direction and a large width w of the magnetic domain wall moving layer 53. 53 and the width w of the magnetic domain wall migration layer 10 10The relationship satisfied therebetween is different from that of the magnetic domain wall moving element 110. The same components as those of the magnetic domain wall moving element 110 are denoted by the same reference numerals, and their description is omitted.

[0187] There is a report that the critical current density of the domain wall moving layer becomes extremely small when the line width is around 70nm (for example, T. Koyama, et al., Nat. Mater. 10, 194 (2011)). Therefore, when the line width of the domain wall moving layer is 70nm or more, the width w of the domain wall moving layer 53 is 53 It is preferable that the width w of the magnetic domain wall migration layer 10 is greater than 10 When the line width of the domain wall moving layer is less than 70 nm, the width w of the domain wall moving layer 53 is 53 It is preferable that the width w of the magnetic domain wall migration layer 10 is greater than 10 By satisfying the above relationship, the critical current density of the domain wall migration layer 53 can be lower than the critical current density of the domain wall migration layer 10. Here, the width in the y direction of the domain wall migration layer refers to the average of the widths in the y direction of the upper surface and the lower surface.

[0188] Furthermore, the tilt angle θ2 of the domain wall moving layer 53 may be larger than the tilt angle θ1 of the domain wall moving layer 10. By satisfying this relationship, the critical current density Jc1 of the domain wall moving element 100 and the critical current density Jc2 of the domain wall moving element 113 satisfy the relationship Jc2 < Jc1.

[0189] (Variation 3)

[0190] Figure 21 The cross-section of the magnetic domain wall moving element 100 and 114 is taken along the xz plane passing through the y-direction center of the magnetic domain wall moving layers 10 and 54, respectively. The magnetic domain wall moving element 114 belongs to the second element group. In Modification 3, the structure of the magnetic domain wall moving element 114 differs from that of the first embodiment. In Modification 3, components identical to those of the first embodiment are designated by the same reference numerals, and their descriptions are omitted.

[0191] The magnetic domain wall moving element 114 includes a magnetic domain wall moving layer 54, a nonmagnetic layer 70, and a ferromagnetic layer 60. The magnetic domain wall moving layer 54 includes a ferromagnetic layer 541, a spacer layer 543, and a ferromagnetic layer 542. The spacer layer 543 is sandwiched between the ferromagnetic layer 541 and the ferromagnetic layer 542 in the z-direction. The ferromagnetic layers 541 and 542 can be made of the same material as the magnetic domain wall moving layer 10 and the ferromagnetic layer 20. The spacer layer 543 can contain at least one selected from the group consisting of Ru, Ir, and Rh.

[0192] The ferromagnetic layer 541 and the ferromagnetic layer 542 are magnetically coupled. For example, the ferromagnetic layer 541 and the ferromagnetic layer 542 are antiferromagnetically coupled. The ferromagnetic layer 541, the isolation layer 543, and the ferromagnetic layer 542 form a synthetic antiferromagnetic structure (SAF structure). By satisfying this relationship, the critical current density Jc1 of the magnetic domain wall moving element 100 and the critical current density Jc2 of the magnetic domain wall moving element 114 satisfy the relationship Jc2 < Jc1.

[0193] (Variation 4)

[0194] Figure 22 The cross-section of the magnetic domain wall moving elements 105 and 115 is taken along the xz plane passing through the y-center of the magnetic domain wall moving layers 10 and 50, respectively. The magnetic domain wall moving element 115 belongs to the second element group. Modification 4 differs from the first embodiment in that the magnetic domain wall moving elements 105 and 115 each include a wiring layer 40 or a wiring layer 80. In the fifth embodiment, components identical to those in the first embodiment are designated by the same reference numerals, and their descriptions are omitted.

[0195] The wiring layers 40 and 80 are in contact with the domain wall migration layers 10 and 50. The wiring layers 40 and 80 are located between the nonmagnetic layers 30 and 70 along the z-direction, sandwiching the domain wall migration layers 10 and 50. For example, the wiring layers 40 and 80 overlap with the ferromagnetic layers 20 and 60 along the z-direction. Alternatively, the wiring layer 40 may be located between the domain wall migration layers 10 and 50 and the via wiring VL.

[0196] The wiring layers 40 and 80 include any one of metal, alloy, intermetallic compound, metal boride, metal carbide, metal silicide, and metal phosphide having a function of generating a spin current by the spin Hall effect when current flows.

[0197] The wiring layers 40 and 80 contain, for example, a non-magnetic heavy metal as a main element. The main element is the element with the highest proportion among the elements constituting the wiring layers 40 and 80. The wiring layers 40 and 80 contain, for example, a heavy metal having a specific gravity greater than that of yttrium (Y). As for non-magnetic heavy metals, those with an atomic number of 39 or greater have a large atomic number and have d electrons or f electrons on the outermost shell, thereby generating a strong spin-orbit interaction. The spin Hall effect is generated by the spin-orbit interaction. In the wiring layers 40 and 80, spins are easily unevenly distributed, and spin current JS is easily generated. The wiring layers 40 and 80 contain any one selected from the group consisting of, for example, Au, Hf, Mo, Pt, W, and Ta.

[0198] The material constituting the wiring layer 80 has a larger spin Hall angle than the material constituting the wiring layer 40. The "spin Hall angle" is one indicator of the strength of the spin Hall effect, indicating the conversion efficiency of the generated spin current to the current flowing along the wiring layers 40 and 80. Specifically, the larger the absolute value of the spin Hall angle, the greater the amount of spin injected into the domain wall moving layers 10 and 50, imparting a greater spin-orbit torque (SOT) to the magnetization. By satisfying this relationship, the critical current density Jc1 of the domain wall moving element 100 and the critical current density Jc2 of the domain wall moving element 115 satisfy the relationship Jc2 < Jc1.

[0199] In the above-described embodiments and variations, an example is shown in which the lengths of all the magnetic domain wall moving elements belonging to the second element group in the longitudinal direction are longer than the lengths of all the magnetic domain wall moving elements belonging to the first element group in the longitudinal direction, but the present invention is not limited to this example. For example, the first element group and the second element group may both include magnetic domain wall moving elements 100 and magnetic domain wall moving elements 110, and the ratio of magnetic domain wall moving elements 110 in the first element group is higher than the ratio of magnetic domain wall moving elements 110 in the second element group. Such a structure also improves the resolution of the second element group, improves the recognition rate of the neuromorphic device in the first element group, and improves the integration of magnetic domain wall moving elements.

[0200] Furthermore, the aforementioned embodiments and variations assume the reproduction of a neural network NN in which the weights applied to the transmission units are varied more finely as layers closer to the output layer Lout are described. In other words, the reproduction assumes that layers closer to the output layer Lout require higher resolution. However, the present invention is not limited to this example. In neural networks, layers closer to the input layer may require higher resolution. To reproduce such a neural network, the neuromorphic device of this embodiment may also have a structure in which the first element group is located closer to the input layer than the second element group.

[0201] While several embodiments have been presented so far, providing examples of integrated devices and neuromorphic devices, the present invention is not limited to these embodiments, and various modifications are possible without departing from the spirit of the invention.

[0202] For example, Figure 23 FIG is a cross-sectional view of the magnetic domain wall moving element 106 of the first element group and the magnetic domain wall moving element 116 of the second element group of the neuromorphic device of Modification 5. Figure 23 As shown, the domain wall migration layers 10 and 50 may be located farther from the substrate Sb than the ferromagnetic layers 20 and 60 . Figure 23 The ferromagnetic layers 20 and 60 having a high magnetization stability are located on the substrate Sb side, which is referred to as an underpinning structure. Figure 23 The bottom pin structure shown is similar to Figure 8The magnetization stability is high compared to the top pin structure shown.

[0203] In addition, the characteristic structures of the above-mentioned embodiment and modified examples may be combined.

[0204] Explanation of symbols

[0205] 10, 50, 51, 52, 53, 54…magnetic domain wall migration layer

[0206] 20, 60, 61, 63… ferromagnetic layer

[0207] 30, 70, 71, 73…non-magnetic layer

[0208] 15, 16, 75, 76…functional layers

[0209] 40, 80…wiring layer

[0210] 100, 104, 105, 110, 111, 112, 113, 114, 115... magnetic domain wall moving elements

[0211] 120, 121, 130, 131…variable resistance elements

[0212] 543…Isolation layer

[0213] A1…first magnetic sector

[0214] A2…Second magnetic area

[0215] C…chip

[0216] C1…first circuit

[0217] C2…Second circuit

[0218] C3…third circuit

[0219] CL…Common Line

[0220] CW...Connection wiring

[0221] DW…Domain Wall

[0222] E…Electrode

[0223] ID, ID1, ID2, ID3... integrated device

[0224] In, In1, In2, In3, In4…insulators

[0225] Lin…input layer, L m …the middle layer

[0226] Lout…output layer

[0227] LS...Laminated structure

[0228] NN…Neural Network

[0229] RL, RL1, RL2…readout lines

[0230] VL…Through-hole wiring

[0231] VSW…Vertical Switching Element

[0232] θ1, θ2…inclination angle

Claims

1. A neuromorphic device comprising a first element group and a second element group, The first element group and the second element group each include a plurality of magnetic domain wall moving elements. Each of the plurality of magnetic domain wall moving elements includes a magnetic domain wall moving layer, a ferromagnetic layer, and a non-magnetic layer sandwiched between the magnetic domain wall moving layer and the ferromagnetic layer. The length of the magnetic domain wall movement layer of each of the magnetic domain wall movement elements belonging to the first element group in the longitudinal direction is shorter than the length of the magnetic domain wall movement layer of each of the magnetic domain wall movement elements belonging to the second element group in the longitudinal direction. The resistance change rate when a predetermined pulse is inputted is larger in each of the magnetic domain wall moving elements belonging to the first element group than in each of the magnetic domain wall moving elements belonging to the second element group. The first element group and the second element group are in a stacked structure. The stacked structure is stacked on a substrate.

2. The neuromorphic device according to claim 1, wherein In the stacked structure, the second element group is located farther from the substrate than the first element group.

3. The neuromorphic device according to claim 1, wherein The number of the magnetic domain wall moving elements belonging to the first element group is greater than the number of the magnetic domain wall moving elements belonging to the second element group.

4. The neuromorphic device according to any one of claims 1 to 3, wherein: The critical current density required for moving the magnetic domain walls of the magnetic domain wall moving elements of the second element group is smaller than that of the magnetic domain wall moving elements of the first element group.

5. The neuromorphic device according to any one of claims 1 to 3, wherein: When viewed in plan from the stacking direction of the magnetic domain wall movement elements, any one of the magnetic domain wall movement elements in the first element group and any one of the magnetic domain wall movement elements in the second element group at least partially overlap.

6. The neuromorphic device according to any one of claims 1 to 3, wherein: In the longitudinal direction of any of the magnetic domain wall moving elements belonging to the second element group, both ends of any of the magnetic domain wall moving elements in the first element group are located inside both ends of any of the magnetic domain wall moving elements in the second element group.

7. The neuromorphic device according to any one of claims 1 to 3, wherein: When viewed in plan from the stacking direction of the magnetic domain wall movement elements, the magnetic domain wall movement elements of the first element group and the magnetic domain wall movement elements of the second element group do not overlap.

8. The neuromorphic device according to any one of claims 1 to 3, wherein: The surface roughness of the lower surface of the magnetic domain wall moving element belonging to the second element group is greater than the surface roughness of the lower surface of the magnetic domain wall moving element belonging to the first element group.

9. The neuromorphic device according to any one of claims 1 to 3, wherein: A connection line is further provided for connecting any one of the magnetic domain wall moving elements in the first element group and any one of the magnetic domain wall moving elements in the second element group.

10. The neuromorphic device according to any one of claims 1 to 3, wherein: The first element group performs a first product-sum operation, and the second element group performs a second product-sum operation, A total of outputs from a plurality of magnetic domain wall moving elements belonging to the first element group is input to the magnetic domain wall moving element belonging to the second element group.

11. The neuromorphic device according to any one of claims 1 to 3, wherein: A pulse length of a write pulse input to the magnetic domain wall moving element belonging to the second element group is different from a pulse length of a write pulse input to the magnetic domain wall moving element belonging to the first element group.

12. The neuromorphic device according to any one of claims 1 to 3, wherein: The pulse amplitude of the write pulse input to the magnetic domain wall moving element belonging to the second element group is different from the pulse amplitude of the write pulse input to the magnetic domain wall moving element belonging to the first element group.

13. The neuromorphic device according to any one of claims 1 to 3, wherein: The first component group and the second component group are respectively responsible for operations between different layers in the neural network.

14. The neuromorphic device according to any one of claims 1 to 3, wherein: The first component group and the second component group are responsible for operations between different layers of the neural network. The output of the product-sum operation of the first element group is input to the second element group, and the second element group uses the result of the product-sum operation of the first element group as input to perform further product-sum operation.

Images