KR102577592B1 - Neuromorphic device - Google Patents

Abstract

A neuromorphic device is provided. A neuromorphic device according to an embodiment of the present invention, a neuromorphic device according to an embodiment of the present invention for solving the above problem, includes a substrate; a first electrode and a second electrode extending in a vertical direction on the substrate and spaced apart from each other; A laminated structure located between the first electrode and the second electrode and comprising two or more reactive metal layers and one or more insulating layers alternately stacked - wherein the reactive metal layer reacts with oxygen ions in the oxygen retention layer to form an insulating oxide layer. Can do it. - ; the oxygen retaining layer interposed between the laminated structure and the first electrode; and an oxygen diffusion-blocking layer interposed between the laminated structure and the oxygen-retaining layer to prevent oxygen ions from moving from the oxygen-retaining layer to the reactive metal layer, wherein the oxygen-diffusion-blocking layer has a thickness in the vertical direction. It can be variable.

Description

Neuromorphic device {NEUROMORPHIC DEVICE}

This patent document relates to neuromorphic devices that simulate the human nervous system and their applications.

Recently, with the miniaturization, lower power consumption, higher performance, and diversification of electronic devices, there is a demand for technology that can efficiently process large amounts of information. In particular, interest in neuromorphic technology that mimics the human nervous system is increasing. The human nervous system contains hundreds of billions of neurons and synapses, which are junctions between neurons. Neuromorphic technology seeks to implement a neuromorphic device by designing neuron circuits and synapse circuits that respectively correspond to these neurons and synapses. Neuromorphic devices can be used in various fields such as data classification and pattern recognition.

The problem to be solved by embodiments of the present invention is to provide a neuromorphic device including a synapse with improved symmetry and linearity.

A neuromorphic device according to an embodiment of the present invention for solving the above problem includes: a substrate; a first electrode and a second electrode extending in a vertical direction on the substrate and spaced apart from each other; A laminated structure located between the first electrode and the second electrode and comprising two or more reactive metal layers and one or more insulating layers alternately stacked - wherein the reactive metal layer reacts with oxygen ions in the oxygen retention layer to form an insulating oxide layer. Can do it. - ; the oxygen retaining layer interposed between the laminated structure and the first electrode; and an oxygen diffusion-blocking layer interposed between the laminated structure and the oxygen-retaining layer to prevent oxygen ions from moving from the oxygen-retaining layer to the reactive metal layer, wherein the oxygen-diffusion-blocking layer has a thickness in the vertical direction. It can be variable.

In the above embodiment, the thickness of the oxygen diffusion blocking layer may increase from top to bottom. The thickness of the oxygen diffusion blocking layer may decrease from top to bottom. The second electrode may include a plurality of second electrodes spaced apart from each other, and the laminated structure may include a plurality of laminated structures corresponding to each of the plurality of second electrodes and spaced apart from each other. It may further include an interlayer insulating layer buried between the plurality of second electrodes and between the plurality of laminate structures. It may further include a slit penetrating the interlayer insulating layer and extending in a direction crossing from the first electrode to between the plurality of second electrodes and between the plurality of laminated structures. It may further include a buried layer that buries the slit and includes the same material as at least one of the oxygen retention layer and the oxygen diffusion blocking layer. The thickness of the reactive metal layer may be different. The thickness of the reactive metal layer may increase from top to bottom. The thickness of the reactive metal layer may decrease from top to bottom. The oxygen retention layer may surround a side of the first electrode. The oxygen diffusion blocking layer may surround a side of the oxygen retention layer. When a reset voltage of the first polarity is applied through the first electrode and the second electrode, a reset operation in which an insulating oxide layer is created or its thickness increases in each of the reactive metal layers at the interface with the oxygen diffusion prevention layer is performed. When a set voltage of a second polarity different from the first polarity is applied through the first electrode and the second electrode, a set operation in which the insulating oxide layer disappears or its thickness decreases may be performed. . The insulating oxide layer can be created most quickly within the reactive metal layer corresponding to the area where the oxygen diffusion blocking layer has the smallest thickness. The thicknesses of the reactive metal layers are different from each other, and the insulating oxide layer can be created most quickly within the reactive metal layer having the smallest thickness. As the number or thickness of the insulating oxide layers increases, conductivity may decrease, and as the number or thickness of the insulating oxide layers decreases, conductivity may increase. During the reset operation, as the number of pulses of the reset voltage increases, the number or thickness of the insulating oxide layers increases, and during the set operation, as the number of pulses of the set voltage increases, the number or thickness of the insulating oxide layers increases. may decrease. During the reset operation, the pulse of the reset voltage may have a constant width and a constant size, and during the set operation, the pulse of the set voltage may have a constant width and a constant size. The oxygen diffusion blocking layer may have a thickness that does not completely block the movement of oxygen ions. The oxygen diffusion blocking layer may include an insulating material, a semiconductor material, or a combination thereof.

According to the above-described embodiments of the present invention, the symmetry and linearity of synapses can be improved, and thus the operating characteristics of the neuromorphic device can be improved.

1 is a diagram for explaining a neuromorphic device and its operating method according to an embodiment of the present invention.
Figures 2A to 2D are diagrams for explaining characteristics required for each synapse in Figure 1.
Figure 3A is a cross-sectional view for explaining the synapse of the comparative example, and Figures 3B and 3C are diagrams for explaining the characteristics of the synapse of Figure 3A.
4A and 4B are diagrams for explaining a neuromorphic device according to an embodiment of the present invention.
Figures 5A to 5C are cross-sectional views for explaining the reset operation of the synapse of Figures 4A and 4B.
Figure 6 is a plan view showing a neuromorphic device according to another embodiment of the present invention.
FIGS. 7A to 10B are diagrams for explaining an example of a method of manufacturing the neuromorphic device of FIGS. 4A and 4B.
Figure 11 is a cross-sectional view showing a neuromorphic device according to another embodiment of the present invention.
Figure 12 is a cross-sectional view showing a neuromorphic device according to another embodiment of the present invention.
Figure 13 is a graph showing the change in current depending on the contact area and/or opposing area of the reactive metal layer and the oxygen retaining layer.
Figure 14 is a cross-sectional view showing a neuromorphic device according to another embodiment of the present invention.
Figure 15 is an example of a pattern recognition system according to an embodiment of the present invention.

Below, various embodiments are described in detail with reference to the attached drawings.

The drawings are not necessarily drawn to scale, and in some examples, the proportions of at least some of the structures shown in the drawings may be exaggerated to clearly show features of the embodiments. When a multi-layer structure having two or more layers is disclosed in the drawings or detailed description, the relative positional relationship or arrangement order of the layers as shown only reflects a specific embodiment and the present invention is not limited thereto, and the relative positions of the layers Relationships and arrangement order may vary. Additionally, drawings or detailed descriptions of multi-story structures may not reflect all layers present in a particular multi-story structure (eg, one or more additional layers may exist between the two layers shown). For example, when a first layer is on a second layer or on a substrate in a multilayer structure in the drawings or detailed description, it indicates that the first layer can be formed directly on the second layer or directly on the substrate. In addition, it may also indicate the case where one or more other layers exist between the first layer and the second layer or between the first layer and the substrate.

1 is a diagram for explaining a neuromorphic device and its operating method according to an embodiment of the present invention.

Referring to FIG. 1, a neuromorphic device according to an embodiment of the present invention includes a plurality of presynaptic neurons (10), a plurality of postsynaptic neurons (20), and a plurality of presynaptic neurons ( 10) and a synapse 30 that provides each connection between a plurality of post-synaptic neurons 20.

The neuromorphic device of this embodiment includes 4 pre-synaptic neurons 10, 4 post-synaptic neurons 20, and 16 synapses 30, but these numbers may vary. The number of pre-synaptic neurons 10 is N (where N is a natural number of 2 or more), and the number of post-synaptic neurons 20 is M (where M is a natural number of 2 or more and may be equal to or different from N). In this case, N*M synapses 30 may be arranged in a matrix form. To this end, a wiring 12 connected to each of the plurality of pre-synaptic neurons 10 and extending in a first direction, for example, the horizontal direction, and a wiring 12 connected to each of the plurality of post-synaptic neurons 20 and intersecting the first direction. The wiring 22 may be provided extending in two directions, for example, vertically. Hereinafter, for convenience of explanation, the wiring 12 extending in the first direction will be referred to as a row line, and the wiring 22 extending in the second direction will be referred to as a column line. A plurality of synapses 30 may be disposed at each intersection of the row wiring 12 and the column wiring 22 to connect the corresponding row wiring 12 and the corresponding column wiring 22 to each other.

The pre-synaptic neuron 10 generates a signal, for example, a signal corresponding to specific data and sends it to the row wiring 12, and the post-synaptic neuron 20 transmits the synaptic signal that has passed through the synaptic element 30 to the column wiring. It can perform the role of receiving and processing through (22). The row wiring 12 may correspond to the axon of the pre-synaptic neuron 10, and the column wiring 22 may correspond to the dendrite of the post-synaptic neuron 20. However, whether a neuron is pre-synaptic or postsynaptic can be determined by its relative relationship to other neurons. For example, when the pre-synaptic neuron 10 receives synaptic signals in relationship with other neurons, it can function as a post-synaptic neuron. Similarly, a postsynaptic neuron 20 can function as a pre-synaptic neuron if it sends signals in relationships with other neurons. The pre-synaptic neuron 10 and the post-synaptic neuron 20 may be implemented with various circuits such as CMOS.

The connection between the pre-synaptic neuron (10) and the post-synaptic neuron (20) may be made through the synapse (30). Here, the synapse 30 is an element whose electrical conductance or weight changes depending on electrical pulses, such as voltage or current, applied to both ends.

The synapse 30 may include, for example, a variable resistance element. A variable resistance element is an element that can switch between different resistance states depending on the voltage or current applied to both ends, and is made of various materials that can have multiple resistance states, such as transition metal oxides and perovskite systems. It may have a single-layer structure or a multi-layer structure including metal oxides such as metal oxides, phase change materials such as chalcogenide-based materials, ferroelectric materials, and ferromagnetic materials. The operation in which the variable resistance element and/or synapse 30 changes from a high resistance state to a low resistance state is called a set operation, and the operation in which it changes from a low resistance state to a high resistance state can be called a reset operation. .

However, the synapse 30 of the neuromorphic device, unlike variable resistance elements used in memory devices such as RRAM, PRAM, FRAM, and MRAM, does not have an abrupt change in resistance during set and reset operations, and the input It can be implemented to have various characteristics that are distinct from variable resistance elements in memory, such as showing analog behavior in which conductivity gradually changes depending on the number of electrical pulses. This is because the characteristics required for the variable resistance element in the memory and the characteristics required for the synapse 30 in the neuromorphic device are different from each other. For reference, it is desirable for the variable resistance element used in memory to maintain its conductivity before a set operation or reset operation is performed even if electrical pulses are repeatedly applied. This is to store different data by clearly distinguishing between low-resistance states and high-resistance states. The characteristics of the synapse 30 suitable for a neuromorphic device will be described in more detail with reference to FIGS. 2A to 2D described later.

Meanwhile, an exemplary explanation of the learning operation of the above neuromorphic device is as follows. For convenience of explanation, the row wiring 12 is referred to as a first row wiring 12A, a second row wiring 12B, a third row wiring 12C, and a fourth row wiring 12D from the top, and is called a column wiring. (22) can be referred to as the first column wiring 22A, the second column wiring 22B, the third column wiring 22C, and the fourth column wiring 22D from the left.

In the initial state, all of the synapses 30 may be in a high-resistance state. If at least a portion of the synapse 30 is in a low-resistance state, an additional initialization operation may be required to bring them into a high-resistance state. Each synapse 30 may have a predetermined threshold. More specifically, when a voltage or current smaller than a predetermined threshold is applied to both ends of each synapse 30, the conductivity of the synapse 30 does not change, and when a voltage or current greater than a predetermined threshold is applied to the synapse 30, the conductivity of the synapse 30 does not change. The conductance of synapse 30 may change.

In this state, in order to perform an operation of learning specific data to a specific column wire 22, an input signal corresponding to specific data may enter the row wire 12. The input signal may appear as the application of an electrical pulse to each row wire 12. As an example, when an input signal corresponding to data of '0011' is input to the row wiring 12, the row wiring 12 corresponding to '0', for example, the first and second row wirings 12A and 12B, have electrical A pulse is not applied, and an electrical pulse may be applied only to the row wiring 12 corresponding to '1', for example, the third and fourth row wirings 12C and 12D.

At this time, the column wiring 22 may be driven with an appropriate voltage or current for learning.

As an example, when the column wire 22 to learn specific data has already been determined, the synapse 30 located at the intersection of the column wire 22 with the row wire 12 corresponding to '1' performs the set operation. The synapse 30 may be driven to receive a voltage greater than the required voltage (hereinafter referred to as set voltage), and the remaining column wiring 22 may be driven to allow the remaining synapses 30 to receive a voltage smaller than the set voltage. For example, if the size of the set voltage is Vset and the column wire 22 from which data of '0011' is to be learned is set to the third column wire 22C, the third column wire 22C and the third and fourth rows The size of the electrical pulse applied to the third and fourth row wires 12C and 12D so that the first and second synapses 30A and 30B located at the intersection with the wires 12C and 12D receive a voltage greater than Vset. may be greater than Vset and the voltage applied to the third column wiring 22C may be 0V. Accordingly, the first and second synapses 30A and 30B may be in a low resistance state. The conductivity of the first and second synapses 30A and 30B in a low-resistance state may gradually increase as the number of electrical pulses increases. Except for the first and second synapses 30A and 30B, the remaining synapses 30 are connected to the remaining column wirings, that is, the first, second and fourth column wirings 22A, 22B and 22D, so that a voltage smaller than Vset is applied. The applied voltage may have a value between 0V and Vset, for example, 1/2Vset. Accordingly, the resistance state of the remaining synapses 30 except for the first and second synapses 30A and 30B may not change.

When the row wiring 12 and the column wiring 22 are driven in the above manner, the synapse 30 that receives the electrical pulse, for example, the third and fourth row wirings 12C and 12D and the third column wiring 22C As the conductivity of the first and second synapses 30A and 30B located at the intersection gradually increases, the current flowing through them to the third column wiring 22C may increase. By measuring the current flowing in the third column wire 22C, when this current reaches a predetermined threshold current, the third column wire 22C is a 'column wire that has learned specific data', for example, a 'column wire that has learned data of '0011'. It can be column wiring.

As another example, the column wiring 22 for learning specific data may not be determined. In this case, while applying an electrical pulse corresponding to specific data to the row wire 12, the current flowing through each column wire 22 is measured, and the column wire 22 that reaches a predetermined threshold current first learns the specific data. It may be a column wiring 22.

By using the method described above, different data can be learned on different column wires 22, respectively.

Meanwhile, in the above learning process, only the set operation for changing the synapse 30 in the high-resistance state to the low-resistance state and the operation for increasing its conductance were explained, but if necessary, the synapse 30 in the low-resistance state can be changed. A reset operation to change the high resistance state again and an operation to reduce the conductivity may be required. For example, this is the case with the initialization operation described above. The polarity of the pulse applied during the set operation and conductance increase of the synapse 30 may be opposite to the polarity of the pulse applied during the reset operation and conductance decrease of the synapse 30. The operation of increasing the conductance of the synapse 30 can be referred to as a potentiation operation, and the operation of decreasing the conductance of the synapse 30 can be referred to as a depression operation.

Meanwhile, the characteristics of the synapse 30 suitable for the above neuromorphic device will be described in more detail with reference to FIGS. 2A to 2D below.

FIGS. 2A to 2D are diagrams for explaining characteristics required for each synapse 30 in FIG. 1 .

Specifically, FIGS. 2A and 2B show the conductance (G) of the synapse 30 according to the number of electrical pulses input to the synapse 30. Figure 2c shows the synapse weight change rate (△W) according to the resistance (R) or conductance (G) of the synapse 30. Figure 2d shows the weight change rate (ΔW) of the synapse 30 according to the magnitude of the voltage (V) applied to the synapse 30.

Referring to FIGS. 2A and 2B, when a voltage pulse of the first polarity having a predetermined threshold value or more, for example, a negative voltage pulse, is repeatedly applied to the synapse 30 in a low-resistance state, the synapse 30 Conductivity (G) can gradually increase. The direction in which the conductance (G) of the synapse 30 increases can be referred to as the G+ direction or the potentiation direction.

When a voltage pulse of a second polarity, for example, a positive voltage pulse, having a magnitude greater than the reset voltage is applied to the synapse 30, a reset operation in which the resistance state of the synapse 30 changes to a high resistance state may be performed.

When voltage pulses of the second polarity are repeatedly applied to the synapse 30 in a high-resistance state, the conductance (G) of the synapse 30 may gradually decrease. The direction in which the conductance (G) of the synapse 30 decreases can be referred to as the G- direction or the depression direction.

When a voltage pulse of the first polarity greater than the set voltage is applied to the synapse 30 again, a set operation in which the resistance state of the synapse 30 changes to a low resistance state may be performed.

At this time, even if the size and width of the pulse are constant, it may be desirable for the conductance (G) of the synapse 30 to be substantially symmetrical and the rate of change of the conductance (G) to be substantially constant in the strengthening and suppressing operations. In other words, it may be desirable for the conductance (G) of the synapse 30 to have linearity and symmetry in the strengthening and suppressing operations. This is to prevent sudden changes in resistance of the synapse 30 during set and reset operations. It is also possible to consider securing the linearity and symmetry of the synapse 30 as described above by varying the size and/or width of the pulse, but since it requires the implementation of additional circuits to generate various pulses, it is disadvantageous in terms of area and power. can do. Therefore, it is desirable that the size and width of the pulse when driving the main synapse 30 are controlled to be constant.

In strengthening and suppressing operations, the conductance (G) of the synapse 30 is required to have linearity and symmetry regardless of whether the weight change rate is small (see △W1) or the weight change rate is large (see △W2). It can be. However, if the size or width of the pulse is not sufficient, the conductivity (G) of the synapse 30 may not change regardless of the number of pulses.

Referring to Figure 2c, regardless of the current state of the synapse 30, that is, regardless of the current resistance (R) or current conductance (G) of the synapse 30, the weight change rate (△W) of the synapse 30 It may be desirable for it to be substantially constant.

Referring to FIG. 2D, the weight and/or conductance of the synapse 30 may not be changed at a voltage below a predetermined threshold, that is, V3. That is, the weight change rate (△W) of the synapse 30 may be 0. On the other hand, at a voltage greater than a predetermined threshold, for example, a voltage greater than V4, the weight change rate (ΔW) of the synapse 30 may increase. At this time, the weight change rate (ΔW) of the synapse 30 may increase substantially in proportion to the magnitude of the voltage.

In summary, the synapse 30 of the neuromorphic device has the characteristic that the conductance (G) increases or decreases substantially in proportion to the number of electrical pulses, regardless of the current state of the synapse 30, and during strengthening operation. It is desirable for the conductance (G) and the conductance (G) during suppression operation to be substantially symmetrical. Here, it is desirable that the change in conductance (G) of the synapse 30 is possible only above a predetermined threshold. This is because the closer the characteristics of the synapse 30 are, the better the operation characteristics can be, such as increasing the learning and recognition accuracy of the neuromorphic device.

In this embodiment, we aim to implement a synapse that can satisfy the above characteristics as much as possible. Before explaining the present embodiment, the synapse of the comparative example will first be described.

Figure 3A is a cross-sectional view for explaining the synapse of the comparative example, and Figures 3B and 3C are diagrams for explaining the characteristics of the synapse of Figure 3A.

Referring to FIG. 3A, the synapse 100 of the comparative example includes a first electrode 110, a second electrode 140, and an oxygen retention layer 120 interposed between the first electrode 110 and the second electrode 140. ), and a reactive metal layer 130 interposed between the oxygen retention layer 120 and the second electrode 140 and capable of reacting with the oxygen of the oxygen retention layer 120.

The first and second electrodes 110 and 140 are ends of the synapse 100 for applying voltage or current, and may be formed of various conductive materials such as metal and metal nitride. The first electrode 110 may be connected to any one of the row wiring 12 and the column wiring 22 of FIG. 1, and the second electrode 140 may be connected to the row wiring 12 and the column wiring 22 of FIG. 1. By being connected to another one of the synapses 100, the synapse 100 can be driven with an electrical pulse. At least one of the first and second electrodes 110 and 140 may be omitted, and in this case, the row wiring 12 or the column wiring 22 replaces the first electrode 110 or the second electrode 140. can do.

The oxygen retention layer 120 is a layer containing oxygen and may include oxides of various metals, such as oxides of transition metals such as Ti, Ni, Al, Nb, Hf, and V, and perovskite-based oxides such as PCMO and LCMO. You can.

The reactive metal layer 130 is a layer that can react with oxygen ions to form an insulating oxide, and may include metals such as Al, Ti, Ta, Mo, or nitrides of these metals.

In the initial state, the synapse 100 may be in a relatively low resistance state. In order to operate the above-described neuromorphic device, an initialization operation that places the synapse 100 in a high-resistance state may be necessary.

When a voltage pulse of a predetermined polarity is applied to the synapse 100 in a low-resistance state through the first and second electrodes 110 and 140, oxygen ions in the oxygen retention layer 120 move toward the reactive metal layer 130 and become reactive. By reacting with the metal included in the metal layer 130, an insulating oxide layer can be formed at the interface with the reactive metal layer 130. As a result, the resistance state of the synapse 100 may be changed to a high resistance state. As the number of voltage pulses increases, the thickness of the insulating oxide layer increases, so the conductivity of the synapse 100 may gradually decrease.

Conversely, when a voltage pulse of opposite polarity is applied to the synapse 100 in a high resistance state, the oxygen ions in the oxygen retention layer 120 move to the opposite side of the reactive metal layer 130, so the thickness of the previously formed insulating oxide layer is thin. You can lose. As a result, the resistance state of the synapse 100 may be changed to a low resistance state. As the number of voltage pulses increases, the thickness of the insulating oxide layer decreases, so the conductivity of the synapse 100 may gradually increase.

In this way, since the thickness of the oxide layer gradually increases or decreases due to the voltage pulse while the synapse 100 switches between the high-resistance state and the low-resistance state, the synapse 100 in each of the high-resistance state and the low-resistance state Analog behavior can be seen where the conductivity changes gradually. However, it may be insufficient to satisfy the synapse characteristics described in FIGS. 2A to 2D. This will be explained with reference to FIGS. 3B and 3C.

Referring to Figure 3b, when a voltage pulse of the first polarity is applied to the synapse 100 in a low-resistance state, it can be seen that the conductivity (G) of the synapse 100 gradually increases as the number of voltage pulses increases. there is. However, it can be seen that the rate of increase in conductivity (G) is very large at the beginning of the set operation when it changes to a low resistance state, and decreases over time. Therefore, there is a problem that the linearity of the synapse 100 is not satisfied.

Additionally, when a voltage pulse of the second polarity having a magnitude greater than the reset voltage is applied to the synapse 100 in a low-resistance state, a reset operation in which the resistance state of the synapse 100 changes to a high-resistance state may be performed. As the number of voltage pulses applied to the synapse 100 in a high-resistance state increases, the conductivity (G) of the synapse 100 may gradually decrease. However, it can be seen that a rapid decrease in conductivity (G) occurs during the reset operation. Additionally, it can be seen that the rate of decrease in conductivity (G) is very large at the beginning of the reset operation and decreases over time. The degree of decrease in conductivity (G) at the beginning of the reset operation may be much greater than the degree of increase in conductivity (G) at the beginning of the set operation. Therefore, there is a problem in that the linearity and symmetry of the synapse 100 are not satisfied.

Referring to FIG. 3C, it can be seen that the weight change rate (ΔW) of the synapse 100 is not constant, depending on the current resistance (R) of the synapse 100. In the G+ direction, when the current resistance (R) of the synapse 100 is relatively large, for example, when it has a value of R5 or R6, the weight change rate (ΔW) of the synapse 100 may increase. In other words, the rate of change in conductivity of the synapse 100 may be large at the beginning of the set operation when the resistance (R) is relatively large. Conversely, in the G- direction, when the current resistance (R) of the synapse 100 is relatively small, for example, when it has a value of R1, the weight change rate (△W) of the synapse 100 may increase. In other words, the rate of change in conductivity of the synapse 100 may be large at the beginning of the reset operation when the resistance (R) is relatively small. This shows that the linearity of the synapse 100 is not satisfied.

In addition, it can be seen that the weight change rate (△W) in the G- direction is greater than the weight change rate (△W) in the G+ direction at the beginning of the operation. In other words, it shows that the symmetry of the synapse 100 is not satisfied.

The reason why the above problem occurs is that the rate of change in resistance of the synapse 100 according to the number of pulses is large at the beginning of each of the set and reset operations, and the reset operation in which the interface oxide layer is created compared to the set operation in which the interface oxide layer disappears. This is because the speed is much greater.

This embodiment seeks to provide a neuromorphic device that can solve the problems of the comparative example.

FIGS. 4A and 4B are diagrams for explaining a neuromorphic device according to an embodiment of the present invention, with FIG. 4A showing a plan view and FIG. 4B showing a cross-sectional view taken along line A-A' of FIG. 4A.

Referring to FIGS. 4A and 4B, the neuromorphic device according to an embodiment of the present invention includes a substrate 200, extends on the substrate 200 in a direction substantially perpendicular to the upper surface of the substrate 200, and extends on the substrate 200 in a direction substantially perpendicular to the upper surface of the substrate 200. A first electrode 210 and a second electrode 240 arranged to be spaced apart, a plurality of reactive metal layers 230 and a plurality of reactive metal layers 230 located between the first electrode 210 and the second electrode 240 and on the substrate 200. A stacked structure (ST) in which insulating layers 235 are alternately stacked, and oxygen positioned between the stacked structure (ST) and the first electrode 210 and extending substantially vertically along the first electrode 210. It may include a retention layer 220 and an oxygen diffusion blocking layer 250. Here, the oxygen diffusion blocking layer 250 is formed relatively adjacent to the stacked structure (ST) compared to the oxygen holding layer 220, and the oxygen holding layer 220 is formed relatively close to the oxygen diffusion blocking layer 250. 1 It may be formed adjacent to the electrode 210.

First, the substrate 200 may include a required lower structure (not shown). The lower structure may include wiring connected to the first electrode 210 and/or the second electrode 240, a neuron circuit, etc.

The first and second electrodes 210 and 240 are both ends of a synapse (SN) for applying voltage or current, and may include various conductive materials such as metal and metal nitride. The first electrode 210 may be connected to any one of the row wiring 12 and the column wiring 22 of FIG. 1, and the second electrode 240 may be connected to the row wiring 12 and the column wiring 22 of FIG. 1. By connecting with another one, the synapse (SN) can be driven with an electrical pulse. To this end, row wiring 12 and/or column wiring 22 may be formed in the substrate 200 to connect to the bottom of the first electrode 210 and/or the second electrode 240 and drive them. . As another example, the row wiring 12 and/or the column wiring 22 may be formed to connect to the top of the first electrode 210 and/or the second electrode 240.

The first and second electrodes 210 and 240 may have a pillar shape extending in a vertical direction from the substrate 200. In the plan view of FIG. 4A, the first electrode 210 is shown to have a circular shape and the second electrode 240 is shown to have a square shape. However, this embodiment is not limited thereto, and the first electrode ( The planar shapes of 210) and the second electrode 240 may be modified in various ways. In addition, in the top view of FIG. 4A, one first electrode 210 and four second electrodes 240 that face the one first electrode 210 in four directions and are separated from each other are shown, but in this embodiment The example is not limited to this, and the number of first electrodes 210 and second electrodes 240 may be varied in various ways. When one first electrode 210 and a plurality of second electrodes 240 face each other, the plurality of second electrodes 240 may be separated from each other.

The reactive metal layer 230 is a layer that can react with oxygen ions to form an insulating oxide, and may include, for example, metals such as Al, Ti, Ta, Mo, or nitrides of these metals. In the cross-sectional view of FIG. 4B, three reactive metal layers 230 stacked in a vertical direction are shown. However, this embodiment is not limited to this, and the number of reactive metal layers 230 may be variously modified provided that there is a plurality. . The thickness of the plurality of reactive metal layers 230 may be substantially the same. Accordingly, the contact areas of the plurality of reactive metal layers 230 and the oxygen diffusion blocking layer 250 may be substantially the same. An insulating layer 235 may be interposed between reactive metal layers 230 adjacent to each other in the vertical direction to electrically separate the adjacent reactive metal layers 230 from each other. The insulating layer 235 may include various insulating materials such as silicon oxide, silicon nitride, or a combination thereof. Two or more reactive metal layers 230 and an insulating layer 235 interposed between reactive metal layers 230 adjacent to each other in the vertical direction may form a stacked structure (ST).

The laminated structure ST may be formed between the first electrode 210 and the second electrode 240 to correspond to each of the second electrodes 240 . For example, in the plan view of FIG. 4A, four stacked structures (ST) may be formed, one side of the four stacked structures (ST) may be in contact with the four second electrodes 240, respectively, and the four stacked structures (ST) may be formed. The other side of (ST) may face one first electrode 210 in four directions. The four stacked structures (ST) can be separated from each other.

In order to electrically separate the plurality of second electrodes 240 and the plurality of stacked structures (ST), one second electrode 240 and the corresponding stacked structure (ST) and another second electrode adjacent thereto The area between 240 and the corresponding stacked structure ST may be filled with an interlayer insulating layer 270 . Furthermore, for more reliable separation of the second electrode 240 and the stacked structure ST, a slit S may be formed in the interlayer insulating layer 270. This slit (S) is located in a direction crossing between one of the second electrodes 240 and the corresponding laminated structure (ST) and the other second electrode 240 and the corresponding laminated structure (ST) adjacent thereto. can be formed. For example, when the four second electrodes 240 and the stacked structure ST have a shape extending in the horizontal and vertical directions from the first electrode 210, the slit S extends diagonally from the first electrode 210. It may have a shape extending to . A buried layer 280 containing an insulating material or the like may be formed within the slit S.

The oxygen retaining layer 220 is a layer containing oxygen. For example, it may include oxides of various metals, such as oxides of transition metals such as Ti, Ni, Al, Nb, Hf, and V, and perovskite-based oxides such as PCMO and LCMO. In the plan view of FIG. 4A, the oxygen retaining layer 220 is shown surrounding the side of the first electrode 210, but the present embodiment is not limited thereto, and the oxygen retaining layer 220 is surrounded by the first electrode 210. ) It is sufficient that it extends in a substantially vertical direction along the side surface and is interposed only between the stacked structure ST and the first electrode 210. In other words, in another embodiment, the oxygen retaining layer 220 may be divided into four pieces that face each of the stacked structures ST and are in contact with the first electrode 210. The thickness of the oxygen retention layer 220 may be substantially constant. However, depending on the process, the thickness of the oxygen retaining layer 220 may vary depending on the height.

The oxygen diffusion blocking layer 250 may serve to prevent oxygen ions from moving from the oxygen retention layer 220 to the reactive metal layer 230. The oxygen diffusion blocking layer 250 may be formed of various insulating or semiconductor materials, such as oxide, nitride, or a combination thereof. In the top view of FIG. 4A, the oxygen diffusion-blocking layer 250 is shown surrounding the side of the oxygen-retaining layer 220, but the present embodiment is not limited thereto, and the oxygen-diffusion-blocking layer 250 is an oxygen-retaining layer. It is sufficient that it extends in the vertical direction along the side of 220 and is interposed only between the stacked structure ST and the oxygen retention layer 220. In other words, in another embodiment, the oxygen diffusion blocking layer 250 may be separated into four pieces that face each of the stacked structures ST and are in contact with the oxygen retention layer 220.

The oxygen diffusion blocking layer 250 impedes the movement of oxygen ions on the premise that it does not completely block the movement of oxygen ions, and an insulating oxide layer is created in the reactive metal layer 230 at the interface with the oxygen diffusion blocking layer 250. can reduce the formation rate. The oxygen diffusion blocking layer 250 may have a thin thickness that does not completely block the movement of oxygen ions, for example, less than 10 nm.

In particular, in this embodiment, the thickness of the oxygen diffusion blocking layer 250 may increase from top to bottom. Accordingly, depending on the upper and lower positions of the reactive metal layer 230, the thickness of the oxygen diffusion blocking layer 250 between the reactive metal layer 230 and the oxygen retention layer 220 may be different. In other words, depending on the upper and lower positions of the reactive metal layer 230, the generation rate and/or thickness of the insulating oxide layer generated in the reactive metal layer 230 at the interface with the oxygen diffusion prevention layer 250 may be different. For this reason, the linearity and/or symmetry of the synapse can be improved, which will be described in more detail with reference to FIGS. 5A to 5C below. The thickest part of the oxygen diffusion blocking layer 250 may also have a thin thickness that does not completely block the movement of oxygen ions.

One first electrode 210, one second electrode 240 opposing it, a laminated structure (ST) interposed between them, an oxygen diffusion blocking layer 250, and an oxygen retaining layer 220 form one A synapse (SN) can be formed. Accordingly, the plan view of FIG. 4A shows four synapses (SNs), and these synapses (SNs) can be separated from each other and operate independently. However, the present embodiment is not limited to this, and the number of synapses (SNs) may be varied in various ways.

The operation method of the above synapse (SN) will be exemplarily explained with reference to FIGS. 5A to 5C below.

Figures 5A to 5C are cross-sectional views for explaining the reset operation of the synapse of Figures 4A and 4B. Figure 5A shows the beginning of the reset operation, Figure 5B shows the middle stage of the reset operation, and Figure 5C shows the end of the reset operation. there is. For convenience of explanation, only one synapse (SN) is shown. Before the synapse (SN) starts operating, that is, in the initial state, as shown in FIG. 4B, the insulating oxide layer is not formed between the oxygen retention layer 220 and the reactive metal layer 230, so the synapse ( SN) may be in a low-resistance state.

When a voltage pulse of a predetermined polarity having a magnitude greater than the reset voltage is applied to the synapse (SN) in a low-resistance state through the first and second electrodes 210 and 240, oxygen ions in the oxygen retention layer 220 prevent oxygen diffusion. It may move through layer 250 toward reactive metal layer 230. As a result, an insulating oxide layer is formed in the reactive metal layer 230 at the interface with the oxygen diffusion blocking layer 250, so that a reset operation in which the resistance state of the synapse SN is changed to a high resistance state can be performed. Here, a relatively positive voltage may be applied to the second electrode 240 compared to the first electrode 210. For example, a positive voltage having a magnitude greater than the reset voltage may be applied to the second electrode 240, and a ground voltage may be applied to the first electrode 210. When a reset voltage pulse is repeatedly applied, the size and width of the reset voltage pulse may be substantially constant.

In this reset operation, referring to FIG. 5A, at the beginning of the reset operation, within the reactive metal layer 230 corresponding to the portion with the smallest thickness of the oxygen diffusion prevention layer 250, that is, within the reactive metal layer 230 located at the top. First, the insulating oxide layer 260 may begin to be formed to a thin thickness. This is because the thickness of the oxygen diffusion blocking layer 250 is small, so the reaction between oxygen ions and the reactive metal layer 230 occurs most quickly.

Referring to FIG. 5B, when the reset operation proceeds, that is, when the number of reset voltage pulses increases, the thickness of the oxygen diffusion blocking layer 250 is thicker and the corresponding reactive metal layer 230 is located in the middle portion. The insulating oxide layer 260 may begin to be formed to a thin thickness even within the reactive metal layer 230. At this time, the thickness of the insulating oxide layer 260 formed in the uppermost reactive metal layer 230 may further increase compared to the initial stage of the reset operation in FIG. 5A.

Referring to FIG. 5C, when the reset operation progresses further, that is, when the number of reset voltage pulses increases, the reactive metal layer 230 corresponding to the portion with the greatest thickness of the oxygen diffusion blocking layer 250, that is, at the bottom. The insulating oxide layer 260 may begin to be formed to a thin thickness within the reactive metal layer 230 located therein. At this time, the thickness of the insulating oxide layer 260 formed in the uppermost and middle reactive metal layer 230 may further increase compared to the middle stage of the reset operation in FIG. 5B.

During such a reset operation, the production rate of the insulating oxide layer 260 is basically reduced by the oxygen diffusion blocking layer 250, so the reset operation speed is overall reduced, thereby preventing a sudden change in resistance at the beginning of the reset operation. there is. In particular, as the reset operation progresses, the size and/or volume of the insulating oxide layer 260 created within the synapse (SN) increases, so the rate of resistance increase is small at the beginning of the reset operation, and the rate of resistance increase increases as the reset operation progresses. You can. Therefore, as in the comparative example, a rapid increase in resistance occurs at the beginning of the reset operation, and the phenomenon in which the resistance change rate is greater as the reset operation progresses compared to the set operation can be compensated for. In other words, the linearity and symmetry of the synapse (SN) can be secured.

Meanwhile, as described above, in the neuromorphic device of FIGS. 4A and 4B, the number of second electrodes 240 and stacked structures (ST) for one first electrode 210, that is, the number of synapses (SN) ) can be modified in various ways. This will be exemplarily explained with reference to FIG. 6 .

Figure 6 is a plan view showing a neuromorphic device according to another embodiment of the present invention. Since the cross-sectional view taken along line A-A' of FIG. 6 may be substantially the same as that of FIG. 4B, its illustration will be omitted. Additionally, the description will focus on differences from the above-described embodiment.

Referring to FIG. 6, in a neuromorphic device according to another embodiment of the present invention, two second electrodes 240' and two stacked structures (ST') are provided for one first electrode 210. You can face it. That is, two synapses (SNs) sharing one first electrode 210 may be formed. In this regard, the neuromorphic device of FIG. 4A includes one first electrode 210, four second electrodes 240 opposing it, and four stacked structures (ST), thereby forming four synapses (SN). It may be different from what is included. In this embodiment, one second electrode 240' and the corresponding stacked structure ST', and another second electrode 240' located opposite to them and the corresponding stacked structure ST' ) may be filled with an interlayer insulating layer 270'. Furthermore, for more reliable separation of the second electrode 240' and the stacked structure ST', a slit S' may be formed in the interlayer insulating layer 270'. This slit (S') includes one second electrode 240' and a corresponding laminated structure (ST'), and another second electrode 240' located opposite to them and a laminated structure corresponding thereto ( ST') can be formed in a direction crossing the space. For example, when the two second electrodes 240' and the stacked structure ST' have a shape extending horizontally from the first electrode 210, the slit S' extends vertically from the first electrode 210. It may have a shape extending in one direction. A buried layer 280' containing an insulating material or the like may be formed within the slit S'.

However, of course, in addition to what is described above, the number and planar shape of the second electrode and the second stacked structure opposing the first electrode 210 can be varied in various ways.

FIGS. 7A to 10B are diagrams for explaining an example of a method of manufacturing the neuromorphic device of FIGS. 4A and 4B, where each figure A represents a top view, and each figure B represents a cross-sectional view taken along line A-A' of each figure A.

Referring to FIGS. 7A and 7B , the initial stacked structure STa may be formed by alternately stacking a plurality of reactive metal materials 231 and a plurality of insulating materials 236 on the substrate 200.

Subsequently, the initial stacked structure STa is selectively etched to form a first hole H1 penetrating the initial stacked structure STa, and then the first hole H1 is filled with a conductive material to form a pillar-shaped first hole H1. Two electrode materials 242 can be formed. The side surface of the second electrode material 242 may be in contact with all of the plurality of reactive metal materials 231, and the number and position of the second electrode material 242 are determined by considering the number and position of the second electrode, which will be described later. It can be adjusted.

Referring to FIGS. 8A and 8B , the initial stacked structure (STa) and the second electrode material 242 may be selectively etched and patterned to have a shape similar to a cross shape on a plane. As a result, a laminated structure ST having a center and branches extending in four directions from the center, and a second electrode 240 connected to each of the four branches of the laminated structure ST can be formed. The stacked structure ST may include an alternating stacked structure of a reactive metal layer 230 and an insulating layer 235 . However, since this is for forming the neuromorphic device of FIGS. 4A and 4B, the patterning of the initial stacked structure (STa) and the second electrode material 242 may be modified in various ways. For example, to form the neuromorphic device of FIG. 6, an initial stacked structure (STa) and two second electrode materials penetrating it are formed, and the initial stacked structure (STa) and the second electrode material have a straight shape. It can be patterned to have. Alternatively, when the number of the stacked structure and the second electrode facing one first electrode is not four, but three or five or more, the initial stacked structure (STa) and the second electrode material extend in several directions from the center and the center. It can be patterned to have three or five or more slender branches.

Subsequently, after depositing an insulating material covering the stacked structure (ST) and the second electrode 240 on the substrate 200, a flattening process, such as CMP (Chemical Mechanical Processing), is performed until the upper surface of the stacked structure (ST) is exposed. By performing a polishing process, etc., the interlayer insulating layer 270 that fills the space between the stacked structure (ST) and the second electrode 240 can be formed.

Referring to FIGS. 9A and 9B , the stacked structure ST may be selectively etched to form a second hole H2 that overlaps the center of the stacked structure ST and penetrates the stacked structure ST. The second hole H2 may provide a space for forming an oxygen holding layer, an oxygen diffusion blocking layer, and a first electrode, which will be described later.

Here, as in the plan view of FIG. 9A, when the second hole H2 has a width larger than the center of the stacked structure ST, the four branches of the stacked structure ST are connected to each other by the second hole H2. can be completely separated. In this case, the process of forming the slit S may be omitted. However, for convenience of explanation, the plan view of FIG. 9A shows a case in which the slit S is formed together with the second hole H2. The slit S may be formed to penetrate the interlayer insulating layer 270 by selectively etching the interlayer insulating layer 270. The slit (ST) may have a direction crossing between the four branches of the laminated structure (ST). The slit S forming process may be performed simultaneously using the same mask and etching process as the second hole H2 forming process.

Meanwhile, for reference, as shown in the plan view of FIG. 9C, when the second hole H2 has a width smaller than the center of the stacked structure ST, the four branches of the stacked structure ST may not be completely separated. there is. In this case, a process of forming a slit (S) may be required to separate the four branches of the laminated structure (ST). The slit S may be formed to penetrate the interlayer insulating layer 270 and the stacked structure ST by selectively etching the interlayer insulating layer 270 and the stacked structure ST.

Referring to FIGS. 10A and 10B, an oxygen diffusion prevention layer 250 is formed on the sidewall of the second hole H2, and the oxygen diffusion prevention layer 250 is formed in a shape whose thickness increases from top to bottom by adjusting the process conditions ( 250) can be formed.

As an example, the oxygen diffusion blocking layer 250 is formed by depositing an oxygen diffusion blocking material that fills the second hole H2, and then dry etching the oxygen diffusion blocking material using a mask that exposes the center of the oxygen diffusion blocking material. It can be formed in this way. Due to the nature of dry etching, as etching progresses, the thickness of the oxygen diffusion blocking layer 250 may increase from top to bottom due to polymer accumulating on the sidewall of the etching object.

Or, as another example, the oxygen diffusion blocking layer 250 is formed by depositing an oxygen diffusion blocking material along the entire surface of the resulting product in which the second hole H2 is formed to a thickness that does not completely fill the second hole H2, and then depositing the oxygen diffusion blocking material to a thickness that does not completely fill the second hole H2. It can also be formed by etching the entire surface. Due to the nature of the entire etching, the etching amount of the portion located above the etching object is relatively large, so the thickness of the oxygen diffusion blocking layer 250 may increase from top to bottom.

Meanwhile, in the process of depositing the oxygen diffusion blocking material, the slit S having a narrow width may be filled with the oxygen diffusion blocking material. Because of the narrow width of the slit (S), the oxygen diffusion blocking material in the slit (S) may not be removed even during the subsequent etching process of the oxygen diffusion blocking material. As a result, a buried layer 280 containing a material substantially the same as the oxygen diffusion blocking material may be formed within the slit S. Even if the slit (S) is not completely filled with an oxygen diffusion blocking material, the remaining space of the slit (S) may be filled with an oxygen retaining material during the process of forming the oxygen retaining layer 220, which will be described later. In this case, the buried layer 280 may include a material that is substantially the same as the oxygen diffusion blocking material and the oxygen retaining material.

Next, referring again to FIGS. 4A and 4B, the oxygen retention layer 220 may be formed along the sidewall of the oxygen diffusion blocking layer 250 in the second hole H2. The oxygen retention layer 220 may be formed by depositing an oxygen retention material along the entire surface of the resulting product where the oxygen diffusion blocking layer 250 is formed, and then etching the entire surface.

Subsequently, the first electrode 210 may be formed to fill the remaining space of the second hole H2 in which the oxygen diffusion blocking layer 250 and the oxygen retention layer 220 are formed. After depositing a conductive material with a thickness that sufficiently fills the second hole H2 in which the oxygen diffusion blocking layer 250 and the oxygen retaining layer 220 are formed, the first electrode 210 is formed on the upper surface of the stacked structure ST. This may be done by performing a flattening process until it is revealed.

As a result, the neuromorphic device of FIGS. 4A and 4B can be manufactured. However, it goes without saying that this manufacturing method can be modified in various ways.

Meanwhile, the neuromorphic device of FIGS. 4A and 4B can have various cross-sectional shapes as long as the thickness of the oxygen diffusion blocking layer 250 between the reactive metal layer 230 and the oxygen retention layer 220 is varied depending on the height. You can. This will be exemplarily explained with reference to FIGS. 11 and 12.

Figure 11 is a cross-sectional view showing a neuromorphic device according to another embodiment of the present invention.

Referring to FIG. 11, a neuromorphic device according to another embodiment of the present invention includes a substrate 300, a first electrode 310 and a second electrode extending in the vertical direction on the substrate 300 and arranged to be spaced apart from each other. (340), a stacked structure (ST) located between the first electrode 310 and the second electrode 340 and in which a plurality of reactive metal layers 330 and a plurality of insulating layers 335 are alternately stacked, a stacked structure ( an oxygen retention layer 320 located between the ST) and the first electrode 310 and extending vertically along the first electrode 310, and an oxygen retention layer 320 located between the stacked structure ST and the oxygen retention layer 320 and extending vertically along the first electrode 310. It may include an oxygen diffusion blocking layer 350 extending in this direction.

Here, the oxygen diffusion blocking layer 350, unlike the embodiment of FIGS. 4A and 4B, may decrease in thickness from top to bottom. Since the oxygen diffusion blocking layer 350, the oxygen holding layer 320, and the first electrode 310 are formed in a hole having a substantially vertical sidewall, the first electrode ( 310) may have a shape where the width increases from top to bottom. The oxygen retention layer 320 may have a substantially constant thickness regardless of height.

Figure 12 is a cross-sectional view showing a neuromorphic device according to another embodiment of the present invention.

Referring to FIG. 12, a neuromorphic device according to another embodiment of the present invention includes a substrate 400, a first electrode 410 and a second electrode extending in the vertical direction on the substrate 400 and arranged to be spaced apart from each other. (440), a stacked structure (ST) located between the first electrode 410 and the second electrode 440 and in which a plurality of reactive metal layers 430 and a plurality of insulating layers 435 are alternately stacked, a stacked structure ( an oxygen retention layer 420 located between the ST) and the first electrode 410 and extending vertically along the first electrode 410, and an oxygen retention layer 420 located between the stacked structure ST and the oxygen retention layer 420 and extending vertically along the first electrode 410. It may include an oxygen diffusion blocking layer 450 extending in this direction.

Here, the hole in which the oxygen diffusion blocking layer 450, the oxygen retention layer 420, and the first electrode 410 are formed has a width that decreases from top to bottom and an inclination, unlike the embodiment of FIGS. 4A and 4B. May have side walls. One side wall of the oxygen diffusion blocking layer 450 may be in contact with the inclined side wall of the hole, and the other side wall may be substantially vertical. Accordingly, the thickness of the oxygen diffusion blocking layer 250 may decrease from top to bottom. The first electrode 410 and the oxygen retaining layer 420 may have a substantially constant width and/or thickness regardless of height.

In summary, the thickness of the oxygen diffusion blocking layer 450 may vary depending on the height. On the other hand, the width and/or thickness of the first electrode 410 and the oxygen retention layer 420 may or may not be variable depending on the height.

Meanwhile, in the above embodiments, it was explained that by providing an oxygen diffusion blocking layer, the overall speed of the reset operation can be reduced and symmetry with the set operation can be secured. Furthermore, in the above embodiments, the resistance of the synapse may be more gradually varied due to the sequential creation and/or thickness change of the insulating oxide layer depending on the thickness of the oxygen diffusion-blocking layer between the reactive metal layer and the oxygen retention layer. , It was explained that, accordingly, the linearity of the synapse can be secured.

In addition, the rate of creation and/or disappearance of the insulating oxide layer between the reactive metal layer and the oxygen retaining layer may vary depending on the contact area and/or opposing area of the reactive metal layer and the oxygen retaining layer. Using this, the linearity and symmetry of the synapse can be further secured. This will be exemplarily explained in more detail with reference to FIGS. 13 and 14 .

Figure 13 is a graph showing the change in current depending on the contact area and/or opposing area of the reactive metal layer and the oxygen retaining layer.

Referring to FIG. 13, it can be seen that as the contact area and/or opposing area between the reactive metal layer and the oxygen retaining layer increases, the set current and reset current increase. In other words, it can be seen that the disappearance rate of the insulating oxide layer increases as the contact area and/or opposing area between the reactive metal layer and the oxygen retaining layer increases during the set operation, and the contact area and/or opposing area between the reactive metal layer and the oxygen retaining layer during the reset operation increases. /Or, it can be seen that as the opposing area increases, the production rate of the insulating oxide layer decreases.

Figure 14 is a cross-sectional view showing a neuromorphic device according to another embodiment of the present invention.

Figure 14 is a cross-sectional view showing a neuromorphic device according to another embodiment of the present invention.

Referring to FIG. 14, a neuromorphic device according to another embodiment of the present invention includes a substrate 600, a first electrode 610 and a second electrode extending in the vertical direction on the substrate 600 and arranged to be spaced apart from each other. (640), a stacked structure (ST) located between the first electrode 610 and the second electrode 640 and in which a plurality of reactive metal layers 630 and a plurality of insulating layers 635 are alternately stacked, a stacked structure ( an oxygen retention layer 620 located between ST) and the first electrode 610 and extending vertically along the first electrode 610, and an oxygen retention layer 620 located between the stacked structure ST and the oxygen retention layer 620 and extending vertically along the first electrode 610. It may include an oxygen diffusion blocking layer 650 extending in this direction.

Here, the thickness of the oxygen diffusion blocking layer 650 may vary depending on the upper and lower positions of the reactive metal layer 630. As an example, the thickness of the oxygen diffusion blocking layer 650 may decrease from top to bottom. In this case, as the reset operation progresses, an insulating oxide layer is created from within the reactive metal layer 630 located at the bottom and its thickness increases, and the insulating oxide layer within the reactive metal layer 630 located at the top is generated last. The thickness may also be the thinnest. As described above, the thicker the oxygen diffusion blocking layer 650 is, the slower the insulating oxide layer is created.

Furthermore, the thickness of the reactive metal layer 630 may also be different depending on the upper and lower positions of the reactive metal layer 630, and accordingly, the opposing areas of the reactive metal layer 630 and the oxygen retention layer 620 may be different. In particular, the thickness of the reactive metal layer 630 may be determined by considering the thickness of the oxygen diffusion prevention layer 650. That is, in areas where the thickness of the oxygen diffusion blocking layer 650 is relatively thick, the thickness of the reactive metal layer 630 may also be relatively thick. As an example, when the thickness of the oxygen diffusion blocking layer 650 decreases from top to bottom, the thickness of the reactive metal layer 630 may also decrease from top to bottom. Accordingly, the opposing area of the reactive metal layer 630 and the oxygen retention layer 620 may decrease from top to bottom. In this case, as the reset operation progresses, an insulating oxide layer is created from within the reactive metal layer 630 located at the bottom and its thickness increases, and the insulating oxide layer within the reactive metal layer 630 located at the top is generated last. The thickness may also be the thinnest. As described above, this is because the larger the opposing area between the reactive metal layer 630 and the oxygen retention layer 620, the slower the production rate of the insulating oxide layer.

As a result, according to this embodiment, the resistance change rate can be further reduced as the reset operation progresses, thereby further securing the linearity and symmetry of the synapse (SN).

The neuromorphic devices of the above-described embodiments may be used in a variety of devices or systems. This will be illustratively explained with reference to FIG. 15 .

Figure 15 is an example of a pattern recognition system according to an embodiment of the present invention. The pattern recognition system of this embodiment may be a system for recognizing various types of patterns, such as a speech recognition system or an image recognition system. The pattern recognition system of this embodiment can be implemented with the neuromorphic devices of the above-described embodiments.

Referring to FIG. 15, the pattern recognition system 800 of this embodiment includes a central processing unit (CPU) 810, a memory device 820, a communication control device 830, a network 840, a pattern output device 850, It may include a pattern input device 860, an analog-to-digital converter (ADC, 470), a neuromorphic device 880, a bus line 890, etc.

The central processing unit 810 generates and transmits various signals for learning of the neuromorphic device 880, and processes various signals to recognize patterns such as voice and video according to the output from the neuromorphic device 880. and functions can be performed. This central processing unit 810 provides a bus line 890 to the memory device 820, communication control device 830, pattern output device 850, analog-to-digital converter 870, and neuromorphic device 880. can be connected through.

The memory device 820 can store various information required to be stored in the pattern recognition system 800, and for this purpose, it can include different types of memories. For example, the memory device 820 may include ROM (ROM) 422, RAM (RAM) 424, etc. The ROM 822 may perform the function of storing various programs or data used in the central processing unit 810 to process and control learning, pattern recognition, etc. of the neuromorphic device 880. The ROM 824 may download and store programs or data of the ROM 822, or may store data such as voice and video converted and analyzed by the analog-to-digital converter 870.

The communication control device 830 can exchange recognized data, such as voice and video, with other communication control devices through the network 840.

The pattern output device 850 can output recognized data such as voice and video in various ways. For example, the pattern recognition device 850 may include a printer, a display unit, etc., and may display voice as a waveform or display an image.

The pattern input device 860 is a part that receives analog audio, video, etc., and may include a microphone, a camera, etc.

The analog-to-digital converter 870 can convert analog data input from the pattern input device 860 into digital data and analyze the digital data.

The neuromorphic device 880 can perform learning and recognition using data output from the analog-to-digital converter 870, and output data corresponding to the recognized pattern. Neuromorphic device 880 may include one or more of the neuromorphic devices of the embodiments described above. For example, the neuromorphic device 880 according to an embodiment of the present invention for solving the above problem includes: a substrate; a first electrode and a second electrode extending in a vertical direction on the substrate and spaced apart from each other; A laminated structure located between the first electrode and the second electrode and comprising two or more reactive metal layers and one or more insulating layers alternately stacked - wherein the reactive metal layer reacts with oxygen ions in the oxygen retention layer to form an insulating oxide layer. Can do it. - ; the oxygen retaining layer interposed between the laminated structure and the first electrode; and an oxygen diffusion-blocking layer interposed between the laminated structure and the oxygen-retaining layer to prevent oxygen ions from moving from the oxygen-retaining layer to the reactive metal layer, wherein the oxygen-diffusion-blocking layer has a thickness in the vertical direction. It can be variable. Through this, the symmetry and linearity of the synapse can be secured. As a result, the operating characteristics of the neuromorphic device 880 may be improved, and accordingly, the operating characteristics of the pattern recognition system 800 may be improved.

In addition, the pattern recognition system 800 may further include other components necessary to properly perform its functions. For example, a keyboard, a mouse, etc. may be further included as input units for inputting various parameters or setting conditions for driving the pattern recognition system 800.

Although various embodiments for the problem to be solved have been described above, it is clear that various changes and modifications can be made within the scope of the technical idea of the present invention by those skilled in the art. .

SN: synapse 210: first electrode
220: oxygen retention layer 230: reactive metal layer
235: insulating layer 240: second electrode
250: Oxygen diffusion blocking layer

Claims (20)

Board;
a first electrode and a second electrode extending in a vertical direction on the substrate and spaced apart from each other;
A laminated structure located between the first electrode and the second electrode and comprising two or more reactive metal layers and one or more insulating layers alternately stacked - wherein the reactive metal layer reacts with oxygen ions in the oxygen retention layer to form an insulating oxide layer. Can do it. - ;
the oxygen retaining layer interposed between the laminated structure and the first electrode; and
Comprising an oxygen diffusion blocking layer interposed between the laminated structure and the oxygen retaining layer to prevent oxygen ions from moving from the oxygen retaining layer to the reactive metal layer,
The thickness of the oxygen diffusion blocking layer is variable in the vertical direction.
Neuromorphic device.
◈Claim 2 was abandoned upon payment of the setup registration fee.◈ According to claim 1,
The thickness of the oxygen diffusion blocking layer increases from top to bottom.
Neuromorphic device. ◈Claim 3 was abandoned upon payment of the setup registration fee.◈ According to claim 1,
The thickness of the oxygen diffusion blocking layer decreases from top to bottom.
Neuromorphic device.
◈Claim 4 was abandoned upon payment of the setup registration fee.◈ According to claim 1,
The second electrode includes a plurality of second electrodes spaced apart from each other,
The laminated structure includes a plurality of laminated structures corresponding to each of the plurality of second electrodes and spaced apart from each other.
Neuromorphic device.
◈Claim 5 was abandoned upon payment of the setup registration fee.◈ According to clause 4,
Further comprising an interlayer insulating layer buried between the plurality of second electrodes and between the plurality of laminate structures.
Neuromorphic device.
◈Claim 6 was abandoned upon payment of the setup registration fee.◈ According to clause 5,
Further comprising a slit penetrating the interlayer insulating layer and extending in a direction transverse from the first electrode to between the plurality of second electrodes and between the plurality of laminate structures.
Neuromorphic device.
◈Claim 7 was abandoned upon payment of the setup registration fee.◈ According to clause 6,
Further comprising a buried layer that buries the slit and includes the same material as at least one of the oxygen retention layer and the oxygen diffusion blocking layer.
Neuromorphic device.
◈Claim 8 was abandoned upon payment of the setup registration fee.◈ According to claim 1,
The thickness of the reactive metal layer is different from each other.
Neuromorphic device.
◈Claim 9 was abandoned upon payment of the setup registration fee.◈ According to clause 2,
The thickness of the reactive metal layer increases from top to bottom.
Neuromorphic device.
◈Claim 10 was abandoned upon payment of the setup registration fee.◈ According to clause 3,
The thickness of the reactive metal layer decreases from top to bottom.
Neuromorphic device.
◈Claim 11 was abandoned upon payment of the setup registration fee.◈ According to claim 1,
The oxygen retention layer surrounds the side surface of the first electrode.
Neuromorphic device.
◈Claim 12 was abandoned upon payment of the setup registration fee.◈ According to claim 11,
The oxygen diffusion blocking layer surrounds the side of the oxygen retention layer.
Neuromorphic device.
◈Claim 13 was abandoned upon payment of the setup registration fee.◈ According to claim 1,
When a reset voltage of the first polarity is applied through the first electrode and the second electrode, a reset operation in which an insulating oxide layer is created or its thickness increases in each of the reactive metal layers at the interface with the oxygen diffusion prevention layer is performed. carried out,
When a set voltage of a second polarity different from the first polarity is applied through the first electrode and the second electrode, a set operation is performed in which the insulating oxide layer disappears or its thickness is reduced.
Neuromorphic device.
◈Claim 14 was abandoned upon payment of the setup registration fee.◈ According to claim 13,
The insulating oxide layer is created most quickly within the reactive metal layer corresponding to the area where the oxygen diffusion blocking layer has the smallest thickness.
Neuromorphic device.
◈Claim 15 was abandoned upon payment of the setup registration fee.◈ According to claim 13,
The thickness of the reactive metal layer is different from each other,
The insulating oxide layer is created most quickly within the reactive metal layer having the smallest thickness.
Neuromorphic device.
◈Claim 16 was abandoned upon payment of the setup registration fee.◈ According to claim 13,
As the number or thickness of the insulating oxide layers increases, conductivity decreases, and as the number or thickness of the insulating oxide layers decreases, conductivity increases.
Neuromorphic device.
◈Claim 17 was abandoned upon payment of the setup registration fee.◈ According to claim 13,
During the reset operation, as the number of pulses of the reset voltage increases, the number or thickness of the insulating oxide layers increases,
During the set operation, as the number of pulses of the set voltage increases, the number or thickness of the insulating oxide layer decreases.
Neuromorphic device.
◈Claim 18 was abandoned upon payment of the setup registration fee.◈ According to claim 17,
During the reset operation, the pulse of the reset voltage has a constant width and constant size,
During the set operation, the pulse of the set voltage has a constant width and constant size.
Neuromorphic device.
◈Claim 19 was abandoned upon payment of the setup registration fee.◈ According to claim 1,
The oxygen diffusion blocking layer is,
Having a thickness that does not completely block the movement of the oxygen ions
Neuromorphic device.
◈Claim 20 was abandoned upon payment of the setup registration fee.◈ According to claim 1,
The oxygen diffusion blocking layer is,
Containing insulating materials, semiconducting materials, or combinations thereof
Neuromorphic device.