KR20230029220A - Neuromorphic device based on memristor device and neuromorphic system using the same - Google Patents

Abstract

The present invention relates to a neuromorphic device using a memristor device and a neuromorphic system using the same, and to a technology that provides a neuromorphic device and a neuromorphic system using the same, for implementing both characteristics of biological synapses and biological neurons, which are responsible for memory and information transfer in human brain, using a memristor device. According to one embodiment of the present invention, the neuromorphic device comprises: a first electrode; a second electrode disposed opposite to the first electrode; a switching layer formed of amorphous material between the first electrode and the second electrode; and a source layer formed of alloy material between the switching layer and the first electrode and controlling either drift or diffusion of metal ions in the switching layer, wherein any one characteristic of artificial neuron integral characteristics due to volatility or artificial synapse characteristics due to non-volatility can be implemented, depending on a size of maximum current density based on a voltage applied through the first electrode.

Description

Neuromorphic device based on memristor device and neuromorphic system using the same

The present invention relates to a neuromorphic device using a memristor device and a neuromorphic system using the same, and to biological synapse characteristics and biological neurons responsible for memory and information transmission in the human brain using the memristor device It relates to a technology for providing a neuromorphic device that implements all neuron characteristics and a neuromorphic system using the same.

Computers used today are all adopting the von Neumann architecture.

The von Neumann structure according to the prior art is separated into a central processing unit (CPU) and memory, and reads instructions or reads and writes data through a bus connecting the two. do.

Thanks to the design of separating the roles of the central processing unit and memory, today's computers have a great advantage in that they can process multiple data only with software without hardware changes.

However, the von Neumann structure has a fatal disadvantage in that the bandwidth between the central processing unit and the memory is low.

In other words, since the program memory and data memory communicate with the central processing unit through one bus without physical distinction, the central processing unit cannot access commands and data simultaneously and reads and writes the listed commands one at a time. ) will inevitably occur.

Deep learning, which has recently been in the limelight in the field of artificial intelligence, requires large-scale parallel processing, and implementing deep learning by applying the von Neumann structure reduces efficiency in terms of data processing, transmission speed, and energy consumption.

Therefore, a neuromorphic architecture that mimics the neural structure of the human brain with hardware has been proposed as a new alternative.

The neuromimetic structure is composed of a device that plays a role of a neuron, a nerve cell in the brain, and a device that plays a role of a biological synapse existing between neurons.

This structure is capable of parallel computing, so that a large amount of memory and calculation can be performed simultaneously.

Recently, active research on hardware-based artificial synapses capable of mimicking the functions of synapses is being conducted.

The human brain has about 100 billion neurons and about 100 trillion synapses connected in parallel.

Each neuron receives input signals from other neurons through dendrites and sends signals to other neurons through axons.

At this time, the synapse serves to connect neurons and plays an important role in determining how much weight the signal received from the axon of another neuron is to be transmitted.

It mimics the characteristics of long-term potentiation/long-term depression, which is a characteristic of biological synapses, by using oxide-based memristors. When constant pulses are continuously applied, synaptic weights are linear and symmetric , and tends to increase or decrease gradually.

The paper (Hirokjyoti Kalita et al., Artificial Neuron using Vertical MoS2/Graphene Threshold Switching Memristors, Scientific Reports, 2019) discloses research on mimicking the properties of biological neurons using MoS ₂ /Graphene-based volatile memristors.

In addition, in the thesis (Min-Kyu Kim et al., Short-Term Plasticity and Long-term Potentiation in Artificial Biosynapses with Diffusive Dynamics, ACS Nano, 2018), short-term memory and long-term memory using memristors (long-term memory).

On the other hand, in the paper (Donguk Lee et al., Various Threshold Switching Devices for Integrate and Fire Neuron Applications, Advanced Electronic Materials, 2019), integration characteristics are implemented using volatile memristors, and a process for implementing coupling characteristics is exemplified. .

However, research on mimicking characteristics of neurons and synapses using a memristor device according to the prior art has a fatal disadvantage in terms of integration because the size of a capacitor is 1000 F ² or more in a CMOS process.

In addition, an artificial synaptic device using an electrolyte gate transistor having three terminals has problems in integration and CMOS process compatibility in that an electrolyte gate is used.

Korean Patent Registration No. 10-2249655, "Neurons and neuromorphic systems including them" Korean Patent Publication No. 10-2021-0061903, "Non-volatile memory device and operation method thereof" Korean Patent Publication No. 10-2021-0063196, "Three-terminal resistance change element-based ignition-type neuron circuit and its operation method" Korean Patent Registration No. 10-1922049, "Artificial synaptic device and its manufacturing method"

An object of the present invention is to provide a neuromorphic device that implements both biological synapse characteristics and biological neuron characteristics responsible for memory and information transmission in the human brain by using a memristor device.

As the present invention implements both biological synapse characteristics and biological neuron characteristics using a single memristor device, the types of materials used in the fabrication of hardware-based artificial neurons and synaptic device arrays are greatly reduced, thereby reducing the process It aims to achieve simplification.

An object of the present invention is to secure the characteristics of synaptic characteristics of the human brain, such as pair pulse facilitation and short-term memory to long-term memory, by using a memristor device.

An object of the present invention is to solve both the integration problem and the CMOS process compatibility problem of implementing a neuromorphic device using a memristor device based on CuTe and GeS ₂ having a size of 4F ² or less.

An object of the present invention is to provide a neuromorphic system using a neuromorphic device that implements both biological synapse characteristics and biological neuron characteristics using a memristor device.

A neuromorphic device according to an embodiment of the present invention includes a first electrode, a second electrode disposed opposite to the first electrode, a switching layer formed of an amorphous material between the first electrode and the second electrode, and the switching layer. A source layer formed of an alloy material between the layer and the first electrode to control one of drift and diffusion of metal ions in the switching layer, wherein a voltage applied through the first electrode Depending on the size of the maximum current density based on , it is possible to implement any one of the integral characteristics of the artificial neuron according to volatility and the characteristics of artificial synapses according to non-volatile.

The neuromorphic device implements the integration characteristics of the artificial neuron when the magnitude of the maximum current density is 400 A/cm ² or less, and the magnitude of the maximum current density exceeds 400 A/cm ² and 2000 A/cm ² Hereinafter, the artificial synaptic properties may be implemented.

In the neuromorphic device, when the maximum current density is 400 A/cm ² or less and no additional voltage is applied through the first electrode, the self resistance returns to an initial state by the diffusion in the switching layer. Indicates volatility that is reset (self-reset), and when the magnitude of the maximum current density exceeds 400 A / cm ² and no additional voltage is applied through the first electrode at 2000 A / cm ² or less, the self-reset may indicate non-volatility.

In the neuromorphic device, when the magnitude of the maximum current density is 400 A/cm ² or less and the amplitude of the voltage applied through the first electrode increases, the magnitude of the output current related to the integral characteristic of the artificial neuron increases. can increase

The neuromorphic device generates an output related to the integration characteristics of the artificial neuron when the pulse width of the voltage applied through the first electrode increases when the magnitude of the maximum current density is 400 A/cm ² or less. The magnitude of the current may increase.

The neuromorphic device generates an output related to the integral characteristics of the artificial neuron when the pulse interval of the voltage applied through the first electrode increases when the maximum current density is 400 A/cm ² or less. The magnitude of the current may decrease.

In the neuromorphic device, the number of times a voltage having the same pulse width, magnitude, and pulse interval is repeatedly applied through the first electrode when the magnitude of the maximum current density exceeds 400 A/cm ² and is 2000 A/cm ² or less Among the short-term memory operation in which the size of the output current is reduced to the initial current size after a predetermined time based on the long-term memory operation in which the size of the current output is maintained after the predetermined time Any one memory operation can be performed.

In the neuromorphic device, when the maximum current density exceeds 400 A/cm ² and is 2000 A/cm ² or less, a voltage having the same pulse width, size, and pulse interval is repeatedly applied through the first electrode. may exhibit a paired pulse facilitation (PPF) characteristic in which the magnitude of the output current repeatedly increases.

The first electrode and the second electrode have a thickness of 30 nm, the switching layer and the source layer have a thickness of 10 nm, the first electrode, the second electrode, the switching layer and the source layer A crossbar-type memristor structure may be formed.

The switching layer may be formed of the amorphous material including germanium sulfide (GeS ₂ ), and the source layer may be formed of the alloy material including copper telluride (CuTe).

The first electrode and the second electrode are platinum (Pt), tungsten (W), titanium nitride (TiN), tantalum nitride (TaN), gold (Au), rubidium (Ru), iridium (Ir), It may be formed of at least one metal material selected from palladium (Pd), titanium (Ti), hafnium (Hf), molybdenum (Mo), and niobium (Nb).

According to one embodiment of the present invention, the neuromorphic system implements artificial synapse characteristics according to non-volatility when the magnitude of the maximum current density based on the applied voltage exceeds 400 A / cm ² and is less than 2000 A / cm ² An artificial synaptic unit including at least one first neuromorphic element and an artificial neuron electrically connected to the artificial synaptic unit and subject to volatility when a maximum current density based on an applied voltage is 400 A/cm ² or less. It may include an artificial neuron unit that performs an integration operation and an ignition operation by including at least one second neuromorphic element that implements an integration characteristic of .

According to an embodiment of the present invention, the neuromorphic system may further include a control unit that controls conductance of the artificial synaptic unit by detecting a firing signal based on the performed firing motion.

The at least one second neuromorphic device includes a first electrode, a second electrode disposed opposite to the first electrode, a switching layer formed of germanium sulfide (GeS ₂ ) between the first electrode and the second electrode, and A source layer formed of copper telluride (CuTe) between the switching layer and the first electrode to control any one of drift and diffusion of metal ions in the switching layer, wherein the maximum current density is 400 A When an additional voltage is not applied through the first electrode at /cm ² or less, self-reset volatility in which resistance returns to an initial state due to the diffusion in the switching layer may be exhibited.

In the at least one second neuromorphic device, when the magnitude of the voltage applied through the first electrode increases when the magnitude of the maximum current density is 400 A/cm ² or less, the magnitude of current related to the integral characteristics of the artificial neuron increases, and when the pulse width of the voltage applied through the first electrode increases when the magnitude of the maximum current density is 400 A / cm ² or less, the magnitude of the output current related to the integral characteristic of the artificial neuron increases, , When the maximum current density is 400 A/cm ² or less and the pulse interval of the voltage applied through the first electrode increases, the magnitude of the output current related to the integration characteristics of the artificial neuron may decrease.

The at least one first neuromorphic device includes a first electrode, a second electrode disposed opposite to the first electrode, a switching layer formed of germanium sulfide (GeS ₂ ) between the first electrode and the second electrode, and A source layer formed of copper telluride (CuTe) between the switching layer and the first electrode to control any one of drift and diffusion of metal ions in the switching layer, wherein the maximum current density is 400 A / cm ² and less than 2000 A/cm ² , when no additional voltage is applied through the first electrode, non-volatility that is not self-reset may be exhibited.

In the at least one first neuromorphic device, a voltage having the same pulse width, size, and pulse interval is applied through the first electrode when the maximum current density exceeds 400 A/cm ² and is 2000 A/cm ² or less. Performs any one of a short-term memory operation in which the output current size is reduced to the initial current size after a certain period of time based on the number of times of repetitive application and a long-term memory operation in which the output current level is maintained after a certain period of time can do.

In the at least one first neuromorphic device, a voltage having the same pulse width, size, and pulse interval is applied through the first electrode when the maximum current density exceeds 400 A/cm ² and is 2000 A/cm ² or less. When repeatedly applied, the output current may exhibit a dual-pulse acceleration characteristic in which the magnitude of the output current repeatedly increases.

According to one embodiment, the present invention provides a neuromorphic device that implements both biological synapse characteristics and biological neuron characteristics responsible for memory and information transmission in the human brain by using a memristor device. can

As the present invention implements both biological synapse characteristics and biological neuron characteristics using a single memristor device, the types of materials used in the fabrication of hardware-based artificial neurons and synaptic device arrays are greatly reduced, thereby reducing the process simplification can be achieved.

According to the present invention, by using a memristor device, it is possible to secure the characteristics of synaptic properties of the human brain, such as facilitation of paired pulses and the change from short-term memory to long-term memory.

The present invention can solve both the integration problem and the CMOS process compatibility problem of implementing a neuromorphic device using a memristor device based on CuTe and GeS ₂ having a size of 4F ² or less.

The present invention can provide a neuromorphic system using a neuromorphic device that implements both biological synapse characteristics and biological neuron characteristics using a memristor device.

1 is a diagram for explaining a neuromorphic device based on a memristor device according to an embodiment of the present invention.
2A and 2B are diagrams for explaining electrical characteristics of a neuromorphic device based on a memristor device according to an embodiment of the present invention.
3 is a diagram for explaining a neuromorphic system using a neuromorphic device according to an embodiment of the present invention.
4 is a diagram for explaining an artificial neuron circuit using a neuromorphic device according to an embodiment of the present invention.
5 to 6B are diagrams for explaining the characteristics of an artificial neuron of a neuromorphic device according to an embodiment of the present invention.
7A to 9 are views for explaining artificial synapse characteristics of a neuromorphic device according to an embodiment of the present invention.
10 is a diagram illustrating an artificial synaptic element array using a memristor element-based artificial synaptic element according to an embodiment of the present invention.

Hereinafter, various embodiments of this document will be described with reference to the accompanying drawings.

Examples and terms used therein are not intended to limit the technology described in this document to specific embodiments, and should be understood to include various modifications, equivalents, and/or substitutes of the embodiments.

In the following description of various embodiments, if it is determined that a detailed description of a related known function or configuration may unnecessarily obscure the subject matter of the invention, the detailed description will be omitted.

In addition, terms to be described below are terms defined in consideration of functions in various embodiments, and may vary according to intentions or customs of users or operators. Therefore, the definition should be made based on the contents throughout this specification.

In connection with the description of the drawings, like reference numerals may be used for like elements.

Singular expressions may include plural expressions unless the context clearly dictates otherwise.

In this document, expressions such as "A or B" or "at least one of A and/or B" may include all possible combinations of the items listed together.

Expressions such as "first," "second," "first," or "second," may modify the corresponding components regardless of order or importance, and are used to distinguish one component from another. It is used only and does not limit the corresponding components.

When a (e.g., first) component is referred to as being "(functionally or communicatively) connected" or "connected" to another (e.g., second) component, a component refers to said other component. It may be directly connected to the element or connected through another component (eg, a third component).

In this specification, "configured to (or configured to)" means "suitable for," "having the ability to," "changed to" depending on the situation, for example, hardware or software ," can be used interchangeably with "made to," "capable of," or "designed to."

In some contexts, the expression "device configured to" can mean that the device is "capable of" in conjunction with other devices or components.

For example, the phrase "a processor configured (or configured) to perform A, B, and C" may include a dedicated processor (eg, embedded processor) to perform the operation, or by executing one or more software programs stored in a memory device. , may mean a general-purpose processor (eg, CPU or application processor) capable of performing corresponding operations.

Also, the term 'or' means 'inclusive or' rather than 'exclusive or'.

That is, unless otherwise stated or clear from the context, the expression 'x employs a or b' means any one of the natural inclusive permutations.

In the above-described specific embodiments, components included in the invention are expressed in singular or plural numbers according to the specific embodiments presented.

However, singular or plural expressions are selected appropriately for the presented situation for convenience of explanation, and the above-described embodiments are not limited to singular or plural components, and even components expressed in plural are composed of a singular number or , Even components expressed in the singular can be composed of plural.

Meanwhile, in the description of the invention, specific embodiments have been described, but various modifications are possible without departing from the scope of the technical idea contained in the various embodiments.

Therefore, the scope of the present invention should not be limited to the described embodiments and should not be defined, but should be defined by not only the claims to be described later, but also those equivalent to these claims.

1 is a diagram for explaining a neuromorphic device based on a memristor device according to an embodiment of the present invention.

1 illustrates the structure and operation of a neuromorphic device based on a memristor device according to an embodiment of the present invention.

Referring to FIG. 1 , a neuromorphic device 100 according to an embodiment of the present invention includes a second electrode 110, a switching layer 120, a source layer 130, and a first electrode 140. .

For example, the neuromorphic device 100 includes a first electrode 140, a second electrode 110 disposed opposite to the first electrode 140, and between the first electrode 140 and the second electrode 110. Between the switching layer 120 formed of an amorphous material and the switching layer 120 and the first electrode 140 formed of an alloy material, during drift and diffusion of metal ions in the switching layer 120 It includes a source layer 130 that controls either one.

For example, either drift or diffusion of metal ions in the switching layer 120 may be controlled according to whether a voltage applied through the first electrode 140 is a set voltage or a reset voltage. .

According to an embodiment of the present invention, the neuromorphic device 100 generates an artificial neuron according to volatility according to the magnitude of the maximum current density based on the voltage applied through the first electrode 140. Any one of artificial synapse characteristics according to integration characteristics and non-volatile characteristics may be implemented.

That is, the neuromorphic device 100 can implement both integral characteristics of an artificial neuron according to volatility and characteristics of an artificial synapse according to non-volatility with one device.

As an example, the neuromorphic device 100 implements the integrated characteristics of artificial neurons at a maximum current density of 400 A/cm ² or less, and has a maximum current density of 400 A/cm ² . and 2000 A/cm ² or less can implement artificial synaptic properties.

Here, the characteristics of the artificial neuron and the implementation of the characteristics of the artificial synapse according to the magnitude of the maximum current density will be supplementarily described using FIGS. 2A and 2B.

According to one embodiment of the present invention, the first electrode 140 and the second electrode 110 may have a thickness of 30 nm, and the switching layer 120 and the source layer 130 may have a thickness of 10 nm.

For example, the neuromorphic device 100 includes a crossbar-type memristor in the formation structure of the first electrode 140, the second electrode 110, the switching layer 120, and the source layer 130. (memristor) structure.

In addition, the neuromorphic device 100 may have a crossbar-type memristor structure having a cell size of 5 μm × 5 μm.

According to one embodiment of the present invention, the switching layer may be formed of an amorphous material including germanium sulfide (GeS ₂ ), and the source layer may be formed of an alloy material including copper telluride (CuTe).

For example, the first electrode 140 and the second electrode 110 may include platinum (Pt), tungsten (W), titanium nitride (TiN), tantalum nitride (TaN), gold (Au), and rubidium (Ru). ), iridium (Ir), palladium (Pd), titanium (Ti), hafnium (Hf), molybdenum (Mo) and niobium (Nb) may be formed of at least one metal material selected from among.

Preferably, the first electrode 140 and the second electrode 110 may be formed of titanium (Ti).

For example, the first electrode 140 may be referred to as a top electrode (TE), and the second electrode 110 may be referred to as a bottom electrode (BE).

The neuromorphic device 100 according to an embodiment of the present invention utilizes amorphous germanium sulfide (GeS ₂ ) with a large band gap as a switching layer to have a high resistance value initially, and a metal filament such as copper (Cu). After the filament is generated, it may exhibit volatile characteristics of self-diffusion.

In addition, the neuromorphic device 100 uses copper telluride (CuTe) instead of copper (Cu) for the source layer 120 to secure gradual current-voltage characteristics, so that copper is germanium sulfide (GeS _{2 )} . It is possible to control the amount of drift and diffusion in the switching layer formed of ).

According to one embodiment of the present invention, in the neuromorphic device 100, as a voltage is applied through the first electrode 140, metal filaments are gradually generated in the switching layer 120, and thus conductance gradually increases. characteristics that may change.

For example, the neuromorphic device 100 may increase the bending of the metal filament by increasing the compliance current to 500 μA.

Through this, the neuromorphic device 100 can be implemented as a memory device having non-volatile characteristics by controlling the movement of self-diffusing Cu atoms.

In addition, the neuromorphic device 100 is characterized by conversion from paired pulse facilitation (PPF) and short-term memory to long-term memory corresponding to the synaptic characteristics of the human brain can represent

According to one embodiment of the present invention, the neuromorphic device 100 may be a memristor device and may be operated as an artificial synaptic device or an artificial neuron device.

Accordingly, the present invention can provide a neuromorphic device that implements both biological synapse characteristics and biological neuron characteristics responsible for memory and information transmission in the human brain by using a memristor device.

In addition, the present invention can solve both the integration problem and the CMOS process compatibility problem of implementing a neuromorphic device using a memristor device based on CuTe and GeS ₂ having a size of 4F ² or less.

2A and 2B are diagrams for explaining electrical characteristics of a neuromorphic device based on a memristor device according to an embodiment of the present invention.

2A illustrates electrical characteristics related to artificial neuron characteristics of a neuromorphic device based on a memristor device according to an embodiment of the present invention, and FIG. 2B illustrates a neuromorphic device based on a memristor device according to an embodiment of the present invention. The electrical characteristics related to the artificial synaptic characteristics of the pick device are exemplified.

Referring to the graph 200 of FIG. 2A , current and voltage characteristics are shown when the maximum current density based on the voltage applied to the neuromorphic device is 400 A/cm ² .

A current density of 400 A/cm ² or less exhibits a self-reset characteristic, and the self-reset characteristic may be related to a volatile characteristic.

Here, self-reset refers to a phenomenon in which resistance of a neuromorphic device returns to an initial state by diffusion when a thin metal filament is formed in a switching layer of the neuromorphic device and the power is turned off.

In addition, when the power is turned off, it may correspond to a case where no voltage is applied to the neuromorphic element through the first electrode.

In addition, when the compliance current is 100 μA or less, the neuromorphic device exhibits a forming voltage of about 1.20 V, a set voltage of about 1.10 V, and a volatile characteristic with a selectivity of about 2.39 × 10 ⁴ can indicate

The neuromorphic device according to an embodiment of the present invention may implement the integrated characteristics of the artificial neuron at a maximum current density of 400 A/cm ^{2 or} less.

For example, the neuromorphic device resists by diffusion in the switching layer when no additional voltage is applied through the first electrode at a maximum current density of 400 A/cm ² or less. This may indicate self-reset volatility that returns to the initial state.

Referring to the graph 210 of FIG. 2B, current and voltage characteristics when the maximum current density based on the voltage applied to the neuromorphic device according to an embodiment of the present invention is 2000 A/cm ² indicates

For example, the neuromorphic device may exhibit non-volatile as a memory characteristic rather than self-reset when the maximum current density exceeds 400 A/cm ² and is less than 2000 A/cm ² . there is.

In other words, the neuromorphic device self-resets when the maximum current density exceeds 400 A/cm ² and no additional voltage is applied through the first electrode at 2000 A/cm ² or less. It can indicate non-volatile that is not self-reset.

According to one embodiment of the present invention, the neuromorphic device has a forming voltage of about 2.24 V, a set voltage of about 1.25 V, and a reset of about -1.0 V when the compliance current exceeds 100 μA. (reset) voltage, and non-volatile characteristics with a memory margin of about 2.71 Х 10 ¹ .

For example, the neuromorphic device is a memristor device having volatility at a maximum current density of 400 A/cm ² or less, and may mimic the integral characteristics of an artificial neuron.

In addition, the neuromorphic device has memory characteristics of a maximum current density of 2000 A/cm ² and may mimic artificial synapse characteristics.

For example, a neuromorphic device can implement both neuron characteristics and synaptic characteristics by adjusting a maximum current density using devices having the same structure.

Therefore, as the present invention implements both biological synapse characteristics and biological neuron characteristics using a single memristor device, the types of materials used in the fabrication of hardware-based artificial neurons and synaptic device arrays are greatly reduced. The simplification of the process can be achieved.

3 is a diagram for explaining a neuromorphic system using a neuromorphic device according to an embodiment of the present invention.

3 illustrates a configuration in which an artificial synapse unit and an artificial neuron unit are implemented using a neuromorphic device according to an embodiment of the present invention, and conductance of the artificial synapse unit is updated through a control unit.

Referring to FIG. 3 , a neuromorphic system 300 according to an embodiment of the present invention includes an artificial synapse unit 310, a converter unit 320, an artificial neuron unit 330, and a control unit 340.

According to one embodiment of the present invention, in the neuromorphic system 300, when an input spike signal in the form of a voltage is applied to the artificial synapse unit 310, current is output, and the output current is converted to the converter unit 320. ), and the converted voltage is applied to the artificial neuron unit 330 to perform an integral operation and a fire operation, and a signal based on the fire operation is sent to the control unit 340. Feedback is given, and the control unit 340 increases (potentiation) or decreases (depression) the conductivity of the artificial synapse unit 310 according to a signal based on a fire operation.

For example, the neuromorphic system 300 performs an integral operation and a fire operation while being applied to the artificial neuron unit 330, and outputs a signal based on the fire operation to an output spike signal. signal) can be output.

According to one embodiment of the present invention, the artificial synaptic unit 310 is non-volatile when the magnitude of the maximum current density based on the applied voltage exceeds 400 A/cm ² and is 2000 A/cm ² or less ( It may include at least one first neuromorphic element implementing artificial synaptic characteristics according to non-volatile). Here, the applied voltage may be an input spike signal in the form of a voltage.

As an example, the artificial neuron unit 330 implements the integral characteristics of the artificial neuron according to volatile when the maximum current density (max current density) based on the applied voltage is 400 A / cm ² or less At least one Including the second neuromorphic element of the integrated (integrate) operation and ignition (fire) operation can be performed. Here, the applied voltage may be a voltage obtained by converting a current output from the artificial synapse unit 310 through the converter unit 320 .

According to one embodiment of the present invention, the control unit 340 can control the conductance of the artificial synaptic unit 310 by detecting a fire signal based on a fire operation performed by the artificial neuron unit 330. .

For example, the artificial synapse unit 310 may output current according to the controlled conductivity.

For example, the at least one second neuromorphic device includes a first electrode, a second electrode disposed opposite to the first electrode, a switching layer formed of germanium sulfide (GeS ₂ ) between the first electrode and the second electrode, and a switching layer. A source layer formed of copper telluride (CuTe) between the layer and the first electrode to control either drift or diffusion of metal ions in the switching layer, and a maximum current density ) When no additional voltage is applied through the first electrode at a size of 400 A / cm ² or less, the resistance returns to the initial state by the diffusion in the switching layer Self-reset volatile (volatile) can represent

According to an embodiment of the present invention, at least one second neuromorphic device has a maximum current density of 400 A/cm ² or less, and the amplitude of the voltage applied through the first electrode is When it is increased, the magnitude of the current related to the integration characteristics of the artificial neuron increases, and the pulse width of the voltage applied through the first electrode increases when the maximum current density is 400 A/cm ² or less. When it increases, the magnitude of the output current related to the integral characteristics of the artificial neuron increases, and the pulse interval (pulse interval) of the voltage applied through the first electrode when the magnitude of the maximum current density is 400 A/cm ² or less interval) may decrease the magnitude of the output current related to the integral characteristics of the artificial neuron.

For example, at least one first neuromorphic device includes a first electrode, a second electrode disposed opposite to the first electrode, and a switching layer formed of germanium sulfide (GeS ₂ ) between the first electrode and the second electrode. and a source layer formed of copper telluride (CuTe) between the switching layer and the first electrode to control any one of drift and diffusion of metal ions in the switching layer, wherein the maximum current density (max current density) exceeds 400 A/cm ² and no additional voltage is applied through the first electrode at 2000 A/cm ² or less, non-volatile that does not self-reset can represent

According to an embodiment of the present invention, at least one first neuromorphic device has a maximum current density exceeding 400 A/cm ² and a pulse width equal to or less than 2000 A/cm ² . ), magnitude (amplitude), and short term memory in which the magnitude of the output current is reduced to the initial current magnitude after a certain time based on the number of times the voltage of the pulse interval is repeatedly applied through the first electrode. ) operation and a long term memory operation in which the size of the output current is maintained after a certain period of time may be performed.

In addition, at least one first neuromorphic device has the same pulse width and amplitude when the maximum current density exceeds 400 A/cm ² and is 2000 A/cm ² or less. and a paired pulse facilitation (PPF) characteristic in which the magnitude of an output current repeatedly increases when a voltage of a pulse interval is repeatedly applied through the first electrode.

4 is a diagram for explaining an artificial neuron circuit using a neuromorphic device according to an embodiment of the present invention.

4 illustrates an artificial neuron circuit related to an artificial neuron unit using a neuromorphic device according to an embodiment of the present invention.

Referring to FIG. 4 , the artificial neuron circuit 400 according to an embodiment of the present invention may be implemented by connecting peripheral circuits using the integration characteristics of the neuromorphic device described in FIG. 1 .

For example, the artificial neuron circuit 400 includes an input stage 401, a non-inverting three-state buffer 402, a buffer 403, a pull-down resistor 404, a neuromorphic device 405, a comparator 406, and an SR latch It includes a flip-flop 407, a D flip-flop 408 and a 4-bit synchronization counter 409.

The firing signal of the artificial neuron circuit 400 may be checked at the measuring point 410 .

According to one embodiment of the present invention, the neuromorphic device 405 may be referred to as a selector.

For example, the artificial neuron circuit 400 is based on the characteristic that the resistance of the neuromorphic element 405 returns to an initial state if no pulse is applied for 3 ms after 200 voltage pulses.

Accordingly, the artificial neuron circuit 400 senses the resistance of the neuromorphic element 405 and uses a comparator that generates a fire signal, receives the fire signal as feedback, and increases the input voltage for 3 ms. It could be a circuit that blocks it.

In addition, the artificial neuron circuit 400 bit-fills the Q node of the D flip-flop 408 through the firing signal, and this signal activates the 4-bit synchronization counter 409 to perform counting.

At the same time, in the artificial neuron circuit 400, the Q_bar signal of the D flip-flop 409 turns off the non-inverting 3-state buffer 402 to block the applied input voltage.

In addition, the artificial neuron circuit 400 adjusts the cycle of the clock signal of the 4-bit synchronization counter 409 so that the time point at which D0 and D1 of the 4-bit synchronization counter 409 are simultaneously high is 3 ms or more. Adjust the duty cycle.

In addition, the artificial neuron circuit 400 performs bit-filling on the output Q of the D flip-flop 409 when D0 and D1 of the 4-bit synchronization counter 409 become high at the same time, resulting in a neuromorphic device. An integration operation and an ignition operation may be performed while the input voltage is applied again.

5 to 6B are diagrams for explaining the characteristics of an artificial neuron of a neuromorphic device according to an embodiment of the present invention.

5 to 6B illustrate measurement results according to mimicking integral characteristics of neurons of a neuromorphic device according to an embodiment of the present invention.

Referring to the graph 500 of FIG. 5 , a result measured by applying a voltage bias to the upper electrode of the neuromorphic device according to an embodiment of the present invention and grounding the lower electrode is illustrated.

The graph 500 integrates a square pulse with a pulse width of 3 ms and a pulse interval of 300 us at the upper electrode while increasing the voltage amplitude from 0.94 V to 1.10 V at 0.04 V intervals. It shows the result of measuring the characteristics.

Specifically, the measurement result shows that 200 pulses are applied for one cycle, and there is an interval of 3 ms between each cycle.

The graph 500 is the same as that of FIG. 4 by adding a sense amplifier or a comparator that detects the decrease in resistance and decreases as voltage is applied to the neuromorphic device according to an embodiment of the present invention. Indicates that artificial neuron circuits can be implemented.

The graph 500 shows how the change width of the integral and the magnitude of the output current increase as the magnitude of the voltage pulse increases from 0.94 V to 1.10 V at intervals of 0.04 V.

Therefore, in the neuromorphic device, when the amplitude of the voltage applied through the upper electrode is increased at a maximum current density of 400 A/cm ² or less, the output current related to the integration characteristics of the artificial neuron The size and range of variation may increase.

In addition, the graph 500 shows that after applying 200 pulses having a voltage pulse size of 0.94 V to 1.10 V, and then applying a pulse having 0 V once, the current returns to the initial state, and this is repeated a total of 5 times. It shows that it exhibits uniform integral characteristics when measured.

Referring to the graph 600 of FIG. 6A, a voltage bias is applied to the upper electrode of the neuromorphic device according to an embodiment of the present invention, and the pulse width is measured by setting the lower electrode to the ground. Illustrates the integral characteristics of neurons according to changes.

The graph 600 shows that when a voltage with a pulse width fixed at 300 μs and a magnitude 1.10 V and a pulse width varied by 0.5 ms from 2 ms to 4.0 ms is applied to the upper electrode of the neuromorphic device, the magnitude of the output current gradually increases. shows an increasing trend.

The graph 600 shows that the integral change width is the smallest when the pulse width is 2 ms and the integral change width is the largest when the pulse width is 4 ms.

Therefore, when the pulse width of the voltage applied through the upper electrode increases at a maximum current density of 400 A/cm ² or less, the neuromorphic device has integrated characteristics of an artificial neuron and The magnitude of the associated output current may increase.

Referring to the graph 610 of FIG. 6B, as a result of applying a voltage bias to the upper electrode of the neuromorphic device according to an embodiment of the present invention and grounding the lower electrode, the pulse interval Illustrates the integral characteristics of neurons according to changes.

The graph 610 shows that when a voltage with a pulse width fixed at 3 ms and a magnitude 1.10 V and a pulse interval varied by 200 μs from 100 μs to 900 μs is applied to the upper electrode of the neuromorphic device, the magnitude of the output current gradually increases. shows a decreasing trend.

The graph 610 shows that the integral change width is the smallest when the pulse interval is 900 μs and the integral change width is the largest when the pulse interval is 100 μs.

Therefore, the neuromorphic device has the integrated characteristics of artificial neurons when the pulse interval of the voltage applied through the upper electrode increases at a maximum current density of 400 A/cm ² or less. The magnitude of the associated output current may be reduced.

7A to 9 are views for explaining artificial synapse characteristics of a neuromorphic device according to an embodiment of the present invention.

7A and 7B are diagrams illustrating paired pulse facilitation (PPF) characteristics among artificial synaptic characteristics of a neuromorphic device according to an embodiment of the present invention.

7A and 7B show non-volatile characteristics in which the maximum current density based on the voltage applied to the neuromorphic device according to an embodiment of the present invention is 2000 A/cm ² or more. Artificial synaptic properties according to

Referring to the graph 700 of FIG. 7A, the change in current is shown when a voltage pulse having a pulse width of 80 ms, a pulse interval of 500 μs, and a voltage magnitude of 1.1 V is applied 50 times to the upper electrode of the neuromorphic device.

In the graph 700, the current increases from about 100 μA to 500 μA in the form of short-term plasticity (STP) until the 25th voltage pulse is applied, but after that, the current saturates at about 700 μA. ) and shows a change in the form of long-term plasticity (LTP).

Referring to the graph 710 of FIG. 7B , the change in current is shown when a voltage pulse having a pulse width of 80 ms, a pulse interval of 500 μs, and a voltage magnitude of 1.1 V is applied to the upper electrode of the neuromorphic device 5 times.

A graph 710 shows that the neuromorphic device exhibits paired pulse facilitation (PPF) characteristics.

The graph 710 shows that the initial current value is about 200 μA, and the current value increases to about 400 μA as five voltage pulses are applied.

As an example, the neuromorphic device has the same pulse width, amplitude and pulse interval (max current density) of more than 400 A/cm ² and less than 2000 A/cm ² ( When a voltage of a pulse interval is repeatedly applied through the upper electrode, a paired pulse facilitation (PPF) characteristic in which the magnitude of an output current repeatedly increases may be exhibited.

8 is a diagram illustrating memory characteristics among artificial synapse characteristics of a neuromorphic device according to an embodiment of the present invention.

8 illustrates experimental results for confirming short-term memory operating characteristics and long-term memory operating characteristics according to artificial synaptic characteristics of a neuromorphic device according to an embodiment of the present invention. do.

Referring to the graph 800 of FIG. 8 , voltage pulses having a pulse width of 80 ms, a magnitude of 1.1 V, and a pulse interval of 500 μs are applied to the upper electrode of the neuromorphic device according to an embodiment of the present invention, respectively. It shows the result of the experiment progress by checking the current value after applying 50 times and rechecking the current value after 60 seconds.

According to one embodiment of the present invention, the neuromorphic device has the same pulse width and amplitude when the maximum current density exceeds 400 A/cm ² and is 2000 A/cm ² or less. ) and a short term memory operation in which the size of the output current is reduced to the initial current size after a certain time based on the number of times the voltage of the pulse interval is repeatedly applied through the upper electrode and the certain time Afterwards, any one memory operation among long term memory operations in which the size of the current output is maintained may be performed.

First, in the graph 800, when a total of five voltage pulses are applied, the current value increases from about 1.9 Х 10 ^-4 A to 3.8 Х 10 ^-4 A, and after about 60 seconds, the current value decreases to the initial value. Illustrates short-term memory operating characteristics.

On the other hand, in the graph 800, when a total of 50 voltage pulses were applied, the current value increased from about 2.0 Х 10 ^-4 A to 7.2 Х 10 ^-4 A, and after about 60 seconds, the current value increased to 7.0 Х 10 ^-4 The long-term memory operating characteristics almost equal to the current value after 50 voltage pulses are applied to A are exemplified.

The graph 800 shows that short-term memory is formed when a small number of pulses are applied, but changes to long-term memory when 50 pulses are applied, and the neuromorphic device perfectly reproduces the biological synaptic properties in the human brain. It confirms imitation.

That is, the neuromorphic device performs a short term memory operation when the number of times it is repeatedly applied through the upper electrode is 5, and long term memory when the number of times it is applied is 50 times. action can be performed.

Therefore, the present invention can secure the characteristics of pair pulse facilitation and the change from short-term memory to long-term memory, which are synaptic characteristics of the human brain, by using the memristor device.

9 is a diagram explaining a linear change characteristic of conductivity among artificial synapse characteristics of a neuromorphic device according to an embodiment of the present invention.

9 illustrates that normalized conductance has a characteristic of linearly changing by applying 100 voltage pulses to an upper electrode of a neuromorphic device according to an embodiment of the present invention.

Referring to the graph 900 of FIG. 9, the long-term potentiation and long-term suppression measurement results of the neuromorphic device are shown. As a potentiation pulse and a read pulse are continuously applied 100 times, the conductivity level value gradually increases. The strengthening characteristics of synapses that increase are exemplified, and the suppression characteristics of synapses in which the conductance level gradually decreases by successively applying a depression pulse and a lead pulse 100 times are illustrated.

Specifically, the enhancement pulse has a positive voltage of 0.8V, a pulse width of about 1 ms, and the suppression pulse has a negative voltage of -0.8 V, a pulse width of about 1 ms, and a lead pulse. has the same pulse width as 1ms and the voltage may be 0.3V.

An interval (inverval) of about 100 μs exists between the enhancement pulse and the lead pulse, and an interval (inverval) of about 100 μs exists between the suppression pulse and the lead pulse.

It can be seen that the curve fitting result of reinforcement according to the reinforcement pulse has a linearity of 3.271, and the curve fitting result of suppression according to a suppression pulse has a linearity of 5.772, and the linearity converges to 0, the better the linearity. .

10 is a diagram illustrating an artificial synaptic element array using a neuromorphic element based on a memristor element according to an embodiment of the present invention.

10 illustrates an artificial synaptic element array including a plurality of neuromorphic elements according to an embodiment of the present invention. Here, the neuromorphic device may be an artificial synaptic device exhibiting synaptic characteristics based on a maximum current density.

Referring to FIG. 10 , an artificial synaptic device array using a plurality of neuromorphic devices according to an embodiment of the present invention has a structure including neuromorphic devices between word lines and bit lines.

The artificial synaptic element array 1000 using a plurality of neuromorphic elements according to an embodiment of the present invention includes a first word line 1010, a second word line 1011, and a third word line when 4 Х 4 1013, the fourth word line 1014 includes a word line, and the first bit line 1030, the second bit line 1031, the third bit line 1033, and the fourth bit line 1034 Bit lines may be included.

For example, the neuromorphic device 1020 can operate in the same way as the neuromorphic device implementing the synaptic characteristics described in FIG. 1 and the like.

The artificial synaptic element array 1000 using a plurality of neuromorphic elements includes a first word line 1010, a second word line 1011, a third word line 1013, and a fourth word line 1014. When the input 1040 is applied, different outputs 1050 are output through the first bit line 1030, the second bit line 1031, the third bit line 1033, and the fourth bit line 1034. can do.

For example, input 1040 can be an input spike signal and output 1050 can be an output spike signal.

According to an embodiment of the present invention, in the neuromorphic device 1020, when the same voltage of any one of a set voltage and a reset voltage is repeatedly applied to the first electrode through each of a plurality of word lines, the value of the output current gradually increases. may have synaptic properties that change to

In addition, in the neuromorphic device 1020, when a set voltage having the same pulse width, pulse interval, and voltage size is repeatedly applied to the first electrode through each of a plurality of word lines, the value of the output current gradually increases to form a plurality of It may have a conductivity level.

In addition, in the neuromorphic device 1020, when a reset voltage having the same pulse width, pulse interval, and voltage level is repeatedly applied to the first electrode through each of a plurality of word lines, the value of the output current gradually decreases, thereby generating a plurality of It may have a conductivity level.

Also, the artificial synaptic element array 1000 may be referred to as a cross point array.

In the artificial synaptic element array 1000, voltage pulses are applied in the form of spikes at random times to input spike signals applied through word lines.

The artificial synaptic element array 1000 may be multiplied and added in parallel with the conductivities of the memristors and transferred to the bit line in the form of current.

The current of the bit line is converted into a voltage through a converter that converts the current into a voltage, and the integrated characteristics of the artificial neuron unit described in FIG. 3 may be induced through the voltage.

In addition, the artificial neuron unit is an artificial neuron circuit, and can drive an ignition operation by sensing resistance changes of peripheral circuits and neuromorphic elements within the artificial neuron circuit described in FIG. 4 .

Therefore, the neuromorphic device according to an embodiment of the present invention embodies the characteristics of both biological synapses and neurons responsible for memory and signal transmission in the human brain, and has both characteristics of biological synapses and neurons as a single type of device. By implementing it, the process simplification can be achieved by reducing the types of materials used when manufacturing hardware-based artificial neurons and synaptic element arrays.

As described above, although the embodiments have been described with limited drawings, those skilled in the art can make various modifications and variations from the above description. For example, the described techniques may be performed in an order different from the method described, and/or components of the described system, structure, device, circuit, etc. may be combined or combined in a different form than the method described, or other components may be used. Or even if it is replaced or substituted by equivalents, appropriate results can be achieved.

Therefore, other implementations, other embodiments, and equivalents of the claims are within the scope of the following claims.

100: neuromorphic element 110: second electrode
120: switching layer 130: source layer
140: first electrode

Claims (18)

a first electrode;
a second electrode disposed opposite to the first electrode;
A switching layer formed of an amorphous material between the first electrode and the second electrode, and
A source layer formed of an alloy material between the switching layer and the first electrode to control one of drift and diffusion of metal ions in the switching layer,
Integrate characteristics of artificial neurons according to volatility and artificial synapse characteristics according to non-volatile according to the magnitude of the maximum current density based on the voltage applied through the first electrode Characterized in that it implements any one of the characteristics
neuromorphic device. According to claim 1,
The neuromorphic device implements the integration characteristics of the artificial neuron when the magnitude of the maximum current density is 400 A/cm ² or less, and the magnitude of the maximum current density exceeds 400 A/cm ² and 2000 A/cm ² Characterized in implementing the artificial synaptic properties below
neuromorphic device. According to claim 2,
In the neuromorphic device, when the maximum current density is 400 A/cm ² or less and no additional voltage is applied through the first electrode, the self resistance returns to an initial state by the diffusion in the switching layer. Indicates volatility that is reset (self-reset), and when the magnitude of the maximum current density exceeds 400 A / cm ² and no additional voltage is applied through the first electrode at 2000 A / cm ² or less, the self-reset Characterized in that it exhibits non-volatility that does not become
artificial synaptic device. According to claim 2,
In the neuromorphic device, when the magnitude of the maximum current density is 400 A/cm ² or less and the amplitude of the voltage applied through the first electrode increases, the magnitude of the output current related to the integral characteristic of the artificial neuron increases. characterized by increasing
neuromorphic device. According to claim 2,
The neuromorphic device generates an output related to the integration characteristics of the artificial neuron when the pulse width of the voltage applied through the first electrode increases when the magnitude of the maximum current density is 400 A/cm ² or less. Characterized by an increase in the magnitude of the current
neuromorphic device. According to claim 2,
The neuromorphic device generates an output related to the integral characteristics of the artificial neuron when the pulse interval of the voltage applied through the first electrode increases when the maximum current density is 400 A/cm ² or less. Characterized by a decrease in the magnitude of the current
neuromorphic device. According to claim 2,
In the neuromorphic device, the number of times a voltage having the same pulse width, magnitude, and pulse interval is repeatedly applied through the first electrode when the magnitude of the maximum current density exceeds 400 A/cm ² and is 2000 A/cm ² or less Among the short-term memory operation in which the size of the output current is reduced to the initial current size after a predetermined time based on the long-term memory operation in which the size of the current output is maintained after the predetermined time Characterized in that performing any one memory operation
neuromorphic device. According to claim 2,
In the neuromorphic device, when the maximum current density exceeds 400 A/cm ² and is 2000 A/cm ² or less, a voltage having the same pulse width, size, and pulse interval is repeatedly applied through the first electrode. Characterized in that it exhibits a paired pulse facilitation (PPF) characteristic in which the magnitude of the output current repeatedly increases in
neuromorphic device. According to claim 1,
The first electrode and the second electrode have a thickness of 30 nm,
The switching layer and the source layer have a thickness of 10 nm,
Characterized in that the first electrode, the second electrode, the switching layer and the source layer form a crossbar-type memristor structure
neuromorphic device. According to claim 1,
The switching layer is formed of the amorphous material including germanium sulfide (GeS ₂ ),
Characterized in that the source layer is formed of the alloy material containing copper telluride (CuTe)
neuromorphic device. According to claim 1,
The first electrode and the second electrode are platinum (Pt), tungsten (W), titanium nitride (TiN), tantalum nitride (TaN), gold (Au), rubidium (Ru), iridium (Ir), Characterized in that it is formed of at least one metal material selected from palladium (Pd), titanium (Ti), hafnium (Hf), molybdenum (Mo) and niobium (Nb)
neuromorphic device. When the magnitude of the maximum current density based on the applied voltage is greater than 400 A/cm ² and less than 2000 A/cm ² Artificial synapse including at least one first neuromorphic device implementing artificial synaptic characteristics according to non-volatility synaptic division and
At least one second neuromorphic device that is electrically connected to the artificial synapse unit and implements the integral characteristics of the artificial neuron according to volatility when the magnitude of the maximum current density based on the applied voltage is 400 A / cm ² or less Characterized in that it comprises an artificial neuron unit that performs an integral operation and an ignition operation, including
Neuromorphic system. According to claim 12,
Characterized in that it further comprises a control unit for controlling conductance of the artificial synapse unit by detecting an ignition signal based on the performed firing operation.
Neuromorphic system. According to claim 12,
The at least one second neuromorphic device includes a first electrode, a second electrode disposed opposite to the first electrode, a switching layer formed of germanium sulfide (GeS ₂ ) between the first electrode and the second electrode, and A source layer formed of copper telluride (CuTe) between the switching layer and the first electrode to control any one of drift and diffusion of metal ions in the switching layer, wherein the maximum current density is 400 A When no additional voltage is applied through the first electrode at / cm ² or less, the resistance returns to an initial state by the diffusion in the switching layer. Characterized in that it exhibits self-reset volatility.
Neuromorphic system. According to claim 14,
In the at least one second neuromorphic device, when the magnitude of the voltage applied through the first electrode increases when the magnitude of the maximum current density is 400 A/cm ² or less, the magnitude of current related to the integral characteristics of the artificial neuron increases, and when the pulse width of the voltage applied through the first electrode increases when the magnitude of the maximum current density is 400 A / cm ² or less, the magnitude of the output current related to the integral characteristic of the artificial neuron increases, , When the pulse interval of the voltage applied through the first electrode increases when the magnitude of the maximum current density is 400 A / cm ² or less, the magnitude of the output current related to the integral characteristic of the artificial neuron decreases. doing
Neuromorphic system. According to claim 12,
The at least one first neuromorphic device includes a first electrode, a second electrode disposed opposite to the first electrode, a switching layer formed of germanium sulfide (GeS ₂ ) between the first electrode and the second electrode, and A source layer formed of copper telluride (CuTe) between the switching layer and the first electrode to control any one of drift and diffusion of metal ions in the switching layer, wherein the maximum current density is 400 A / cm ² When no additional voltage is applied through the first electrode at more than 2000 A / cm ² or less, it is characterized in that it exhibits non-volatility that is not self-reset.
Neuromorphic system. According to claim 16,
In the at least one first neuromorphic device, a voltage having the same pulse width, size, and pulse interval is applied through the first electrode when the maximum current density exceeds 400 A/cm ² and is 2000 A/cm ² or less. Performs any one of a short-term memory operation in which the output current size is reduced to the initial current size after a certain period of time based on the number of times of repetitive application and a long-term memory operation in which the output current level is maintained after a certain period of time characterized by
Neuromorphic system. According to claim 16,
In the at least one first neuromorphic device, a voltage having the same pulse width, size, and pulse interval is applied through the first electrode when the maximum current density exceeds 400 A/cm ² and is 2000 A/cm ² or less. Characterized in that it exhibits a double-pulse promotion characteristic in which the magnitude of the output current repeatedly increases when repeatedly applied
Neuromorphic system.

Images