Spiking artificial neural network-based nanoelectromechanical neuron device

Abstract

The present invention relates to a spiking artificial neural network-based nanoelectromechanical neuron device, the device comprising: a first gate electrode receiving a first voltage; a second gate electrode disposed in parallel with the first gate electrode while being spaced apart from the first gate electrode at a predetermined interval, and receiving a second voltage; a drain electrode disposed to be spaced apart from one end of the first gate electrode at a predetermined interval and receiving a spike current signal; a movable beam disposed between the first gate electrode and the second gate electrode and bent in the direction of the drain electrode according to the spike current signal; and an anchor fixing the movable beam to a lower layer. Therefore, the present invention can implement an integrate & fire (IF) neuron by using a nanoelectromechanical (NEM) relay switch, thereby providing high integration and low power compared to existing complementary metal oxide semiconductor (CMOS)-based neurons.

Description

Nano-electromechanical neuron device based on spiking artificial neural network

[0001] The present invention relates to a technology for implementing a spiking artificial neural network hardware, and more specifically, to a nano electro mechanical neuron device based on a spiking artificial neural network using a nano electro mechanical (NEM) relay switch.

[0002]

[0003] Recent advancements in AI technologies, such as autonomous driving, image processing, and ChatGPT, that rival human intelligence, have become a hot topic. However, while the human brain consumes only about 20W of energy during cognitive processes, high-performance AI demands massive data processing, inevitably leading to significant power consumption when processed on hardware. Therefore, neuromorphic computing, based on spiking neural networks (SNNs), has garnered significant attention to mimic the ultra-low-power operation of the human brain.

[0004] The human brain consists of a vast number of neurons interconnected by synapses. Neurons accumulate electrical charge on their membranes in response to incoming stimuli. When the voltage generated by this membrane charge exceeds a certain threshold, a spike is emitted, transmitting an electrical signal to the next neuron connected to the synapse. SNNs operate very similarly to the human brain, and active research is underway to efficiently implement neuronal circuits that accumulate membrane charge and generate spikes.

[0005] However, until recently, hardware-based SNNs have relied solely on CMOS technology, which consists solely of transistors to implement neuron operations. CMOS-based neuron circuits utilize a large number of transistors, approximately 10 , which occupies a large area, hindering high integration, and suffer from significant energy loss due to leakage current.

[0006] Therefore, it is necessary to develop next-generation neural circuits that consume as little power as possible.

[0007]

[0008] [Prior Art Literature]

[0009] [Patent Document]

[0010] (Patent Document 1) Korean Patent No. 10-2107784 (April 28, 2020)

[0011]

[0012] One embodiment of the present invention provides a nano-electromechanical neuron device based on a spiking artificial neural network that can achieve area gain by replacing a large number of transistors for implementing a conventional CMOS-based neuron with a single nano-electromechanical relay switch.

[0013] One embodiment of the present invention provides a nano-electromechanical neuron device based on a spiking artificial neural network capable of an ultra-high integration structure in which a neuron circuit is implemented on a synapse array, since a nano-electromechanical relay switch is three-dimensionally integrated on a metal wiring layer on a CMOS logic circuit.

[0014] One embodiment of the present invention provides a nano-electromechanical neuron device based on a spiking artificial neural network capable of ultra-low power operation by eliminating dynamic power consumption occurring in a transistor and preventing leakage current from occurring due to a gap in a nano-electromechanical device.

[0015]

[0016] Among the embodiments, a nano-electromechanical neuron device based on a spiking artificial neural network includes a first gate electrode receiving a first voltage input; a second gate electrode disposed parallel to and spaced apart from the first gate electrode and receiving a second voltage input; a drain electrode disposed spaced apart from one end of the first gate electrode by a certain distance and receiving a spike current signal input; a movable beam disposed between the first gate electrode and the second gate electrode and bent in the direction of the drain electrode according to the spike current signal; and an anchor for fixing the movable beam to a lower layer.

[0017] The above nano-electromechanical neuron device comprises a membrane capacitor (C) connected to the drain electrode. mem ); and an output resistance (R) connected to the above movable beam out ) may be included.

[0018] The above nano-electromechanical neuron device has a leakage resistor (R) connected in parallel to the same node as the membrane capacitor. leak ) may be included.

[0019] The above nano-electromechanical neuron device has a leakage resistance (R) when the spike current signal is not input. leak ) can cause the charge charged in the membrane capacitor to leak, thereby performing a leaky IF (Leaky Integrate and Fire) operation.

[0020] The above nano-electromechanical neuron device can be integrated in a three-dimensional form on a metal wiring layer on top of a CMOS logic circuit.

[0021] The nano-electromechanical neuron device may include a plurality of operating states, one of which may correspond to an initial state in which the movable beam is parallel to the gate electrodes between the first gate electrode and the second gate electrode; an integrate state in which the movable beam is bent in the direction of the drain electrode; and a fire state in which the movable beam is in contact with the drain electrode.

[0022] The above initial state and integrated state may correspond to a state in which the movable beam and the drain electrode are physically separated.

[0023] The above nano-electromechanical neuron device can control the threshold potential at which an output spike occurs by applying a constant voltage to the first gate electrode and the second gate electrode.

[0024] The above nano-electromechanical neuron device can perform an excitatory operation in which an electrostatic force is generated in the direction of the drain electrode when a positive voltage is applied to the first gate electrode, thereby reducing the threshold potential for spike generation.

[0025] The above nano-electromechanical neuron device can perform an inhibitory operation in which an electrostatic force is generated in the opposite direction of the drain electrode when a positive voltage is applied to the second gate electrode, thereby increasing the threshold potential.

[0026] The above movable beam can be formed of any one selected from copper (Cu), aluminum (Al), molybdenum (Mo), cobalt (Co), and combinations thereof.

[0027] In one embodiment, a spiking artificial neural network-based nano-electromechanical neuron device includes a nano-electromechanical (NEM) relay switch that applies a spike current signal to a drain electrode to perform a pull-in operation of a movable beam with respect to the drain electrode, and in which the spike current signal is accumulated and fired during the pull-in operation of the movable beam; a membrane capacitor connected to the drain electrode and charged in an accumulation state of the NEM relay switch and discharged in an firing state of the NEM relay switch; and an output resistor connected to the movable beam and outputting a spike voltage signal by a discharge of the membrane capacitor.

[0028] The above-mentioned movable beam is disposed between the first gate electrode and the second gate electrode, and one end is fixed through an anchor and the other end can be horizontally moved so as to be in contact with the drain electrode while not in contact with the first gate electrode and the second gate electrode.

[0029] The above NEM relay switch can be formed on a metal wiring layer on top of a CMOS logic circuit.

[0030] The above NEM relay switch can perform the pull-in operation in which an air gap between the drain electrode and the movable beam is charged by a spike current signal applied to the drain electrode, an electrostatic force is generated between the drain electrode and the movable beam, and the movable beam is bent in the direction of the drain electrode.

[0031] The above NEM relay switch is such that when the potential of the drain electrode is higher than a preset threshold potential, the movable beam comes into contact with the drain electrode, so that the charge stored in the air gap between the drain electrode and the movable beam is discharged, and a restoring force is applied to the movable beam so that the movable beam can return to its initial position from the drain electrode.

[0032] The above NEM relay switch can control the threshold potential that determines spike generation by applying a specific voltage to the first gate electrode or the second gate electrode, thereby enhancing (excitatory) or suppressing (inhibitory) the spike generation.

[0033] Among the embodiments, a spiking artificial neural network-based nano-electromechanical neuron device includes a nano-electromechanical (NEM) relay switch that applies a spike current signal to a drain electrode to perform a pull-in operation of a movable beam with respect to the drain electrode and accumulates and fires the spike current signal during the pull-in operation of the movable beam; a membrane capacitor connected to the drain electrode and charged in an accumulation state of the NEM relay switch and discharged in an firing state of the NEM relay switch; a leakage resistor connected in parallel to the membrane capacitor and leaking charges charged in the membrane capacitor; and an output resistor connected to the movable beam and outputting a spike voltage signal by a discharge of the membrane capacitor.

[0034] The above NEM relay switch can perform a pull-in operation of the movable beam by leaking the charge stored in the membrane capacitor through the leakage resistor when a spike current signal is not applied to the drain electrode.

[0035] The above movable beam can be formed of any one selected from copper (Cu), aluminum (Al), molybdenum (Mo), cobalt (Co), and combinations thereof.

[0036]

[0037] The disclosed technology may have the following effects. However, this does not mean that a particular embodiment must include all or only the following effects, and thus the scope of the disclosed technology should not be construed as being limited thereby.

[0038] A nano-electromechanical neuron device based on a spiking artificial neural network according to one embodiment of the present invention can achieve the effect of high integration implementation by using a single nano-electromechanical (NEM) relay switch integrated into a metal wiring layer, unlike a conventional neuron circuit using a large number of transistors.

[0039] In addition, the nano-electromechanical neuron device based on a spiking artificial neural network according to one embodiment of the present invention operates as an open circuit between the movable beam and the drain electrode before a spike occurs, so that leakage current, which was a concern in the existing transistor-based neuron circuit, does not occur, and thus an effect of enabling implementation of an ultra-low-power neuron circuit can be obtained.

[0040]

[0041] FIG. 1 is a schematic diagram illustrating a nano-electromechanical neuron device based on a spiking artificial neural network according to one embodiment of the present invention.

[0042] FIG. 2 and FIG. 3 are diagrams illustrating a nano-electromechanical neuron device based on a spiking artificial neural network according to one embodiment of the present invention.

[0043] FIGS. 4a-4c are perspective views (a) and plan views (b) of each state of a nano-electromechanical neuron device based on a spiking artificial neural network according to one embodiment of the present invention.

[0044] FIG. 5 is a diagram illustrating the operation process of a nano-electromechanical neuron device based on a spiking artificial neural network according to one embodiment of the present invention and an equivalent circuit and timing diagram thereof.

[0045] Figure 6 is a graph showing the results of an HSPICE simulation performed to verify the operation of the nano-electromechanical neuron device of Figure 2.

[0046] Figure 7 is a graph showing the results of an HSPICE simulation performed to verify the operation of a nano-electromechanical neuron device that implements the leaky IF (Leaky Integrate and Fire) operation of Figure 3.

[0047] Figure 8 is a diagram showing the results of ANSYS simulation for verifying the operation of a nano-electromechanical neuron device based on a spiking artificial neural network.

[0048]

[0049] The description of the present invention is merely an example for structural and functional explanation, and therefore, the scope of the present invention should not be construed as being limited by the embodiments described in the text. That is, since the embodiments can be modified in various ways and can take various forms, the scope of the present invention should be understood to include equivalents that can realize the technical idea. In addition, the purposes or effects presented in the present invention do not mean that a specific embodiment must include all of them or only such effects, and therefore, the scope of the present invention should not be construed as being limited thereby.

[0050] Meanwhile, the meaning of the terms described in this application should be understood as follows.

[0051] Terms such as "first" and "second" are intended to distinguish one component from another, and the scope of the rights should not be limited by these terms. For example, the first component may be referred to as the second component, and similarly, the second component may also be referred to as the first component.

[0052] When a component is said to be "connected" to another component, it should be understood that while it may be directly connected to that other component, there may also be other components intervening. Conversely, when a component is said to be "directly connected" to another component, it should be understood that there are no other intervening components. Similarly, other expressions describing relationships between components, such as "between" and "directly between," or "adjacent to" and "directly adjacent to," should be interpreted similarly.

[0053] Singular expressions should be understood to include plural expressions unless the context clearly indicates otherwise, and terms such as "comprises" or "have" should be understood to specify the presence of a feature, number, step, operation, component, part or combination thereof, but not to exclude the possibility of the presence or addition of one or more other features, numbers, steps, operations, components, parts or combinations thereof.

[0054] For each step, the identifiers (e.g., a, b, c, etc.) are used for convenience of explanation and do not describe the order of the steps. The steps may occur in a different order than stated unless the context clearly dictates a specific order. That is, the steps may occur in the same order as stated, may be performed substantially simultaneously, or may be performed in the opposite order.

[0055] Unless otherwise defined, all terms used herein have the same meaning as commonly understood by those of ordinary skill in the art to which this invention pertains. Terms defined in commonly used dictionaries should be interpreted to be consistent with their meaning within the context of the relevant technology, and should not be interpreted as having an idealized or overly formal meaning unless explicitly defined herein.

[0056] Hereinafter, preferred embodiments of the present invention will be described in more detail with reference to the attached drawings. Identical components in the drawings will be denoted by the same reference numerals, and redundant descriptions of identical components will be omitted.

[0057]

[0058] Recent hardware-based artificial neural network technologies require ultra-low power consumption and high energy efficiency to process massive amounts of data. Hardware-based SNNs capable of low-power operation require neuron circuits for spiking.

[0059] In the present invention, a neuron circuit utilizing the low power and high integration characteristics of a nano electromechanical relay switch is proposed.

[0060]

[0061] FIG. 1 is a schematic diagram illustrating a nano-electromechanical neuron device based on a spiking artificial neural network according to one embodiment of the present invention.

[0062] Referring to FIG. 1, a spiking artificial neural network device can be composed of synaptic and IF (Integrate and Fire) neuron circuits.

[0063] In a spiking neural network (SNN), a single neuron processes information from multiple inputs, generating a single spike output signal and transmitting it to the next neuron. A spike occurs when the accumulated signal from the previous input exceeds a certain potential. The probability of a neuron generating a spike increases with excitatory inputs and decreases with inhibitory inputs. Implementing this in hardware requires an IF (integrate & fire) circuit, which performs the same function as the neuron.

[0064] Here, the IF neuron circuit can be implemented as a nano electromechanical neuron device using a nano electromechanical (NEM) relay switch. The NEM relay switch can include two gate electrodes (110, 120), a movable beam (130) located between the gate electrodes (110, 120), and one drain electrode (140). The NEM relay switch can perform an IF operation in which a spike current signal is applied to the drain electrode (140) to perform a pull-in operation of the movable beam (130) with respect to the drain electrode (140), and the accumulation and firing of the spike current signal occur during the pull-in operation of the movable beam (130). The nano electromechanical neuron device can be implemented as an output resistor (R) at a node such as the movable beam (130). out ) can be connected. The nano-electromechanical neuron device has a membrane capacitor (C) connected to the drain electrode (140). mem) can be connected. The nano-electromechanical neuron device can receive spike signals from synapses, where the spike signals can correspond to current. The nano-electromechanical neuron device can include multiple operating states. The multiple operating states can include an initial state, an integrate state, and a fire state. The multiple operating states can each be determined by the position of the movable beam (130).

[0065]

[0066] FIG. 2 and FIG. 3 are diagrams illustrating a nano-electromechanical neuron device based on a spiking artificial neural network according to one embodiment of the present invention.

[0067] Referring to FIGS. 2 and 3, the nano-electromechanical neuron device is applied with a first voltage (V G1 ) is input, the first gate electrode (210) is arranged parallel to the first gate electrode (210) and spaced apart from the second voltage (V G2 ) are positioned at a certain distance from the second gate electrode (220) and the first gate electrode (210) receiving the spike current signal (I spike ) is configured to include a drain electrode (240) that receives input.

[0068] The movable beam (230) is arranged between the first gate electrode (210) and the second gate electrode (220). One end of the movable beam (230) is fixed through an anchor, and the other end can be horizontally moved so as to be in contact with the drain electrode (240) while not in contact with the first gate electrode (210) and the second gate electrode (220). The movable beam (230) receives a spike current signal (I spike) is bent in the direction of the drain electrode (240), and the voltage applied to the drain electrode (240) induces a pull-in operation of the movable beam (230). Here, the movable beam (230) may be formed of any one selected from copper (Cu), aluminum (Al), molybdenum (Mo), cobalt (Co), and a combination thereof. In one embodiment, the larger the Young's modulus, which is a mechanical characteristic indicating the rigidity of a metal material, the more difficult the pull-in of the movable beam (230). The Young's modulus according to the type of metal is as shown in Table 1 below.

[0069] [Table 1]

[0070]

[0071] Additionally, the output resistance (260, R) to the node such as the movable beam (230) out ) is connected, and a membrane capacitor (250, C) is connected to the drain electrode (240). mem ) is connected. The membrane capacitor (250) is connected to the drain electrode (240) and can be charged in the accumulation state of the NEM relay switch and discharged in the firing state of the NEM relay switch. The output resistor (260) is connected to the movable beam (230) and generates a spike voltage signal (V) by the discharge of the membrane capacitor (250). out ) can be output. And, it can be configured as an anchor that fixes the movable beam (230) to the lower layer.

[0072] Also, as shown in Fig. 3, a membrane capacitor (250, C mem ) and leakage resistance (270, R) at the same node leak ) can be connected in parallel to implement a leaky IF (Leaky Integrate and Fire) neuron circuit. A nano-electromechanical neuron device with implemented leaky IF characteristics generates a spike current signal (I spike ) is not entered, the leakage resistance (270, R leak ) through a membrane capacitor (250, Cmem ) is leaked. The NEM relay switch can perform a pull-in operation of the movable beam (230) by leaking the charge charged in the membrane capacitor (250) through the leakage resistor (270) when a spike current signal is not applied to the drain electrode (240).

[0073] Here, nano-electromechanical neuron devices can be integrated in a three-dimensional form on top of a metal wiring layer on top of a CMOS logic circuit.

[0074]

[0075] FIGS. 4a to 4c illustrate perspective views (a) and plan views (b) of each state of a nano-electromechanical neuron device based on a spiking artificial neural network according to one embodiment of the present invention, and are intended to explain the initial state, the integrate state, and the fire state of the nano-electromechanical neuron device.

[0076] First, referring to FIG. 4A, the initial state of the nano-electromechanical neuron device refers to a state in which no charge is stored in the membrane capacitor (450) and no force is applied between the drain electrode (440) and the movable beam (430). That is, the movable beam (430) is maintained in a state parallel to the gate electrodes (410, 420) between the first gate electrode (410) and the second gate electrode (420).

[0077] Referring to FIG. 4b, when a current spike signal is input from the front end to the drain electrode (440) in the integrated state of the nano-electromechanical neuron device, the corresponding charge is stored in the membrane capacitor (450) and the drain electrode (440) has a potential. At this time, an electrostatic force is applied to the movable beam (430) and the drain electrode (440), so that the movable beam (430) receives a force toward the fixed drain electrode (440) and gradually bends. Here, the movable beam (430) is bent but is not in contact with the drain electrode (440). The nano-electromechanical neuron device can perform a pull-in operation in which an air gap between the drain electrode (440) and the movable beam (430) is charged by a spike current signal applied to the drain electrode (440), and an electrostatic force is generated between the drain electrode (440) and the movable beam (430), thereby bending the movable beam (430) toward the drain electrode (440).

[0078] Referring to FIG. 4C, when the potential of the drain electrode (440) exceeds a certain threshold in the firing state of the nano-electromechanical neuron device, the movable beam (430) comes into contact with the drain electrode (440). In the nano-electromechanical neuron device, when the potential of the drain electrode (440) is higher than the preset threshold potential, the movable beam (430) comes into contact with the drain electrode (440), so that the charge stored in the gap between the drain electrode (440) and the movable beam (430) is discharged, and a restoring force is applied to the movable beam (430), so that the movable beam (430) can return to the initial position from the drain electrode (440).

[0079] That is, when the movable beam (430) exists between the first gate electrode (410) and the second gate electrode (420) in a non-stressed state, it is an initial state; when the movable beam (430) is bent by force toward the drain electrode (440) but does not come into contact with the drain electrode (440), it is an integrated state; and when the movable beam (430) comes into contact with the drain electrode (440) by receiving sufficient force, it is a fire state. This performs the role of a switch that configures an open and short circuit between two nodes by utilizing the principle of mechanical movement.

[0080]

[0081] FIG. 5 illustrates the operation process of a nano-electromechanical neuron device based on a spiking artificial neural network according to one embodiment of the present invention and an equivalent circuit and timing diagram thereof, and represents the operation process for the operation states of the nano-electromechanical neuron device as a planar circuit diagram.

[0082] First, referring to (a) of Fig. 5, the initial state is illustrated, in which the movable beam (530) does not contact any of the first gate electrode (510), the second gate electrode (520), or the drain electrode (540). At this time, when a current spike generated in the previous stage is applied to the drain electrode (540), the capacitor (C) by the movable beam (530) and the drain electrode (540) gap , C mem ) and a positive voltage (V) is applied to the drain electrode (540). D ) is applied, and the movable beam is slowly pulled toward the drain electrode as shown in Fig. 5(b). The current spike signal (I) generated at the front end spike ) corresponds to a kind of integration operation that converts the potential energy of the movable beam.

[0083] Referring to Fig. 5(c), when a voltage higher than the threshold potential is applied to the drain electrode, the movable beam comes into contact with the drain electrode. The gap existing between the drain electrode and the movable beam is transformed into a short circuit between the metals, resulting in a fire operation in which the charge accumulated in the drain electrode is instantly discharged to the movable beam.

[0084] Next, referring to Fig. 5(d), the discharged charge generates a spike current, which is connected in series to the output resistor (R) of the movable beam. out ) spike voltage signal (V) according to Ohm's law out ) is generated. The charge stored in the drain electrode is instantly discharged, and only the restoring force in the opposite direction of the drain electrode acts on the movable beam, returning it to its original initial state.

[0085] In the operation process of such a nano-electromechanical neuron device, the threshold potential that determines spike generation can be controlled by applying a specific voltage to the first gate electrode (510) or the second gate electrode (520), thereby enhancing or suppressing spike generation. That is, the positive voltage (V) applied to the first gate electrode (510) G1 ) applies an electrostatic force toward the drain electrode (540) to the movable beam (530), thereby assisting the movable beam (530) to contact the drain electrode (540) and perform a fire operation even at a lower drain voltage. This corresponds to an excitatory operation that reduces the threshold potential for spike generation in the SNN. Conversely, the positive voltage (V) applied to the second gate electrode (520) G2 ) applies an electrostatic force to the movable beam (530) in the opposite direction to the drain electrode (540), thereby causing a fire operation at a higher drain voltage. This corresponds to an inhibitory operation that increases the threshold potential for spike generation.

[0086]

[0087] FIG. 6 is a graph showing the operation verification through HSPICE simulation of a nano-electromechanical neuron device based on a spiking artificial neural network according to one embodiment of the present invention.

[0088] Fig. 6(a) shows a spike current signal input to the drain electrode of a nano-electromechanical neuron device based on a spiking artificial neural network. Fig. 6(b) shows the position of the movable beam, Fig. 6(c) shows the voltage of the drain electrode, and Fig. 6(d) shows the output spike voltage signal over time.

[0089] In Fig. 6(b) to Fig. 6(d), when there is no gate voltage (A), when voltage is applied to the first gate electrode (B), and when voltage is applied to the second gate electrode (C) are shown, respectively.

[0090] As can be seen in Fig. 6, a spike current accumulates in the drain electrode, and when a certain threshold voltage is exceeded, a pull-in operation of the movable beam occurs, generating an output spike voltage. Furthermore, it can be confirmed that the threshold voltage for generating an output spike can be controlled by applying voltage to the first and second gate electrodes.

[0091]

[0092] Figure 7 is a graph showing the results of an HSPICE simulation of a nano-electromechanical neuron device based on a spiking artificial neural network that implements the leaky IF (Leaky Integrate and Fire) operation of Figure 3.

[0093] Figures 7(a), 7(b), and 7(c) respectively show the spike current signal (I spike signal), membrane capacitor voltage (C mem node voltage) and output resistance voltage (R outvoltage). As can be seen in Fig. 7, the HSPICE simulation results show that the leakage resistance (R) is leak ) can be used to confirm that the charge stored in the membrane capacitor is leaking.

[0094]

[0095]

[0096] Figure 8 is a diagram showing an ANSYS simulation to verify the operation of a nano-electromechanical neuron device based on a spiking artificial neural network.

[0097] Figure 8 shows the verification of the operation of the nano-electromechanical neuron device through ANSYS simulation using finite element analysis. The integrated dynamic occurs due to the voltage applied to the drain electrode, and the fire operation can be confirmed at a drain voltage of approximately 2 V. In the case of excitatory voltage applied to the first gate electrode, the fire operation can be confirmed to occur at a drain voltage of 1.5 V, which is a lower voltage.

[0098]

[0099] As described above, if a neuron circuit is implemented using the electrostatic force and restoring force that are major factors in relay switch operation in a spiking artificial neural network-based nano-electromechanical neuron device, the neuron circuit has the advantage of high integration because it uses a single nano-electromechanical relay switch integrated in a metal wiring layer, unlike a conventional neuron circuit that uses a large number of transistors.

[0100] Additionally, before a spike occurs, the movable beam and drain electrode operate as an open circuit, which has the great advantage of preventing leakage current, which has been a concern in transistor-based neuron circuits.

[0101] It is expected that Korea will be able to preempt and lead the next-generation low-power and high-energy-efficiency edge AI market by utilizing the low-power / high-integration characteristics of the nano-electromechanical neuron device based on a spiking artificial neural network according to one embodiment of the present invention. In addition, the nano-electromechanical relay switch of the neuron has characteristics that are resistant to radiation and cosmic rays compared to silicon-based devices, and thus, in the fields of military applications of artificial neural networks and application of space technology, it is possible to possess core components and circuit technology that can operate normally even in extreme situations that cannot be handled by existing transistor-based technologies.

[0102] This spiking artificial neural network-based nano-electromechanical neuron device is a technology that can accelerate the implementation of a new concept of high-energy-efficiency hardware artificial neural network, and is expected to secure competitiveness in the rapidly growing ultra-low-power neuromorphic computing field and contribute to leading the overall next-generation low-power and high-energy-efficiency artificial intelligence semiconductor market.

[0103]

[0104] Although the present invention has been described above with reference to preferred embodiments thereof, it will be understood by those skilled in the art that various modifications and changes may be made to the present invention without departing from the spirit and scope of the present invention as set forth in the claims below.

[0105]

[0106] [Explanation of symbols]

[0107] 110,210,410,510: First gate electrode

[0108] 120,220,420,520: Second gate electrode

[0109] 130,230,430,530: Mobile beam

[0110] 140,240,440,540: Drain electrode

[0111] 150,250: Membrane capacitor

[0112] 260: Output resistance

[0113] 270: Leakage resistance

Claims

1. A first gate electrode receiving a first voltage; A second gate electrode arranged parallel to the first gate electrode and spaced apart from the first gate electrode and receiving a second voltage; A drain electrode positioned at a certain distance from the first gate electrode and receiving a spike current signal; A movable beam disposed between the first gate electrode and the second gate electrode and bent in the direction of the drain electrode according to the spike current signal; and A nano-electromechanical neuron device based on a spiking artificial neural network comprising anchors for fixing the above-mentioned movable beam to a lower layer.

2. In the first paragraph, the nano-electromechanical neuron device A membrane capacitor (C) connected to the above drain electrode mem ); and Output resistance (R) connected to the above movable beam out ) is further characterized by a spiking artificial neural network-based nano-electromechanical neuron device.

3. In the second paragraph, the nano-electromechanical neuron device A leakage resistor (R) connected in parallel to the same node as the above membrane capacitor leak ) is further characterized by a spiking artificial neural network-based nano-electromechanical neuron device.

4. In the third paragraph, the nano-electromechanical neuron device When the above spike current signal does not come in, the leakage resistance (R leak ) is characterized by a spiking artificial neural network-based nano-electromechanical neuron device that performs a leaky IF (Leaky Integrate and Fire) operation by leaking the charge charged in the membrane capacitor.

5. In the first paragraph, the nano-electromechanical neuron device A nano-electromechanical neuron device based on a spiking artificial neural network characterized by being integrated in a three-dimensional form on a metal wiring layer on top of a CMOS logic circuit.

6. In the first paragraph, the nano-electromechanical neuron device Contains multiple operating states, One of the above multiple operating states An initial state in which the above-mentioned movable beam is parallel to the gate electrodes between the first gate electrode and the second gate electrode; An integrated state in which the above-mentioned movable beam is bent in the direction of the drain electrode; and A nano-electromechanical neuron device based on a spiking artificial neural network, characterized in that the above-mentioned movable beam corresponds to a fire state in contact with the drain electrode.

7. In the 6th paragraph, the initial state and the integrated state are A nano-electromechanical neuron device based on a spiking artificial neural network, characterized in that the above-mentioned movable beam and the above-mentioned drain electrode are physically separated.

8. In the first paragraph, the nano-electromechanical neuron device A nano-electromechanical neuron device based on a spiking artificial neural network, characterized in that a threshold potential at which an output spike occurs is controlled by applying a constant voltage to the first gate electrode and the second gate electrode.

9. In the first paragraph, the nano-electromechanical neuron device A nano-electromechanical neuron device based on a spiking artificial neural network, characterized in that when a positive voltage is applied to the first gate electrode, an electrostatic force is generated in the direction of the drain electrode, thereby performing an excitatory operation that reduces the threshold potential for spike generation.

10. In the first paragraph, the nano-electromechanical neuron device A nano-electromechanical neuron device based on a spiking artificial neural network, characterized in that when a positive voltage is applied to the second gate electrode, an electrostatic force is generated in the opposite direction of the drain electrode, thereby performing an inhibitory operation to increase the threshold potential.

11. In paragraph 1, the movable beam A nano-electromechanical neuron device based on a spiking artificial neural network, characterized in that it is formed of one selected from copper (Cu), aluminum (Al), molybdenum (Mo), cobalt (Co), and combinations thereof.

12. A nano electro mechanical (NEM) relay switch that performs a pull-in operation of a movable beam with respect to the drain electrode by applying a spike current signal to the drain electrode, and in which accumulation and firing of the spike current signal occur during the pull-in operation of the movable beam; A membrane capacitor connected to the drain electrode and charged in the accumulation state of the NEM relay switch and discharged in the firing state of the NEM relay switch; and A nano-electromechanical neuron device based on a spiking artificial neural network, comprising an output resistor connected to the above-described movable beam and outputting a spike voltage signal by the discharge of the membrane capacitor.

13. In paragraph 12, the movable beam A nano-electromechanical neuron device based on a spiking artificial neural network, characterized in that the nano-electromechanical neuron device is arranged between a first gate electrode and a second gate electrode, one end of which is fixed via an anchor and the other end of which is horizontally movable so as to be in contact with the drain electrode while not in contact with the first gate electrode and the second gate electrode.

14. In the 12th paragraph, the NEM relay switch A nano-electromechanical neuron device based on a spiking artificial neural network characterized by being formed on a metal wiring layer on top of a CMOS logic circuit.

15. In the 12th paragraph, the NEM relay switch A nano-electromechanical neuron device based on a spiking artificial neural network, characterized in that a spike current signal applied to the drain electrode charges an air gap between the drain electrode and the movable beam, an electrostatic force is generated between the drain electrode and the movable beam, and the pull-in operation of bending the movable beam toward the drain electrode is performed.

16. In the 15th paragraph, the NEM relay switch A spiking artificial neural network-based nano-electromechanical neuron device, characterized in that when the potential of the drain electrode is higher than a preset threshold potential, the movable beam comes into contact with the drain electrode, so that charges accumulated in the air gap between the drain electrode and the movable beam are discharged, and a restoring force is applied to the movable beam so that the movable beam returns to the initial position from the drain electrode.

17. In the 13th paragraph, the NEM relay switch A nano-electromechanical neuron device based on a spiking artificial neural network, characterized in that the spike generation is strengthened (excitatory) or suppressed (inhibitory) by controlling the threshold potential that determines the spike generation by applying a specific voltage to the first gate electrode or the second gate electrode.

18. A nano electro mechanical (NEM) relay switch that performs a pull-in operation of a movable beam with respect to the drain electrode by applying a spike current signal to the drain electrode, and in which accumulation and firing of the spike current signal occur during the pull-in operation of the movable beam; A membrane capacitor connected to the drain electrode and charged in the accumulation state of the NEM relay switch and discharged in the firing state of the NEM relay switch; A leakage resistor connected in parallel with the membrane capacitor and leaking the charge charged in the membrane capacitor; and A nano-electromechanical neuron device based on a spiking artificial neural network, comprising an output resistor connected to the above-described movable beam and outputting a spike voltage signal by the discharge of the membrane capacitor.

19. In the 18th paragraph, the NEM relay switch A nano-electromechanical neuron device based on a spiking artificial neural network, characterized in that the pull-in operation of the movable beam is performed by leaking the charge stored in the membrane capacitor through the leakage resistor when a spike current signal is not applied to the drain electrode.

20. In paragraph 18, the movable beam A nano-electromechanical neuron device based on a spiking artificial neural network, characterized in that the device is formed of one selected from copper (Cu), aluminum (Al), molybdenum (Mo), cobalt (Co), and combinations thereof.

Images