Synaptic and neuromorphic devices

Abstract

The application relates to the technical field of neuromorphic computing, and discloses a synapse device and a neuromorphic device, the synapse device comprising a friction nanogenerator and a load resistor, wherein: the positive electrode of the friction nanogenerator is electrically connected with the first end of the load resistor, and the negative electrode of the friction nanogenerator is electrically connected with the second end of the load resistor. The synapse device based on the friction nanogenerator is simple in structure, the load resistor is easy to adjust, has good synaptic plasticity, can be prepared by using an existing friction nanogenerator, is easy to prepare, and has excellent synaptic performance.

Description

A synaptic device and a neuromorphic device Technical Field

[0001] This invention relates to the field of neuromorphic computing technology, and in particular to a synaptic device and a neuromorphic device. Background Technology

[0002] With the rise of the big data era, the inefficiency of traditional von Neumann architecture general-purpose computers for data-intensive tasks has become more prominent, and the bus connecting memory and processor in von Neumann general-purpose computers has become a bottleneck for data transmission.

[0003] Inspired by the way the human brain works, neuromorphic computing holds promise for simulating the brain to achieve low-power, real-time interactive computing. The human brain is a highly complex neural network and information processing system, composed of approximately 10... 11 10 interconnected neurons and 10 15 It consists of synapses. Neuronal synapses are the basic units of cognitive activity in the brain. In order to realize neuromorphic computing in hardware, neuromorphic devices are the key to realizing neuromorphic hardware computing circuits.

[0004] Currently, synaptic devices are mainly divided into two categories: two-terminal memristors and three-terminal transistors. However, the electrical characteristics of these synaptic devices are difficult to change, and their fabrication process is often complex and requires high-level technology. Summary of the Invention

[0005] This invention provides a synaptic device and a neuromorphic device, which can expand new ideas for the research of synaptic devices and avoid the problems of complex fabrication process and high process requirements.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] This invention provides a synaptic device, including a triboelectric nanogenerator and a load resistor, wherein:

[0008] The positive terminal of the triboelectric nanogenerator is electrically connected to the first end of the load resistor, and the negative terminal of the triboelectric nanogenerator is electrically connected to the second end of the load resistor.

[0009] As an optional implementation, the above-mentioned triboelectric nanogenerator is one of the following:

[0010] Slide mode;

[0011] Contact separation mode;

[0012] Hybrid contact separation and sliding modes.

[0013] As an optional implementation, the resistance value of the load resistor is adjustable.

[0014] The aforementioned synaptic device consists of a triboelectric nanogenerator and a load resistor connected between the positive and negative terminals of the triboelectric nanogenerator. It has a simple structure, the load resistor is easily adjustable, and it exhibits good synaptic plasticity. It can be fabricated using existing triboelectric nanogenerators, making the fabrication process easy, and it also possesses excellent synaptic performance.

[0015] This invention provides a neuromorphic device, which includes an array substrate and a plurality of the above-mentioned synaptic devices, wherein:

[0016] Multiple synaptic devices are arranged in an array structure on an array substrate;

[0017] The load resistor values ​​of each synaptic device are different.

[0018] As an optional implementation, the above-mentioned plurality of synaptic devices are grouped into a single synaptic device group;

[0019] Neuromorphic devices consist of multiple synaptic device groups.

[0020] As an optional implementation, the triboelectric nanogenerator in the above-mentioned synaptic device includes a first electrode layer, a dielectric layer, and a second electrode layer;

[0021] The array substrate has holes and pillars that correspond one-to-one with the synaptic devices;

[0022] For any hole and its corresponding column, a first electrode layer and a dielectric layer are sequentially arranged on the inner wall of the hole, and a second electrode layer is arranged on the periphery of the column. When the column slides along the direction of the hole, the second electrode layer and the dielectric layer generate friction.

[0023] As an optional implementation, a second electrode layer is provided on the first end face of the columnar body away from the array substrate;

[0024] Neuromorphic devices also include test units corresponding to each synaptic device group;

[0025] For each test unit, the first end of the test unit is electrically connected to each of the corresponding first electrode layers, and when the second end of the test unit is located at different positions, it is electrically connected to the second electrode layer on the first end face of the column corresponding to the position.

[0026] The test unit is used to detect the potential difference between the positive and negative electrodes of the triboelectric nanogenerator corresponding to the location.

[0027] As an optional implementation, the above-mentioned test unit includes an electrometer;

[0028] The negative probe of the electrometer is electrically connected to each of the first electrode layers;

[0029] The positive probe of the electrometer is electrically connected to the second electrode layer on the first end face of the column corresponding to the location of the positive probe.

[0030] As an optional implementation, the above-mentioned test unit also includes a frame and a slider that can slide on the frame;

[0031] The frame and the side of the array substrate closest to the first end face of the column are arranged opposite each other;

[0032] The positive probe of the electrometer is connected to the slider so that when the slider slides to the corresponding position on the frame, the positive probe is electrically connected to the second electrode layer on the first end face of the column corresponding to the position.

[0033] As an optional implementation, the above-mentioned column is a square column.

[0034] The aforementioned neuromorphic device differentiates the characteristics of multiple synaptic devices by setting different load resistor values. Because of these differences, the potentials generated by the multiple synaptic devices can be processed as a single set of data, achieving data processing in a dimensionality-reduced manner. Attached Figure Description

[0035] Figure 1 is a schematic diagram of a synaptic device provided in an embodiment of the present invention;

[0036] Figure 2 is a schematic diagram of a basic synaptic signal provided in an embodiment of the present invention;

[0037] Figure 3 is a schematic diagram of a triboelectric nanogenerator provided in an embodiment of the present invention;

[0038] Figure 4 is a schematic diagram of the structure of a neuromorphic device provided in an embodiment of the present invention;

[0039] Figure 5 is a schematic diagram of another synaptic device provided in an embodiment of the present invention;

[0040] Figure 6 is a schematic diagram of the circuit corresponding to a neuromorphic device provided in an embodiment of the present invention;

[0041] Figure 7 is a schematic diagram of synaptic signals of a neuromorphic device provided in an embodiment of the present invention;

[0042] Figure 8 is a structural schematic diagram of a frame provided in an embodiment of the present invention.

[0043] icon:

[0044] 1-First electrode layer; 2-Dielectric layer; 3-Second electrode layer;

[0045] 4-Second substrate layer; 5-First substrate layer; 6-Array substrate;

[0046] 7-Metal layer of slider; 8-Base layer of slider; 9-Frame. Detailed Implementation

[0047] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0048] Over the past few decades, the von Neumann architecture has revolutionized computing, driving technological advancements. However, general-purpose computers based on the von Neumann architecture are inefficient for data-intensive tasks, with the bus connecting memory and processors becoming a bottleneck for data transmission, known as the von Neumann bottleneck. To improve the performance of computing systems, current computing methods must be fundamentally changed. The human brain is a highly complex neural network and information processing system. It consists of approximately 10... 11 10 interconnected neurons and 10 15 It is composed of synapses. Neuroscientists and psychologists around the world have been studying the functional architecture of the human brain, which has inspired data-centric computing approaches such as artificial neural networks (ANNs) and machine learning (ML).

[0049] In recent years, neuromorphic computing has emerged as a promising technology in the post-Moore's Law era to overcome the von Neumann bottleneck. Neuromorphic computing systems offer high connectivity and parallelism with relatively low power consumption and memory usage. To realize neuromorphic systems in hardware, it is crucial to implement artificial neurons that mimic biological neurons and artificial synapses that mimic biological synapses. These artificial neurons and synapses must be highly efficient and scalable, capable of executing relevant learning rules to achieve large-scale neuromorphic functions. Neuronal synapses are the basic units of cognitive activity in the brain; therefore, synaptic devices and neuromorphic devices are fundamental to realizing neuromorphic hardware computing circuits. Currently, synaptic devices are mainly divided into two categories: two-terminal memristors and three-terminal transistors.

[0050] However, once synaptic devices are fabricated, their plasticity is essentially fixed, and neuromorphic devices composed of these synaptic devices often involve complex fabrication processes and demanding technical requirements. Therefore, there is an urgent need for a neuromorphic device with a simple structure and adjustable synaptic plasticity.

[0051] Maxwell's electromagnetic theory laid the foundation for the development of the information age. In this theory, the current generated by surface electrostatics is the theoretical basis for piezoelectric nanogenerators (PENs) and triboelectric nanogenerators (TENs). Nanogenerators (NGs), as self-driven sensors that sense dynamic information triggered by mechanical action, have found wide applications in environmental monitoring, motion detection, the Internet of Things, healthcare, and infrastructure monitoring. Furthermore, the action potential (AP) generated by nanogenerators can serve as a presynaptic current for biomimetic synapse research. This has led to the development of techniques using triboelectric nanogenerators to generate pressure-triggered electrical signals, using reduced graphene oxide embedded in a triboelectric layer as an electron trap to achieve synaptic plasticity, and coupling the piezoelectric potential of piezoelectric nanogenerators with ion-gel-gate graphene transistors to adjust synaptic weight through strain pulses.

[0052] To address the aforementioned problems, this invention proposes a synaptic device based on a triboelectric nanogenerator. This synaptic device features a simple circuit structure and fabrication process, and exhibits excellent synaptic performance. Furthermore, based on this synaptic device, this invention provides a neuromorphic device. This neuromorphic device can effectively alter the weights of the synaptic devices by adjusting the resistance value of the load resistor, thereby improving the accuracy of feature recognition and demonstrating scalability.

[0053] Figure 1 is a schematic diagram of a synaptic device provided in an embodiment of the present invention. As shown in Figure 1, the embodiment of the present invention provides a synaptic device, which includes a triboelectric nanogenerator and a load resistor, wherein:

[0054] The positive terminal of the triboelectric nanogenerator is electrically connected to the first end of the load resistor, and the negative terminal of the triboelectric nanogenerator is electrically connected to the second end of the load resistor.

[0055] The study found that numerous channel proteins exist on the cell membrane of neuronal synapses, including sodium channel proteins, potassium channel proteins, and some normally open channel proteins (such as chloride ions). According to the Hodgkin-Huxley model (HH model), these channel proteins exhibit different permeabilities to ions, essentially acting as a battery and a resistor. The selective permeability of the cell membrane to ions and the difference in ion concentration across the membrane result in a potential difference across the membrane; therefore, the passive nature of the membrane potential is equivalent to capacitance. The Leaky Integrate-and-Fire (LIF) model neglects the specific dynamic process of the presynaptic action potential, considering only the passive nature of the membrane potential. Therefore, the LIF model simplifies the HH model to a parallel connection of capacitance and resistance driven by the input current, and the excitatory postsynaptic current in this model exhibits an exponential decay pattern related to resistance. Since a triboelectric nanogenerator is a high-impedance charge source, its first-order equivalent is essentially a series connection of a charge source and a capacitor. Therefore, after connecting a load resistor between the positive and negative terminals of the triboelectric nanogenerator, it has a high degree of similarity to the LIF model. Furthermore, the potential difference between the positive and negative terminals exhibits an exponential decay mode related to resistance, consistent with the LIF model, as shown in Figure 2. The potential difference between the positive and negative electrodes decays exponentially after rapidly reaching its maximum value. Therefore, after connecting a load resistor between the positive and negative terminals of the triboelectric nanogenerator, it can be equivalent to an ideal synaptic device.

[0056] In some possible implementations, the resistance value of the load resistor in the aforementioned synaptic device is adjustable, specifically by using an adjustable resistor as the load resistor. In practice, the potential generated by the synaptic device (i.e., the potential difference between the positive and negative electrodes of the triboelectric nanogenerator) can be adjusted by changing the resistance value of the load resistor. Under the same conditions (such as the contact area of ​​the upper and lower electrodes inside the triboelectric nanogenerator), the larger the resistance value of the load resistor, the larger the potential generated by the synaptic device, and the slower the decay rate of this potential. Therefore, using an adjustable resistor as the load resistor of the aforementioned synaptic device facilitates the adjustment of the weight of the synaptic device, i.e., the potential of the synaptic device.

[0057] In addition, in order to avoid the inability to detect the potential difference at the positive and negative poles of the triboelectric nanogenerator in this embodiment of the invention, the resistance value of the load resistor cannot be set too small, such as at the MΩ level. However, this embodiment of the invention does not limit the upper limit of the load resistor value, and it can be at the GΩ level or even the TΩ level.

[0058] In this embodiment of the invention, the triboelectric nanogenerator in the above-mentioned synaptic device is one of the following:

[0059] Slide mode;

[0060] Contact separation mode;

[0061] Hybrid contact separation and sliding modes.

[0062] Furthermore, the aforementioned triboelectric nanogenerators are not limited to solid-solid triboelectric nanogenerators, but can also be solid-liquid triboelectric nanogenerators and other phase-mode triboelectric nanogenerators.

[0063] When the above-mentioned triboelectric nanogenerator is in sliding mode, a possible structure of the triboelectric nanogenerator is shown in Figure 3. Specifically, the triboelectric nanogenerator includes a first electrode layer 1, a dielectric layer 2, and a second electrode layer 3 as shown in the figure.

[0064] The first electrode layer 1, the dielectric layer 2, and the second electrode layer 3 are stacked to form an electrode-dielectric-electrode sandwich structure; the first electrode layer 1 serves as the negative electrode of the triboelectric nanogenerator, i.e., the lower electrode; the second electrode layer 2 serves as the positive electrode of the triboelectric nanogenerator, i.e., the upper electrode.

[0065] The first electrode layer 1 is bonded to the dielectric layer 2, and the second electrode layer 3 is slidably disposed along the dielectric layer 2, that is, the second electrode layer 3 can slide on the dielectric layer 2.

[0066] In this embodiment of the invention, the dimensions of the above-mentioned synaptic device, the dimensions and size relationship of the first electrode layer, dielectric layer and second electrode layer of the triboelectric nanogenerator in the synaptic device can be freely set. However, since the first electrode layer and the dielectric layer are bonded together, the corresponding contact surfaces in the first electrode layer and the dielectric layer are usually the same size. The size of the contact surface between the second electrode layer and the dielectric layer can be set to be larger than the size of the corresponding contact surface in the dielectric layer, or it can be set to be less than or equal to it. For example, the size of the first electrode layer and the dielectric layer can be set to 1cm×2cm, and the size of the second electrode layer can be set to 1cm×1cm.

[0067] Similarly, the materials of the above-mentioned synaptic devices are not limited in the embodiments of the present invention. The electrode layer of the triboelectric nanogenerator can be selected from copper foil, aluminum, indium tin oxide (ITO) or other conductive materials, and the dielectric layer can be selected from fluorinated ethylene propylene copolymer (FEP), polytetrafluoroethylene (PTFE), polyimide (Capton) or other materials.

[0068] In some possible implementations, as shown in FIG3, the above-mentioned triboelectric nanogenerator further includes a first substrate layer 5 and a second substrate layer 4, wherein:

[0069] The first base layer 5 is disposed on the side of the first electrode layer 1 away from the dielectric layer 2;

[0070] The second substrate layer 4 is disposed on the side of the second electrode layer 3 away from the dielectric layer 2;

[0071] That is, the first base layer 5, the first electrode layer 1, the dielectric layer 2, the second electrode layer 3 and the second base layer 4 are stacked together. The first base layer 5 is used to fix the shape of the first electrode layer 1 and the second base layer 4 is used to fix the shape of the second electrode layer 3.

[0072] In this embodiment of the invention, the size and material of the first base layer 5 and the second base layer 4 are not limited. Their sizes can be the same or different, and their material can be acrylic.

[0073] Based on the above-mentioned synaptic devices, this embodiment of the invention also provides a neuromorphic device, as shown in Figures 4 and 5. This neuromorphic device includes an array substrate 6 and multiple of the above-mentioned synaptic devices (one possible structure of the synaptic device is shown in Figure 5), wherein:

[0074] Multiple synaptic devices are arranged in an array structure on an array substrate;

[0075] The load resistor values ​​of each synaptic device are different.

[0076] It should be noted that Figure 4 shows only one possible structure for a neuromorphic device. In specific implementations, the number of multiple synaptic devices is not limited, and their structure is not limited to the structure shown in Figure 5. Furthermore, the array structure of the aforementioned multiple synaptic devices is not limited to the row and column array structure shown in Figure 4; it can also be a ring-shaped structure. In this embodiment of the invention, no specific array structure is limited.

[0077] In the array arrangement, the spacing between each synaptic device can be freely set, for example, to 0.5 cm, or further expanded. Increasing the spacing helps avoid interference between the triboelectric nanogenerators in multiple synaptic devices. However, when increasing the spacing, the density of synaptic devices needs to be reasonably configured according to the scenario. Furthermore, the spacing between synaptic devices should not be too small to avoid electric field interference between the triboelectric nanogenerators of multiple synaptic devices.

[0078] The size of each synaptic device is not limited in this embodiment of the invention. For example, it can be set to 1cm×1cm. Depending on the specific scenario, it can be further enlarged or reduced, but its size should not be too small, such as 1mm×1mm, so as to avoid the inability to detect the signal.

[0079] In this embodiment of the invention, the selection of the resistance value of the load resistor for each synaptic device is not limited to this; the sliding distance of each synaptic device can also be different. The purpose is to create obvious characteristic differences between the synaptic devices for easy differentiation. In this embodiment, the difference in the resistance value of the load resistor between each synaptic device is used to ensure differentiation. At the same time, to avoid the synaptic device failing to detect the potential, the resistance value of the load resistor cannot be too small, for example, in the MΩ range, but the load resistor can be as large as possible, for example, in the GΩ or TΩ range.

[0080] In some alternative implementations, the plurality of synaptic devices are grouped into a single synaptic device group; neuromorphic devices include multiple synaptic device groups.

[0081] The structures of the synaptic device groups can be the same or different, and the number of synaptic devices and their array structures can also be the same or different. However, in specific settings, the structures, number of synaptic devices, and array structures of multiple synaptic device groups in a neuromorphic device are usually set to be the same. This application does not limit the method of dividing synaptic device groups. The method can be as follows: when all synaptic devices in the neuromorphic device form an M-row × N-column array structure (where M and N are both positive integers), the synaptic device groups can be divided by rows or columns; or when the synaptic devices form a ring-shaped array structure, each ring of synaptic devices can be considered as a synaptic device group, or synaptic devices along a radial direction can be considered as a synaptic device group.

[0082] For example, when synaptic devices are arranged in a 3×3 array structure as shown in Figure 4, the load resistors of synaptic devices in the same column can be set to have the same resistance value, while the load resistors of synaptic devices in the same row can have different resistance values. For example, the load resistor of the first column of synaptic devices can be R1, the load resistor of the second column of synaptic devices can be R2, and the load resistor of the third column of synaptic devices can be R3. The synaptic devices in each row can be grouped as a group of synaptic devices, that is, the neuromorphic device includes three groups of synaptic devices with the same structure.

[0083] The aforementioned neuromorphic device differentiates the characteristics of multiple synaptic devices by setting different load resistor values. Because of these differences, the potentials generated by the multiple synaptic devices can be processed as a single set of data, achieving data processing in a dimensionality-reduced manner.

[0084] Taking the neuromorphic device shown in Figure 4 as an example, when the load resistance values ​​of the nine synaptic devices are all the same, the neuromorphic device needs to process a total of nine data (i.e., the potentials generated by the nine synaptic devices). However, when the load resistance values ​​of each row of synaptic devices are set to be different from each other, and each row of synaptic devices is grouped as a synaptic device with the same group structure, since the characteristics of the synaptic devices in each group are different, the data generated by the synaptic device can be determined based on the data of the synaptic device group. Therefore, each synaptic device group can be treated as a data. At this time, the neuromorphic device needs to process a total of three data (i.e., the data generated by the three synaptic device groups), thus realizing data processing in a "dimensionality reduction" manner.

[0085] The following description, in conjunction with Figures 4 and 5, uses one possible neuromorphic device as an example to illustrate the specific structure of the aforementioned neuromorphic device. It should be noted that the structure of the neuromorphic device shown in the figures is only an example, and the structure of the neuromorphic device in the embodiments of the present invention is not limited to this. Furthermore, the synaptic devices do not necessarily have to be arranged on the array substrate in an embedded manner; they can also be arranged on the array substrate in a surface-mount manner or other methods, as long as the function of the neuromorphic device can be realized.

[0086] In some possible implementations, as shown in Figures 4 and 5, the triboelectric nanogenerator in the above-described synaptic device includes a first electrode layer 1, a dielectric layer 2, and a second electrode layer 3.

[0087] The array substrate 6 is provided with holes and pillars that correspond one-to-one with the synaptic devices;

[0088] For any hole and its corresponding column, a first electrode layer 1 and a dielectric layer 2 are sequentially arranged on the inner wall of the hole, and a second electrode layer 3 is arranged on the periphery of the column (i.e. the side that contacts the dielectric layer of the inner wall of the hole during sliding). When the column slides along the direction of the hole, the second electrode layer 3 and the dielectric layer 2 generate friction.

[0089] In implementation, acrylic with a preset thickness (e.g., 0.5cm) can be used as the array substrate 6, with holes on it that are the same number as the number of synaptic devices. The holes are arranged in an array structure, and the size and shape of the holes can be set by the user (e.g., set as rectangular holes of size 1cm×1cm). The shape of the column corresponds to the holes, and the height of the column can be set by the user, which can be greater than, less than or equal to the thickness of the holes.

[0090] The inner wall of the hole can be sequentially provided with a first electrode layer 1 and a dielectric layer 2 by means of adhesion. The columnar body corresponding to the hole can also be provided with a second electrode layer 3 by means of adhesion. By making the columnar body slide along the direction of the hole, the second electrode layer 3 and the dielectric layer 2 can generate electricity through friction.

[0091] The columnar body can be made of acrylic or other materials, the first electrode layer 1 and the second electrode layer 3 can be made of copper foil, aluminum or other conductive materials, and the dielectric layer 2 can be made of FEP or other materials.

[0092] In some embodiments, the column can be a square column, such as a cuboid or cube, in which case the shape of the hole should be set to a rectangle or square accordingly. The column can also be other shapes, such as a cylinder, in which case the shape of the hole should be set to a circle accordingly.

[0093] To facilitate the testing of synaptic signals in the aforementioned neuromorphic device, the neuromorphic device in this embodiment of the invention further includes a test unit corresponding to each group of synaptic devices. To facilitate the connection between the test unit and the synaptic device, a second electrode layer is provided on the first end face of the columnar body away from the array substrate for electrical connection to the second end of the corresponding test unit. The first end face of the columnar body away from the array substrate is the side closest to the test unit.

[0094] It should be noted that, in addition to setting up corresponding test units for each synaptic device group as described above, a corresponding test unit can also be set up for each neuromorphic device. That is, multiple synaptic device groups in a neuromorphic device can share a test unit, the structure of which is shown in Figure 6. When one end of the test unit (i.e., the small triangle shown in the figure) is moved to different positions, the test unit measures the electrical signals of different synaptic devices.

[0095] Specifically, each test unit includes a first end and a second end. The first end of the test unit is electrically connected to the first electrode layer of each synapse in its corresponding synaptic device group. Alternatively, the first end of the test unit can be configured similarly to the second end, i.e., when located at different positions, it connects to the first electrode layer of the synapse corresponding to that position. However, it is necessary to ensure that its position corresponds to the position of the second end. That is, the first end and the second end are connected to the first electrode layer of the same synapse and the second electrode layer on the first end face of its corresponding column. The position of the second end of the test unit is not fixed. When it is located at different positions, it is electrically connected to the second electrode layer on the first end face of the column corresponding to that position (i.e., the second electrode layer of the corresponding synapse). In addition, during the process of the second end of the test unit moving to different positions, it simultaneously pushes the column corresponding to that position to slide along the direction of the corresponding hole, so that the synapse generates an electrical signal.

[0096] Since the test unit is located at different positions and its second end is connected to the second electrode layer of different synaptic devices, the test unit only measures the electrical signal of one synaptic device at a time. Therefore, the problem of crosstalk between load resistors in multiple synaptic devices in a group of synaptic devices is avoided.

[0097] The aforementioned test unit is used to detect the potential difference between the positive and negative electrodes of the triboelectric nanogenerator corresponding to the position. By moving the second end of the test unit, the electrical signals of each synaptic device in the synaptic device group are measured sequentially. Figure 7 is a schematic diagram of the electrical signals in a possible synaptic device group. As can be seen from Figure 7, there are obvious differences in the electrical signals of each synaptic device in this synaptic device group, that is, the resistance value of the load resistor of each synaptic device in this synaptic device group is set appropriately, and this synaptic device group meets the requirements.

[0098] Specifically, the aforementioned testing unit includes an electrometer;

[0099] The negative probe of the electrometer is electrically connected to the first electrode layer of each synapse device in the synapse device group, and the positive probe of the electrometer is electrically connected to the second electrode layer on the first end face of the column corresponding to the position of the positive probe. The structure and working principle of the positive probe are the same as the second end of the test unit mentioned above, and will not be described again here.

[0100] In addition to the structure described above, the negative probe of the electrometer can also be configured to be similar to the structure of the positive probe, that is, when it is located at different positions, it is connected to the first electrode layer of the synaptic device corresponding to the position. However, it is necessary to ensure that its position corresponds to the position of the positive probe. That is, the positive probe and the negative probe of the electrometer are connected to the first electrode layer of the same synaptic device and the second electrode layer on the first end face of the corresponding columnar body.

[0101] In some alternative implementations, as shown in Figures 4 and 8, the test unit further includes a frame 9 and a slider that slides on the frame. The slider is used to connect to the positive probe of the electrometer to assist in the movement of the positive probe of the electrometer.

[0102] The frame 9 is disposed opposite to the first end face of the array substrate near the columnar body. The slider can slide on the frame 9 to the position corresponding to each columnar body (i.e., synaptic device). In this embodiment of the invention, the specific structure of the frame 9 is not limited. The frame structure shown in Figure 8 is only for illustration.

[0103] The positive probe of the electrometer is connected to the slider so that when the slider slides to the corresponding position on the frame, the positive probe is electrically connected to the second electrode layer on the first end face of the column corresponding to the position.

[0104] In specific implementation, the slider can be configured to assist the positive probe of the electrometer in sliding, so that the positive probe is electrically connected to the second electrode layer on the first end face of the column corresponding to the corresponding position. In this case, the slider can be made of a non-conductive material. In addition, the slider can also be configured to act as a conductor, that is, when the slider slides to the corresponding position on the frame 9, the positive probe of the electrometer is electrically connected to the second electrode layer on the first end face of the column corresponding to the corresponding position through the slider. In this case, the material used for the slider must include conductive materials, such as copper foil.

[0105] It should be noted that the aforementioned slider can be made of one material or multiple materials, and its shape can be a cube, cylinder, or other shapes. Furthermore, the size of the slider is not limited in this invention. One possible structure is shown in Figure 4, consisting of a metal layer 7 and a base layer 8. When the slider slides to a corresponding position on the frame, its metal layer 7 is electrically connected to the second electrode layer on the first end face of the corresponding column. The metal layer 7 is adhered to the base layer 8, which serves to fix the shape of the metal layer 7. The metal layer 7 can be a copper foil with dimensions of 1cm × 0.2cm.

[0106] Utilizing the aforementioned neuromorphic devices for machine learning can improve the efficiency of neuromorphic computation. For example, in feature extraction from "handwriting" data, because the load resistance values ​​of each synapse within a synaptic device group are different (i.e., synaptic weights are different), there is no series interference between the load resistances of each synapse during testing. Therefore, the electrical signal from which synapse is being generated can be clearly analyzed from the collected features, analogous to feature extraction in unsupervised learning. Furthermore, each synaptic device group only needs to be measured once, analogous to dimensionality reduction in unsupervised learning. In addition, since the signals generated by synaptic devices are time-dependent, "handwriting" can also be analyzed.

[0107] Obviously, those skilled in the art can make various modifications and variations to the embodiments of the present invention without departing from the spirit and scope of the invention. Therefore, if these modifications and variations fall within the scope of the claims of the present invention and their equivalents, the present invention also intends to include these modifications and variations.

Claims

1. A neuromorphic device, characterized in that, The neuromorphic device includes an array substrate and multiple synaptic devices, wherein: each synaptic device includes a triboelectric nanogenerator and a load resistor, the positive terminal of the triboelectric nanogenerator is electrically connected to a first terminal of the load resistor, and the negative terminal of the triboelectric nanogenerator is electrically connected to a second terminal of the load resistor; the multiple synaptic devices are arranged in an array structure on the array substrate; the resistance value of the load resistor of each synaptic device is different.

2. The neuromorphic device according to claim 1, characterized in that, The triboelectric nanogenerator is one of the following: sliding mode; contact separation mode; or a mixture of contact separation and sliding mode.

3. The neuromorphic device according to claim 1, characterized in that, The resistance value of the load resistor is adjustable.

4. The neuromorphic device according to claim 1, characterized in that, The plurality of synaptic devices constitute a group of synaptic devices; the neuromorphic device comprises a plurality of synaptic device groups.

5. The neuromorphic device according to claim 1, characterized in that, The triboelectric nanogenerator in the synaptic device includes a first electrode layer, a dielectric layer, and a second electrode layer; the array substrate is provided with holes and pillars corresponding to the synaptic devices; for any hole and its corresponding pillar, the inner wall of the hole is sequentially provided with the first electrode layer and the dielectric layer, and the periphery of the pillar is provided with the second electrode layer; when the pillar slides along the direction of the hole, the second electrode layer and the dielectric layer generate friction.

6. The neuromorphic device according to claim 5, characterized in that, The columnar body has a second electrode layer disposed on its first end face away from the array substrate; the neuromorphic device further includes a test unit corresponding to each synaptic device group; for each test unit, the first end of the test unit is electrically connected to each first electrode layer corresponding to it, and when the second end of the test unit is located at different positions, it is electrically connected to the second electrode layer on the first end face of the columnar body corresponding to the position; the test unit is used to detect the potential difference between the positive and negative electrodes of the triboelectric nanogenerator corresponding to the position.

7. The neuromorphic device according to claim 6, characterized in that, The testing unit includes an electrometer; the negative electrode probe of the electrometer is electrically connected to each of the first electrode layers; the positive electrode probe of the electrometer is electrically connected to the second electrode layer on the first end face of the column corresponding to the position of the positive electrode probe.

8. The neuromorphic device according to claim 7, characterized in that, The test unit further includes a frame and a slider that can slide on the frame; the frame is disposed opposite to the side of the array substrate near the first end face of the column; the positive probe of the electrometer is connected to the slider, so that when the slider slides to the corresponding position on the frame, the positive probe is electrically connected to the second electrode layer on the first end face of the column corresponding to the corresponding position.

9. The neuromorphic device according to any one of claims 5-8, characterized in that, The column is a square column.

Images