Moving direction detection device and detection system

Abstract

The present disclosure relates to a detection device and a detection system for movement direction. The detection device includes: a plurality of sensors, each sensor including a plurality of sensor units linearly arranged along a first direction; a first neuromorphic device including a first synaptic unit, a first dendritic unit, and a first soma unit, the first synaptic unit including a silent synapse and an excitatory synapse, the first soma unit processing the output of the first dendritic unit to generate a first output signal; a second neuromorphic device including a second synaptic unit including only excitatory synapses, a second dendritic unit, and a second soma unit, the second soma unit processing the output of the second dendritic unit to generate a second output signal; and a third neuromorphic device including a third synaptic unit, a third dendritic unit, and a third soma unit, the third synaptic unit including an inhibitory synapse that receives the second output signal and an excitatory synapse that receives the first output signal, the third soma unit processing the output of the third dendritic unit to generate a third output signal.

Description

Technical Field

[0001] The present disclosure relates to the field of neuromorphic neural networks, and in particular to a moving direction detection device and a detection system. Background Art

[0002] The nervous system is considered a collection of neuronal cells interconnected by synapses. From this perspective, cells are not isolated individuals, but rather complex platforms that carry out countless metabolic and signaling processes within an intricate network of interconnected cellular components. In biologically amplified computational processes, neural processing elements (NPCs), such as synapses, dendrites, somata, and axons, play a key role in supporting, connecting, and propagating signals.

[0003] Biological dendrites are nonlinear processors that compute local input information entering their branches. However, most traditional artificial neural networks (ANNs) use dendrites (or axons) as linear connectors connecting weighted synaptic inputs to the soma. Without properly incorporating this nonlinear computation into dendritic devices, the cell-based paradigm commonly used in ANNs may not achieve the ideal functionality of neural processing elements.

[0004] For example, when detecting the direction of movement, dendritic devices cannot directly generate a response to the direction of movement. A CPU or GPU is still required to calculate the direction of movement, which requires a certain amount of power to run the artificial neural network. Artificial neural networks typically have full-node-to-full-node connections, which significantly increases the system's power consumption. Currently, motion detection systems based on convolutional neural networks (CNNs) consume approximately 10-100 W of GPU power for computing due to the deep processing involved in learning a large number of parameters. Spiking neural networks (SNNs) using event-based sensor data on FPGA chips also consume 10-100 mW of operating power. Running a large number of calculations can place a significant burden on the system. Summary of the Invention

[0005] In related technologies, when detecting the moving direction, a certain amount of power is required to run the artificial neural network, resulting in high system power consumption.

[0006] In view of this, the present disclosure provides a device for detecting a moving direction, which can determine the direction of an input signal without an additional processor.

[0007] According to some embodiments of the present disclosure, a device for detecting a moving direction is provided, comprising:

[0008] a plurality of sensors for sensing movement of an object, each sensor comprising a plurality of sensor units linearly arranged along a first direction;

[0009] A first neuromorphic device includes a first synaptic unit, a first dendritic unit, and a first soma unit, wherein the first synaptic unit includes a silent synapse and an excitatory synapse, wherein the silent synapse is activated only when the input signal strength of the excitatory synapse is greater than an activation threshold; the first dendritic unit includes a plurality of dendritic branches, wherein each dendritic branch of at least some of the dendritic branches is connected to a silent synapse and a plurality of excitatory synapses, wherein the silent synapses are located closer to the first soma unit than the excitatory synapses, wherein the plurality of synapses on each dendritic branch are connected to a plurality of sensor units in different sensors in a one-to-one correspondence, wherein the silent synapse and its adjacent excitatory synapses are connected to different sensors, and the first soma unit is configured to process an output of the first dendritic unit to generate a first output signal;

[0010] a second neuromorphic device comprising a second synaptic unit, a second dendritic unit, and a second soma unit, wherein the second synaptic unit comprises an excitatory synapse for receiving the first output signal, the second dendritic unit comprises at least one dendritic branch, each dendritic branch being connected to the excitatory synapse, and the second soma unit is configured to process an output of the second dendritic unit to generate a second output signal;

[0011] A third neuromorphic device includes a third synaptic unit, a third dendritic unit, and a third cell body unit, wherein the third synaptic unit includes an inhibitory synapse and an excitatory synapse, the inhibitory synapse is used to receive the second output signal, and the excitatory synapse is used to receive the first output signal. The third dendritic unit includes at least one dendritic branch, each dendritic branch is connected to an inhibitory synapse and an excitatory synapse, the inhibitory synapse is located closer to the third cell body unit than the excitatory synapse, and the third cell body unit is used to process the output of the third dendritic unit to generate a third output signal.

[0012] In some embodiments, the second output signal is used to indicate that the object moves along a first direction; the third output signal is used to indicate that the object moves along a second direction, and the vector of the first direction forms a first angle with the vector of the second direction.

[0013] In some embodiments, the first neuromorphic device further includes a first axon unit, the first axon unit being configured to convert an output of the first soma unit into a first voltage pulse; the second neuromorphic device further includes a second axon unit, the second axon unit being configured to convert an output of the second soma unit into a second voltage pulse; and the third neuromorphic device further includes a third axon unit, the third axon unit being configured to convert an output of the third soma unit into a third voltage pulse.

[0014] In some embodiments, the first axon unit includes a first current switch, which is configured to turn on when the output of the first cell unit is greater than a first current threshold; the second axon unit includes a second current switch, which is configured to turn on when the output of the second cell unit is greater than a second current threshold; the third axon unit includes a third current switch, which is configured to turn on when the output of the third cell unit is greater than a third current threshold, wherein the third current threshold is less than the second current threshold.

[0015] In some embodiments, the third axon unit further includes: a first voltage switch configured to be turned off when the intensity of the first output signal is greater than a first voltage threshold.

[0016] In some embodiments, the first output signal includes a first direction output signal and a second direction output signal, and when the object moves along the first direction:

[0017] An excitatory synapse on a dendritic branch of the first dendritic unit first receives a pulse signal from a corresponding sensor unit, thereby activating a silent synapse, which then receives a pulse signal from the corresponding sensor unit. When the output of the first soma unit is greater than the first current threshold, the first current switch is turned on, and the first neuromorphic device outputs the first direction output signal.

[0018] The output of the second cell unit is greater than a second current threshold, the second current switch is turned on, and the second neuromorphic device outputs the second output signal;

[0019] The inhibitory synapse receives the second output signal, so that the first direction output signal cannot conduct the third neuromorphic device, and the third neuromorphic device does not output the third output signal.

[0020] In some embodiments, the at least some dendritic branches include a first dendritic branch and a second dendritic branch, and when the object moves along the second direction:

[0021] Some of the excitatory synapses on the first dendritic branch first receive pulse signals from corresponding sensor units;

[0022] Some of the excitatory synapses on the second dendritic branch then receive the pulse signal from the corresponding sensor unit, activating the silent synapses, and the silent synapses then receive the pulse signal from the corresponding sensor unit;

[0023] The output of the first cell unit is greater than the first current threshold, the first current switch is turned on, the first neuromorphic device outputs the second direction output signal, and the intensity of the second direction output signal is less than the first direction output signal;

[0024] The output of the second cell unit is not greater than the second current threshold, the second current switch is not turned on, and the second neuromorphic device does not output the second output signal;

[0025] The output of the third cell unit is greater than the third current threshold, the third current switch is turned on, and the third neuromorphic device outputs the third output signal.

[0026] In some embodiments, the first dendritic unit further includes dendritic branches connected only to excitatory synapses.

[0027] In some embodiments, the detection device further comprises:

[0028] a fourth neuromorphic device comprising a fourth synaptic unit, a fourth dendritic unit, and a fourth soma unit, wherein the fourth synaptic unit comprises an excitatory synapse, the fourth dendritic unit comprises a plurality of dendritic branches, the plurality of excitatory synapses being connected to each dendritic branch of at least some of the dendritic branches, the plurality of synapses on a dendritic branch being connected in a one-to-one correspondence to a plurality of sensor units in different sensors, and the fourth soma unit being configured to process an output of the fourth dendritic unit to generate a fourth output signal;

[0029] a fifth neuromorphic device comprising a fifth synaptic unit, a fifth dendritic unit, and a fifth soma unit, wherein the fifth synaptic unit comprises an inhibitory synapse and an excitatory synapse, the inhibitory synapse being configured to receive the first output signal, the excitatory synapse being configured to receive the fourth output signal, the fifth dendritic unit comprising at least one dendritic branch, each dendritic branch being connected to an inhibitory synapse and an excitatory synapse, the inhibitory synapse being located closer to the fifth soma unit than the excitatory synapse, and the fifth soma unit being configured to process the output of the fifth dendritic unit to generate a fifth output signal;

[0030] a sixth neuromorphic device comprising a sixth synaptic unit, a sixth dendritic unit, and a sixth soma unit, wherein the sixth synaptic unit comprises an excitatory synapse for receiving the fifth output signal, the sixth dendritic unit comprises at least one dendritic branch, each dendritic branch being connected to the excitatory synapse, and the sixth soma unit is configured to process an output of the sixth dendritic unit to generate a sixth output signal;

[0031] A seventh neuromorphic device includes a seventh synaptic unit, a seventh dendritic unit, and a seventh soma unit, wherein the seventh synaptic unit includes an inhibitory synapse and an excitatory synapse, the inhibitory synapse is used to receive the sixth output signal, and the excitatory synapse is used to receive the fifth output signal. The seventh dendritic unit includes at least one dendritic branch, each dendritic branch is connected to an inhibitory synapse and an excitatory synapse, the inhibitory synapse is located closer to the seventh soma unit than the excitatory synapse, and the seventh soma unit is used to process the output of the seventh dendritic unit to generate a seventh output signal.

[0032] In some embodiments, the sixth output signal is used to indicate that the object moves along a third direction, which is opposite to the first direction; the seventh output signal is used to indicate that the object moves along a fourth direction, which is opposite to the second direction.

[0033] In some embodiments, the fourth neuromorphic device further includes a fourth axon unit, which is used to convert the output of the fourth cell unit into a fourth voltage pulse; the fifth neuromorphic device further includes a fifth axon unit, which is used to convert the output of the fifth cell unit into a fifth voltage pulse; the sixth neuromorphic device further includes a sixth axon unit, which is used to convert the output of the sixth cell unit into a sixth voltage pulse; the seventh neuromorphic device further includes a seventh axon unit, which is used to convert the output of the seventh cell unit into a seventh voltage pulse.

[0034] In some embodiments, the fourth axon unit includes a fourth current switch, which is configured to turn on when the output of the fourth cell unit is greater than a fourth current threshold; the fifth axon unit includes a fifth current switch, which is configured to turn on when the output of the fifth cell unit is greater than a fifth current threshold; the sixth axon unit includes a sixth current switch, which is configured to turn on when the output of the sixth cell unit is greater than a sixth current threshold; the seventh axon unit includes a seventh current switch, which is configured to turn on when the output of the seventh cell unit is greater than a seventh current threshold, wherein the seventh current threshold is less than the seventh current threshold.

[0035] In some embodiments, the seventh neuromorphic device further comprises: a second voltage switch configured to be turned off if the strength of the fifth output signal is greater than a second voltage threshold; and / or a third voltage switch configured to be turned off if the strength of the sixth output signal is greater than a third voltage threshold.

[0036] In some embodiments, the fourth output signal includes a third direction output signal and a fourth direction output signal, and when the object moves along the third direction:

[0037] On a dendritic branch of the first dendritic unit, a silent synapse first receives a pulse signal from the corresponding sensor unit, and an excitatory synapse later receives a pulse signal from the corresponding sensor unit, the silent synapse is inactivated, the output of the first soma unit is not greater than the first current threshold, the first current switch is not turned on, and the first neuromorphic device does not output the first output signal;

[0038] An excitatory synapse on a dendritic branch of the fourth dendritic unit successively receives pulse signals from the corresponding sensor unit, the output of the fourth cell unit is greater than the fourth current threshold, the fourth current switch is turned on, and the fourth neuromorphic device outputs the third direction output signal;

[0039] The output of the fifth cell unit is greater than the fifth current threshold, the fifth current switch is turned on, and the fifth neuromorphic device outputs a fifth output signal corresponding to the third direction output signal;

[0040] The output of the sixth cell unit is greater than the sixth current threshold, the sixth current switch is turned on, and the sixth neuromorphic device outputs the sixth output signal;

[0041] The inhibitory synapse of the seventh neuromorphic device receives the sixth output signal, so that the fifth output signal cannot conduct the seventh neuromorphic device, and the seventh neuromorphic device does not output the seventh output signal.

[0042] In some embodiments, at least some of the dendrite branches of the fourth dendrite unit include third dendrite branches and fourth dendrite branches, and when the object moves along the fourth direction:

[0043] On a dendritic branch of the first dendritic unit, a silent synapse first receives a pulse signal from the corresponding sensor unit, and an excitatory synapse later receives a pulse signal from the corresponding sensor unit, the silent synapse is inactivated, the output of the first soma unit is not greater than the first current threshold, the first current switch is not turned on, and the first neuromorphic device does not output the first output signal;

[0044] Some of the excitatory synapses on the third dendritic branch first receive the pulse signal from the corresponding sensor unit;

[0045] Some of the excitatory synapses on the fourth dendritic branch then receive pulse signals from corresponding sensor units;

[0046] The output of the fourth cell unit is greater than the fourth current threshold, the fourth current switch is turned on, and the fourth neuromorphic device outputs the fourth direction output signal, wherein the intensity of the fourth direction output signal is less than the third direction output signal;

[0047] The output of the fifth cell unit is greater than the fifth current threshold, the fifth current switch is turned on, and the fifth neuromorphic device outputs a fifth output signal corresponding to the fourth direction output signal;

[0048] The output of the sixth cell unit is not greater than the sixth current threshold, the sixth current switch is not turned on, and the sixth neuromorphic device does not output the sixth output signal;

[0049] The output of the seventh cell unit is greater than the seventh current threshold, the seventh current switch is turned on, and the seventh neuromorphic device outputs the seventh output signal.

[0050] In some embodiments, when the object moves along the first direction or the second direction: the first neuromorphic device is turned on and outputs the first output signal; when the inhibitory synapse of the fifth neuromorphic device receives the first output signal, the fifth neuromorphic device is turned off and does not output the fifth output signal.

[0051] In some embodiments, in the first dendritic unit, multiple synapses on each dendritic branch are interlacedly connected to multiple sensor units in adjacent sensors; and / or in the fourth dendritic unit, multiple synapses on each dendritic branch are interlacedly connected to multiple sensor units in adjacent sensors.

[0052] In some embodiments, the sensor unit is in the shape of a regular hexagon, the multiple sensors are an array of sensors in which multiple sensor units are densely packed in a regular hexagonal plane, and a center line connecting multiple sensor units in the same sensor is parallel to the first direction.

[0053] In some embodiments, the first direction is a horizontal rightward direction, and the first angle is 30°.

[0054] In some embodiments, the vector of the first direction forms an angle of 60° with the vector of the horizontal right direction, and the first angle is 30°.

[0055] In some embodiments, the second direction is a vertical upward direction, and the first angle is 30°.

[0056] According to other embodiments of the present disclosure, a moving direction detection system is provided, comprising:

[0057] The first detection device includes the detection device described above and is configured to detect movement of the object in a horizontal direction, a direction forming an angle of 150° or 330° to the right of the horizontal direction;

[0058] The second detection device includes the detection device as described above, and is configured to detect movement of the object in a direction that makes an angle of 30°, 60°, 210°, or 240° with respect to the horizontal rightward direction;

[0059] The third detection device, including the detection device described above, is configured to detect movement of the object in a direction forming an angle of 90°, 120°, 270°, or 300° with respect to the horizontal rightward direction.

[0060] In some embodiments, the detection system further includes: a processor configured to determine the moving direction of the object based on the output signal of the first detection device, the output signal of the second detection device, and the output signal of the third detection device.

[0061] In some embodiments, the processor is configured to:

[0062] Using a morphological network to represent each neuromorphic device in the detection system;

[0063] Mapping the morphological network of each neuromorphic device to a topological network of equivalent neural processing elements;

[0064] Embedding the topological network into a hyperbolic space to obtain a hyperbolic embedded network;

[0065] The hyperbolic embedding network is used to identify different functional layers according to the topological connectivity of the nodes of the neural processing element. The functional layers include a mapping layer, a direction selection layer, and an orientation selection layer.

[0066] In the above embodiment, by utilizing the detection device composed of the new neuromorphic device, it is possible to respond to the direction of the input signal without an additional processor and perform spatiotemporal signal processing, thereby significantly reducing the power consumption of the system.

[0067] Other features and advantages of the present disclosure will become apparent from the following detailed description of embodiments of the present disclosure with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments of the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0069] The present disclosure can be more clearly understood from the following detailed description with reference to the accompanying drawings, in which:

[0070] Figure 1is a block diagram illustrating a device for detecting a moving direction according to some embodiments of the present disclosure;

[0071] Figure 2 is a morphological schematic diagram illustrating a neuromorphic circuit according to some embodiments of the present disclosure;

[0072] Figure 3A is a block diagram of a first neuromorphic device according to some embodiments of the present disclosure;

[0073] Figure 3B is a cross-sectional view illustrating a first neuromorphic transistor according to some embodiments of the present disclosure;

[0074] Figure 4 is a morphological diagram illustrating a first neuromorphic transistor according to some embodiments of the present disclosure;

[0075] Figure 5A is a block diagram of a second neuromorphic device according to some embodiments of the present disclosure;

[0076] Figure 5B is a structural diagram illustrating a second neuromorphic transistor according to some embodiments of the present disclosure;

[0077] Figure 6A is a block diagram of a third neuromorphic device according to some embodiments of the present disclosure;

[0078] Figure 6B is a logic diagram showing input-output signals of a neuromorphic transistor with excitatory synapses and inhibitory synapses located at different positions;

[0079] Figure 7 is a schematic diagram of a model circuit of a neuromorphic device circuit according to some embodiments of the present disclosure;

[0080] Figure 8 is a morphological schematic diagram illustrating a device for detecting a moving direction according to some embodiments of the present disclosure;

[0081] Figure 9 is a block diagram illustrating a device for detecting a moving direction according to other embodiments of the present disclosure;

[0082] Figure 10 is a morphological schematic diagram illustrating a neuromorphic circuit according to some other embodiments of the present disclosure;

[0083] Figure 11A This is a simulation diagram showing how the output signal generated by a dendritic branch of the first neuromorphic transistor changes with the input signal;

[0084] Figure 11B This is a simulation diagram of the output signal generated by a dendritic branch of the fourth neuromorphic transistor changing with the input signal;

[0085] Figure 12 is a schematic diagram of a model circuit of a neuromorphic device circuit according to some other embodiments of the present disclosure;

[0086] Figure 13 is a morphological schematic diagram illustrating a moving direction detection device according to other embodiments of the present disclosure;

[0087] Figure 14 is a schematic diagram of a mapping method according to some embodiments of the present disclosure;

[0088] Figure 15 is a schematic diagram of a hyperbolic mapping method according to some embodiments of the present disclosure;

[0089] Figure 16 is a hyperbolic NPC network graph according to some embodiments of the present disclosure;

[0090] Figure 17 is a block diagram illustrating a system for detecting a moving direction according to some embodiments of the present disclosure;

[0091] Figure 18 is a schematic diagram illustrating a connection relationship between different detection devices and sensors in a moving direction detection system according to some embodiments of the present disclosure;

[0092] Figure 19 is a schematic diagram showing output signals of a moving direction detection system for different exemplary input signals according to some embodiments of the present disclosure;

[0093] Figure 20 is a block diagram illustrating a system for detecting a moving direction according to other embodiments of the present disclosure;

[0094] Figure 21 is a cluster diagram illustrating a hyperbolic NPC map of a movement direction detection system according to some embodiments of the present disclosure.

[0095] It should be understood that the size of each part shown in the drawings is not drawn according to the actual proportional relationship.In addition, the same or similar reference numerals represent the same or similar components. DETAILED DESCRIPTION

[0096] Various embodiments of the present disclosure will now be described in detail with reference to the accompanying drawings. The description of the embodiments is merely illustrative and is in no way intended to limit the present disclosure, its application, or use. The present disclosure can be implemented in many different forms and is not limited to the embodiments described herein. These embodiments are provided to make the present disclosure thorough and complete and to fully convey the scope of the present disclosure to those skilled in the art. It should be noted that unless otherwise specifically stated, the relative arrangement of the components and steps described in these embodiments should be interpreted as being merely illustrative and not as limiting.

[0097] The terms "first", "second" and similar terms used in this disclosure do not indicate any order, quantity or importance, but are only used to distinguish different parts. The terms "include" and similar terms mean that the elements preceding the term include the elements listed after the term, and do not exclude the possibility of also including other elements. The terms "upper", "lower", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0098] In the present disclosure, when a specific element is described as being located between a first element and a second element, an intervening element may or may not exist between the specific element and the first element or the second element.

[0099] All terms (including technical or scientific terms) used in this disclosure have the same meaning as understood by one of ordinary skill in the art to which this disclosure belongs, unless otherwise specifically defined. It should also be understood that terms defined in, for example, commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant art, and should not be interpreted in an idealized or highly formal sense, unless explicitly defined as such herein.

[0100] Technologies, methods, and equipment known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods, and equipment should be considered part of the specification.

[0101] In a detection device based on a traditional artificial neural network, a CPU or a GPU is used to detect the direction, and the power consumption of the system is relatively high.

[0102] In view of this, the present disclosure provides a device for detecting a moving direction, which can respond to the direction of an input signal without an additional processor, thereby significantly reducing the power consumption of the system.

[0103] first, Figure 1 is a block diagram illustrating an apparatus for detecting a moving direction according to some embodiments of the present disclosure.

[0104] like Figure 1As shown, the moving direction detection device DD1 includes multiple sensors SE, a first neuromorphic device N1, a second neuromorphic device N2, and a third neuromorphic device N3.

[0105] The plurality of sensors are used to sense the movement of an object, and each sensor comprises a plurality of sensor units linearly arranged along a first direction.

[0106] The first neuromorphic device N1 , the second neuromorphic device N2 , and the third neuromorphic device N3 may constitute a neuromorphic circuit C1 .

[0107] Figure 2 is a morphological schematic diagram illustrating a neuromorphic circuit C1 according to some embodiments of the present disclosure.

[0108] like Figure 2 As shown, the first neuromorphic device N1 includes a first synaptic unit SU1 , a first dendrite unit DU1 , and a first soma unit CU1 .

[0109] The first synaptic unit DU1 is configured to receive sensing signals from multiple sensors SE for sensing object movement. These sensors include silent synapses SS and excitatory synapses ES. Silent synapses SS activate only when the input signal strength of excitatory synapses ES exceeds an activation threshold. This activation threshold represents the signal strength required to activate the excitatory synapse in order for the voltage in the first dendritic unit to meet the threshold for silent synapse activation. The input signal strength is primarily determined by the amplitude, pulse width, and pulse period of the input voltage pulse.

[0110] That is to say, the silent synapse SS will be activated only when the input of multiple excitatory synapses ES makes the dendritic unit DU1 meet the activation conditions of the silent synapse SS.

[0111] The first dendritic unit DU1 is configured to integrate input from the first synaptic unit SU1 and includes multiple dendritic branches. At least some of these branches are connected to a silent synapse and multiple excitatory synapses. The silent synapses are located closer to the first cell body unit CU1 than the excitatory synapses. The multiple synapses on each dendritic branch are connected in a one-to-one correspondence to multiple sensor units in different sensors. A silent synapse and its adjacent excitatory synapse are connected to different sensors.

[0112] The first cell body unit CU1 is used to process the output of the first dendritic unit DU1 to generate a first output signal. The first output signal includes an excitation output N1E.

[0113] The second neuromorphic device N2 includes a second synaptic unit SU2 , a second dendrite unit DU2 , and a second soma unit CU2 .

[0114] The second synaptic unit SU2 includes an excitatory synapse ES for receiving an excitatory output N1E in the first output signal. The second dendritic unit DU2 includes at least one dendritic branch, each of which is connected to the excitatory synapse ES. The second cell body unit CU2 is configured to process the output of the second dendritic unit DU2 to generate a second output signal. The second output signal includes an inhibitory output N2I and an excitatory output N2E. The excitatory output N2E in the second output signal is configured to indicate movement of the object in the first direction.

[0115] The third neuromorphic device N3 includes a third synaptic unit SU3 , a third dendritic unit DU3 , and a third soma unit CU3 .

[0116] The third synapse unit SU3 includes an inhibitory synapse IS and an excitatory synapse ES. The inhibitory synapse IS is used to receive the inhibitory output N2I in the second output signal, and the excitatory synapse ES is used to receive the excitatory output N1E in the first output signal.

[0117] The third dendritic unit DU3 is used to integrate input from the third synaptic unit SU3 and includes at least one dendritic branch. Each dendritic branch is connected to an inhibitory synapse IS and an excitatory synapse ES, and the inhibitory synapse IS is located closer to the third cell body unit CU3 than the excitatory synapse ES.

[0118] The third soma unit CU3 is configured to process the output of the third dendritic unit DU3 to generate a third output signal of the third neuromorphic device. The third output signal includes a third excitation output N3E indicating that the object is moving in a second direction. The vector in the first direction forms a first angle with the vector in the second direction.

[0119] It should be understood that Figure 2 The first dendritic unit DU1 shown in FIG. 1 is only for illustration. The first dendritic unit may include one silent synapse or multiple silent synapses.

[0120] In some embodiments, the first neuromorphic device N1 further includes a first axon unit AU1, such as Figure 2 The first axon unit AU1 is used to convert the output of the first cell unit CU1 into a first voltage pulse. In some embodiments, the first axon unit includes a voltage pulse generator for achieving the above output conversion.

[0121] The voltage pulse generator may include a positive voltage pulse generator and a negative voltage pulse generator. The positive voltage pulse generator may convert the output of the cell body unit into an excitatory output, and the negative voltage pulse generator may convert the output of the cell body unit into an inhibitory output.

[0122] In some other embodiments, the first axon unit further includes a first current switch configured to be turned on when the output of the first soma unit is greater than a first current threshold.

[0123] The second neuromorphic device N2 may further include a second axon unit AU2. The second axon unit AU2 is configured to convert the output of the second soma unit CU2 into a second voltage pulse. Similarly, the second axon unit may include a voltage pulse generator to implement the above-mentioned output conversion.

[0124] In some embodiments, the second axon unit includes a second current switch configured to be turned on if the output of the second soma unit is greater than a second current threshold.

[0125] The third neuromorphic device N3 may further include a third axon unit AU3. The third axon unit AU3 is configured to convert the output of the third soma unit into a third voltage pulse. Similarly, the third axon unit may include a voltage pulse generator to achieve the above output conversion.

[0126] In some embodiments, the third axon unit further includes a third current switch configured to be turned on when the output of the third soma unit is greater than a third current threshold, wherein the third current threshold is less than the second current threshold.

[0127] The aforementioned current switch can enhance the directional selectivity of the neuromorphic device. On one hand, the current switch in the first neuromorphic device enables the first neuromorphic device to conduct only when the generated current exceeds a certain current threshold. In other words, by using a suitable current switch and configuring the appropriate current threshold, the first neuromorphic device can be made to conduct only for input signals of a specific direction. The specific implementation method will be described in detail below.

[0128] On the other hand, by setting the third current threshold lower than the second current threshold, detection device DD1 can further discern the direction of the input signal. When the input signal is directed in a specific direction, the outputs of multiple sensors are received by the synapses of the same dendritic branch, creating a superposition effect within that dendritic branch, causing the first neuromorphic device to produce a stronger output. By setting the second current threshold of the second current switch and the third current threshold of the third current switch, the direction of input signals that can turn on the first neuromorphic device but have different output intensities can be determined.

[0129] In some other embodiments, the third axon unit further includes a first voltage switch configured to be turned off if the intensity of the excitation output N1E in the first output signal exceeds a first voltage threshold. This switch directly determines the output intensity of the first neuromorphic device and turns off when the intensity exceeds a certain threshold (i.e., the third neuromorphic device should not output), thereby reducing noise output by the third neuromorphic device.

[0130] The above describes the overall structure of the neuromorphic device circuit. Figures 3A to 4 Introduce the structure and principle of the first neuromorphic device.

[0131] Figure 3A is a block diagram of a first neuromorphic device according to some embodiments of the present disclosure.

[0132] like Figure 3A As shown, the first neuromorphic device N1 includes a first neuromorphic transistor T1 , a first pulse generator PG1 , and a first current switch CSW1 .

[0133] The first pulse generator PG1 is configured to convert the current output by the first neuromorphic transistor T1 into a first voltage pulse. A first current switch CSW1 is electrically connected to the first pulse generator PG1 and configured to turn on when the output current of the first neuromorphic transistor T1 exceeds a first current threshold. The first pulse generator PG1 and the first current switch CSW1 collectively correspond to the first axon unit described above.

[0134] The main function of the first neuromorphic device is realized by the first neuromorphic transistor included therein, which will be described in detail below.

[0135] Figure 3B is a cross-sectional view illustrating a first neuromorphic transistor according to some embodiments of the present disclosure.

[0136] like Figure 3B As shown, the first neuromorphic transistor T1 includes a first gate T16, a resistive switch T15, and a first source T18. The first gate T16 corresponds to the first synaptic unit SU1 and includes a first-type gate T16A and a second-type gate T16B. The first-type gate T16A corresponds to the excitatory synapse. The second-type gate T16B is closer to the first source T18 than the first-type gate T16A. The resistive switch T15 is only turned on when the input signal strength of the excitatory synapse exceeds the activation threshold. The resistive switch T15 is electrically connected to the second-type gate T16B and the entire device corresponds to the silent synapse.

[0137] like Figure 3BAs shown, the first neuromorphic transistor T1 further includes: a first source region T11 and a first drain region T12 , corresponding first source electrode T18 and first drain electrode T19 , a first channel region T13 , and an ion-doped first dielectric layer T14 .

[0138] like Figure 3B As shown, the first channel region T13 is located between the first source region T11 and the first drain region T12. In some embodiments, the material of the channel region is doped nanowires, such as silicon nanowires. The nanowires can be formed by etching on the top layer of the silicon substrate or by gas-phase or liquid-phase catalytic growth.

[0139] In some embodiments, the nanowires in the channel region T13 are arranged in a honeycomb shape between the source region T11 and the drain region T12. The honeycomb-shaped nanowires can effectively sense potential changes on their surfaces.

[0140] The first source region T11 and the first drain region T12 , the corresponding first source electrode T18 and the first drain electrode T19 , and the first channel region T13 as a whole correspond to the first cell unit CU1 .

[0141] The first dielectric layer T14 covers the first channel region T13, extends along the length of the first channel region, and contacts the first type gate T16A. The ion-doped first dielectric layer T14 corresponds to the first dendrite unit DU1.

[0142] In some embodiments, the first dielectric layer includes an ion-doped sol-gel film, for example, a composite material including silicon dioxide, containing copper ions Cu 2+ 、Nickel ion Ni 2+ and chloride ions Cl - Ion-doped sol-gel membranes, due to the presence of mobile ions, can simulate ion diffusion in neuronal membranes.

[0143] In some embodiments, the above-mentioned resistance switch T15 is located between the second type gate T16B and the first dielectric layer T14, and is a three-terminal resistance switch, including a first end, a second end and a third end. The first end is connected to the second type gate T16B, the second end is connected to the first dielectric layer T14, and the third end is connected to the interface between the first channel region and the first dielectric layer.

[0144] The resistive switch is conductive only when the voltage applied to it exceeds a threshold. The specific implementation of the resistive switch is not limited. For example, the resistive switch material layer T15 can be formed by patterning and sputtering using soft lithography techniques. The resistive switch material layer T15 can be made of a resistive switch material such as TaO2 or HfO2.

[0145] That is, the silent synapse SS will be activated only when the sol-gel membrane potential (equivalent to the postsynaptic condition in the dendritic unit DU1 ) determined by the input signal strength of the plurality of first-type gates T16A meets the activation condition of the silent synapse SS.

[0146] In some embodiments, the neuromorphic transistor T1 further includes a first substrate T10. Figure 3B As shown, the first substrate T10 includes a first region T10A and a second region T10B. The first region T10A and the second region T10B have no overlapping region. The first substrate T10 may be a lightly doped semiconductor substrate, such as a P or N type silicon substrate with a low doping concentration.

[0147] like Figure 3B As shown, the first source region T11 and the first drain region T12 are located in the first region T10A. In some embodiments, the first source region T11 and the first drain region T12 are semiconductor regions with a doping type opposite to that of the first substrate T10. In other words, the first source region T11, the first substrate T10, and the first drain region T12 may form two back-to-back PN junctions.

[0148] In other embodiments, the first source region T11 and the first drain region T12 are semiconductor regions having the same doping type as the first channel region T13 but a different doping concentration. For example, the first source region T11 and the first drain region T12 are heavily doped N-type semiconductors, while the first channel region T13 is lightly doped N-type semiconductors. In this way, the first source region T11, the first drain region T12, and the first channel region T13 can form a junctionless transistor, thereby improving the sensitivity of the transistor.

[0149] The first source electrode T18 and the first drain electrode T19 are located on the first source region T11 and the first drain region T12, respectively. In some embodiments, the first source electrode T18 and the first drain electrode T19 are parallel strip electrodes. The first source electrode T18 and the first drain electrode T19 can be electrodes formed of metals such as aluminum or platinum.

[0150] like Figure 3B As shown, the first portion T14A of the first dielectric layer T14 is located above the first channel region T13 and covers the first channel region T13. The second portion T14B is located above the second region T10B. The second portion T14B of the first dielectric layer T14 includes a first position P1 and a second position P2. In some embodiments, the distance between the first position P1 and the first source electrode T18 is smaller than the distance between the second position P2 and the first source electrode T18. Figure 3B shown.

[0151] A resistive switch T15 and a second-type gate T16B are located at a first position P1 of the second portion T14B of the first dielectric layer T14. A first-type gate T16A is located at a second position P2 of the second portion T14B. These gates may be electrodes formed of a metal such as platinum or gold. For example, depending on the material type of the resistive switch material layer T15, the second-type gate T16B may be made of silver or copper.

[0152] Figure 3B Multiple first-type gates T16A are shown, and their corresponding positions can be P3, P4, and P5, which are gradually farther away from the first source T18. That is, the distance between the second-type gate T16B and the first source T18 is smaller than the distance between the first-type gate T16A and the first source T18.

[0153] The structure of the first neuromorphic transistor is introduced above, and the first neuromorphic transistor can enable the first neuromorphic device to achieve direction selectivity. Figure 4 The functions that can be achieved by the above structure of the first neuromorphic transistor are introduced.

[0154] Figure 4 FIG. 4 is a morphological diagram illustrating a dendrite branch of a first neuromorphic transistor according to some embodiments of the present disclosure.

[0155] like Figure 4 As shown, the morphological representation of the first neuromorphic transistor includes: a first synaptic unit SU1 , a dendrite branch br1 of a first dendrite unit, and a first cell body unit CU1 .

[0156] The first synaptic unit SU1 includes an excitatory synapse ES and a silent synapse SS, wherein the excitatory synapse ES is Figure 4 Shown as a solid triangle, silent highlight SS in Figure 4 Shown as a hollow triangle.

[0157] Different locations on the dendritic branch br1 of the first dendritic unit are connected to different synapses, for example, one-to-one with multiple synapses in the first synaptic unit, to receive and integrate the inputs of the synaptic unit. This integration is typically nonlinear. The first cell body unit CU1 processes the outputs of the multiple dendritic branches of the dendritic unit, including linearly integrating the outputs of different dendritic branches.

[0158] The function of the first synaptic unit SU1 is to receive input and introduce the input into the dendrite branch br1 , which is equivalent to the first gate receiving a pulse signal and introducing the influence of the pulse signal into the ion-doped dielectric layer.

[0159] Excitatory synapses ES directly direct input into dendritic branch br1, whereas silent synapses SS direct input into dendritic branch br1 only when postsynaptic conditions (e.g., the state in the dendritic unit) meet a threshold.

[0160] Silent synapses in biology remain silent until the internal dendritic membrane potential (i.e., the postsynaptic condition) induced by other synaptic inputs reaches a certain threshold. Without an increase in internal membrane voltage, the silent synapse will not activate, even with high input strength. Similar to silent synapses in biology, the silent synapse SS in the disclosed embodiments similarly directs input into dendritic branch br1 only when the voltage induced by other synaptic inputs in that branch meets the threshold.

[0161] Correspondingly, the first type gate directly introduces the voltage signal into the dielectric layer, while the second type gate connected to the resistance switch introduces the voltage signal into the dielectric layer only when the resistance switch is turned on.

[0162] The activation threshold is used to characterize the input signal strength required to the first type gate in order for the voltage in the first dendrite unit to meet the threshold for the resistance switch to turn on. The activation threshold is determined by the material of the resistance switch and the positions of the first type gate and the second type gate in the first dielectric layer.

[0163] Depend on Figure 4 As can be seen from the morphological representation of the first synaptic unit, the different synapses have a certain positional relationship. The location of the silent synapse SS corresponds to the first position (1), and the location of the excitatory synapse ES corresponds to multiple second positions (2, 3, 4, 5). The spatial position of synapses along the dendrite plays a crucial role in spatiotemporal information processing and synaptic regulation because it affects the nonlinear integration coefficient of the dendritic unit.

[0164] After the synapse introduces the input into the dendritic unit, the dendritic unit must integrate the input and output it to the cell body unit. The dendritic unit itself (equivalent to the ion-doped dielectric layer) has a certain resistance, and its resistance is positively correlated with its length. Therefore, the farther the synapse is from the cell body unit, the smaller the contribution of the synaptic input to the dendritic unit output.

[0165] The following will refer to Figure 4 The morphological representation diagram of FIG. 1 describes the operation process of the first neuromorphic transistor included in some embodiments of the present disclosure to illustrate how the first neuromorphic transistor responds to the direction of the input signal.

[0166] Figure 4 Also shown is a sensor array configured to sense the motion of an object. The sensor array includes a plurality of sensors. When a sensor senses an object, a voltage pulse is sent to a synapse connected thereto, i.e., a corresponding gate on the first neuromorphic device.

[0167] Figure 4 Five sensors are shown: sensor S5, sensor S4, sensor S3, sensor S2, and sensor S1. This example is used below to describe how a neuromorphic device responds to the direction of an input signal.

[0168] For example, when an object moves from below to above, it passes through sensors in the order of 5, 4, 3, 2, and 1. Accordingly, sensors S5, S4, S3, S2, and S1, corresponding to positions 5, 4, 3, 2, and 1, sequentially send voltage pulses. For example, sensor S5 first sends a voltage pulse to the corresponding silent synapse, followed by sensors S4, S3, S2, and finally, S1.

[0169] The direction in which the object moves from top to bottom can be called a first direction, for example, the outward direction. Conversely, when the object moves from top to bottom, the direction in which it passes through the sensor in the order of positions 1, 2, 3, 4, and 5 can be called an inward direction, i.e., the direction opposite to the first direction.

[0170] As described above, the silent synapse SS in the first neuromorphic transistor activates and accepts input only when the postsynaptic condition meets a threshold. Therefore, if an object passes the sensor in the order 5, 4, 3, 2, and 1 (outbound), since the silent synapse is the first to receive input, the conditions for activation in dendritic branch br1 are not met, and the silent synapse does not accept input. If dendritic branch br1 does not receive the input from position 5, which contributes the most, the output of the first dendritic unit DU1 is insufficient. Even if the first neuromorphic transistor T1 is turned on, the output of the first soma unit CU1 cannot reach the current threshold required to turn on the first current switch CSW1, meaning that the first axon unit AU1 will not output. In other words, when the input signal is outbound, the first neuromorphic device is inactive.

[0171] Conversely, when an object passes through the sensor in the order 1, 2, 3, 4, and 5 (in the forward direction), the excitatory synapse of the first neuromorphic transistor receives the input first and directs it to dendritic branch br1. At this point, when the silent synapse SS receives the input, the conditions in dendritic branch br1 are already in place for the silent synapse to activate. Therefore, the silent synapse directs the input at position 5 to dendritic branch br1, increasing the output of the first dendritic unit. The output of the first soma unit CU1 reaches the threshold that turns on the first current switch CSW1, causing the first axon unit AU1 to output. In other words, when the input signal is in the forward direction, the first neuromorphic device turns on.

[0172] In summary, the first neuromorphic device according to the embodiment of the present disclosure is turned on and outputs only when receiving input in a specific direction. Through the cooperation of neuromorphic transistors and current switches, directional selectivity is achieved without an additional processor.

[0173] The structure and principle of the first neuromorphic device in the present disclosure are introduced above. The operation process of the first neuromorphic transistor is described to illustrate how the first neuromorphic device achieves directionally selective output.

[0174] Figure 5A is a structural diagram of a second neuromorphic device according to some embodiments of the present disclosure.

[0175] like Figure 5A As shown, similar to the first neuromorphic device, the second neuromorphic device N2 includes a second neuromorphic transistor T2, a second pulse generator PG2, and a second current switch CSW2. The second pulse generator PG2 is configured to convert the current output by the second neuromorphic transistor T2 into a second voltage pulse. The second current switch CSW2 is electrically connected to the second pulse generator PG2 and configured to turn on when the current output by the second neuromorphic transistor N2 exceeds a second current threshold. The second pulse generator PG2 and the second current switch CSW2 collectively correspond to the second axon unit described above.

[0176] In some embodiments, the second axon unit may further include an additional voltage switch electrically connected to the second pulse generator PG2 and configured to be turned on when the equivalent gate voltage of the second neuromorphic transistor N2 exceeds a first voltage threshold.

[0177] With the above configuration, by utilizing the equivalent gate voltage in the neuromorphic transistor to further determine whether the output strength of the first neuromorphic transistor meets the conditions, noise caused by the generated current instability can be reduced, thereby improving the stability of the system.

[0178] The second neuromorphic transistor T2 will be introduced below. Figure 5B is a diagram illustrating the structure of a second neuromorphic transistor according to some embodiments of the present disclosure. The structure of the second neuromorphic transistor is similar to that of the first neuromorphic transistor, except that the second gate of the second neuromorphic transistor includes at least one first-type gate, but does not include a second-type gate.

[0179] like Figure 5B As shown, the second neuromorphic transistor T2 includes a second gate T26 and a second source T28, wherein the second gate T26 corresponds to the second synaptic unit SU2, which only includes the first type gate T26A. The gate can be an electrode formed of metal such as platinum or gold.

[0180] like Figure 5B As shown, the second neuromorphic transistor T2 further includes a second source region T21 and a second drain region T22 , corresponding second source electrode T28 and second drain electrode T29 , a second channel region T23 , and an ion-doped second dielectric layer T24 .

[0181] like Figure 5B As shown, the second channel region T23 is located between the first source region T21 and the first drain region T22. In some embodiments, the material of the channel region is doped nanowires, such as silicon nanowires. The nanowires can be formed by etching on the top layer of the silicon substrate or by gas-phase or liquid-phase catalytic growth.

[0182] In some embodiments, the nanowires in the channel region T23 are arranged in a honeycomb shape between the source region T21 and the drain region T22 . The honeycomb-shaped nanowires can effectively sense potential changes on their surfaces.

[0183] The second source region T21 and the second drain region T22 , the corresponding second source electrode T28 and the second drain electrode T29 , and the second channel region T23 as a whole correspond to the second cell unit CU2 .

[0184] The second dielectric layer T24 corresponds to the second dendrite unit DU2. The second dielectric layer T24 covers the second channel region T23, extends along the length of the second channel region T23, and contacts the second gate T26. The second neuromorphic transistor T2 does not include a second-type gate or a resistive switch.

[0185] In some embodiments, the second dielectric layer includes an ion-doped sol-gel film, for example, a composite material including silicon dioxide, containing copper ions Cu 2+ 、Nickel ion Ni 2+ and chloride ions Cl - Ion-doped sol-gel membranes, due to the presence of mobile ions, can simulate ion diffusion in neuronal membranes.

[0186] Similar to the first neuromorphic transistor, in some embodiments, the second neuromorphic transistor T2 further includes a first substrate T20. Figure 5B As shown, the second substrate T20 includes a first region T20A and a second region T20B. The first region T20A and the second region T20B have no overlapping region. The first substrate T20 can be a lightly doped semiconductor substrate, such as a P or N type silicon substrate with a low doping concentration.

[0187] like Figure 5BAs shown, the second source region T21 and the second drain region T22 are located in the first region T20A. In some embodiments, the second source region T21 and the second drain region T22 are semiconductor regions with a doping type opposite to that of the second substrate T20. In other words, the second source region T21, the second substrate T20, and the second drain region T22 can form two back-to-back PN junctions.

[0188] In other embodiments, the first source region T11 and the first drain region T12 are semiconductor regions having the same doping type as the first channel region T13 but a different doping concentration. For example, the first source region T11 and the first drain region T12 are heavily doped N-type semiconductors, while the first channel region T13 is lightly doped N-type semiconductors. In this way, the first source region T11, the first drain region T12, and the first channel region T13 can form a junctionless transistor, thereby improving the sensitivity of the transistor.

[0189] The second source electrode T28 and the second drain electrode T29 are located on the second source region T21 and the second drain region T22, respectively. In some embodiments, the second source electrode T28 and the second drain electrode T29 are parallel strip electrodes. The second source electrode T28 and the second drain electrode T29 can be electrodes formed of metals such as aluminum or platinum.

[0190] like Figure 5B As shown, the first portion T24A of the second dielectric layer T24 is located above the first channel region T23 and covers the first channel region T23, while the second portion T24B is located above the second region T20B.

[0191] As described above, the main function of the second neuromorphic device is to determine whether the output strength of the first neuromorphic device reaches a threshold. Therefore, the above structure makes the conduction of the second neuromorphic device depend on whether the strength of the input signal can cause the second neuromorphic transistor to generate a current exceeding the second current threshold.

[0192] Figure 6A is a block diagram of a third neuromorphic device according to some embodiments of the present disclosure.

[0193] like Figure 6A As shown, the third neuromorphic device N3 includes a third neuromorphic transistor T3, a third pulse generator PG3, and a third current switch CSW3. The structure of the third neuromorphic device N3 is basically the same as that of the second neuromorphic device, and the difference lies mainly in the input.

[0194] The third neuromorphic transistor T3 includes a third gate and a third source. The third gate corresponds to a third synaptic unit and includes multiple first-type gates but no second-type gates. The multiple first-type gates include a first sub-gate connected to the first source and a second sub-gate connected to the second source. The first sub-gate is closer to the third source than the second sub-gate. The first sub-gate corresponds to an inhibitory synapse, and the second sub-gate corresponds to an excitatory synapse. The gates may be electrodes formed of a metal such as platinum or gold.

[0195] The third pulse generator PG3 is configured to convert the current output by the third source into a third voltage pulse. The third current switch CSW3 is electrically connected to the third pulse generator PG3 and is configured to be turned on when the current output by the third neuromorphic transistor N3 exceeds a third current threshold. The third pulse generator PG3 and the third current switch CSW3 as a whole correspond to the third axon unit. Figure 6B Introducing the functions of excitatory and inhibitory synapses of neuromorphic transistors.

[0196] Figure 6B is a logic diagram showing the input-output signals of a neuromorphic transistor with excitatory and inhibitory synapses located at different locations.

[0197] The inhibiting input I means that the input signal is negative, in other words, the input pulse signal is a negative voltage pulse signal. The exciting input E means that the input signal is positive, in other words, the input pulse signal is a positive voltage pulse signal.

[0198] exist Figure 6B In (i), the excitatory synapse receiving the excitatory input E is located at a position farther from the cell body, while the inhibitory synapse receiving the inhibitory input I is located closer to the cell body, that is, the conduction path from the excitatory input E to the cell body passes through the synapse receiving the inhibitory input I, and the inhibitory input I inhibits the excitatory input E. In (ii) and (iii), the synapse receiving the excitatory input E is located closer to the cell body, while the inhibitory synapse receiving the inhibitory input I is located farther from the cell body, that is, the conduction path from the excitatory input E to the cell body does not pass through the synapse receiving the inhibitory input I.

[0199] like Figure 6B As shown in (i), when the inhibitory input I is in the path of the excitatory input E, the neuromorphic transistor can perform a "NAND" logic operation—in other words, an "E AND (NOT)" operation. For example, when both the excitatory and inhibitory inputs are activated, E and I are both 1. In this case, the neuromorphic transistor is off because the inhibitory input I suppresses the excitatory input E. The neuromorphic transistor only conducts when E is 1 and I is 0. When the neuromorphic transistor is on, the logic output X is 1.

[0200] When the inhibitory input I is outside the path of the excitatory input E, this logical operation no longer applies. Figure 6B As shown, in case (ii), the amplitude of the inhibitory input I is not greater than the amplitude of the excitatory input E ( ), as long as there is an excitatory input in the system (E=1), the neuromorphic transistor is turned on; in the case of (iii), when the amplitude of the inhibitory input I is greater than the amplitude of the excitatory input E ( ), since the inhibitory input I will inhibit the excitatory input E, the neuromorphic transistor will also embody the "E and (not I)" logic operation similar to (i).

[0201] References above Figures 3A to 6B The neuromorphic devices according to some embodiments of the present disclosure are introduced. Figure 7 A circuit diagram of a neuromorphic device circuit C1 according to some embodiments of the present disclosure is introduced.

[0202] Figure 7 is a schematic diagram of a model circuit of a neuromorphic device according to some embodiments of the present disclosure.

[0203] like Figure 7 As shown, the model circuit of the neuromorphic circuit C1 includes: a voltage switch vsw, an equivalent resistor R h11 - R h55 、R h2E 、R h3E 、R h3I , ion-doped dielectric layer SGF1-SGF3, source S, resistor R I , inherent resistance R B , drain D, current switches csw (N1), csw (N2), csw (N3), voltage switches vsw (N2), vsw (N3), positive voltage pulse generator v B(E) , negative voltage pulse generator v B(I) . Figure 7 Some circuit related parameters are also shown, such as v GS,11 -v GS,55 , N1E, N2E, N2I, N3E, Th1-3, V Th1-3 、V DS , I DS,N11-N15 , I DS,N1 , I DS,N2 , I DS,N3、 v gS1-3 .

[0204] Figure 7 The ion-doped dielectric layer SGF is equivalent to the dendrite unit, the source S, the drain D, and the resistor R I The whole is equivalent to the cell body unit, the voltage pulse generator v B(E)、v B(I) , inherent resistance R B And the corresponding current switch csw is equivalent to the axon unit.

[0205] The above-mentioned current switches csw (N1), csw (N2), and csw (N3) correspond to the first current switch, the second current switch, and the third current switch, respectively, and are configured to be turned on when the output current of the corresponding neuromorphic transistor exceeds the current threshold. The voltage switch vsw (N3) corresponds to the above-mentioned first voltage switch and is configured to be turned off when the intensity of the excitation output N1E in the first output signal is greater than the first voltage threshold. The voltage switch vsw (N2) corresponds to the above-mentioned additional voltage switch and is configured to be turned on when the equivalent gate voltage v of the second neuromorphic transistor is greater than the equivalent gate voltage v gS2 The switch can also reduce the noise output in neuromorphic devices.

[0206] At the same time, SGF1 is also equivalent to Figure 3B The dielectric layer T14, SGF2 is also equivalent to Figure 5B The dielectric layer T24 in the source electrode S is equivalent to Figure 3B The source T18 and Figure 5B The source T28 and drain D are equivalent to Figure 3B The drain T19 and Figure 5B Drain T29 in. Figure 7 The resistance R in the circuit I It is an analog parameter necessary for measuring current.

[0207] like Figure 7 As shown, the pulse signal v GS,11 -v GS,15 The input signals received by multiple gates are equivalent to the synaptic unit receiving the sensing signal from the sensor that senses the movement of the object, which is also equivalent to Figure 3B The input signal received by multiple gates in the pulse signal v GS,11 -v GS,14 、v GS,21 -v GS,24 、v GS,31 -v GS,34 、v GS,41 -v GS,44 、v GS,51 -v GS,54 is the input signal received from the excitatory synapse, the pulse signal v GS,15 、v GS,25 、v GS,35 、v GS,45 、v GS,55 is the input signal received from the silent synapse.

[0208] I DS,N11-N15are the source and drain currents generated by the dendritic branches br1-br5 of the first neuromorphic device N1. The outputs of different branches of the dendritic unit are integrated together through the cell body unit. The integration between the dendritic branches is usually linear integration. I DS,N1 , I DS,N2 , I DS,N3 are the total source-drain currents generated in the first neuromorphic device N1, the second neuromorphic device N2, and the third neuromorphic device N3, respectively.

[0209] Figure 7 The voltage switch vsw in the circuit is equivalent to the functional mechanism of silent synapses, which is also equivalent to Figure 3B The resistor switch T15 in the voltage pulse v GS,11 -v GS,14 When the integration of the dendritic unit reaches the threshold, the voltage switch vsw is turned on to introduce the input signal received by the silent synapse. GS,11 -v GS,14 When the integration of the dendritic unit is below the threshold, the voltage switch vsw is not turned on and the input signal received from the silent synapse is not introduced.

[0210] Figure 7 The equivalent resistance R in the circuit h11 - R h55 、R h2E 、R h3E - R h3I It is equivalent to the impedance that the input signal of each synapse needs to experience when it is transmitted through the dendritic unit to the cell body unit.

[0211] In some embodiments, the equivalent resistance can be calculated according to the following formula:

[0212] .

[0213] r hi is a constant resistance parameter related to the position of the synapse on the dendritic unit. Since the resistance is related to the length of the ion migration path, r hi is a parameter based on the distance from the cell body to the synapse. That is, r hi It is determined by the distance between the synaptic site i and the cell body unit and the material of the dendritic unit. The farther the distance between the synaptic site i and the cell body unit, the longer the ion migration path required for input signal conduction, and r hi The bigger.

[0214] V A,i is the input voltage pulse v GS,i The amplitude of v gS It represents the output of the dendrite unit after integrating the input from the synaptic unit, which is reflected as the equivalent gate voltage v of the transistor gS, and v GS,j It changes nonlinearly. When the equivalent gate voltage v gS When the voltage is greater than the membrane threshold voltage, the voltage switch vsw is turned on.

[0215] V DS V represents the potential difference between the source S and the drain D, that is, the source-drain voltage. Th1-3 It represents the threshold voltage corresponding to the conduction of the neuromorphic transistor, reflecting the minimum equivalent gate voltage required to turn the neuromorphic transistor from the off state to the on state.

[0216] Based on the above description of the model circuit diagram of the neuromorphic device circuit, the following is combined with Figure 8 The operation process of the detection device including the above circuit is further described. Figure 8 3 is a morphological schematic diagram illustrating a device for detecting a moving direction according to some embodiments of the present disclosure.

[0217] like Figure 8 As shown, the moving direction detection device DD1 includes the neuromorphic circuit C1 as described above and a plurality of sensors SE, wherein the plurality of sensors constitute a sensor array.

[0218] The plurality of sensors are used to sense movement of an object, and each sensor includes a plurality of sensor units linearly arranged along a first direction. The first neuromorphic device includes a plurality of dendrite branches br1-br3.

[0219] In the detection device DD1 , multiple synapses of the first neuromorphic device N1 in the neuromorphic circuit C1 are connected to sensor units in the sensor array in a one-to-one correspondence, wherein a silent synapse SS and its adjacent excitatory synapse are connected to different sensors.

[0220] like Figure 8 As shown, in the detection device DD1, the sensor units in the sensor are linearly arranged along a first direction O1, that is, arranged horizontally to the right. Therefore, five sensor units in the horizontal direction constitute a sensor, for example, sensor units S11, S12, S13, S14, and S15 constitute a sensor S1.

[0221] The multiple synapses on each dendritic branch of the first dendritic unit can be interleaved with multiple sensor units in adjacent sensors. For example, the multiple synapses on dendritic branch br2 are connected to sensor units S31, S22, S33, S24, and S35 in the second sensor S2 and the third sensor S3, respectively. The silent synapse SS is connected to sensor unit S35 of the third sensor S3, while the excitatory synapse adjacent to the silent synapse SS is connected to sensor unit 24 of the second sensor S2. Other dendritic branches can be connected in a similar manner, for example, dendritic branch br1 is connected to S21, S12, S23, S14, and S25.

[0222] In some embodiments, the first dendritic unit further includes dendritic branches connected only to excitatory synapses.

[0223] exist Figure 8 In the detection device DD1 shown, when an object moves along a first direction O1 (0° in this embodiment, i.e., horizontally to the right), an excitatory synapse on a dendritic branch of the first synaptic unit first receives a pulse signal from the corresponding sensor unit, activating a silent synapse, which then receives a pulse signal from the corresponding sensor unit.

[0224] Since the sensor is composed of a plurality of sensor units linearly arranged along a first direction, movement of the object along the first direction will activate all sensor units connected to the synapses on the corresponding dendrite units.

[0225] For example, in Figure 8 In the example of movement along the first direction O1 shown in FIG, the excitatory synapses ES on dendritic branch br2 first receive pulse signals from their corresponding sensor units S31, S22, S33, and S24. This causes dendritic branch br2 to meet the conditions for activating the silent synapse SS. The silent synapse then receives a pulse signal from its corresponding sensor unit S35. All synapses on dendritic branch br2 receive the pulse signal.

[0226] Due to the contribution of the silent synapse SS closest to the cell body, the current generated in the first neuromorphic transistor meets the first current threshold of the first current switch, and the first neuromorphic device is turned on.

[0227] In this scenario, all synapses on the same dendritic branch receive input from the sensor unit and direct it to the dendritic unit. Due to the nonlinear integration of synaptic input signals by dendritic branches described above, the superposition of multiple input signals from the same branch results in a stronger current in the neuromorphic device. Consequently, the pulses generated by the pulse generator in the neuromorphic device are wider. Therefore, in this scenario, the first neuromorphic device N1 outputs a first-directional output signal, including an excitation signal N1E with a stronger signal strength.

[0228] The second neuromorphic device N2 receives as input an excitation signal N1E from the first directional output signal. When the excitation signal N1E is strong, the current output by the second soma unit meets the second current threshold of the second current switch, causing the second neuromorphic device to conduct and output a second output signal. This second output signal includes an inhibition signal N2I and an excitation signal N2E.

[0229] The third neuromorphic device N3 receives as input the excitation output N1E in the first direction output signal and the inhibition output N2I in the second direction output signal. As described above, the presence of the inhibition output N2I prevents the excitation output N1E from turning on the third neuromorphic device N3, i.e., the third neuromorphic device N3 does not output the third output signal.

[0230] Therefore, when the object moves in the first direction, the second neuromorphic device is turned on and outputs N2E, indicating that the object moves in the first direction; while the third neuromorphic device is not turned on and has no output.

[0231] In the above embodiment, when the object moves along the second direction O2 (in this embodiment, a direction forming an angle of 315° with the horizontal rightward direction), some of the excitatory synapses on the first dendritic branch first receive the pulse signal from the corresponding sensor unit; some of the excitatory synapses on the second dendritic branch then receive the pulse signal from the corresponding sensor unit, activating the silent synapses, which then receive the pulse signal from the corresponding sensor unit.

[0232] For example, in Figure 8 In the example of movement along the second direction O2 shown in Figure 1, some of the excited synapses on dendritic branch br1 first receive pulse signals from corresponding sensor units S12 and S23, respectively. Meanwhile, some of the excited synapses on dendritic branch br3 receive pulse signals from corresponding sensor unit S34. This causes dendritic branch br3 to meet the conditions for activating silent synapses SS. The silent synapses then receive pulse signals from corresponding sensor unit S45.

[0233] Due to the contribution of the silent synapse SS closest to the cell body, the current generated in the first neuromorphic transistor meets the first current threshold of the first current switch, and the first neuromorphic device is turned on.

[0234] In the above scenario, the sensor unit's input is dispersed across the different dendritic branches of the dendritic unit. Similarly, based on the nonlinear integration of synaptic input signals by dendritic branches as described above, the current generated in the neuromorphic device is relatively weak due to the relatively dispersed input pulse signal. Consequently, the pulses generated by the pulse generator in the neuromorphic device are relatively narrow. Consequently, in the above scenario, the first neuromorphic device N1 outputs a second-directional output signal, including an excitation signal N1E with a weaker signal strength. That is, the second-directional output signal has a lower strength than the first-directional output signal.

[0235] The second neuromorphic device N2 receives the excitation signal N1E in the second direction output signal as input. If the excitation signal N1E is weak, the output of the second soma unit does not meet the second current threshold of the second current switch, the second current switch is not turned on, and the second neuromorphic device does not output the second output signal.

[0236] At this point, the third neuromorphic device N3 receives only the excitation output N1E from the second direction output signal as input. Because there is no inhibitory synapse, the output of the third soma unit meets the third current threshold of the third current switch, turning the third neuromorphic device on. The third neuromorphic device N3 then outputs a third output signal, including the excitation output N3E.

[0237] Therefore, when the object moves in the second direction, the third neuromorphic device is turned on and outputs an excitation output N3E, indicating that the object moves in the second direction; while the second neuromorphic device is turned off and has no output.

[0238] In the above embodiment, when the object moves in a direction different from the first direction and the second direction, for example Figure 8 The direction O shown in FIG. 1 is 180° from the horizontal right direction. d The silent synapse of the first synaptic unit first receives the pulse signal from the corresponding sensor unit, and the excited synapse of the first synaptic unit later receives the pulse signal from the corresponding sensor unit. At the same time, the silent synapse no longer receives the pulse signal.

[0239] exist Figure 8 In the example of movement shown in Figure 1, silent synapse SS first receives a pulse signal from sensor unit S35. At this point, dendrite branch br2 does not meet the conditions for silent synapse activation, and silent synapse SS is unable to direct input to dendrite branch br2. Excitatory synapse ES then receives pulse signals from sensor units S24, S33, S22, and S31, respectively.

[0240] In this case, due to the lack of contribution from the silent synapse SS closest to the cell body, the current generated in the first neuromorphic transistor does not meet the first current threshold of the first current switch, and the first neuromorphic device does not conduct. The first neuromorphic device N1 does not output the first output signal, so the second and third neuromorphic devices are both off, and the detection device DD1 does not output.

[0241] The direction O that is 180° to the right of the horizontal direction d For example input only, in the first detection device DD1 according to the embodiment of the present disclosure, a first current threshold can be set so that input signals different from the first direction and the second direction cannot turn on the first neuromorphic device, thereby enabling the detection device DD1 to detect the first direction and the second direction.

[0242] Therefore, when the object moves in a direction different from the first direction and the second direction, the first neuromorphic device has no output and the detection device has no output.

[0243] It can be seen that the moving direction detection device DD1 in the above embodiment can generate different outputs for different directions of the moving signal without a processor, thereby significantly reducing the power consumption of the system.

[0244] The above describes a detection device DD1 for detecting movement in a first direction and a second direction. The detection device DD1 can detect two directions with a certain angle. According to other embodiments of the present disclosure, the present disclosure further proposes a detection device DD2 for detecting movement in four directions, based on the above detection device DD1. The detection device DD2 can detect the first direction, the second direction, and a third and fourth direction, where the third direction is opposite to the first direction, and the fourth direction is opposite to the second direction.

[0245] Figure 9 is a block diagram illustrating a device for detecting a moving direction according to other embodiments of the present disclosure. Figure 9 and Figure 1 The difference is that the detection device DD2 further includes a fourth neuromorphic device N4, a fifth neuromorphic device N5, a sixth neuromorphic device N6, and a seventh neuromorphic device N7.

[0246] like Figure 9 As shown, the detection device DD2 includes a plurality of sensors S, a first neuromorphic device N1, a second neuromorphic device N2, a third neuromorphic device N3, a fourth neuromorphic device N4, a fifth neuromorphic device N5, a sixth neuromorphic device N6, and a seventh neuromorphic device N7.

[0247] Similar to detection device DD1, the first neuromorphic device N1 through the seventh neuromorphic device N7 of detection device DD2 can constitute a neuromorphic circuit C2 in detection device DD2. The following further describes the first neuromorphic device N1 through the seventh neuromorphic device N7 and their connection methods using a morphological representation of neuromorphic circuit C2.

[0248] Figure 10 is a morphological schematic diagram illustrating a neuromorphic circuit C2 according to some other embodiments of the present disclosure.

[0249] like Figure 10 As shown, the first neuromorphic device N1 in the neuromorphic circuit C2 is substantially the same as that in the above neuromorphic circuit C1, with the only difference being that the first output signal includes not only an excitation output N1E but also an inhibition output N1I.

[0250] The second neuromorphic device N2 and the third neuromorphic device N3 are consistent with those in the above neuromorphic circuit C1.

[0251] A fourth neuromorphic device N4 includes a fourth synaptic unit SU4, a fourth dendritic unit DU4, and a fourth cell soma unit CU4. The fourth synaptic unit CU4 includes multiple excitatory synapses ES. The fourth dendritic unit DU4 includes multiple dendritic branches, each of which is connected to the multiple excitatory synapses. The multiple synapses on a dendritic branch are connected one-to-one to multiple sensor units in different sensors. The fourth cell soma unit CU4 is configured to process the output of the fourth dendritic unit to generate a fourth output signal. This fourth output signal includes an excitatory output N4E.

[0252] In some embodiments, the fourth neuromorphic device N4 further includes a fourth axon unit AU4. The fourth axon unit AU4 is configured to convert the output of the fourth soma unit CU4 into a fourth voltage pulse. In some embodiments, the fourth axon unit includes a voltage pulse generator to implement the above output conversion.

[0253] The voltage pulse generator may include a positive voltage pulse generator and a negative voltage pulse generator. The positive voltage pulse generator may convert the output of the cell body unit into an excitatory output, and the negative voltage pulse generator may convert the output of the cell body unit into an inhibitory output.

[0254] In some other embodiments, the fourth axon unit further includes a fourth current switch, and the fourth current switch is configured to be turned on when the output of the fourth soma unit is greater than a fourth current threshold.

[0255] The fifth neuromorphic device N5 includes a fifth synaptic unit SU5, a fifth dendritic unit DU5, and a fifth soma unit CU5. The fifth synaptic unit CU5 includes an inhibitory synapse IS and an excitatory synapse ES. The inhibitory synapse is used to receive the inhibitory output N1I in the first output signal, and the excitatory synapse is used to receive the excitatory output N4E in the fourth output signal. The fifth dendritic unit DU5 includes at least one dendritic branch, each of which is connected to an inhibitory synapse and an excitatory synapse. The inhibitory synapse is located closer to the fifth soma unit CU5 than the excitatory synapse. The fifth soma unit is used to process the output of the fifth dendritic unit to generate a fifth output signal. The fifth output signal includes the excitatory output N5E.

[0256] The fifth neuromorphic device N5 further includes a fifth axon unit AU5. The fifth axon unit AU5 is configured to convert the output of the fifth soma unit CU5 into a fifth voltage pulse. Similarly, the fifth axon unit may include a voltage pulse generator to implement the above output conversion.

[0257] In some embodiments, the fifth axon unit includes a fifth current switch configured to be turned on when the output of the fifth soma unit is greater than a fifth current threshold.

[0258] The sixth neuromorphic device N6 includes a sixth synaptic unit SU6, a sixth dendritic unit DU6, and a sixth cell soma unit CU6. The sixth synaptic unit SU6 includes an excitatory synapse ES for receiving an excitatory output N5E from the fifth output signal. The sixth dendritic unit DU6 includes at least one dendritic branch, each dendritic branch connected to the excitatory synapse. The sixth cell soma unit CU6 is configured to process the output of the sixth dendritic unit to generate a sixth output signal. The sixth output signal includes an excitatory output N6E and an inhibitory output N6I.

[0259] The sixth neuromorphic device N6 further includes a sixth axon unit AU6. The sixth axon unit AU6 is configured to convert the output of the sixth soma unit CU6 into a sixth voltage pulse. Similarly, the sixth axon unit may include a voltage pulse generator to implement the aforementioned output conversion.

[0260] In some embodiments, the sixth axon unit includes a sixth current switch, and the sixth current switch is configured to be turned on when the output of the sixth soma unit is greater than a sixth current threshold.

[0261] The seventh neuromorphic device N7 includes a seventh synaptic unit SU7, a seventh dendritic unit DU7, and a seventh cell soma unit CU7. The seventh synaptic unit SU7 includes an inhibitory synapse IS and an excitatory synapse ES. The inhibitory synapse is used to receive the inhibitory output N6I in the sixth output signal, and the excitatory synapse is used to receive the excitatory output N5E in the fifth output signal. The seventh dendritic unit includes at least one dendritic branch, each of which is connected to an inhibitory synapse and an excitatory synapse. The inhibitory synapse is located closer to the seventh cell soma unit CU7 than the excitatory synapse. The seventh cell soma unit CU7 is used to process the output of the seventh dendritic unit to generate a seventh output signal. The seventh output signal includes the excitatory output N7E.

[0262] The seventh neuromorphic device N7 further includes a seventh axon unit AU7. The seventh axon unit AU7 is configured to convert the output of the seventh soma unit CU7 into a seventh voltage pulse. Similarly, the seventh axon unit may include a voltage pulse generator to implement the above-mentioned output conversion.

[0263] In some embodiments, the seventh axon unit includes a seventh current switch, and the seventh current switch is configured to be turned on when the output of the seventh soma unit is greater than a seventh current threshold, wherein the seventh current threshold is less than the sixth current threshold.

[0264] The function of the above-mentioned current switch is consistent with that of the current switch in the detection device DD1, and can enhance the directional selectivity of the neuromorphic device.

[0265] In some other embodiments, the seventh axon unit of the seventh neuromorphic device further includes a second voltage switch, wherein the second voltage switch is configured to be turned off when the intensity of the excitation output N5E in the fifth output signal exceeds a second voltage threshold. The second voltage switch functions in the same manner as the first voltage switch.

[0266] In some further embodiments, the seventh synaptic unit of the seventh neuromorphic device further includes a third voltage switch, where the third voltage switch is configured to be turned off when the intensity of the excitation output N6E in the sixth output signal is greater than a third voltage threshold.

[0267] The third voltage switch described above functions similarly to the second voltage switch. When the sixth neuromorphic device is conducting and outputting, the seventh neuromorphic device should not output. Therefore, the addition of the third voltage switch further reduces noise in the seventh neuromorphic device's output by disconnecting the seventh neuromorphic device's input when the sixth neuromorphic device is outputting.

[0268] In the above-mentioned second detection device DD2, the excited output N2E in the second output signal is used to indicate that the object moves along the first direction; the excited output N3E in the third output signal is used to indicate that the object moves along the second direction; the excited output N6E in the sixth output signal is used to indicate that the object moves along the above-mentioned third direction; the excited output N7E in the seventh output signal is used to indicate that the object moves along the above-mentioned fourth direction.

[0269] The above article introduces the overall structure of neuromorphic device circuits. The following will specifically introduce the structure and principle of each neuromorphic device.

[0270] The structure of the fourth neuromorphic device is essentially identical to that of the second neuromorphic device, differing primarily in the synapses that receive input signals. The synapses of the second neuromorphic device only need to receive the first excitation output N1E of the first output signal, so the number of synapses is unlimited. However, like the first neuromorphic device, the fourth neuromorphic device needs to receive sensing signals from multiple sensors that detect object movement. Therefore, multiple gates of multiple dendritic branches must be connected to the sensors in a one-to-one correspondence to receive input signals.

[0271] Figure 11A 、 11B are simulation diagrams showing how the output signals generated by one dendrite branch of the first neuromorphic transistor and one dendrite branch of the fourth neuromorphic transistor vary with the input signal according to some embodiments of the present disclosure. Specifically, Figure 11A This is a simulation diagram showing how the output signal generated by a dendritic branch of the first neuromorphic transistor changes with the input signal; Figure 11B This is a simulation diagram of the output signal generated by a dendritic branch of the fourth neuromorphic transistor varying with the input signal.

[0272] from Figure 11A 、 Figure 11B It can be seen that when receiving an inbound signal, the outputs of the first neuromorphic transistor and the fourth neuromorphic transistor are consistent; when receiving an outbound signal, the synaptic input of the first neuromorphic transistor is blocked, and the input at position 5 cannot be introduced into the first neuromorphic transistor. Therefore, even if the first neuromorphic transistor is turned on, the current it generates is small and insufficient to reach the first current threshold that turns on the first current switch. However, the synaptic input of the fourth neuromorphic transistor is not blocked, so the current generated by turning on the fourth neuromorphic transistor is large and may reach the fourth current threshold that turns on the fourth current switch.

[0273] Figure 11A 、 Figure 11B It reflects the influence of input signals in different directions on output signals, especially the neuromorphic devices with silent synapses fully demonstrate the directional selectivity of input signals.

[0274] It can be seen that the fourth neuromorphic device differs from the first neuromorphic device in that it cannot respond to the direction of the input signal within the same dendritic branch. Based on the above description of the operation of the first neuromorphic device, by setting the current threshold of the current switch, the fourth neuromorphic device can be turned on and output only when the input signal direction is the first direction, the second direction, the third direction, or the fourth direction.

[0275] The fifth neuromorphic transistor has the same structure as the third neuromorphic transistor and receives the inhibitory output N1I of the first neuromorphic device and the excitatory output N4E of the fourth neuromorphic device.

[0276] The sixth neuromorphic transistor has the same structure as the second neuromorphic transistor and receives the excitation output N5E of the fifth neuromorphic device.

[0277] In some embodiments, the sixth axon unit may further include an additional voltage switch to further reduce noise in the neuromorphic device. The voltage switch is electrically connected to the sixth pulse generator and configured to turn on when the equivalent gate voltage of the sixth neuromorphic transistor exceeds a first voltage threshold.

[0278] The seventh neuromorphic transistor has the same structure as the third neuromorphic device and receives the excitation output N5E of the fifth neuromorphic device and the inhibition output N6I of the sixth neuromorphic device.

[0279] Based on the above Figure 9-11B The morphology of the neuromorphic device circuit C2 in the detection device DD2 is introduced, and the structure of each neuromorphic device in C2 is also described in detail. Figure 12 Introduce the circuit structure of neuromorphic device circuit C2.

[0280] Figure 12 is a schematic diagram of a model circuit of a neuromorphic device circuit C2 according to some other embodiments of the present disclosure.

[0281] Figure 12 The circuit components shown in FIG have some Figure 7 The basic consistency shown in Figure 12 and Figure 7 The difference is that it also includes circuit structures corresponding to the fourth neuromorphic device N4 to the seventh neuromorphic device N7.

[0282] like Figure 12 As shown, the model circuit of the neuromorphic circuit C2 includes: a voltage switch vsw, an equivalent resistor R h111 -Rh155 、R h2E 、R h3E 、R h3I 、R h411 -R h455 、R h5E、 R h5I、 R h6E 、R h7E 、R h7I , ion-doped dielectric layer SGF, source S, resistor R I , inherent resistance R B , drain D, current switch csw (N1-N7), voltage switch vsw (N2), vsw (N3), vsw (N6), vsw (N7), vsw (N5E), positive voltage pulse generator v B(E) , negative voltage pulse generator v B(I) . Figure 12 Some circuit related parameters are also shown, such as v GS,11 -v GS,55 , N1E, N1I, N2E, N2I, N3E, N4E, N5E, N6E, N6I, N7E, Th1-7, V Th1-7 、V DS , I DS,N1-N7 、v gS1-7 .

[0283] I DS,N41-N45 The source and drain currents of the dendritic branches br1-br5 of the fourth neuromorphic device N4 are respectively. The outputs of different branches of the dendritic unit are integrated together through the cell body unit.

[0284] Figure 12 The ion-doped dielectric layer SGF is equivalent to the dendrite unit, the source S, the drain D, and the resistor R I The whole is equivalent to the cell body unit, the voltage pulse generator v B(E) 、v B(I) , inherent resistance R B And the corresponding current switch csw and voltage switch vsw are equivalent to the axon unit.

[0285] The current switches csw ( N1 - N7 ) correspond to the first to seventh current switches, respectively, and are configured to be turned on when the output current of the corresponding neuromorphic transistor exceeds a current threshold.

[0286] The voltage switch vsw (N3) corresponds to the above-mentioned first voltage switch and is configured to be turned off when the intensity of the first excitation output N1E is greater than the first voltage threshold. The voltage switch vsw (N7) corresponds to the above-mentioned second voltage switch and is configured to be turned off when the intensity of the fifth excitation output N5E is greater than the second voltage threshold. The voltage switch vsw (N5E) corresponds to the above-mentioned third voltage switch and is configured to be turned off when the intensity of the sixth excitation output N6E is greater than the third voltage threshold. The voltage switches vsw (N2) and vsw (N6) correspond to the above-mentioned additional voltage switches and are configured to be turned off when the equivalent gate voltage v of the corresponding neuromorphic transistor is greater than the equivalent gate voltage v gS The switch can also reduce the noise output in neuromorphic devices.

[0287] Based on the above description of the neuromorphic circuit C2, Figure 13 Let's further describe the detection device DD2 designed based on the above-mentioned neuromorphic circuit C2. Figure 13 3 is a morphological schematic diagram showing a device for detecting a moving direction according to other embodiments of the present disclosure.

[0288] like Figure 13 As shown, the moving direction detection device DD2 includes the neuromorphic circuit C2 as described above and a plurality of sensors SE, wherein the plurality of sensors constitute a sensor array. In some embodiments, the sensors are retinal sensors.

[0289] like Figure 13 As shown, the sensor unit of the detection device DD2 is in the shape of a regular hexagon, and the multiple sensors are an array of sensors with multiple sensor units densely packed in a regular hexagonal plane, and the center line of multiple sensor units in the same sensor is parallel to the first direction O1. For example, Figure 13 The sensor units S31, S33, and S35 constitute a sensor, and the center line thereof is parallel to the first direction O1; the sensor units S32 and S34 constitute another sensor, and the center line thereof is parallel to the center line of the sensor units S31, S33, and S35.

[0290] In some embodiments, as Figure 13 As shown, multiple synapses on each dendritic branch in the first dendritic unit are connected to multiple sensor units in the adjacent sensor in an interlaced manner; multiple synapses on each dendritic branch in the fourth dendritic unit are connected to multiple sensor units in the adjacent sensor in an interlaced manner.

[0291] like Figure 13As shown, multiple synapses on the dendritic branch br2 of the first neuromorphic device and multiple synapses on the fourth neuromorphic device br2' are interlacedly connected to sensor units {S32, S34} and sensor units {S31, S33, S35}, respectively. That is, two adjacent synapses are connected to different sensors. Silent synapse SS is connected to sensor unit S35. Other dendritic branches can adopt similar connection methods, for example, dendritic branch br1 is connected to S21 through S25.

[0292] The connection manner between the synapses and the sensors of the fourth neuromorphic device may be consistent with the connection manner between the synapses and the sensors of the first neuromorphic device.

[0293] exist Figure 13 In the second detection device DD2 shown, similar to the first detection device, when an object moves in a first direction O1 (0° in this embodiment, i.e., horizontally to the right), an excitatory synapse on a dendritic branch of the first dendritic unit first receives a pulse signal from the corresponding sensor unit, activating a silent synapse. The silent synapse then receives a pulse signal from the corresponding sensor unit. An excitatory synapse on a dendritic branch of the fourth dendritic unit then receives a pulse signal from the corresponding sensor unit.

[0294] Since the sensor is composed of a plurality of sensor units linearly arranged along a first direction, movement of the object along the first direction will activate all sensors connected to the synapses on the corresponding dendrite units.

[0295] For example, in Figure 13 In the example of movement along the first direction O1 shown in FIG, the excitatory synapses ES on the dendritic branch br2 of the first neuromorphic device first receive pulse signals from their corresponding sensor units S31 to S34. This causes the dendritic branch br2 to meet the conditions for activating the silent synapse SS. The silent synapse then receives a pulse signal from its corresponding sensor unit S35. All synapses on the dendritic branch br2 receive the pulse signal.

[0296] The dendrite branch br2' of the fourth neuromorphic device also receives the pulse signal from the corresponding sensor units S31 to S35. All synapses of the dendrite branch br2' receive the pulse signal.

[0297] Due to the contribution of the silent synapse SS closest to the cell body, the current generated in the first neuromorphic transistor meets the first current threshold of the first current switch, and the first neuromorphic device is turned on. Similarly, due to the contribution of the synapse closest to the cell body, the fourth neuromorphic device is turned on.

[0298] As described above, all synapses on the same dendritic branch receive input from the sensor unit and direct it to the dendritic unit. Due to the nonlinear integration of synaptic input signals by dendritic branches described above, the superposition of multiple input signals from the same branch results in a stronger current in the neuromorphic device, and thus a wider pulse generated by the pulse generator in the neuromorphic device. Therefore, in the above scenario, the first neuromorphic device N1 outputs a first-directional output signal, including an excitation signal N1E with a stronger signal strength.

[0299] As described above for the detection device DD1, in the above situation, the second neuromorphic device is turned on and outputs a second output signal, including an excited output N2E, indicating that the object moves along the first direction; the third neuromorphic device is not turned on and has no output.

[0300] The inhibitory synapse of the fifth neuromorphic device receives the inhibitory output N1I from the first output signal, and the excitatory synapse receives the excitatory output N4E from the fourth output signal. However, due to the inhibitory synapse, the fifth neuromorphic device is not conducting and does not output the fifth output signal. Therefore, the sixth and seventh neuromorphic devices have no inputs or outputs.

[0301] Therefore, when the object moves in the first direction, the second neuromorphic device is turned on and outputs N2E, indicating that the object moves in the first direction; while the other neuromorphic devices are not turned on and have no output.

[0302] In the above embodiment, when the object moves along the second direction O2 (330° in this embodiment), some of the excitatory synapses on the first dendritic branch of the first neuromorphic device first receive the pulse signal from the corresponding sensor unit; some of the excitatory synapses on the second dendritic branch then receive the pulse signal from the corresponding sensor unit, activating the silent synapses, which then receive the pulse signal from the corresponding sensor unit.

[0303] Some of the excitatory synapses on the third dendritic branch of the fourth neuromorphic device first receive the pulse signal from the corresponding sensor unit; some of the excitatory synapses on the fourth dendritic branch then receive the pulse signal from the corresponding sensor unit.

[0304] For example, in Figure 13 In the example of movement along the second direction O2 shown in FIG, some excitatory synapses on dendritic branch br1 of the first neuromorphic device first receive pulse signals from corresponding sensor units S22 and S23. Some excitatory synapses on dendritic branch br2 then receive pulse signals from corresponding sensor unit S34. This causes dendritic branch br2 to meet the conditions for activation of silent synapses SS. The silent synapses then receive pulse signals from corresponding sensor unit S35.

[0305] Some of the excitatory synapses on the dendritic branch br1 ′ of the fourth neuromorphic device first receive pulse signals from the corresponding sensor units S22 and S23 ; some of the excitatory synapses on the dendritic branch br2 ′ receive pulse signals from the corresponding sensor units S34 and S35 .

[0306] Due to the contribution of the silent synapse SS closest to the cell body, the current generated in the first neuromorphic transistor meets the first current threshold of the first current switch, and the first neuromorphic device is turned on. Similarly, due to the contribution of the synapse closest to the cell body, the fourth neuromorphic device is turned on.

[0307] As described above, the sensor unit's input is dispersed across the different dendritic branches of the dendritic unit. Similarly, due to the nonlinear integration of synaptic input signals by the dendritic branches described above, the current generated in the neuromorphic device is relatively weak due to the relatively dispersed input pulse signal. Consequently, the pulse width of the pulses generated by the pulse generator in the neuromorphic device is relatively low. Therefore, in the above scenario, the first neuromorphic device N1 outputs a second-direction output signal, including an excitation signal N1E with a relatively weak signal strength. That is, the second-direction output signal has a lower strength than the first-direction output signal.

[0308] As described above for the detection device DD1, in the above situation, the third neuromorphic device is turned on and outputs a third output signal, including an excited output N3E, indicating that the object moves in the second direction; the second neuromorphic device is turned off and has no output.

[0309] Similarly, the inhibitory synapse of the fifth neuromorphic device receives the inhibitory output N1I from the first output signal, and the excitatory synapse receives the excitatory output N4E from the fourth output signal. However, due to the inhibitory synapse, the fifth neuromorphic device is not conducting and does not output the fifth output signal. Therefore, the sixth and seventh neuromorphic devices have no inputs or outputs.

[0310] Therefore, when the object moves in the second direction, the third neuromorphic device is turned on and outputs N3E, indicating that the object moves in the second direction; while the other neuromorphic devices are not turned on and have no output.

[0311] In the above embodiment, when the object moves in the third direction O3 (i.e., the direction opposite to the first direction, 180 degrees in this embodiment), the silent synapse on a dendritic branch of the first dendritic unit first receives the pulse signal from the corresponding sensor unit, and the excitatory synapse then receives the pulse signal from the corresponding sensor unit, and the silent synapse is inactive. The excitatory synapse on a dendritic branch of the fourth dendritic unit receives the pulse signal from the corresponding sensor unit.

[0312] For example, in Figure 13 In the example of movement along the third direction O3 shown in the figure, the silent synapse on the dendritic branch br2 of the first neuromorphic device first receives a pulse signal from the corresponding sensor unit S35. At this time, the silent synapse is not activated and cannot import the input into the first dendritic unit; the excitatory synapse ES then receives pulse signals from the corresponding sensor units S34 to S31 respectively.

[0313] The dendrite branch br2 ′ of the fourth neuromorphic device also receives the pulse signal from the units S35 to S31 of the corresponding sensor.

[0314] Due to the lack of contribution from the silent synapse SS closest to the soma, the output of the first soma is not greater than the first current threshold, the first current switch is not turned on, the first neuromorphic device is not turned on, and the first neuromorphic device does not output the first output signal.

[0315] Since the fourth dendritic unit still has the contribution of the synapse closest to the soma unit, the output of the fourth soma unit is greater than the fourth current threshold, the fourth current switch is turned on, and the fourth neuromorphic device is turned on.

[0316] In this case, all synapses on the same dendritic branch receive input from the sensor unit and direct it to the dendritic unit. Therefore, in this case, the fourth neuromorphic device N4 outputs a third directional output signal, including an excitation signal N4E with a stronger signal strength.

[0317] The fifth neuromorphic device N5 receives as input the excitation signal N4E from the third-directional output signal. In the absence of the inhibition signal N1I, the output of the fifth soma unit exceeds the fifth current threshold, the fifth current switch turns on, and the fifth neuromorphic device turns on and outputs a fifth output signal corresponding to the strength of the excitation signal N4E. The fifth output signal includes the excitation signal N5E.

[0318] The sixth neuromorphic device N6 receives the excitation signal N5E from the fifth output signal as input. When the excitation signal N5E is strong, the output of the sixth cell unit exceeds the sixth current threshold, the sixth current switch turns on, and the sixth neuromorphic device outputs a sixth output signal. This sixth output signal includes an inhibition output N6I and an excitation output N6E.

[0319] The seventh neuromorphic device N7 receives as input the excitatory output N5E of the fifth output signal and the inhibitory output N6I of the sixth output signal. As described above, the inhibitory synapse of the seventh neuromorphic device receives the inhibitory output N6I of the sixth output signal, causing the excitatory output N5E of the fifth output signal to be unable to conduct to the seventh neuromorphic device, and the seventh neuromorphic device does not output the seventh output signal.

[0320] Therefore, when the object moves along the third direction, the sixth neuromorphic device is turned on and outputs N6E, indicating that the object moves along the third direction; while the other neuromorphic devices are not turned on and have no output.

[0321] In the above embodiment, when the object moves along the fourth direction O4 (i.e., the direction opposite to the second direction, which is 150 degrees in this embodiment), the silent synapse on a dendritic branch of the first dendritic unit of the first neuromorphic device first receives the pulse signal from the corresponding sensor unit, and the excitatory synapse then receives the pulse signal from the corresponding sensor unit, and the silent synapse is inactivated.

[0322] Some of the excitatory synapses on the third dendritic branch of the fourth neuromorphic device first receive the pulse signal from the corresponding sensor unit; some of the excitatory synapses on the fourth dendritic branch then receive the pulse signal from the corresponding sensor unit.

[0323] For example, in the example of movement along the fourth direction O4 shown in FIG11 , the silent synapses on the dendritic branch br2 of the first neuromorphic device first receive a pulse signal from the corresponding sensor unit S35. At this time, the silent synapses are inactive and cannot transmit input to the first dendritic unit. Some of the excited synapses on the dendritic branch br2 then receive a pulse signal from the corresponding sensor unit S34. Some of the excited synapses on the dendritic branch br1 then receive pulse signals from the corresponding sensor units S23 and S22, respectively.

[0324] Some of the excited synapses on the third dendritic branch br2 ′ of the fourth neuromorphic device first receive pulse signals from corresponding sensor units S35 and S34 , and some of the excited synapses on the fourth dendritic branch br1 ′ then receive pulse signals from corresponding sensor units S23 and S22 .

[0325] Due to the lack of contribution from the silent synapse SS closest to the soma, the output of the first soma is not greater than the first current threshold, the first current switch is not turned on, the first neuromorphic device is not turned on, and no first output signal is output.

[0326] Since the fourth dendritic unit still has the contribution of the synapse closest to the soma unit, the output of the fourth soma unit is greater than the fourth current threshold, the fourth current switch is turned on, and the fourth neuromorphic device is turned on.

[0327] In the above case, the input of the sensor unit is dispersed to different dendrite branches of the dendrite unit. Therefore, in the above case, the fourth neuromorphic device N4 outputs a fourth directional output signal, in which the signal strength of the excitation signal N4E included is relatively weak.

[0328] The fifth neuromorphic device N5 receives the excitation signal N4E from the fourth direction output signal as input. In the absence of the inhibition signal N1I, the output of the fifth soma unit exceeds the fifth current threshold, the fifth current switch turns on, and the fifth neuromorphic device turns on and outputs a fifth output signal corresponding to the strength of the excitation signal N4E. The fifth output signal includes the excitation signal N5E.

[0329] The sixth neuromorphic device N6 receives the excitation signal N5E in the fifth output signal as input. When the excitation signal N5E is weak, the output of the sixth cell unit is less than the sixth current threshold, the sixth neuromorphic device is not turned on, and the sixth neuromorphic device does not output the sixth output signal.

[0330] The seventh neuromorphic device N7 receives only the excitation output N5E of the fifth output signal as input. Since there is no inhibitory synapse, the output of the seventh soma unit exceeds the seventh current threshold, the seventh current switch turns on, the seventh neuromorphic device turns on, and outputs the seventh output signal, which includes the excitation output N7E.

[0331] Therefore, when the object moves along the fourth direction, the seventh neuromorphic device is turned on and outputs N7E, indicating that the object moves along the fourth direction; while the other neuromorphic devices are not turned on and have no output.

[0332] It can be seen that the first detection device for the moving direction in the above embodiment can generate different outputs for different directions of the moving signal without a processor, thereby significantly reducing the power consumption of the system.

[0333] Building on the movement direction detection device described above, this disclosure also proposes a mapping method based on neural processing components (NPCs). This mapping method can transform a morphological diagram into an NPC network diagram, highlighting the topological logic of the detection device or system, and facilitating the analysis and understanding of functional mechanisms related to the connectivity of complex systems.

[0334] Figure 14 is a schematic diagram of a mapping method according to some embodiments of the present disclosure.

[0335] like Figure 14 As shown, Figure 14The neuromorphic device circuit C2 in (a) can be mapped as Figure 14 (b) NPC network diagram, where NPC nodes are grouped according to the cell-based paradigm, and each dotted box represents a neuromorphic device, including dendrites, cell bodies, and axons. Arrows between nodes indicate signal directions.

[0336] In the NPC network diagram, triangles with plus signs represent excitatory synapses, triangles with minus signs represent inhibitory synapses, and triangles with SS symbols represent silent synapses; each diamond represents a dendritic branch; circles represent cell body units; and squares represent axon units.

[0337] The present disclosure also proposes a hyperbolic mapping method, which can further organize the above-mentioned NPC network diagram and organize the NPCs into groups according to their functions.

[0338] Figure 15 is a schematic diagram of a hyperbolic mapping method according to some embodiments of the present disclosure.

[0339] like Figure 15 As shown, Figure 14 The NPC network graph in (b) is embedded in a hyperbolic space to obtain a hyperbolic embedded network graph. The above embedding is achieved, for example, by agglomerative embedding.

[0340] In a hyperbolic embedding network, nodes that are closer are more likely to be connected to each other, and the nodes can be arranged into functional communities in a hierarchical structure. In this case, the connectivity (e.g. Figure 15 The clustering is done based on different embedding analysis and different initial value and output analysis. As mentioned above, this clustering is done based on the function of NPC, which is different from Figure 14 The grouping of NPCs in (b) is different.

[0341] Figure 16 is a hyperbolic NPC network diagram according to some embodiments of the present disclosure.

[0342] like Figure 16 As shown in the hyperbolic NPC network diagram, NPCs are clustered according to their functions based on the topological connectivity of NPC nodes, thereby identifying different functional layers, such as Figure 16 The input layer, direction selection layer, and orientation selection layer are shown.

[0343] The input layer includes multiple synaptic units for receiving input from multiple sensors. The direction selection layer includes an inbound direction selection layer and an outbound direction selection layer. The inbound direction selection layer determines whether the direction is the first or second direction, while the outbound direction selection layer determines whether the direction is the third or fourth direction. The azimuth selection layer performs further determinations and includes an inbound azimuth selection layer and an outbound azimuth selection layer. The inbound azimuth selection layer determines whether the direction is the first or second direction, while the outbound azimuth selection layer determines whether the direction is the third or fourth direction.

[0344] Furthermore, the role of a single node becomes important when its centrality is high, where the centrality of a node is determined based on the number of other nodes connected to it. Figure 16 The centrality in is expressed as the size of the node.

[0345] The NPC network graph, hyperbolic embedded network graph, and hyperbolic NPC network graph obtained by the mapping method of the present disclosure are helpful in modifying the topological logic of the network, thereby further improving the analysis performance of the network.

[0346] Based on the above-described moving direction detection device, the present disclosure further proposes a moving direction detection system by analyzing and improving the above-mentioned mapping method.

[0347] Figure 17 is a block diagram illustrating a system for detecting a moving direction according to some embodiments of the present disclosure.

[0348] like Figure 17 As shown, the detection system Sys1 includes a first detection device Dev1, a second detection device Dev2, and a third detection device Dev3.

[0349] The first detection device Dev1 is configured to detect movement of an object in the horizontal direction, in a direction that forms an angle of 150° or 330° with respect to the horizontal right direction. In the first detection device Dev1, the first direction is the horizontal right direction, i.e., 0°; the first angle is 30°; the second direction forms an angle of 330° with respect to the horizontal right direction; the third direction is opposite to the first direction, that is, the horizontal left direction; and the fourth direction is opposite to the second direction, that is, forming an angle of 150° with respect to the horizontal right direction.

[0350] The second detection device Dev2 is configured to detect movement of an object in a direction that forms an angle of 60°, 30°, 240°, or 210° with the horizontal right direction. In the second detection device Dev2, a vector of a first direction forms an angle of 60° with a vector of a horizontal right direction, a first angle is 30°, a second direction forms an angle of 30° with the horizontal right direction, a third direction is opposite to the first direction and forms an angle of 240° with the horizontal right direction, and a fourth direction is opposite to the second direction and forms an angle of 210° with the horizontal right direction.

[0351] The third detection device Dev3 is configured to detect movement of an object in a vertical direction, a direction at an angle of 120° to the horizontal right direction, or a direction at an angle of 300° to the horizontal right direction. In the third detection device Dev3, the first angle is 30°, the first direction is at an angle of 120° to the horizontal right direction, the second direction is a vertical upward direction at an angle of 90° to the horizontal right direction, the third direction is opposite to the first direction and at an angle of 300° to the horizontal right direction, and the fourth direction is opposite to the second direction and is a vertical downward direction.

[0352] Figure 18 Schematic diagram showing the connection relationship between different detection devices and sensors in a moving direction detection system according to some embodiments of the present disclosure.

[0353] like Figure 18 As shown, the first detection device Dev1, the second detection device Dev2, and the third detection device Dev3 are connected to the sensor array, which is composed of the same multiple sensors, in different ways. The connection method of detection device Dev1 to the sensor array is the same as that of detection device DD2 described above; the connection method of detection device Dev2 to the sensor array is the connection method of detection device Dev1 rotated 120 degrees clockwise around the center of the sensor array; the connection method of detection device Dev3 to the sensor array is the connection method of detection device Dev1 rotated 120 degrees counterclockwise around the center of the sensor array.

[0354] The above connection method can reduce the number of sensors required by the detection system Sys1 and further improve the processing efficiency of the system.

[0355] It can be seen that the above three devices working together can achieve 360° surround detection in a two-dimensional plane.

[0356] Figure 19 1 is a schematic diagram showing output signals of a moving direction detection system for different exemplary input signals according to some embodiments of the present disclosure.

[0357] Figure 19 The horizontal axis represents time, and the vertical axis represents the amplitude of each input pulse. For example, the amplitude of each input pulse is 9V, and different output signals have different pulse widths. For example, Figure 19As shown, when the input signal is that the object moves in the 0° direction, that is, the first direction of the first detection device Dev1, the system outputs N2E, indicating that the object moves in the 0° direction, and the other neuromorphic devices have no output; when the input signal is that the object moves in the 30° direction, that is, the second direction of the second detection device Dev2, the system outputs N3E', indicating that the object moves in the 30° direction; when the input signal is that the object moves in the 270° direction, that is, the third direction of the third detection device Dev3, the system outputs N6E'', indicating that the object moves in the 270° direction.

[0358] As can be seen, the detection system Sys1 can generate corresponding outputs for different directions of a two-dimensional motion signal without a processor, significantly reducing power consumption. Furthermore, the detection system does not require full-node-to-full-node connections, and has sparse connectivity, reducing its complexity.

[0359] Figure 20 is a block diagram illustrating a system for detecting a moving direction according to other embodiments of the present disclosure.

[0360] like Figure 20 As shown, the moving direction detection system Sys2 further includes a processor Pro1. The processor Pro1 is configured to determine the moving direction of the object based on the output signal of the first detection device Dev1, the output signal of the second detection device Dev2, and the output signal of the third detection device Dev3. For example, the processor can determine the moving direction of the object based on the output signal of the first detection device Dev1, the output signal of the second detection device Dev2, and the output signal of the third detection device Dev3. Figure 19 The corresponding relationship between the input signal and the output signal is used to determine the moving direction of the object in the input signal. If the moving direction is different from the above direction, the system will not produce a valid output.

[0361] In addition, the system's sensitivity to direction detection can be improved by setting appropriate current thresholds, voltage thresholds, and using an appropriate number and type of sensor arrays.

[0362] and Figures 14 to 16 Similarly, the above mapping method disclosed in the present invention can also be used in a moving direction detection system, such as Figure 21 shown. Figure 21 is a cluster diagram illustrating a hyperbolic NPC map of a movement direction detection system according to some embodiments of the present disclosure.

[0363] In the clustering diagram, each five-pointed star represents a cluster of NPCs according to their functions, corresponding to the input layer, direction selection layer, and orientation selection layer mentioned above, which can clearly reflect the design topology logic of the detection system.

[0364] In the above embodiment, the processor P1 included in the detection system Sys2 is configured to use a morphological network to represent each neuromorphic device in the detection system; map the morphological network of each neuromorphic device to a topological network of an equivalent neural processing element; embed the topological network into a hyperbolic space to obtain a hyperbolic embedding network; and use the hyperbolic embedding network to identify different functional layers according to the topological connectivity of the nodes of the neural processing element, wherein the functional layers include a mapping layer, a direction selection layer, and an orientation selection layer.

[0365] The NPC network graph, hyperbolic embedded network graph, and hyperbolic NPC network graph obtained by the mapping method of the present disclosure are helpful in modifying the topological logic of the network, thereby further improving the analysis performance of the network.

[0366] In summary, the motion direction detection device and detection system according to the embodiments of the present disclosure can generate corresponding outputs for different directions of a two-dimensional motion signal without a processor, significantly reducing power consumption. Furthermore, the detection device and detection system do not require full-node-to-full-node connections, exhibiting sparse connectivity and reducing complexity.

[0367] Thus far, various embodiments of the present disclosure have been described in detail. To avoid obscuring the concept of the present disclosure, some details known in the art have not been described. Based on the above description, those skilled in the art can fully understand how to implement the technical solutions disclosed herein.

[0368] Although some specific embodiments of the present disclosure have been described in detail through examples, those skilled in the art will understand that the above examples are for illustration only and are not intended to limit the scope of the present disclosure. Those skilled in the art will understand that the above embodiments may be modified or some technical features may be replaced with equivalents without departing from the scope and spirit of the present disclosure. The scope of the present disclosure is defined by the appended claims.

Claims

1. A device for detecting a moving direction, comprising: a plurality of sensors for sensing movement of an object, each sensor comprising a plurality of sensor units linearly arranged along a first direction; A first neuromorphic device includes a first synaptic unit, a first dendritic unit, and a first soma unit, wherein the first synaptic unit includes a silent synapse and an excitatory synapse, wherein the silent synapse is activated only when the input signal strength of the excitatory synapse is greater than an activation threshold; the first dendritic unit includes a plurality of dendritic branches, wherein each dendritic branch of at least some of the dendritic branches is connected to a silent synapse and a plurality of excitatory synapses, wherein the silent synapses are located closer to the first soma unit than the excitatory synapses, wherein the plurality of synapses on each dendritic branch are connected to a plurality of sensor units in different sensors in a one-to-one correspondence, wherein the silent synapse and its adjacent excitatory synapses are connected to different sensors, and the first soma unit is configured to process an output of the first dendritic unit to generate a first output signal; a second neuromorphic device comprising a second synaptic unit, a second dendritic unit, and a second soma unit, wherein the second synaptic unit comprises an excitatory synapse for receiving the first output signal, the second dendritic unit comprises at least one dendritic branch, each dendritic branch being connected to the excitatory synapse, and the second soma unit is configured to process an output of the second dendritic unit to generate a second output signal, wherein the second output signal is configured to indicate that the object moves in a first direction; A third neuromorphic device includes a third synaptic unit, a third dendritic unit, and a third cell body unit, wherein the third synaptic unit includes an inhibitory synapse and an excitatory synapse, the inhibitory synapse is used to receive the second output signal, and the excitatory synapse is used to receive the first output signal. The third dendritic unit includes at least one dendritic branch, each dendritic branch is connected to an inhibitory synapse and an excitatory synapse, the inhibitory synapse is located closer to the third cell body unit than the excitatory synapse, and the third cell body unit is used to process the output of the third dendritic unit to generate a third output signal, wherein the third output signal is used to indicate that the object moves along a second direction, and the vector of the first direction forms a first angle with the vector of the second direction.

2. The detection device according to claim 1, wherein: The first neuromorphic device further includes a first axon unit configured to convert an output of the first soma unit into a first voltage pulse; The second neuromorphic device further includes a second axon unit configured to convert an output of the second soma unit into a second voltage pulse; The third neuromorphic device further includes a third axon unit configured to convert an output of the third soma unit into a third voltage pulse.

3. The detection device according to claim 2, wherein: The first axon unit includes a first current switch, wherein the first current switch is configured to be turned on when the output of the first soma unit is greater than a first current threshold; The second axon unit includes a second current switch, and the second current switch is configured to be turned on when the output of the second soma unit is greater than a second current threshold; The third axon unit includes a third current switch configured to be turned on when the output of the third soma unit is greater than a third current threshold, wherein the third current threshold is less than the second current threshold.

4. The detection device according to claim 3, wherein: The third axon unit further comprises: The first voltage switch is configured to be turned off when the intensity of the first output signal is greater than a first voltage threshold.

5. The detection device according to claim 4, wherein: The first output signal includes a first direction output signal or a second direction output signal. When the object moves along a first direction: An excitatory synapse on a dendritic branch of the first dendritic unit first receives a pulse signal from a corresponding sensor unit, thereby activating a silent synapse, and the silent synapse then receives a pulse signal from the corresponding sensor unit; The output of the first soma unit is greater than the first current threshold, the first current switch is turned on, and the first neuromorphic device outputs the first direction output signal; The output of the second cell unit is greater than a second current threshold, the second current switch is turned on, and the second neuromorphic device outputs the second output signal; The inhibitory synapse receives the second output signal, so that the first direction output signal cannot conduct the third neuromorphic device, and the third neuromorphic device does not output the third output signal.

6. The detection device according to claim 5, wherein: The at least part of the dendrite branches includes a first dendrite branch and a second dendrite branch, and when the object moves along the second direction: Some of the excitatory synapses on the first dendritic branch first receive pulse signals from corresponding sensor units; Some of the excitatory synapses on the second dendritic branch then receive the pulse signal from the corresponding sensor unit, activating the silent synapses, and the silent synapses then receive the pulse signal from the corresponding sensor unit; The output of the first cell unit is greater than the first current threshold, the first current switch is turned on, the first neuromorphic device outputs the second direction output signal, and the intensity of the second direction output signal is less than the first direction output signal; The output of the second cell unit is not greater than the second current threshold, the second current switch is not turned on, and the second neuromorphic device does not output the second output signal; The output of the third cell unit is greater than the third current threshold, the third current switch is turned on, and the third neuromorphic device outputs the third output signal.

7. The detection device according to claim 1, wherein: The first dendritic unit further includes dendritic branches connected only to excitatory synapses.

8. The detection device according to claim 3, further comprising: a fourth neuromorphic device comprising a fourth synaptic unit, a fourth dendritic unit, and a fourth soma unit, wherein the fourth synaptic unit comprises an excitatory synapse, the fourth dendritic unit comprises a plurality of dendritic branches, the plurality of excitatory synapses being connected to each dendritic branch of at least some of the dendritic branches, the plurality of synapses on a dendritic branch being connected in a one-to-one correspondence to a plurality of sensor units in different sensors, and the fourth soma unit being configured to process an output of the fourth dendritic unit to generate a fourth output signal; a fifth neuromorphic device comprising a fifth synaptic unit, a fifth dendritic unit, and a fifth soma unit, wherein the fifth synaptic unit comprises an inhibitory synapse and an excitatory synapse, the inhibitory synapse being configured to receive the first output signal, the excitatory synapse being configured to receive the fourth output signal, the fifth dendritic unit comprising at least one dendritic branch, each dendritic branch being connected to an inhibitory synapse and an excitatory synapse, the inhibitory synapse being located closer to the fifth soma unit than the excitatory synapse, and the fifth soma unit being configured to process the output of the fifth dendritic unit to generate a fifth output signal; a sixth neuromorphic device, comprising a sixth synaptic unit, a sixth dendritic unit, and a sixth soma unit, wherein the sixth synaptic unit includes an excitatory synapse for receiving the fifth output signal, the sixth dendritic unit includes at least one dendritic branch, each dendritic branch connected to the excitatory synapse, and the sixth soma unit is configured to process an output of the sixth dendritic unit to generate a sixth output signal, wherein the sixth output signal is configured to indicate that the object moves in a third direction, the third direction being opposite to the first direction; A seventh neuromorphic device includes a seventh synaptic unit, a seventh dendritic unit, and a seventh soma unit, wherein the seventh synaptic unit includes an inhibitory synapse and an excitatory synapse, the inhibitory synapse is used to receive the sixth output signal, and the excitatory synapse is used to receive the fifth output signal. The seventh dendritic unit includes at least one dendritic branch, each dendritic branch is connected to an inhibitory synapse and an excitatory synapse, the inhibitory synapse is located closer to the seventh soma unit than the excitatory synapse, and the seventh soma unit is used to process the output of the seventh dendritic unit to generate a seventh output signal, wherein the seventh output signal is used to indicate that the object moves along a fourth direction, which is opposite to the second direction.

9. The detection device according to claim 8, wherein: The fourth neuromorphic device further includes a fourth axon unit, the fourth axon unit being configured to convert an output of the fourth soma unit into a fourth voltage pulse; The fifth neuromorphic device further includes a fifth axon unit, the fifth axon unit being configured to convert an output of the fifth soma unit into a fifth voltage pulse; The sixth neuromorphic device further includes a sixth axon unit, the sixth axon unit being configured to convert an output of the sixth soma unit into a sixth voltage pulse; The seventh neuromorphic device further includes a seventh axon unit configured to convert an output of the seventh soma unit into a seventh voltage pulse.

10. The detection device according to claim 9, wherein: The fourth axon unit includes a fourth current switch, and the fourth current switch is configured to be turned on when the output of the fourth soma unit is greater than a fourth current threshold; The fifth axon unit includes a fifth current switch, wherein the fifth current switch is configured to be turned on when the output of the fifth soma unit is greater than a fifth current threshold; The sixth axon unit includes a sixth current switch, and the sixth current switch is configured to be turned on when the output of the sixth soma unit is greater than a sixth current threshold; The seventh axon unit includes a seventh current switch, and the seventh current switch is configured to be turned on when the output of the seventh soma unit is greater than a seventh current threshold, wherein the seventh current threshold is less than the seventh current threshold.

11. The detection device according to claim 10, wherein: The seventh neuromorphic device further includes: a second voltage switch configured to be turned off when the intensity of the fifth output signal is greater than a second voltage threshold; and / or The third voltage switch is configured to be turned off when the intensity of the sixth output signal is greater than a third voltage threshold.

12. The detection device according to claim 11, wherein: The fourth output signal includes a third direction output signal or a fourth direction output signal. When the object moves along the third direction: On a dendritic branch of the first dendritic unit, a silent synapse first receives a pulse signal from the corresponding sensor unit, and an excitatory synapse later receives a pulse signal from the corresponding sensor unit, the silent synapse is inactivated, the output of the first soma unit is not greater than the first current threshold, the first current switch is not turned on, and the first neuromorphic device does not output the first output signal; An excitatory synapse on a dendritic branch of the fourth dendritic unit successively receives pulse signals from the corresponding sensor unit, the output of the fourth cell unit is greater than the fourth current threshold, the fourth current switch is turned on, and the fourth neuromorphic device outputs the third direction output signal; The output of the fifth cell unit is greater than the fifth current threshold, the fifth current switch is turned on, and the fifth neuromorphic device outputs a fifth output signal corresponding to the third direction output signal; The output of the sixth cell unit is greater than the sixth current threshold, the sixth current switch is turned on, and the sixth neuromorphic device outputs the sixth output signal; The inhibitory synapse of the seventh neuromorphic device receives the sixth output signal, so that the fifth output signal cannot conduct the seventh neuromorphic device, and the seventh neuromorphic device does not output the seventh output signal.

13. The detection device according to claim 12, wherein: At least part of the dendrite branches of the fourth dendrite unit include the third dendrite branch and the fourth dendrite branch, and when the object moves along the fourth direction: On a dendritic branch of the first dendritic unit, a silent synapse first receives a pulse signal from the corresponding sensor unit, and an excitatory synapse later receives a pulse signal from the corresponding sensor unit, the silent synapse is inactivated, the output of the first soma unit is not greater than the first current threshold, the first current switch is not turned on, and the first neuromorphic device does not output the first output signal; Some of the excitatory synapses on the third dendritic branch first receive the pulse signal from the corresponding sensor unit; Some of the excitatory synapses on the fourth dendritic branch then receive pulse signals from corresponding sensor units; The output of the fourth cell unit is greater than the fourth current threshold, the fourth current switch is turned on, and the fourth neuromorphic device outputs the fourth direction output signal, wherein the intensity of the fourth direction output signal is less than the third direction output signal; The output of the fifth cell unit is greater than the fifth current threshold, the fifth current switch is turned on, and the fifth neuromorphic device outputs a fifth output signal corresponding to the fourth direction output signal; The output of the sixth cell unit is not greater than the sixth current threshold, the sixth current switch is not turned on, and the sixth neuromorphic device does not output the sixth output signal; The output of the seventh cell unit is greater than the seventh current threshold, the seventh current switch is turned on, and the seventh neuromorphic device outputs the seventh output signal.

14. The detection device according to claim 8, wherein: In the case where the object moves in the first direction or the second direction: The first neuromorphic device is turned on to output the first output signal; When the inhibitory synapse of the fifth neuromorphic device receives the first output signal, the fifth neuromorphic device turns off and does not output the fifth output signal.

15. The detection device according to claim 8, wherein: In the first dendritic unit, a plurality of synapses on each dendritic branch are interlacedly connected to a plurality of sensor units in an adjacent sensor; and / or In the fourth dendrite unit, a plurality of synapses on each dendrite branch are connected to a plurality of sensor units in an adjacent sensor in an interlaced manner.

16. The detection device according to claim 15, wherein the sensor unit is in the shape of a regular hexagon, the multiple sensors are an array of sensors in which multiple sensor units are densely packed in a regular hexagonal plane, and the center line connecting the multiple sensor units in the same sensor is parallel to the first direction.

17. The detection device according to claim 16, wherein: The first direction is a horizontal rightward direction, and the first angle is 30°.

18. The detection device according to claim 16, wherein: The vector in the first direction forms an angle of 60° with the vector in the horizontal right direction, and the first angle is 30°.

19. The detection device according to claim 16, wherein: The second direction is a vertical upward direction, and the first angle is 30°.

20. A moving direction detection system, comprising: The first detection device comprises the detection device according to claim 17, configured to detect movement of the object in a horizontal direction, a direction forming an angle of 150° or 330° to the right of the horizontal direction; a second detection device comprising the detection device according to claim 18, configured to detect movement of the object in a direction forming an angle of 30°, 60°, 210°, or 240° with respect to a horizontal rightward direction; The third detection device comprises the detection device according to claim 19, and is configured to detect movement of the object in a direction forming an angle of 90°, 120°, 270°, or 300° with respect to the horizontal rightward direction.

21. The detection system according to claim 20, further comprising: The processor is configured to determine the moving direction of the object based on the output signal of the first detection device, the output signal of the second detection device, and the output signal of the third detection device.

22. The detection system according to claim 21, wherein: The processor is configured to: Using a morphological network to represent each neuromorphic device in the detection system; Mapping the morphological network of each neuromorphic device to a topological network of equivalent neural processing elements; Embedding the topological network into a hyperbolic space to obtain a hyperbolic embedded network; The hyperbolic embedding network is used to identify different functional layers according to the topological connectivity of the nodes of the neural processing element. The functional layers include a mapping layer, a direction selection layer, and an orientation selection layer.

Images