Neuron circuit of artificial neural network, having function of compensating for threshold voltage variations

Abstract

Disclosed are a neuron circuit of an artificial neural network and a method for operating the neuron circuit. The neuron circuit includes: a threshold voltage storage unit for storing a threshold voltage of an output transistor in a threshold voltage capacitor; a threshold voltage compensation unit for compensating the potential of the threshold voltage capacitor for the intrinsic threshold voltage of the output transistor; and an output unit for transmitting a constant output value to the next synapse by using the compensated threshold voltage.

Description

Artificial neural network neuron circuit with threshold voltage fluctuation compensation function

[0001] The present disclosure relates to an artificial neural network neuron circuit and a method for driving the same, and more specifically, to an artificial neural network neuron circuit that compensates for threshold voltage fluctuations due to element mutations and external environments and a method for driving the same.

[0002] Artificial neural networks (ANNs) are computer algorithms that mimic the neural networks of the human brain. ANNs are a key concept used in machine learning and artificial intelligence. An ANN consists of multiple neurons that accept input data, multiply weights, and apply activation functions to generate output.

[0003] In artificial neural networks (ANNs), driver transistors are actual hardware elements involved in the integration and output of information to membrane capacitors. In other words, driver transistors are a key hardware component in the operation of artificial neurons.

[0004] However, in artificial neural networks (ANNs), fluctuations in the threshold voltage of driving transistors, a key component of the artificial neuron circuit, can impact the performance and stability of the artificial neuron, leading to a decline in inference accuracy. Fluctuations in the transistor threshold voltage, or undesired changes, can make weight adjustments difficult and prevent accurate processing of input data. Key factors that cause fluctuations in the transistor threshold voltage in these artificial neuron circuits include manufacturing process variations, temperature fluctuations, and voltage fluctuations.

[0005] Much research is being conducted to address the issue of threshold voltage fluctuations in transistors in artificial neural networks (ANNs). Conventional techniques for addressing this issue include methods that compensate only for fluctuations due to external environmental factors, such as temperature and time. Other techniques compensate only for fluctuations due to device variations. Research is needed to develop methods that compensate for all of these variations.

[0006] The present disclosure proposes an artificial neural network neuron circuit and driving method that performs accurate activation function operation by compensating for threshold voltage fluctuations of a device.

[0007] The present disclosure proposes an artificial neural network neuron circuit and driving method capable of compensating for both threshold voltage fluctuations due to external environmental changes and threshold voltage fluctuations due to device mutations.

[0008] The purpose of the present disclosure is not limited to the purposes mentioned above, and other purposes not mentioned will be clearly understood by those skilled in the art from the description below.

[0009] An artificial neural network neuron circuit according to one aspect of the present invention; the threshold voltage of the output transistor is determined by a threshold voltage capacitor (C vth ) and a threshold voltage storage unit; and the threshold voltage capacitor (C vth ) includes a threshold voltage compensation unit that compensates for the potential of the output transistor with the unique threshold voltage; and an output unit that transmits a constant output value to the next synapse through the compensated threshold voltage.

[0010] According to one or more embodiments of the present invention, the neuron circuit can provide a constant output value by compensating for both threshold voltage fluctuations due to element mutations and changes in the external environment.

[0011] According to one or more embodiments of the present invention, a storage capacitor (C) is provided to receive the output value of a previous synapse mem) further comprising a synaptic storage unit storing the threshold voltage capacitor (C vth ) and the above storage capacitor (C mem ) may be serially connected with a node in between where the output value of the previous synapse is input.

[0012] According to one or more embodiments of the present invention, the threshold voltage compensation unit comprises a source of the output transistor and the threshold voltage capacitor (C vth ) may include a fifth transistor (T5), a second transistor (T2), and a third transistor (T3) formed between the first node connected to the first electrode.

[0013] According to one or more embodiments of the present invention, the synaptic storage unit comprises the storage capacitor (C mem ) may include a fourth transistor (T4) connected in parallel.

[0014] According to one or more embodiments of the present invention, the output unit comprises the storage capacitor (C mem ) can pass an output value corresponding to the stored value.

[0015] According to one or more embodiments of the present invention, the threshold voltage compensation unit comprises the threshold voltage capacitor (C vth ) may further include a transistor formed between a second node connected to the other electrode and the gate of the output transistor.

[0016] According to one or more embodiments of the present invention, the threshold voltage capacitor (C vth ) may include a reference voltage terminal provided at a second node connected to the other electrode, and a third transistor (T3) located between the second node and the reference voltage terminal.

[0017] According to one or more embodiments of the present invention, a sixth transistor (T6) may be included for connecting an intermediate base voltage to a source of the output transistor.

[0018] A neuromorphic device according to another aspect of the present invention; is a neuromorphic computing device including an array structure in which a plurality of synaptic elements are connected to each other, and includes an artificial neural network neuron circuit that receives an output value of a previous synaptic element among the synaptic elements, compensates for a threshold voltage variation, and transmits the output value to a next synapse, and the artificial neural network neuron circuit may be an artificial neural network neuron circuit according to one aspect of the present invention.

[0019] A method for driving an artificial neural network neuron circuit having a compensation function for threshold voltage fluctuation according to one aspect of the present invention; is a threshold voltage capacitor (C) for driving a gate of an output transistor in an artificial neural network neuron circuit. vth ) an initial value setting step for storing the initial threshold voltage; a threshold voltage compensation step for detecting the inherent threshold voltage of the output transistor and compensating the voltage of the initial value to the inherent threshold voltage; a storage capacitor (C mem ) to initialize the potential of the storage capacitor (C) to 0 mem ) initialization step; integrate the input value from the synapse and store it in the storage capacitor (C mem ) storage step; and driving the output transistor based on the threshold voltage compensated in the threshold voltage compensation step to store the storage capacitor (C mem ) may include an output stage that transmits an output value corresponding to the charge;

[0020] According to one or more embodiments of the present invention, the initial value setting step comprises: vth ) and the reference voltage applied to one electrode of the storage capacitor (C mem ) by the base voltage applied to one electrode of the storage capacitor (C mem ) and the above threshold voltage capacitor (C vth ) may be a step to set the initial value to an adjustable range.

[0021] According to one or more embodiments of the present invention, the output stage comprises a threshold voltage capacitor (C vth ) by driving the gate of the output transistor by the compensated threshold voltage of the storage capacitor (C mem ) may be a step in which the output value according to the charge is transmitted to the next synapse.

[0022] According to the present invention described above, there is an effect of compensating for both element mutations and threshold voltage fluctuations due to external environmental changes in an artificial neural network neuron circuit.

[0023] Furthermore, the present invention configures a display compensation circuit using an analog-based neuron circuit, enabling dynamic compensation of the threshold voltage of transistors within the circuit. Furthermore, the circuit proposed in the present invention enables real-time compensation for both element variations and threshold voltage fluctuations, effectively compensating for variations occurring during operation.

[0024] When using the circuit proposed in the present invention, all neurons can compensate for threshold voltage variations occurring in the external environment or process, so that uniform activation function results can always be transmitted to the hidden layer, thereby maintaining constant inference accuracy.

[0025] The concept of the present invention can be applied to systems such as artificial neural networks, spiking neural networks, and circuits or sensors that require a specific result value to be maintained at a constant level.

[0026] FIG. 1 is a circuit diagram of an artificial neural network neuron circuit according to one embodiment of the present invention.

[0027] FIG. 2 is a circuit diagram of an initial value setting step of an artificial neural network neuron circuit according to one embodiment of the present invention and a signal timing diagram in the initial value setting step.

[0028] FIG. 3 is a circuit diagram of a threshold voltage compensation step of an artificial neural network neuron circuit according to one embodiment of the present invention and a signal timing diagram in the threshold voltage compensation step.

[0029] FIG. 4 is a circuit diagram of a storage capacitor (Cmem) initialization step of an artificial neural network neuron circuit according to one embodiment of the present invention and a signal timing diagram in the storage capacitor (Cmem) initialization step.

[0030] FIG. 5 is a circuit diagram of a storage step of an artificial neural network neuron circuit according to one embodiment of the present invention and a signal timing diagram in the storage step.

[0031] FIG. 6 is a circuit diagram of an output stage of an artificial neural network neuron circuit according to one embodiment of the present invention and a signal timing diagram at the output stage.

[0032] The embodiments of this disclosure are provided for the purpose of illustrating the technical concepts of this disclosure. The scope of rights under this disclosure is not limited to the embodiments presented below or the specific descriptions of these embodiments.

[0033] All technical and scientific terms used in this disclosure, unless otherwise defined, have the meanings commonly understood by those of ordinary skill in the art to which this disclosure pertains. All terms used in this disclosure have been selected for the purpose of more clearly explaining this disclosure and are not intended to limit the scope of rights under this disclosure.

[0034] Expressions such as “including,” “comprising,” “having,” and the like used in this disclosure should be understood as open-ended terms that imply the possibility of including other embodiments, unless otherwise stated in the phrase or sentence in which the expression is included.

[0035] The singular forms described in this disclosure may include plural meanings unless otherwise stated, and the same applies to the singular forms described in the claims.

[0036] Hereinafter, embodiments of the present disclosure will be described with reference to the attached drawings. Furthermore, in the description of the embodiments below, duplicate descriptions of identical or corresponding components may be omitted. However, even if descriptions of components are omitted, this does not mean that such components are not included in any embodiment.

[0037] Hereinafter, embodiments of the present invention will be described with reference to the attached drawings. In this process, the thickness of lines and the sizes of components depicted in the drawings may be exaggerated for clarity and convenience. Furthermore, in the description of the embodiments below, redundant descriptions of identical or corresponding components may be omitted. However, omission of a description of a component does not imply that such component is not included in any embodiment.

[0038] In addition, the examples below do not limit the scope of the present invention, but are merely exemplary of the components presented in the claims of the present invention, and examples that include components that are included in the technical idea throughout the specification of the present invention and can be substituted as equivalents in the components of the claims may be included in the scope of the present invention.

[0039] The present invention relates to an artificial neural network neuron circuit having a compensation function for threshold voltage fluctuation.

[0040] In the case of the compensation circuits of conventional and similar artificial neural networks, it was often possible to compensate for only one of the variations due to the process and the variations due to the external environment.

[0041] However, the artificial neural network neuron circuit according to the present invention can compensate for changes in the threshold voltage of the device that occur due to deterioration of device characteristics over time, temperature, and external environment during the process or by using a threshold voltage capacitor that stores the threshold voltage value of the output transistor, thereby maintaining the threshold voltage value at a constant value, and can sense the threshold voltage value of the output transistor, which plays the most important role in performing the activation function operation through the compensation step.

[0042] Furthermore, the neuron circuit according to the present invention can set the threshold voltage compensation cycle to a desired cycle through signal regulation. Since circuit operation can be controlled through external signals, controllability, such as changing the activation function operating point, can be increased.

[0043] In addition, according to one embodiment of the neuron circuit according to the present invention, the number of compensations per input data can be optimized and controlled according to the size of the capacitor or the manufactured circuit, as well as the method of compensation for each input.

[0044] In addition, according to one embodiment of the neuron circuit according to the present invention, since it can be composed of an n-type transistor using a-IGZO (Amorphous Indium Gallium Zinc Oxide), the circuit can be manufactured using a low-temperature process. In addition, due to the IGZO material characteristic of very low leakage current, the standby power consumption of the neuron can be greatly reduced.

[0045] Furthermore, in one embodiment of the present invention, the structure of a display pixel compensation circuit is introduced into an analog-based neuron circuit, enabling dynamic compensation of the threshold voltage of transistors within the circuit. This allows dynamic adjustment of the compensated threshold voltage value based on voltage changes depending on the situation.

[0046] The analog-based neural circuit of the present invention is suitable for processing continuous values, making it ideal for precisely calibrating pixel threshold voltages. Furthermore, its high precision allows for real-time data processing and real-time pixel compensation.

[0047] Hereinafter, embodiments of the present invention will be described in more detail based on the circuit diagrams illustrated in FIGS. 1 to 6.

[0048] Below, each component is described in detail with reference to FIGS. 1 to 6, but the circuit diagrams shown below are merely examples, and can be replaced with any other structure within the scope of the technical idea of ​​the present invention.

[0049] As illustrated in FIG. 1, an artificial neural network neuron circuit having a threshold voltage fluctuation compensation function according to one embodiment of the present invention may include a synaptic storage unit (100), an output unit (200), a threshold voltage storage unit (300), and a threshold voltage compensation unit (400).

[0050] Here, the synaptic accumulation unit (100) is a storage capacitor (C) that integrates and stores the synaptic output value. mem ) and the output section (200) includes the storage capacitor (C mem ) includes an output transistor (T1) that transmits an output value corresponding to the potential of the charge accumulated in the output transistor (T1), and the threshold voltage storage unit (300) includes a threshold voltage capacitor (C) that stores the gate threshold voltage (potential of the threshold voltage) for driving the output transistor (T1). vth ) may be included.

[0051] The threshold voltage compensation unit (400) detects the unique threshold voltage of the output transistor (T1) and compensates the threshold voltage capacitor (C) vth ) can be compensated to match the potential (voltage) of the initial charge accumulated in the above-mentioned unique threshold voltage.

[0052] The above compensation is the threshold voltage capacitor (Cvth ) can be achieved by a fifth transistor (T5) connecting an electrical path between one electrode (first node) of the transistor and the source of the output transistor (T1).

[0053] The above threshold voltage capacitor (C vth ) can be connected to a first node, and the first node can mean a lower node connected to a threshold voltage capacitor on the circuit diagram presented in Fig. 1. The threshold voltage capacitor (C vth ) can be connected to a second node, which is the upper node connected to the threshold voltage capacitor in the circuit diagram presented in Fig. 1.

[0054] In addition, the storage capacitor (C mem ) can be connected to the opposite side of the threshold voltage capacitor of the first node. The storage capacitor (C mem ) of the other electrode is the base voltage terminal (V SS ) can be connected to.

[0055] The synaptic output value mentioned above may be a value received from a previous neuron circuit or a value received from the outside.

[0056] The above components are explained in detail as follows:

[0057] The above synaptic accumulation unit (100) is the storage capacitor (C mem ) and the above storage capacitor (C mem ) by switching the two electrodes to the same potential to store the capacitor (C mem ) may include a fourth transistor (T4) that initializes the transistor.

[0058] The above output unit (200) may include the above output transistor (T1).

[0059] The above threshold voltage storage unit (300) is the threshold voltage capacitor (C vth ) may be included.

[0060] The above threshold voltage compensation unit (400) is a storage capacitor (C) that stores the threshold voltage. mem ) may include switching transistors that control the voltage at both ends. In the embodiment of FIG. 1, the switching transistors may be the second transistor (T2), the third transistor (T3), and the fifth transistor (T5).

[0061] The above threshold voltage compensation unit comprises the threshold voltage capacitor (C vth ) may include a second transistor (T2) located between the second node connected to the other electrode of the second transistor (T1) and the gate of the output transistor (T1), and a third transistor (T3) connected to the opposite side of the second transistor of the second node. In addition, the threshold voltage capacitor (C vth ) may include a first node connected to one electrode of the output transistor (T1) and a fifth transistor (T5) located between the node located between the output transistor (T1) of the output unit and the output switch.

[0062] The above second transistor can perform the role of switching the gate voltage to the output transistor (T1) of the output section.

[0063] In addition, the threshold voltage compensation unit (400) can compensate the initial threshold voltage to the unique threshold voltage of the output transistor (T1) when a turn-on voltage higher than the threshold voltage is applied to the gate of the output transistor (T1) and the drain-source channel is opened. At this time, the threshold voltage capacitor (C vth ) is connected to the source of the output transistor (T1), the channel of the fifth transistor (T5) is opened, and the source of the output transistor (T1) and the threshold voltage capacitor (C vth ) can be switched to the same ground potential.

[0064] In this state, the voltage of the first node, which is the source voltage of the output transistor (T1), increases, and the voltage applied to the gate of the output transistor (T1) can be constant. After this, the threshold voltage capacitor (C vth ) increases the potential of the first node connected to the threshold voltage capacitor (C vth ) may be closed when the voltage stored in the output transistor (T1) drops to the inherent threshold voltage of the output transistor (T1). As a result, the threshold voltage capacitor (C vth ) can be driven in such a way that the voltage of the output transistor (T1) has a compensated threshold voltage value that matches the inherent threshold voltage of the output transistor (T1).

[0065] In Figure 1, the input value coming from the previous synapse array is stored in the storage capacitor (C) of the synapse accumulation unit (100). mem ) can be connected to the first node connected to one electrode of the storage capacitor (C mem ) of the other electrode is the base voltage terminal (V SS ) can be directly connected to.

[0066] The source and drain of the fourth transistor (T4) are connected to the line through which the input value enters the first node and the base voltage terminal (V ss ) between the storage capacitor (C mem ) can be connected to form a parallel structure. In this case, when the fourth transistor is turned on in the threshold voltage compensation step, the storage capacitor (C mem ) can be initialized to 0.

[0067] In the above output section (200), the drain of the output transistor (T1) is connected to the driving voltage terminal (V DD ) is connected to the output transistor (T1), and the source of the output transistor (T1) is connected to the intermediate base voltage terminal (V SS2) can be optionally connected to the output transistor. A sixth transistor can be optionally located between the source of the output transistor and the intermediate base voltage terminal.

[0068] Meanwhile, the second transistor (T2) of the threshold voltage compensation unit (400) is the threshold voltage capacitor (C vth ) may be located between the other electrode and the gate of the output transistor (T1). At this time, the second transistor may be located between the threshold voltage capacitor (C vth ) and the signal of the electrical path between the gate of the output transistor (T1) can be turned on and off.

[0069] In the above threshold voltage storage unit (300), the threshold voltage capacitor (C vth ) is connected to the first node of the storage capacitor (C mem ) can be connected to the first node. In addition, the threshold voltage capacitor (C vth ) is connected to the second node, where the reference voltage (V ref ) can be connected to a reference voltage terminal to which the voltage is applied.

[0070] Referring to Fig. 1, in the threshold voltage compensation unit (400), the fifth transistor (T5) may be connected between the first node located between the threshold voltage capacitor and the storage capacitor and the third node located between the source of the output transistor (T1) and the output switch. The third node may also be connected to the sixth transistor described above.

[0071] FIG. 2 is a circuit diagram of an initial value setting step and a signal timing diagram in the initial value setting step according to one embodiment of the present invention.

[0072] The above initial value setting step is to set the threshold voltage capacitor (C) for driving the gate of the output transistor (T1). vth ) is the step of storing the initial threshold voltage.

[0073] The above initial value setting step is a process of setting the value stored in the capacitor within the circuit as the initial value. In this step, the storage capacitor (C mem ) can also be performed at the same time. When the fourth transistor (T4) is turned on, the storage capacitor (C mem ) The positive node value is V SS becomes reset to potential 0, and the threshold voltage capacitor (C vth ) is the base voltage terminal (V SS ) is connected.

[0074] According to the signal diagram, the third transistor (T3) is on, so the threshold voltage capacitor (C vth ) The voltage value of the other electrode is V ref , the threshold voltage capacitor (C vth ) in V ref -V ss is saved and V ref -V ss can be a voltage value to initialize to a constant value before storing the threshold voltage of the actual transistor.

[0075] FIG. 3 is a circuit diagram of a threshold voltage compensation step and a signal timing diagram in the threshold voltage compensation step according to one embodiment of the present invention.

[0076] The above threshold voltage compensation step detects the unique threshold voltage of the output transistor (T1) and the threshold voltage capacitor (C vth ) may be a step of compensating the initial threshold voltage of the output transistor with the unique threshold voltage of the output transistor.

[0077] In the above threshold voltage compensation step, the threshold voltage capacitor (C vth ) is connected to the second node and the potential applied to the gate of the output transistor (T1) is the reference voltage V ref It can be. V is applied to the gate of the above output transistor (T1). refis applied, the output transistor (T1) is turned on, current flows, and thus the storage capacitor (C mem ) and threshold voltage capacitor (C vth ) can gradually increase in potential of the first node. The potential value of the first node is V ref -V th When the gate-source voltage (V) of the output transistor (T1) gs ) This V th The moment the value becomes 0, the output transistor (T1) turns off and the threshold voltage capacitor (C vth ) has a potential of V th (unique threshold voltage) can be adjusted to the value.

[0078] In the above threshold voltage compensation step, the potential of the first node is V ref -V th And the potential of the second node is V SS A storage capacitor (C) is located between the two mem ) has V ref -V th -V SS can be saved.

[0079] Figure 4 is a storage capacitor (C) according to one embodiment of the present invention. mem ) Circuit diagram of the initialization stage and storage capacitor (C mem ) is the signal timing diagram during the initialization phase.

[0080] The above storage capacitor (C mem ) The initialization step is to initialize the storage capacitor (C) before receiving input from the previous synapse. mem ) may be the initialization step.

[0081] The above storage capacitor (C mem ) In the initialization phase, the storage capacitor (C mem ) The node values ​​at both ends are V SScan be initialized to 0. At this stage, the fourth transistor (T4) is turned on and the two electrodes of the storage capacitor become the same potential, so the storage capacitor (C mem ) can be initialized.

[0082] FIG. 5 is a circuit diagram of a storage step and a signal timing diagram in the storage step according to one embodiment of the present invention.

[0083] The above storage step receives the output value of the previous synapse as an input value and stores it in the storage capacitor (C mem ) is the step of integrating and accumulating.

[0084] In the above storage step, when the current comes in as an input value from the previous synapse array, the storage capacitor (C) connected to it mem ) can be integrated and stored. At this time, the threshold voltage capacitor is placed between the first node and V mem The value is V in the second node th +V mem Values ​​can be formed.

[0085] FIG. 6 is a circuit diagram of an output stage and a signal timing diagram in the output stage according to one embodiment of the invention.

[0086] In the above output stage, the threshold voltage (V) compensated in the threshold voltage compensation stage th ) to drive (turn on) the output transistor (T1) based on the storage capacitor (C mem ) can output an output value corresponding to the voltage.

[0087] Specifically, in this step, V is added to the second node. mem +V th is applied and the same potential can be formed at the gate of the output transistor (T1) connected thereto. At this time, the output value of the output transistor (T1) is the gate-source voltage (V gs )-V th is. That is, V gs -V th =(V mem +Vth -V SS2 )-V th Therefore V th are offset by V mem -V SS2 As a result, the output value is V mem and V SS2 is a value dependent on , and this output value is V mem -V SS2 Occurs when V becomes greater than 0 mem The larger the value, the more it can increase.

[0088] As described in detail above, the present invention enables the implementation of an artificial neural network neuron circuit that performs accurate activation function operation (ReLU) by compensating for threshold voltage fluctuations of an output transistor (T1). This invention can compensate for threshold voltage fluctuations due to semiconductor device variations and external environmental changes, and is beneficial for the development of systems for improving artificial neural network inference accuracy.

[0089] The above description is merely an illustrative example of the technical idea of ​​the present invention, and those skilled in the art will appreciate that various modifications and variations can be made without departing from the essential characteristics of the present invention. Therefore, the embodiments disclosed in the present invention are intended to illustrate rather than limit the technical idea of ​​the present invention, and the scope of the technical idea of ​​the present invention is not limited by these embodiments. The scope of protection of the present invention should be interpreted by the following claims, and all technical ideas within a scope equivalent thereto should be interpreted as being included in the scope of the rights of the present invention.

[0090] <Explanation of symbols>

[0091] 100: Synaptic accumulation zone

[0092] 200: Output section

[0093] 300: Threshold voltage storage unit

[0094] 400: Threshold voltage compensation unit

Claims

1. A threshold voltage storage unit that stores the threshold voltage of the output transistor in a threshold voltage capacitor; and A threshold voltage compensation unit that compensates the potential of the above threshold voltage capacitor to the unique threshold voltage of the output transistor; An output section that transmits a constant output value to the next synapse through the above compensated threshold voltage; An artificial neural network neuron circuit with a compensation function for threshold voltage fluctuations.

2. In paragraph 1, The above neuron circuit is, It provides a constant output value by compensating for both threshold voltage fluctuations due to device mutations and changes in the external environment. An artificial neural network neuron circuit with a compensation function for threshold voltage fluctuations.

3. In paragraph 1, It further includes a synaptic storage unit that receives the output value of the previous synapse and stores it in a storage capacitor; The above threshold voltage capacitor and the above storage capacitor are connected in series with the node where the output value of the previous synapse is input in between. An artificial neural network neuron circuit with a compensation function for threshold voltage fluctuations.

4. In paragraph 1, The threshold voltage compensation unit includes a fifth transistor formed between the source of the output transistor and the first node connected to one electrode of the threshold voltage capacitor. An artificial neural network neuron circuit with a compensation function for threshold voltage fluctuations.

5. In paragraph 3, The above synaptic accumulation area including a fourth transistor connected in parallel with the storage capacitor; An artificial neural network neuron circuit with a compensation function for threshold voltage fluctuations.

6. In paragraph 1, The above output section Transmitting an output value corresponding to the stored value of the above storage capacitor, An artificial neural network neuron circuit with a compensation function for threshold voltage fluctuations.

7. In paragraph 1, The above threshold voltage compensation unit Further comprising a transistor formed between a second node connected to the other electrode of the threshold voltage capacitor and the gate of the output transistor. An artificial neural network neuron circuit with a compensation function for threshold voltage fluctuations.

8. In paragraph 1, A reference voltage terminal is provided at the second node connected to the other electrode of the above threshold voltage capacitor, Including a third transistor located between the second node and the reference voltage terminal, An artificial neural network neuron circuit with a compensation function for threshold voltage fluctuations.

9. In paragraph 1, A sixth transistor comprising a base voltage connected to the source of the output transistor; An artificial neural network neuron circuit with a compensation function for threshold voltage fluctuations.

10. A neuromorphic computing device including an array structure in which a plurality of synaptic elements are connected to each other, comprising an artificial neural network neuron circuit that receives an output value of a previous synaptic element among the synaptic elements, compensates for a threshold voltage fluctuation, and transmits it to the next synapse, The above artificial neural network neuron circuit is the artificial neural network neuron circuit of the first clause. A neuromorphic device having a threshold voltage variation compensation function.

11. An initial value setting step for storing an initial threshold voltage in a threshold voltage capacitor for driving the gate of an output transistor in an artificial neural network neuron circuit; A threshold voltage compensation step that detects the unique threshold voltage of an output transistor and compensates the voltage of the initial value to the unique threshold voltage; Storage capacitor initialization step that initializes the potential of the storage capacitor to 0; A storage step for integrating the input value from the synapse and accumulating it in the storage capacitor; and An output step for driving an output transistor based on the threshold voltage compensated in the threshold voltage compensation step to transmit an output value corresponding to the charge of the storage capacitor; including; A method for driving an artificial neural network neuron circuit having a compensation function for threshold voltage fluctuations.

12. In paragraph 11, The above initial value setting step is A step of setting an initial value in an adjustable range for values ​​stored in the storage capacitor and the threshold voltage capacitor by a reference voltage applied to the other electrode of the threshold voltage capacitor and a base voltage applied to one electrode of the storage capacitor. A method for driving an artificial neural network neuron circuit having a compensation function for threshold voltage fluctuations.

13. In paragraph 11, The above initial value setting step is Further comprising the step of initializing the storage capacitor by connecting the two electrodes of the storage capacitor to each other by the fourth transistor to make them have the same potential. A method for driving an artificial neural network neuron circuit having a compensation function for threshold voltage fluctuations.

14. In paragraph 11, The above output steps are A step in which the gate of the output transistor is driven by the compensated threshold voltage of the threshold voltage capacitor, and the output value according to the charge of the storage capacitor is transmitted to the next synapse. A method for driving an artificial neural network neuron circuit having a compensation function for threshold voltage fluctuations.

Images