CN112906880B - Adaptive neuron circuit based on memristor - Google Patents

Abstract

The invention provides a memristor-based adaptive neuron circuit, which comprises: the excitation pulse, the nonvolatile memristor and the capacitor form a charging loop; the volatile memristor, the capacitor and the resistor form a discharge loop; voltage signals at two ends of the resistor are used as output pulses of the neuron circuit; the working process is as follows: after the nonvolatile memristor receives an excitation pulse, a capacitor is charged through a charging loop, the voltage of the capacitor is gradually increased, when the voltage at two ends of the volatile memristor is smaller than the threshold voltage, the output pulse is 0, when the voltage at two ends of the volatile memristor is larger than or equal to the threshold voltage, the capacitor discharges through a discharging loop, and the voltage at two ends of the resistor serves as the output pulse to generate an action potential; under the action of the excitation pulse, the resistance value of the nonvolatile memristor is gradually increased, so that the generated action potential frequency is reduced gradually, and the characteristic that the neuron gradually adapts to constant external stimulation is simulated. The invention simulates the adaptive function of the neuron.

Description

A Memristor-Based Adaptive Neuron Circuit

technical field

The invention belongs to the field of brain-like bionics, and more particularly, relates to an adaptive neuron circuit based on a memristor.

Background technique

It is an important way to realize artificial intelligence to build a bionic neural computing system by simulating the physiological structure and working mechanism of the human brain. The human brain is composed of about ten billion neurons and about one trillion synapses. A single neuron is interconnected with multiple protrusions to form a complex network structure. Therefore, the premise of simulating the structure of the human brain is that the hardware realizes the There are two basic units of synapse. Traditional artificial neurons and artificial synapses are constructed based on CMOS circuits. The realization of a single neuron or synapse often requires dozens or even hundreds of transistors. The circuit structure is complex, the area is large, and the power consumption is high, which is not conducive to large-scale integration. . Therefore, realizing artificial neurons and synapses with smaller power consumption and lower power consumption based on novel devices is an important research direction in the field of neuromorphic computing.

The advent of the memristor provided an opportunity to solve this problem. At present, memristor-based artificial synapses have made great progress. A single memristor device has been able to simulate multiple functions of synapses. At the same time, memristor-based artificial neurons have also made important breakthroughs. For example, the literature "An Artificial Neuron Based on aThresholdSwitching Memristor" uses a simple circuit structure and combines volatile memristive devices to realize various functions of neurons. However, these works are often based on simplified IF neuron models, ignoring most biological neuron dynamics. For example, some biological neurons will show a certain ability to adapt when receiving constant stimulation, which is manifested in the production of The frequency of action potentials decreased.

Therefore, there is a need to implement an adaptive neuron circuit based on a simple circuit structure to simulate the ability of neurons to adapt to constant stimulation.

SUMMARY OF THE INVENTION

In view of the defects of the prior art, the purpose of the present invention is to provide a memristor-based adaptive neuron circuit, which aims to solve the problem that the prior art cannot simulate the ability of neurons to adapt to constant stimulation.

To achieve the above object, the present invention provides a memristor-based adaptive neuron circuit, including: a nonvolatile memristor, a volatile memristor, a capacitor and a resistor;

The first end of the nonvolatile memristor is connected to the excitation pulse input end, and the second end is respectively connected to the first end of the capacitor and the first end of the volatile memristor; if the nonvolatile memristor When the first end of the device receives the excitation pulse, the resistance value of the non-volatile memristor will gradually increase; when the voltage across the volatile memristor is less than its threshold voltage, the volatile memristor will in a high resistance state; when the voltage across the volatile memristor is greater than or equal to its threshold voltage, the volatile memristor is in a low resistance state;

The second end of the volatile memristor is connected to the first end of the resistor; the second end of the capacitor and the second end of the resistor are both grounded; the excitation pulse, the nonvolatile memristor and the capacitor constitute a charging circuit; the volatile memristor, the capacitor and the resistance constitute a discharge circuit; the excitation pulse is used as the input pulse of the neuron circuit, and the voltage signal at both ends of the resistance is used as the output pulse of the neuron circuit;

The working process of the neuron circuit is as follows: after the first end of the non-volatile memristor receives the excitation pulse, the capacitor is charged through the charging circuit, and the voltage across the capacitor gradually increases. When the voltage is less than the threshold voltage, the output pulse is 0, and this process is a neuron integration process; when the voltage across the volatile memristor is greater than or equal to the threshold voltage, the capacitor discharges through the discharge circuit, and the voltage across the resistor is generated as an output pulse Action potential, this process is the firing process of neurons;

The adaptive process of the neuron circuit is as follows: under the action of the excitation pulse, the resistance of the non-volatile memristor gradually increases, resulting in a gradual increase in the time constant of the charging loop in the circuit, and the capacitance is in a constant state. The charging speed under the excitation pulse will become slower and slower, and the time required to reach the threshold voltage of the volatile memristor will gradually increase, and finally the frequency of the action potential generated will become smaller and smaller, simulating the neuron's constant The characteristic of gradual adaptation to external stimuli.

In an optional example, the capacitor is a fixed value capacitor or a variable capacitor, and the capacitance value ranges from 10fF to 10μF.

In an optional example, the resistance is a fixed-value resistance or a variable resistance, and the resistance value is greater than the on-state resistance of the volatile memristor, and is smaller than the off-state resistance of the volatile memristor.

In an optional example, the excitation pulse is a voltage pulse or a current pulse.

In an optional example, the first end of the volatile memristor is an active electrode end, and the second end is an inert electrode end.

In an optional example, a voltage is pre-applied to the non-volatile memristor to bring it into a low resistance state prior to applying the excitation pulse to the non-volatile memristor.

In general, compared with the prior art, the above technical solutions conceived by the present invention have the following beneficial effects:

The invention provides an adaptive neuron circuit based on a memristor, which utilizes the threshold transition characteristics of a volatile memristor and combines with a peripheral circuit to realize basic functions such as neuron integral discharge. In the present invention, in the state of constant excitation input, the resistance value of the non-volatile memristor in the low-resistance state gradually increases, which causes the time constant of the charging loop in the circuit to gradually increase, so that the corresponding speed of the neuron to the input gradually decreases. small, so as to realize the adaptive function of neurons.

The present invention provides an adaptive neuron circuit based on a memristor, which can adjust the overall firing frequency of neurons based on different resistance windows by changing the types of non-volatile memristors, making it suitable for different scenarios . Compared with the traditional adaptive neuron based on CMOS circuit, the neuron circuit has small area and low energy consumption, which is favorable for integration in an array.

Description of drawings

1 is a schematic structural diagram of a memristor-based adaptive neuron circuit provided by an example of the present invention;

FIG. 2 is a schematic diagram of the resistance value change of the nonvolatile memristor under the action of an electric pulse provided by an embodiment of the present invention;

3 is a schematic diagram of I-V characteristics of a volatile memristor provided by an embodiment of the present invention;

4 is a schematic diagram of the input and output of an action potential generated by a neuron circuit according to an embodiment of the present invention;

FIG. 5 is a schematic diagram of an adaptive function of a neuron circuit provided by an embodiment of the present invention under constant electrical pulse excitation.

Detailed ways

In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, but not to limit the present invention.

FIG. 1 is a memristor-based adaptive neuron circuit according to an embodiment of the present invention. As shown in Figure 1, the neuron circuit contains excitation input, capacitance C, resistance _RO , non-volatile memristor and volatile memristor. in:

The excitation input, nonvolatile memristor, and capacitor C form the charging loop. The upper end of the excitation input is connected to the left end of the nonvolatile memristor, the upper end of the capacitor C is connected to the right end of the nonvolatile memristor, and the lower end of the excitation input is connected to the capacitor C. The lower end is interconnected and grounded. Volatile memristor, capacitor C and resistor _RO form a discharge loop. The left end of the volatile memristor is connected to the upper end of the capacitor C, the right end of the volatile memristor is used as an output pulse end, and is connected to the upper end of the resistor _RO , and the lower end of the capacitor C is connected to the resistor The upper end of R _O is connected and grounded.

It should be noted that if the left end of the non-volatile memristor receives the excitation pulse, the resistance value of the non-volatile memristor will gradually increase; the left end of the volatile memristor is the active electrode end, and the right end is the inert end Electrode terminal, when the voltage across the volatile memristor is less than its threshold voltage, the volatile memristor is in a high resistance state; when the voltage across the volatile memristor is greater than or equal to its threshold voltage, the volatile memristor is in a high resistance state. The sex memristor is in a low resistance state.

The neuron circuit provided by this application utilizes the process of charging and discharging the capacitor C to cause the threshold transition behavior of the volatile memristor to realize the basic function of the neuron, and at the same time utilizes the non-volatile memristor to increase the resistance value under the excitation input The process of increasing the RC time constant of the charging loop simulates the gradual adaptation of neurons to constant stimulation.

Based on a simple circuit structure, the invention realizes basic functions such as neuron integral discharge and refractory period, and at the same time simulates the neuron's adaptability to constant stimulation.

Furthermore, the initial resistance state of the volatile memristor is a high resistance state. The device remains in a high-impedance state as the voltage across it increases from 0 to the threshold voltage _Vth . After the voltage across it is greater than the threshold voltage _Vth , the device transitions from a high-impedance state to a low-impedance state. After the voltage across it is gradually reduced to the holding voltage _Vhold , the device recovers from a low-impedance state to a high-impedance state. where the threshold voltage V _th is greater than the hold voltage V _hold .

Furthermore, non-volatile memristors are brought into a low resistance state by applying a voltage in advance. Under the action of the excitation pulse, the resistance state of the device will gradually increase.

Further, the capacitor C is a fixed capacitor or a variable capacitor, and the capacitance value ranges from 10fF to 10μF. The resistance is a fixed-value resistance or a variable resistance, and the resistance value is greater than the on-state resistance of the volatile memristor, and is smaller than the off-state resistance of the volatile memristor at the same time. The excitation pulse is a current pulse or a voltage pulse.

Further, when the excitation pulse is applied, the capacitor C in the circuit is charged through the charging loop. When the voltage across the volatile memristor is less than the threshold voltage V _th , the device is in a high resistance state, and the charging loop time The constant is much smaller than the time constant of the discharge loop, and the voltage across the capacitor C gradually increases. When the capacitor C is charged to a certain level, the voltage across the volatile memristor exceeds the threshold voltage V _th , and the volatile memristor changes to a low-resistance state. At this time, the time constant of the charging loop is much greater than the time of the discharging loop. Constant, the voltage across the resistor R _O is used as the output, generating an action potential. This process realizes the basic integral-discharge function of neurons.

Furthermore, in the process of applying the excitation input, the resistance value of the non-volatile memristor will gradually increase from small to large under the action of the electric pulse. During this process, the charging time constant τ=C*R of the _{non-volatile memristor} will gradually increase, the charging speed of the neuron circuit will become slower, the time will become longer, and the frequency of generating action potentials will decrease. This process simulates the neuron's property of gradually adapting to a constant input, i.e. the neuron's adaptive function.

Before applying an excitation pulse to the non-volatile memristor, the non-volatile memristor is pre-applied with a voltage to make its internal oxygen vacancies migrate to form an oxygen vacancy conductive wire, and the device changes from a high-resistance state to a low-resistance state. Then, under the action of the excitation pulse, the oxygen vacancies migrate and the conductive wire gradually retracts, so that the resistance state of the device will gradually increase. FIG. 2 is a schematic diagram of the resistance value change of the nonvolatile memristor under the action of an electric pulse in an embodiment of the present invention.

3 is a schematic diagram of IV characteristics of a volatile memristor in an embodiment of the present invention, and the initial resistance state of the volatile memristor is a high resistance state. During the process of increasing the voltage across the device from 0 to the threshold voltage _Vth , Ag ions in the upper electrode gradually migrate to the functional layer, but no complete conductive filament is formed. At this time, the current flowing through the device is extremely small, and the device remains The high-impedance state remains unchanged. When the voltage across its two ends is greater than the threshold voltage _Vth , Ag ions migrate to the lower electrode, and a complete metal conductive wire is formed in the device, the current suddenly increases to limit the current, and the device changes from a high-resistance state to a low-resistance state. After the voltage at both ends of the device is gradually reduced to the holding voltage V _hold , the conductive wire in the device breaks spontaneously, the current suddenly decreases, and the device recovers from a low-resistance state to a high-resistance state. where the threshold voltage V _th is greater than the hold voltage V _hold .

The capacitor C is a fixed capacitor or a variable capacitor, and the capacitance value ranges from 10fF to 10μF, and in this embodiment, the capacitance value is 20nF. The resistance is a fixed-value resistance or a variable resistance, and the resistance value is greater than the on-state resistance of the volatile memristor and smaller than the off-state resistance of the volatile memristor, and in this embodiment, the resistance value is 10kΩ. The excitation pulse is a current pulse or a voltage pulse. In this embodiment, the excitation pulse is a voltage pulse.

The process of generating the action potential in the neuron circuit is as follows: when the excitation pulse is applied, the capacitor C in the circuit is charged through the charging circuit, and when the voltage across the volatile memristor is less than the threshold voltage _Vth , the device is in a high resistance At this time, the time constant of the charging loop is much smaller than the time constant of the discharging loop, and the voltage across the capacitor C gradually increases. At the same time, since the resistance R _O is much smaller than the off-state resistance of the volatile memristor, the circuit output is 0. This process is the integration process of neurons. When the capacitor C is charged to a certain level, the voltage across the volatile memristor exceeds the threshold voltage V _th , the volatile memristor changes to a low resistance state, and the resistance is extremely small. At this time, the time constant of the charging loop is very large. Due to the discharge loop time constant, the capacitor is discharged through the discharge loop. At the same time, the resistance R _O is much larger than the on-state resistance of the volatile memristor, so its voltage division is large, so the voltage across the resistor R _O is used as the output voltage to generate an action potential. This process is the firing process of neurons. The above realizes the basic integral-discharge function of neurons. FIG. 4 is a schematic diagram of the input and output of the action potential generated by the neuron circuit according to the embodiment of the present invention. The integration process is the charging process of the capacitor C, and the release process is the process in which the volatile memristor becomes low resistance, causing the capacitor C to discharge through the discharge circuit.

Furthermore, the neuron circuit can simulate the function of neuron adaptation. Taking the applied excitation pulse as a voltage pulse as an example, the non-volatile memristor in the circuit is placed in a low resistance state. In the case of applying a voltage pulse, the resistance of the non-volatile memristor gradually increases. Increase (as shown in Figure 2), resulting in a gradual increase in the time constant of the charging loop in the circuit τ=C*R _{non-volatile memristor} . Therefore, the charging speed of the capacitor C under a constant voltage pulse will become slower and slower, and the time required to reach the threshold voltage of the volatile memristor will gradually increase. The characteristic of the element gradually adapting to constant external stimuli. FIG. 5 is a schematic diagram of an adaptive function of a neuron circuit according to an embodiment of the present invention. As shown in FIG. 5 , when the voltage pulse amplitude and frequency remain unchanged, the frequency of the action potential gradually decreases.

The above is an introduction to the functional principle of memristor-based neurons. Specifically, the specific construction method of the adaptive neuron circuit is as follows: after the prepared volatile memristor is connected in series with the resistor _R0 , it is connected in parallel with the capacitor C, and then the nonvolatile memristor is connected with the capacitor C is in series and the excitation input is applied to the other end of the nonvolatile memristor. During the test, an oscilloscope was used to detect the voltage signal across the resistor R _O to capture the action potential generated by the neuron.

Those skilled in the art can easily understand that the above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements and improvements made within the spirit and principles of the present invention, etc., All should be included within the protection scope of the present invention.

Claims (6)

1. An adaptive neuron circuit based on a memristor, characterized in that, comprising: a nonvolatile memristor, a volatile memristor, a capacitor and a resistor; The first end of the nonvolatile memristor is connected to the excitation pulse input end, and the second end is respectively connected to the first end of the capacitor and the first end of the volatile memristor; if the nonvolatile memristor When the first end of the device receives the excitation pulse, the resistance value of the non-volatile memristor will gradually increase; when the voltage across the volatile memristor is less than its threshold voltage, the volatile memristor will in a high resistance state; when the voltage across the volatile memristor is greater than or equal to its threshold voltage, the volatile memristor is in a low resistance state; The second end of the volatile memristor is connected to the first end of the resistor; the second end of the capacitor and the second end of the resistor are both grounded; the excitation pulse, the nonvolatile memristor and the capacitor constitute a charging circuit; the volatile memristor, the capacitor and the resistance constitute a discharge circuit; the excitation pulse is used as the input pulse of the neuron circuit, and the voltage signal at both ends of the resistance is used as the output pulse of the neuron circuit; The working process of the neuron circuit is as follows: after the first end of the non-volatile memristor receives the excitation pulse, the capacitor is charged through the charging circuit, and the voltage across the capacitor gradually increases. When the voltage is less than the threshold voltage, the output pulse is 0, and this process is a neuron integration process; when the voltage across the volatile memristor is greater than or equal to the threshold voltage, the capacitor discharges through the discharge circuit, and the voltage across the resistor is generated as an output pulse Action potential, this process is the firing process of neurons; The adaptive process of the neuron circuit is: under the action of the excitation pulse, the resistance of the non-volatile memristor gradually increases, resulting in a gradual increase in the time constant of the charging circuit in the neuron circuit, and the capacitance is With a constant excitation pulse, the charging speed will be slower and slower, and the time required to reach the threshold voltage of the volatile memristor will gradually increase, eventually resulting in a lower and lower frequency of the action potential generated, simulating a neuron The characteristic of gradual adaptation to constant external stimuli. 2 . The adaptive neuron circuit according to claim 1 , wherein the capacitor is a fixed-value capacitor or a variable capacitor, and the capacitance value ranges from 10 fF to 10 μF. 3 . 3. The adaptive neuron circuit according to claim 1, wherein the resistance is a fixed-value resistance or a variable resistance, and the resistance value is greater than the on-state resistance of the volatile memristor, and is smaller than that of the volatile memristor. Resistor off-state resistance. 4. The adaptive neuron circuit according to claim 1, wherein the excitation pulse is a voltage pulse or a current pulse. 5 . The adaptive neuron circuit according to claim 1 , wherein the first end of the volatile memristor is an active electrode end, and the second end is an inert electrode end. 6 . 6 . The adaptive neuron circuit according to claim 1 , wherein before applying the excitation pulse to the non-volatile memristor, a voltage is pre-applied to the non-volatile memristor to make it into a low resistance state.

Images