Memristor-based adaptive neuron circuit and application

Abstract

The application discloses a kind of adaptive neuron circuits based on memristor and application, belong to brain-like bionics field, including LIF neuron circuit and adaptive feedback loop;Wherein, LIF neuron circuit is based on the threshold transition characteristics of volatile memristor realizes the impulse emission of neuron, adaptive feedback loop is inhibited by adjusting transistor gate voltage the impulse emission of LIF neuron circuit, and reaches the stable pulse emission frequency after a period of time, to realize the adaptive function of biological neuron;The application only includes volatile threshold transition memristor, transistor M1, M2, capacitor C2 and resistance R L These devices make full use of the electrical characteristics of volatile threshold transition memristor and transistor, with simple structure, small circuit area, low energy consumption, and can realize the adaptive function of biological neuron with high integration.

Description

Technical Field

[0001] This invention belongs to the field of brain-inspired bionics technology, and more specifically, relates to an adaptive neuron circuit based on memristors and its application. Background Technology

[0002] With the rapid development of artificial intelligence, the Internet of Things, cloud computing, and other fields, global information is exploding. Traditional computing systems based on the von Neumann architecture, due to the physical separation of memory and processor, are not conducive to efficient data processing in intelligent systems. In contrast, neuromorphic computing architectures inspired by the human brain, with their high parallelism and event-driven computing modes, are considered to have enormous application prospects in the field of artificial intelligence.

[0003] Neurons are crucial processing units in neuromorphic computing systems. Traditional artificial neurons are built using CMOS circuits, often requiring dozens or even hundreds of transistors to realize a single neuron. This results in complex circuit structures, large areas, and high power consumption, hindering large-scale integration. Fortunately, the advent of memristors has provided an effective solution to this problem. Currently, memristor-based artificial neurons have achieved various biological neuron functions using simple circuit structures. However, these works are often based on simplified LIF neuron models, neglecting the dynamics of most biological neurons. Adaptation is considered a neuronal dynamic behavior with significant potential applications, referring to the adaptive ability of certain biological neurons to receive constant stimuli. Existing memristor adaptive neuron circuits often require large operational amplifiers and comparators, as well as complex feedback systems, resulting in high power consumption and area overhead, hindering high-density large-scale integration.

[0004] Therefore, there is an urgent need for a compact neuronal circuit to enable the adaptive function of biological neurons. Summary of the Invention

[0005] In view of the above-mentioned defects or improvement needs of the prior art, the present invention provides an adaptive neuron circuit and its application based on memristors, so as to solve the technical problem that the prior art cannot achieve the adaptive function of biological neurons with a high degree of integration.

[0006] To achieve the above objectives, in a first aspect, the present invention provides an adaptive neuron circuit based on memristors, comprising: a LIF neuron circuit and an adaptive feedback loop;

[0007] The LIF neuron circuit includes: a volatile threshold switching memristor and a transistor M1; the first terminal of the memristor serves as the input terminal of the excitation pulse, and the second terminal is connected to the drain of the transistor M1; the source of the transistor M1 is grounded.

[0008] The adaptive feedback loop includes: resistor R LTransistor M2 and capacitor C2; resistor R L The first terminal is used to connect to the voltage source V. dd The second terminal is connected to the gate of transistor M1, the drain of transistor M2, and the first terminal of capacitor C2, respectively; the second terminal of capacitor C2 and the source of transistor M2 are both grounded; the gate of transistor M2 is connected to the second terminal of memristor.

[0009] In the initial state, transistor M1 is turned on, transistor M2 is turned off, and the memristor is in a high-resistance state; both transistor M1 and transistor M2 are NMOS transistors.

[0010] More preferably, the threshold voltage of the memristor is in the range of [2.2V, 2.6V], and the holding voltage is in the range of [0.4V, 0.8V].

[0011] More preferably, the volatile threshold switching memristor is a Mott memristor based on Mott phase transition, a diffuse memristor based on a metal conductive filament, or an OTS memristor based on a chalcogenide.

[0012] More preferably, when the voltage across the memristor is higher than or equal to its threshold voltage, the memristor transitions from a high-resistance state to a low-resistance state; when the voltage across the memristor is less than or equal to its holding voltage, the memristor transitions from a low-resistance state to a high-resistance state.

[0013] More preferably, the above-mentioned adaptive neuron circuit further includes a capacitor C1 connected in parallel across the memristor.

[0014] In a second aspect, the present invention provides an artificial neural network comprising multiple neurons, wherein the neurons are the adaptive neuron circuits provided in the first aspect of the present invention.

[0015] Thirdly, the present invention provides an electronic chip including the memristor-based adaptive neuron circuit provided in the first aspect of the present invention.

[0016] In summary, the above-described technical solutions conceived in this invention can achieve the following beneficial effects:

[0017] 1. This invention provides an adaptive neuron circuit based on a memristor, comprising a LIF neuron circuit and an adaptive feedback loop. The LIF neuron circuit utilizes the threshold switching characteristics of a volatile memristor to achieve neuronal pulse firing. The adaptive feedback loop suppresses the pulse firing of the LIF neuron circuit by adjusting the gate voltage of a transistor, reaching a stable pulse firing frequency after a period of time, thereby realizing the adaptive function of the biological neuron. This invention only includes a volatile threshold switching memristor, transistors M1 and M2, capacitor C2, and resistor R. LThese devices fully utilize the electrical characteristics of volatile threshold switching memristors and transistors, resulting in simple structures, small circuit areas, and low power consumption. They can achieve adaptive functions of biological neurons with a high degree of integration.

[0018] 2. Furthermore, in the adaptive neuron circuit provided by the present invention, the threshold voltage of the memristor is in the range of [2.2V, 2.6V], and the holding voltage is in the range of [0.4V, 0.8V]. Under these conditions, the adaptive behavior of the biological neuron is more obvious and stable.

[0019] 3. Furthermore, the adaptive neuron circuit provided by the present invention also includes a capacitor C1 connected in parallel across the memristor to achieve more stable and wider range of pulse firing frequency regulation. Attached Figure Description

[0020] Figure 1 A schematic diagram of the structure of an adaptive neuron circuit based on memristors provided in an embodiment of the present invention;

[0021] Figure 2 A flowchart illustrating the operation of a memristor-based adaptive neuron circuit provided in an embodiment of the present invention;

[0022] Figure 3 This is a schematic diagram of the IV characteristic curve of a volatile threshold switching memristor provided in an embodiment of the present invention;

[0023] Figure 4 The neuron firing frequency and pulse interval provided in the embodiments of the present invention are subject to adaptive control signal V. g A schematic diagram of the change curve;

[0024] Figure 5 The waveform diagram of the adaptive neuron circuit based on memristors simulating the adaptive pulse firing of a neuron is provided in the embodiments of the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the 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 merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0026] To achieve the above objectives, in a first aspect, the present invention provides an adaptive neuron circuit based on memristors, comprising: a LIF neuron circuit and an adaptive feedback loop;

[0027] The LIF neuron circuit includes: a volatile threshold switching memristor and a transistor M1; the first terminal of the memristor serves as the input terminal of the excitation pulse, and the second terminal is connected to the drain of the transistor M1; the source of the transistor M1 is grounded.

[0028] The adaptive feedback loop includes: resistor R L Transistor M2 and capacitor C2; resistor R L The first terminal is used to connect to the voltage source V. dd The second terminal is connected to the gate of transistor M1, the drain of transistor M2, and the first terminal of capacitor C2, respectively; the second terminal of capacitor C2 and the source of transistor M2 are both grounded; the gate of transistor M2 is connected to the second terminal of memristor.

[0029] In the initial state, transistor M1 is turned on, transistor M2 is turned off, the memristor is in a high-resistance state, the initial potential of capacitor C1 is 0, and the initial potential of capacitor C2 is V. dd Both transistors M1 and M2 are NMOS transistors.

[0030] It should be noted that when the voltage across the memristor is higher than or equal to its threshold voltage, the memristor transitions from a high-resistance state to a low-resistance state; when the voltage across the memristor is less than or equal to its holding voltage, the memristor transitions from a low-resistance state to a high-resistance state.

[0031] The resistive state of the memristor affects the resistance of transistor M2 and the voltage across capacitor C2;

[0032] The threshold voltage and protection voltage of the memristor satisfy the following equation during the operation of the adaptive neuron circuit: The time t in the equation has a periodic solution; where R M2 (t) represents the resistance of transistor M2 at time t; v c2 (t) represents the voltage across capacitor C2 at time t.

[0033] Preferably, in one optional embodiment, the threshold voltage of the memristor is in the range of [2.2V, 2.6V], and the holding voltage is in the range of [0.4V, 0.8V].

[0034] In one optional implementation, the adaptive neuron circuit further includes an external capacitor C1 connected in parallel across the memristor. The circuit then operates as follows: Under the action of the excitation pulse, capacitor C1 continuously charges. When the voltage across the memristor is greater than or equal to the threshold voltage, the memristor undergoes a threshold switching behavior, capacitor C1 discharges rapidly, and the voltage across transistor M1 generates an action potential as an output pulse, which is applied to the adaptive feedback loop. Transistor M2 turns on, capacitor C2 begins to discharge, suppressing the pulse firing of the LIF neuron circuit. When the charging and discharging of capacitor C2 reaches a dynamic equilibrium, the pulse firing frequency of the neuron circuit eventually stabilizes, thus realizing the adaptive function of the biological neuron. It should be noted that the above process also applies to adaptive neuron circuits without an external capacitor C1 connected in parallel across the memristor. In this case, capacitor C1 in the above process can be considered as the parasitic capacitance of the memristor itself.

[0035] In one alternative implementation, the volatile threshold switching memristor can be a Mott memristor based on Mott phase transition, a diffuse memristor based on a metal conductive filament, or an OTS memristor based on a chalcogenide.

[0036] To further illustrate the adaptive neuron circuit provided by the present invention, a specific embodiment is described in detail below:

[0037] like Figure 1 As shown, this embodiment uses an adaptive neuron circuit with an external capacitor C1 connected in parallel across the memristor as an example:

[0038] Specifically, Figure 2 This is a flowchart illustrating the operation of a memristor-based adaptive neuron circuit according to an embodiment of the present invention, wherein V m V is the voltage across the memristor. th V is the threshold voltage of the memristor. hold The memristor is kept at a voltage. The neuron circuit operates as follows: Under the action of an excitation pulse, capacitor C1 is continuously charged, which is the integration process of the neuron; when there is no excitation pulse, capacitor C1 continuously discharges until the potential drops to 0, which is the leakage process of the neuron; when the voltage across the memristor is greater than or equal to the threshold voltage, the memristor changes from a high-resistance state to a low-resistance state, capacitor C1 discharges rapidly, and the voltage across transistor M1 generates an action potential as an output pulse, which is the firing process of the neuron; when the voltage across the memristor is less than or equal to the holding voltage, the memristor changes from a low-resistance state to a high-resistance state, capacitor C1 starts charging again, and so on, realizing the oscillating behavior of the neuron circuit.

[0039] The adaptive process of the neuron circuit is as follows: In the initial state, transistor M2 is in a high-resistance state (off), and the initial potential of capacitor C2 is V. ddWhen the neuron fires a pulse, the resistance of transistor M2 drops sharply, and capacitor C2 discharges rapidly. This reduces the voltage applied to the gate of transistor M1, increases the resistance of transistor M1, and consequently reduces the voltage division across the memristor, thus decreasing the neuron's firing frequency. Initially, during the neuron's oscillation, the discharge of capacitor C2 exceeds its charge, causing the voltage applied to the gate of transistor M1 to generally decrease, and the neuron's firing frequency to continuously decrease. After the neuron oscillates for a period of time, the discharge of capacitor C2 equals its charge, and the neuron's firing frequency eventually stabilizes, thus achieving the adaptive function of the biological neuron.

[0040] Specifically, the aforementioned adaptive neuron circuit can be described as an adaptive neuron mathematical model for spiking neural networks.

[0041] First, let's analyze the neuron firing process: In the initial state, the memristor is in a high-resistivity state, using R... ins Let represent its insulation resistance. Then, the transient process of this circuit is controlled by a first-order ordinary differential equation, as shown below:

[0042]

[0043] In the formula, V c1 R represents the voltage across capacitor C1. M1 This represents the resistance of transistor M1. Let the voltage across the capacitor at time t be V. c1 (t), R M1 The resistance is R M1 (t),

[0044]

[0045] Among them, I ds1 (t) represents the drain current of transistor M1 at time t;

[0046] The solution to equation (1-1) can be obtained:

[0047]

[0048] in, This represents the time constant of a neuron circuit.

[0049] The ignition process of a neuron can be simplified to the following formula:

[0050]

[0051] The membrane potential of a neuron is

[0052]

[0053] Next, we analyze the neuron's adaptive process: In the initial state, the memristor is in a high-resistivity state, using R... M2 Let the resistance of NMOS transistor M2 represent the transient process of the circuit, which is then controlled by a first-order ordinary differential equation.

[0054]

[0055] Let the voltage across capacitor C2 at time t be V. c2 (t), R M2 The resistance is R M2 (t),

[0056]

[0057] Among them, I ds2 (t) represents the drain current of transistor M2 at time t;

[0058] The solution to equation (1-6) can be obtained:

[0059]

[0060] in, This represents the time constant of the adaptive feedback loop.

[0061] During the above process, the drain current I of the transistor ds It can be obtained from the following formula:

[0062] When V DS ≤V GS -V th(GS) hour,

[0063]

[0064] When V DS ≥V GS -V th(GS) When >0,

[0065]

[0066] It should be noted that the above process also applies to adaptive neuron circuits where no external capacitor C1 is connected in parallel across the memristor. In this case, capacitor C1 in the above process can be regarded as the parasitic capacitance of the memristor itself.

[0067] Specifically, in this embodiment, capacitor C1 has a value of 100pF, capacitor C2 has a value of 0.5nF, the initial voltage is 3V, and the constant voltage source V... dd 3V, resistor R L The resistance is 15000Ω, and the gate threshold voltage of transistors M1 and M2 is 0.7V.

[0068] Figure 3 This is the IV characteristic curve of the volatile threshold-switching memristor used in this embodiment. Wherein, the threshold voltage V of the memristor... th The voltage is 2.21V, and the high impedance is 20kΩ; the holding voltage V hold The voltage is 0.5V, and the low resistance is 500Ω. When the voltage across the memristor is greater than or equal to the threshold voltage, the memristor transitions from a high resistance state to a low resistance state; when the voltage across the memristor is less than or equal to the holding voltage, the memristor transitions from a low resistance state to a high resistance state.

[0069] Figure 4 The adaptive control signal V in this embodiment of the invention g The effect on neuron firing frequency and pulse interval, including the input excitation voltage V in Keeping it constant, directly changing the gate voltage V applied to transistor M1 g With V g As the resistance R of transistor M1 gradually increases, ds The gradual decrease in capacitance C1 results in faster charging and discharging, meaning a faster pulse firing frequency and shorter intervals between adjacent pulses in the neuron circuit. Therefore, due to... Figure 4 It can be seen that by changing the gate voltage V of NMOS transistor M1 g It can effectively modulate the pulse firing frequency of neuronal circuits.

[0070] Figure 5 The waveform diagram shows the adaptive pulse firing of a neuron simulated by an adaptive neuron circuit based on a memristor, provided in an embodiment of the present invention. In the initial state, the memristor is in a high-resistance state, the initial voltage of capacitor C1 is 0, and the initial voltage of capacitor C2 is V. dd Transistor M1 operates in the variable resistance region, while transistor M2 operates in the cutoff region. It can be seen that at the initial moment of input activation, the neuron circuit fires pulses at a high frequency, and these pulses are simultaneously applied to the adaptive feedback loop, causing the gate voltage V of NMOS transistor M1 to... g The overall trend is downward, and the resistance R of transistor M1 is decreasing. ds As the voltage gradually increases, the pulse firing frequency gradually decreases. After the neuron oscillates for a period of time, the discharge amount of capacitor C2 equals the charge amount, and the gate voltage V of NMOS transistor M1... g By oscillating slightly around a fixed voltage value, the firing frequency of neurons eventually stabilizes, thus achieving the adaptive function of biological neurons.

[0071] This invention provides an adaptive neuron circuit based on memristors, capable of simulating the leakage integral ignition function of neurons and suppressing neuronal excitability through a feedback loop to achieve adaptive behavior. Compared to traditional CMOS-based adaptive neurons and existing memristor-based adaptive neurons, this neuron circuit fully utilizes the electrical characteristics of volatile threshold-switching memristors and transistors, offering advantages such as simple circuit structure and ease of integration, and has great application prospects in the field of artificial intelligence.

[0072] In a second aspect, the present invention provides an artificial neural network comprising multiple neurons, wherein the neurons are the adaptive neuron circuits provided in the first aspect of the present invention.

[0073] The related technical solutions are the same as the adaptive neuron circuit provided in the first aspect of this invention, and will not be described in detail here.

[0074] Thirdly, the present invention provides an electronic chip including the memristor-based adaptive neuron circuit provided in the first aspect of the present invention.

[0075] The related technical solutions are the same as the adaptive neuron circuit provided in the first aspect of this invention, and will not be described in detail here.

[0076] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An adaptive neuron circuit based on memristors, characterized in that, include: LIF neuron circuit and adaptive feedback loop; The LIF neuron circuit includes: a volatile threshold switching memristor and a transistor M1; the first terminal of the volatile threshold switching memristor serves as the input terminal of the voltage excitation pulse, and the second terminal is connected to the drain of the transistor M1; the source of the transistor M1 is grounded. The adaptive feedback loop includes: a resistor R L Transistor M2 and capacitor C2; resistor R L The first terminal is used to connect to the voltage source V. dd The second terminal is connected to the gate of transistor M1, the drain of transistor M2, and the first terminal of capacitor C2, respectively; the second terminal of capacitor C2 and the source of transistor M2 are both grounded; the gate of transistor M2 is connected to the second terminal of the volatile threshold switching memristor. In the initial state, transistor M1 is turned on, transistor M2 is turned off, and the volatile threshold switching memristor is in a high-resistance state; both transistor M1 and transistor M2 are NMOS transistors. The voltage across transistor M1 generates an action potential as an output pulse and is applied to the adaptive feedback loop; the adaptive feedback loop suppresses the pulse firing of the LIF neuron circuit by adjusting the gate voltage of transistor M1.

2. The adaptive neuron circuit according to claim 1, characterized in that, The threshold voltage range of the volatile threshold switching memristor is [2.2V, 2.6V], and the holding voltage range is [0.4V, 0.8V].

3. The adaptive neuron circuit according to claim 1, characterized in that, The volatile threshold switching memristor is a Mott memristor based on Mott phase transition, a diffuse memristor based on a metal conductive filament, or an OTS memristor based on a chalcogenide.

4. The adaptive neuron circuit according to claim 1, characterized in that, When the voltage across the volatile threshold switching memristor is higher than or equal to its threshold voltage, the volatile threshold switching memristor transitions from a high-resistance state to a low-resistance state; when the voltage across the volatile threshold switching memristor is less than or equal to its holding voltage, the volatile threshold switching memristor transitions from a low-resistance state to a high-resistance state.

5. The adaptive neuron circuit according to any one of claims 1-4, characterized in that, It also includes a capacitor C1 connected in parallel across the volatile threshold switching memristor.

6. An artificial neural network, characterized in that, It includes multiple neurons, wherein the neurons are the adaptive neuron circuits described in any one of claims 1-5.

7. An electronic chip, characterized in that, Includes the adaptive neuron circuit described in any one of claims 1-5.

Images