A memristor cell neuron circuit that generates bidirectional action potentials

Abstract

The present invention relates to the field of neuron circuit technology, and in particular to a memristor cell neuron circuit that generates bidirectional action potentials, including replacing linear resistors in a cell neuron model with memristors to construct a memristor cell neuron model; and generating bidirectional spike discharge phase trajectory characteristics in different states by adjusting the hyperbolic tangent function gradient value in the memristor cell neuron model. The present invention solves the problem that the existing neuron network model does not have bidirectional action potential capability, thus has weak encoding capability and cannot perform complex calculations and information processing tasks.

Description

Technical Field

[0001] The present invention relates to the technical field of neuron circuits, and in particular to a memristor cell neuron circuit for generating bidirectional action potentials. Background Art

[0002] In 1988, Chua proposed the Cellular Neural Network (CNN), a large-scale nonlinear analog circuit that can process signals in real time. It has the characteristics of continuous real-time, high-speed parallel computing, and is suitable for implementation in very large-scale integrated circuits.

[0003] The theoretical design and hardware implementation of CNN circuits are based on a simplified simulation of the information processing mechanism of the biological neural network of the human brain. Unlike biological neurons, the connection between CNN cell neurons is mainly controlled by weight templates. Different templates exhibit different nonlinear characteristics. Memristors with memory characteristics can be applied to functional connection points between neurons.

[0004] There are two types of action potentials in memristive neurons: unidirectional action potentials and bidirectional action potentials; unidirectional action potentials refer to action potentials that are only fired upward, while bidirectional action potentials refer to action potentials that are fired both upward and downward.

[0005] However, most existing neuronal circuits can only generate unidirectional action potentials, but cannot produce bidirectional firing behaviors;

[0006] For example: Xie Y, Yao Z, Ren G, et al. Estimate physical reliability in Hindmarsh-Rose neuron. Physics Letters A, uses the Hindmarsh-Rose model to generate unidirectional action potentials; another example is Firing mechanism based on single memristive neuron and double memristivecoupled neurons, which uses a memristive circuit to generate discharge behavior, but the memristive circuit has a complex structure and only generates unidirectional action potentials.

[0007] Compared with classical CNN, memristive CNN has the characteristics of smaller size, lower power consumption and better information processing performance. It is of great significance to manufacture memristive CNN circuits that can generate bidirectional action potentials, which can expand the encoding capacity of neural signals and may perform more complex computing and information processing tasks in the nervous system. Summary of the invention

[0008] In view of the shortcomings of existing circuits, the present invention uses ideal memristors to replace the linear resistors in traditional cell neurons, simulates the changes in cell membrane potential, and uses numerical simulation to analyze the discharge behavior of memristor neuromorphic circuits. An analog circuit is designed for experiments to verify the correctness and feasibility of the numerical simulation.

[0009] The technical solution adopted by the present invention is: a memristive cell neuron circuit that generates bidirectional action potentials: the linear resistor in the cell neuron model is replaced by a memristor to construct a memristive cell neuron model; by adjusting the gradient value of the hyperbolic tangent function in the memristive cell neuron model, bidirectional spike discharge phase trajectory characteristics in different states are generated.

[0010] As a preferred embodiment of the present invention, the memristor cell neuron model has the formula:

[0011]

[0012] In the formula, v x is the neuron state variable, is the internal variable of the memristor, is the memristor function, b is the gradient of the hyperbolic tangent function, I m sin(2πFt) is the sinusoidal current excitation.

[0013] As a preferred embodiment of the present invention, the circuit equation of the memristor cell neuron model is:

[0014]

[0015] Among them, V x is the state variable of the neuron, is the internal variable of the memristor, V x is the state variable of the neuron, is the internal variable of the memristor, is the memristor magnetic induction current, Asin(2πft) is the sinusoidal current excitation, A and f represent the frequency and amplitude of the sinusoidal excitation signal respectively, and g is the multiplier amplification factor.

[0016] As a preferred embodiment of the present invention, the main circuit of the memristor cell neuron model includes: a current source I, a capacitor C2, a hyperbolic tangent function circuit, an operational amplifier U1, a memristor equivalent circuit G M and resistors R1 to R5, the current source I is connected in parallel with the capacitor C2 and the memristor equivalent circuit G MThe output end of the hyperbolic tangent function circuit is connected in series with a resistor R1 and then connected to the inverting input end of the operational amplifier U1; the inverting input end and the output end of the operational amplifier U1 are connected in parallel with a resistor R2; the in-phase end of the operational amplifier U1 is connected to a resistor R4 and then connected to one end of R3, and the output end of the operational amplifier U1 is connected to the other end of the resistor R3; the in-phase end of the operational amplifier U1 is connected to the ground through a resistor R5, and the resistor R3 is also connected to the memristor equivalent circuit G M Connected to the common upper end of capacitor C2.

[0017] As a preferred embodiment of the present invention, the memristor equivalent circuit includes operational amplifiers U2 to U5, a multiplier M1, a hyperbolic cosine function circuit, a multiplier M1, resistors R6 to R 13 , the inverting input terminal of the operational amplifier U2 is respectively connected to the output terminal of the operational amplifier U2 and one end of the input of the multiplier M1, the output terminal of the operational amplifier U2 is connected to one end of the resistor R6, the other end of the resistor R6 is respectively connected to the inverting input terminal of the operational amplifier U3 and one end of the resistor R7, the other end of the resistor R7 is respectively connected to the output terminal of the operational amplifier U3 and one end of the resistor R8, the other end of the resistor R8 is connected to one end of the capacitor C1, the output terminal of the operational amplifier U4 is respectively connected to the other end of the capacitor C1 and the input terminal of the hyperbolic cosine function circuit, the output terminal of the hyperbolic cosine function circuit is connected to the other input terminal of the multiplier M1, the output terminal of the multiplier M1 is connected to one end of the resistor R9, the other end of the resistor R9 is respectively connected to the resistor R 10 One end is connected to the inverting input terminal of the operational amplifier U5, and the resistor R 10 The other end is connected to the output of operational amplifier U5 and resistor R 11 The same-direction input terminal of operational amplifier U5 is connected to the resistor R 11 The other end of the parallel resistor R 12 The same-direction input terminal of operational amplifier U5 is also connected to resistor R 13 Connect in series and then ground.

[0018] As a preferred embodiment of the present invention, the operational amplifiers U1 to U5 are of type AD741JN.

[0019] As a preferred embodiment of the present invention, when the hyperbolic tangent function gradient b=3.4, it is a bidirectional spike discharge with a period of 2; when b=4.268, it is a bidirectional spike discharge with a period of 3; when b=4.96, it is a bidirectional spike discharge with a period of 5; when b=3.3, it is a chaotic bidirectional spike discharge.

[0020] As a preferred embodiment of the present invention, the current source I is a sinusoidal current source.

[0021] Beneficial effects of the present invention:

[0022] 1. Use the two-dimensional memristor cell neuron model to realize the discharge behavior of the neuromorphic form and generate bidirectional action potentials, which can be used in the medical field;

[0023] 2. The construction of the memristor cell neuron model has a scientific theoretical basis, and the design of a simple analog circuit will provide reference value for the research on the implementation of memristor-driven cell neural network circuits. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a schematic diagram of the main circuit of a cell neuron;

[0025] Figure 2 is the schematic diagram of the ideal memristor equivalent circuit;

[0026] Figure 3 The frequency is F = 0.5, the amplitude is I m =1, the phase trajectory diagram and timing diagram of MATLAB numerical simulation under the conditions that the hyperbolic tangent function gradient b is 3.4, 4.268, 4.28 and 4.96 respectively;

[0027] Figure 4 They are the circuit experimental phase trajectory diagram and timing diagram under the conditions of frequency f=50Hz, amplitude A=0.1mA, and hyperbolic tangent function gradient b of 2.308, 2.692, 4.831 and 2.423 respectively. DETAILED DESCRIPTION

[0028] The present invention is further described below in conjunction with the accompanying drawings and embodiments. This figure is a simplified schematic diagram, which only illustrates the basic structure of the present invention in a schematic manner, and therefore it only shows the components related to the present invention.

[0029] A memristor cell neuron circuit for generating bidirectional action potentials comprises:

[0030] The cell neuron main circuit and the memristor equivalent circuit, based on the cell neural network circuit, use ideal memristors to replace the linear resistors in the cell neural network circuit to simulate the changes in cell membrane potential and form a memristor cell neuron model. The memristor cell neuron model is expressed as:

[0031]

[0032] In the formula, v x is the neuron state variable, is the internal variable of the memristor, is the memristor function, b is the gradient of the hyperbolic tangent function, I m sin(2πFt) is the external sinusoidal current excitation.

[0033] like Figure 1,2, the main circuit of the cell neuron includes a sinusoidal AC power supply I, a capacitor C2, and a memristor equivalent circuit input terminal V x 、Hyperbolic tangent function circuit tanh(.) input terminal -V x , operational amplifier U1, the sinusoidal current source I of the cell neuron main circuit is connected in parallel with the capacitor C2 and the memristor equivalent circuit, the output end of the hyperbolic tangent function circuit is connected in series with the resistor R1 and then connected to the inverting input end of the operational amplifier U1; a resistor R2 is connected in parallel between the inverting input end and the output end of the operational amplifier U1; the in-phase terminal of the operational amplifier U1 is connected to the resistor R4 and then connected to one end of R3, and the output end of the operational amplifier U1 is connected to the other end of the resistor R3; the in-phase end of the operational amplifier U1 is connected to the ground via the resistor R5.

[0034] The memristor equivalent circuit includes an operational amplifier U2, an operational amplifier U3, an operational amplifier U4, a multiplier M1, and a hyperbolic cosine function circuit input terminal. The output terminal V of the operational amplifier U2 x The series resistor R6 is connected to the inverting input terminal of the operational amplifier U3, and the inverting input terminal and the output terminal of the operational amplifier U3 are connected in parallel with the resistor R7; the output terminal of the operational amplifier U3 -V x The series resistor R8 is connected to the inverting input terminal of the operational amplifier U4; the inverting input terminal and the output terminal of the operational amplifier U4 are connected in parallel with the capacitor C1; the output terminal of the operational amplifier U4 is connected to the input terminal of the circuit realizing the hyperbolic cosine function cosh(.); the two input terminals x and y of the multiplier M1 are respectively connected to the output terminal of the operational amplifier U2 and the output terminal of the hyperbolic cosine function circuit; the output terminal of the multiplier M1 is connected to the inverting input terminal of the operational amplifier U5 after being connected in series with the resistor R9; the inverting input terminal and the output terminal of the operational amplifier U5 are connected in parallel with the resistor R 10 ; Operational amplifier U5's common-mode termination resistor R 12 After R 11 The output end of the operational amplifier U5 is connected to the resistor R 11 The other end of the operational amplifier U5 is connected to the non-inverting end through the resistor R 13 Connect to ground; the non-inverting input of operational amplifier U2 is connected to V x ; The non-inverting input terminals of operational amplifiers U3 and U4 are both grounded.

[0035] Formula (1) contains two state variables v x and correspond Figure 1 and Figure 2 The two variables V in the circuit state equation x and The circuit state equation corresponding to formula (1) is as follows:

[0036] exist Figure 2In the memristor equivalent circuit, there is an integral channel used to implement the second equation of formula (1); according to Kirchhoff's current law and the electrical characteristics of circuit components, the circuit differential equation of the memristor cell neuron model in formula (1) can be written as:

[0037]

[0038] Among them, V x is the state variable of the neuron, is the internal variable of the memristor, is the memristor magneto-induction current, Asin(2πft) is the external sinusoidal current excitation, and A and f represent the frequency and amplitude of the sinusoidal excitation signal, respectively.

[0039] Taking the time accuracy as 10ms, by comparing equations (1) and (2), we can get:

[0040] R1=R2=R4=R9=R 10 =R 12 =1kΩ; R3=2.5kΩ; R5=3.5kΩ; R6=R7=R8=R 11 =10kΩ;

[0041] R 13 =11kΩ; C1=C2=1μF; g=1; f=50Hz.

[0042] Numerical simulation: When the gradient b of the hyperbolic tangent function is used as the parameter of the system, the MATLAB ODE45Ps algorithm is used to carry out numerical research on the memristor cell neuron model; when the frequency F = 0.5 and the amplitude I m =1, Figure 3 The phase trajectory diagram and time domain diagram under different gradients are depicted. When the gradient b=3.4, it shows bidirectional spike discharge of period 2; when b=4.268, it shows bidirectional spike discharge of period 3; when b=4.96, it shows bidirectional spike discharge of period 5; when b=3.3, it shows chaotic bidirectional spike discharge.

[0043] Experimental verification: This design uses Pspice to complete the circuit construction and simulation operation. The discrete devices use the AD741JN operational amplifier and ideal multiplier with a supply voltage of ±15V. The discrete components use precision adjustable resistors and capacitors. m The adjustment mainly depends on adjusting the amplitude A of the external input excitation sinusoidal signal source. m The numerical relationship between I and A can be expressed as m =10 4 A, for example, when I m=1, the amplitude A of the sinusoidal signal source should be 0.1mA; for the adjustment of the hyperbolic tangent function gradient b, it mainly depends on adjusting the gradient b of the hyperbolic tangent function circuit; when the frequency f=50Hz and the amplitude A=0.1mA, the hyperbolic tangent function circuit gradient b is adjusted. When b is 2.308, 2.692, 4.831 and 2.423 respectively, the captured phase trajectory diagram and time domain diagram v x -t as Figure 3 (a)-3(d); ignoring some minor differences between numerical simulation and hardware circuit experiments caused by calculation errors and parasitic circuit parameters, the experimental results are almost consistent with the numerical simulation. Figure 4 (a)-4(d) It can be seen that with the change of the gradient b of the hyperbolic tangent function, the discharge activity changes from period 2 to period 3 and then to period 5, and finally enters a chaotic state.

[0044] Therefore, the memristor cell neuron model capable of generating bidirectional action potentials constructed by the present invention has a scientific theoretical basis, and the designed analog circuit unit will provide reference value for the research on the realization of memristor-driven cellular neural network circuits.

[0045] Based on the above ideal embodiments of the present invention, the relevant staff can make various changes and modifications without departing from the technical concept of the present invention through the above description. The technical scope of the present invention is not limited to the contents of the specification, and its technical scope must be determined according to the scope of the claims.

Claims

1. A memristor cell neuron circuit that generates bidirectional action potentials, characterized in that: include: The linear resistor in the cell neuron model is replaced by a memristor to construct a memristor cell neuron model; by adjusting the gradient value of the hyperbolic tangent function in the memristor cell neuron model, the bidirectional spike discharge phase trajectory characteristics of different states are generated; The main circuit of the memristor cell neuron model includes: current source I, capacitor C2, hyperbolic tangent function circuit, operational amplifier U1, memristor equivalent circuit G M and resistors R1 to R5, the current source I is connected in parallel with the capacitor C2 and the memristor equivalent circuit G M The output end of the hyperbolic tangent function circuit is connected in series with a resistor R1 and then connected to the inverting input end of the operational amplifier U1; the inverting input end and the output end of the operational amplifier U1 are connected in parallel with a resistor R2; the in-phase terminal of the operational amplifier U1 is connected to a resistor R4 and then connected to one end of a resistor R3, and the output end of the operational amplifier U1 is connected to the other end of the resistor R3; the in-phase end of the operational amplifier U1 is connected to a resistor R5, and the resistor R3 is also connected to the memristor equivalent circuit G M Connected to the common upper end of capacitor C2; The memristor equivalent circuit includes operational amplifiers U2 to U5, multiplier M1, hyperbolic cosine function circuit, multiplier M1, resistors R6 to R 13 The inverting input terminal of the operational amplifier U2 is respectively connected to the output terminal of the operational amplifier U2 and one input terminal of the multiplier M1, the output terminal of the operational amplifier U2 is connected to one end of the resistor R6, the other end of the resistor R6 is respectively connected to the inverting input terminal of the operational amplifier U3 and one end of the resistor R7, the other end of the resistor R7 is connected to one end of the resistor R8, the other end of the resistor R8 is respectively connected to the inverting input terminal of the operational amplifier U4 and one end of the capacitor C1, the output terminal of the operational amplifier U4 is respectively connected to the other end of the capacitor C1 and the input terminal of the hyperbolic cosine function circuit, the output terminal of the hyperbolic cosine function circuit is connected to the other input terminal of the multiplier M1, the output terminal of the multiplier M1 is connected to one end of the resistor R9, the other end of the resistor R9 is respectively connected to the resistor R 10 One end is connected to the inverting input terminal of the operational amplifier U5, and the resistor R 10 The other end is connected to the output of operational amplifier U5 and resistor R 11 The same-direction input terminal of operational amplifier U5 is connected to the resistor R 11 The other end of the parallel resistor R 12 The same-direction input terminal of operational amplifier U5 is also connected to resistor R 13 connect.

2. The memristor cell neuron circuit for generating bidirectional action potential according to claim 1, characterized in that: Memristor cell neuron model, the formula is: In the formula, v x is the neuron state variable, is the internal variable of the memristor, is the memristor function, b is the gradient of the hyperbolic tangent function, I m sin(2πFt) is the sinusoidal current excitation.

3. The memristor cell neuron circuit for generating bidirectional action potential according to claim 1, characterized in that: The circuit equation of the memristor cell neuron model is: Among them, V x is the state variable of the neuron, is the internal variable of the memristor, is the memristor magnetic induction current, Asin(2πft) is the sinusoidal current excitation, A and f represent the frequency and amplitude of the sinusoidal excitation signal respectively, and g is the multiplier amplification factor.

4. The memristor cell neuron circuit for generating bidirectional action potential according to claim 1, characterized in that: The operational amplifiers U1 to U5 are AD741JN.

5. The memristor cell neuron circuit for generating bidirectional action potential according to claim 1, characterized in that: The bidirectional spike discharges in different states include: bidirectional spike discharges in period 2, bidirectional spike discharges in period 3, bidirectional spike discharges in period 5 and chaotic bidirectional spike discharges.

6. The memristor cell neuron circuit for generating bidirectional action potential according to claim 1, characterized in that: The current source I is a sinusoidal current source.

Images