CN117669676A - A memristor-based sensing neuron circuit and its application - Google Patents

Abstract

The invention discloses a sensing neuron circuit based on a memristor and application thereof, and belongs to the technical field of intelligent sensing; an integrated sensing neuron array is constructed through word lines and bit lines, and all sensing neurons in the array convert the sensed external environment information into pulse signals in parallel and output the pulse signals; each sensing neuron in the array comprises a resistance sensor and a threshold-switching memristor connected in series; the sensor serves as an adjustable resistor, so that the memristor works in a local active area to perform normal pulse distribution, the sensing environment information and the voltage division function are achieved, and the coding error of a sensing coding circuit caused by the difference between devices of the memristor when a plurality of sensing circuits are integrated is reduced. In addition, the parasitic capacitance of the memristor itself acts as the necessary capacitance in the memristive neuron circuit; through the design, the sensing neuron circuit is more compact, the integration density is greatly improved, the area is reduced, the hardware is saved, the energy consumption is reduced, and the energy efficiency is improved.

Description

A memristor-based sensing neuron circuit and its application

Technical field

The invention belongs to the field of intelligent sensing technology, and more specifically, relates to a memristor-based sensing neuron circuit and its application.

Background technique

With the continuous development of intelligent sensing computing, the construction of artificial intelligence robots has placed higher demands on hardware systems. For example, when robots work in some complex environments (autonomous driving, intelligent firefighting, deep-sea exploration), it is necessary to be able to perceive and process efficiently in real time.

However, in traditional architecture, the analog data collected by sensors are first converted into digital signals through analog-to-digital converters, then stored in memory and sent to computing units, resulting in high energy consumption and low efficiency. In comparison, neuromorphic perceptual computing systems inspired by biological senses have the advantages of high energy efficiency, strong robustness, high flexibility and low fault tolerance in processing sensory information. However, the current neuromorphic perception processing hardware technology is mainly based on traditional CMOS circuits. With the development of intelligent perception, a large amount of environmental information needs to be collected at the same time and converted into digital signals that can be used for processing, which will inevitably make the circuit more complex. . CMOS circuits have complex structures and poor scalability, which are not conducive to high-density scale integration; and hardware based on CMOS circuits requires a huge number of operations when processing complex multi-modal signals, resulting in very high power consumption; in addition In addition, in many working scenarios, large and multiple data transmissions are required between sensing, storage and processing, which limits the system response time.

Contents of the invention

In view of the above defects or improvement needs of the existing technology, the present invention provides a memristor-based sensing neuron circuit and application to solve the problem of poor integration and high energy consumption of existing neuromorphic sensing processing hardware technology. High technical issues.

In order to achieve the above object, in the first aspect, the present invention provides a memristor-based sensing neuron circuit, including: a sensing neuron array; each sensing neuron in the array converts the external environment information it perceives in parallel. Output pulse signals;

Each sensing neuron in the array includes a resistive sensor and a threshold transition memristor whose anode is connected to the resistive sensor; the other end of the sensor in the same row is connected to the same word line of the array for connecting to the power supply; The negative electrodes of the memristors in the same column are connected to the same bit line of the array and grounded; the initial state of the memristor is a high resistance state;

Among them, in the process of sensing external environmental information, the resistance of the resistive sensor changes with changes in the external environment, which in turn causes the voltage at both ends of the memristor connected to it to change; when the voltage at both ends of the memristor reaches its threshold voltage , the memristor undergoes a threshold transition to a low-resistance state, and begins to discharge, outputting pulses through the positive electrode, until the voltage across the memristor is lower than its holding voltage, and the memristor returns to a high-resistance state.

Further preferably, the resistance sensor is a pressure sensor, which is used to sense pressure changes in the external environment. Its resistance decreases as the pressure in the external environment increases, thereby increasing the pulse frequency output by the sensing neuron.

Further preferably, the material of the above-mentioned memristor is a metal-insulator transition phase change material, or a photosensitive resistive oxide material.

Further preferably, when the material of the memristor is a metal-insulator transition phase change material or a photosensitive resistive oxide material, the memristor is also used to sense temperature changes in the external environment, and its threshold voltage changes with the external environment. The temperature decreases as the temperature increases, which in turn causes the amplitude of the pulses output by the sensory neurons to decrease and the frequency to increase.

Further preferably, when the material of the above-mentioned memristor is a photosensitive resistive variable oxide material, the above-mentioned memristor is also used to sense changes in the intensity of light in the external environment, and its resistance value decreases as the intensity of light in the external environment increases, thereby making The pulse frequency output by the sensory neuron increases but the amplitude remains unchanged.

In a second aspect, the present invention provides a signal recognition circuit, including: a perceptual neuron circuit and a memristive neural network circuit; the perceptual neuron circuit is the perceptual neuron circuit provided in the first aspect of the present invention;

Among them, the memristive neural network circuit includes: a memristive input neuron circuit, a memristive synaptic array and a memristive output neuron circuit connected in sequence;

The memristive input neuron circuit includes a plurality of memristive input neurons, and each memristive input neuron is connected to the positive electrode of the memristor of each sensing neuron in the sensing neuron array in a one-to-one correspondence;

The memristive synapses in the memristive synapse array are non-volatile memristors;

The memristive output neuron circuit includes a plurality of memristive output neurons;

In the training mode, multiple sets of external environment information are preset, and the sensing neuron circuit is used to obtain the pulse signal under each set of external environment information and input it into the memristive neural network circuit for training;

In the recognition mode, the sensing neuron circuit is used to convert the external environment information it perceives into pulse signals, and output them to the memristive neural network circuit to identify the external environment information.

Further preferably, when the resistance sensor in the sensory neuron circuit is a pressure sensor, the above-mentioned external environment information includes external pressure information.

Further preferably, when the material of the memristor is a metal-insulator transition phase change material or a photosensitive resistive oxide material, the external environment information also includes external temperature information.

Further preferably, when the material of the memristor is a photosensitive resistive oxide material, the external environment information also includes external light signal information.

In a third aspect, the present invention provides an electronic chip, including the sensing neuron circuit provided by the first aspect of the present invention, or the signal recognition circuit provided by the second aspect of the present invention.

Generally speaking, through the above technical solutions conceived by the present invention, the following beneficial effects can be achieved:

1. The present invention provides a memristor-based sensing neuron circuit, which constructs an integrated sensing neuron array through word lines and bit lines. Each sensing neuron in the array converts the external environment information it perceives in parallel. To output pulse signals; each sensing neuron in the array includes a series-connected resistance sensor and a threshold transition memristor; the resistance sensor will sense external environmental signals and its resistance will change with the sensed environmental signals, so it can Acting as an adjustable resistor, the memristor works in a local active area to emit normal pulses, and plays a role in sensing environmental information and dividing voltage. It is an adjustable resistor with sensing capabilities; in addition, the resistance sensor also reduces the When integrating a sensing circuit, the coding error of the sensing encoding circuit is caused by the differences between the memristor devices. In addition, the parasitic capacitance of the memristor itself acts as a necessary capacitance in the memristive neuron circuit, playing an integral role to simulate changes in neuron membrane potential, without the need for additional external capacitance; through the above design, the circuit structure can be simplified, making the sensing nerve The element circuit is more compact, greatly improving the integration density, and the reduction in area and saving in hardware also reduces energy consumption and improves energy efficiency.

2. Further, in the sensory neuron circuit provided by the present invention, the material of the memristor is a phase change material of metal-insulator transition, or a light-sensitive resistive oxide material, which can process signals from external sensors. It can also sense ambient light/electricity/heat and other information to realize the coupling of its sensing information with the modal information of the sensor, achieving multi-modal perception. In addition, memristors based on these materials have fast switching speeds, low device energy consumption, and can convert the external information sensed by the sensor into pulse signals without the need for an analog-to-digital converter, which is beneficial to reducing the cost of multi-modal sensing neurons. reduce the energy consumption of the array part and improve calculation accuracy.

3. The present invention also provides a signal recognition circuit, including the perceptual neuron circuit and the memristive neural network circuit provided in the first aspect of the present invention; the perceptual neuron circuit is the perceptual neuron circuit provided in the first aspect of the present invention. ; Among them, the memristive input neuron circuit includes multiple memristive input neurons, each memristive input neuron is connected to the positive electrode of the memristor of each sensing neuron in the sensing neuron array in a one-to-one correspondence, and can sense in parallel at the same time. and process information from multiple signal points. At the same time, the signal output by the cross array can be directly transmitted to the memristive neural network circuit for identification calculation. This parallel perception and calculation mode greatly improves the efficiency of processing large-scale information and further improves the efficiency of processing large-scale information. Energy efficiency, reducing hardware area.

Description of drawings

Figure 1 is a schematic structural diagram of a memristor-based sensing neuron circuit provided by an embodiment of the present invention;

Figure 2 is an I-V characteristic curve diagram of a memristor used for pulse coding provided by an embodiment of the present invention;

Figure 3 is a schematic diagram of the specific structure of a sensory neuron provided by an embodiment of the present invention;

Figure 4 is a diagram illustrating the pulse emission characteristics of a sensory neuron under different resistive sensor resistance values R provided by an embodiment of the present invention; wherein (a)-(e) are respectively the pulse firing characteristics of the sensory neuron under different resistance sensor resistance values R. Schematic diagram of pulse signals; (f) is a statistical diagram of the frequency of pulse signals emitted by sensory neurons under different resistance sensor resistance values R;

Figure 5 is a schematic diagram of the I-V characteristic curve of the memristive device at different temperatures and a schematic structural diagram of the sensor neuron circuit provided by the embodiment of the present invention; (a) is the I-V characteristic curve of the memristive device at different temperatures; (b) ) is a schematic diagram of the specific circuit structure of the sensory neuron unit;

Figure 6 is a diagram of the output pulse characteristics of the sensing neuron under the same pressure and different temperatures in the embodiment of the present invention; wherein (a)-(d) are schematic diagrams of the output pulse signals of the sensing neuron under the same pressure and different temperatures;

Figure 7 is a characteristic diagram of the pulse amplitude and pulse frequency output by the sensory neuron provided by the embodiment of the present invention as a function of temperature and pressure; wherein, (a) is the change of the pulse amplitude and pulse frequency output by the sensory neuron provided by the embodiment of the present invention. Characteristic diagram of changes in temperature and pressure; (b) Characteristic diagram of changes in pulse frequency output by sensory neurons provided by embodiments of the present invention as a function of temperature and pressure;

Figure 8 is a diagram illustrating the application of the signal recognition circuit provided by the embodiment of the present invention to multi-modal signal sensing of temperature and pressure on the palm;

Figure 9 is a diagram showing the process of neuromorphic calculation of pulse signals by the memristive neural network circuit after sensing and the results thereof provided by the embodiment of the present invention; (a) is the processing of the pulse signal by the memristive neural network circuit after sensing. A schematic diagram of the process of performing neuromorphic calculations; (b) is a schematic structural diagram of a memristive neural network circuit; (c) is a curve of the recognition accuracy of the signal recognition circuit provided by the embodiment of the present invention as a function of the cycle.

Detailed ways

In order to make the purpose, technical solutions and advantages of the present invention more clear, 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 here are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

In order to achieve the above object, in the first aspect, as shown in Figure 1, the present invention provides a memristor-based sensing neuron circuit, including: a sensing neuron array; each sensing neuron in the array senses it in parallel The received external environment information is converted into pulse signals for output;

Each sensing neuron in the array includes a resistive sensor and a threshold transition memristor whose anode is connected to the resistive sensor; the other end of the sensor in the same row is connected to the same word line of the array for connecting to the power supply; The negative electrodes of the memristors in the same column are connected to the same bit line of the array and grounded; the initial state of the memristor is a high resistance state;

Among them, in the process of sensing external environmental information, the resistance value of the resistive sensor changes with changes in the external environment, which in turn causes the voltage at both ends of the memristor connected to it to change; when the voltage at both ends of the memristor reaches its threshold voltage , the memristor undergoes a threshold transition to a low-resistance state, and begins to discharge, outputting pulses through the positive electrode, until the voltage across the memristor is lower than its holding voltage, and the memristor returns to a high-resistance state.

The above-mentioned sensory neuron array is an integration of pulse-encoded compact circuits, and the processed output signal V _out is a pulse signal.

The above-mentioned sensing neuron array is a multi-modal sensing neuron cross array, including memristors connected according to a preset number of rows and columns, and sensors connected to the memristors in each row and column respectively. Sensors are used to sense changes in the external multi-modal environment and convert environmental information into electrical changes. The sensors here can be changed according to the required sensing information, that is, different sensors can be used. The unit that senses each specific row and column position in the neuron array (perception neuron) is a 1T1S neuron unit composed of a sensor and a memristor connected in series. The integrated sensing neuron array connects the memristors of each column in the array together through bit lines and connects the sensors in each row through word lines. The word lines and bit lines constitute the connection part of the row and column devices of the sensing neuron array. The sensory neuron array provided by the present invention can simultaneously perceive and process information from multiple signal points in parallel.

Each neuron unit in the sensing neuron array is composed of a memristor and a sensor; one end of the sensor is connected to the power source V _in , and the other end is connected to the positive terminal of the memristor; the positive terminal of the memristor is connected to the sensor, The negative terminal of the other end is connected to ground. Memristive neuron circuits often require a capacitor as an integrator to change the membrane potential. Generally, an external capacitor is used. One end of the capacitor is connected to the positive electrode of the memristor, and the other end is grounded. However, the present invention uses the parasitic function of the memristor. Capacitance C serves as the capacitance in the sensing neuron circuit in the array. Compact neuron circuits based on parasitic capacitance can increase the integration density of the array and reduce energy consumption. Figure 2 shows the IV characteristic curve of the memristor used for pulse coding in the present invention. The memristive device has good cycle uniformity. As can be seen from Figure 2, the memristor used should have the following characteristics:

(1) Having threshold voltage _Vth and holding voltage _Vh ;

(2) It has volatile threshold transition characteristics.

Specifically, when the input voltage gradually increases and exceeds the threshold voltage V _th of the device, the device will switch to a low resistance state; at this time, the input voltage is slowly reduced. When the voltage on the device is lower than the holding voltage V _hold , the device will Automatically restores to high impedance state.

Figure 3 shows a schematic diagram of the specific structure of the sensing neuron provided by the present invention (where C is the parasitic resistance of the memristor itself, which is located in the memristor, and is only schematically illustrated here). The circuit is composed of the above-mentioned resistance sensor and a memristor with threshold transition characteristics connected in series. The resistance sensor is set above the threshold transition memristor and connected to the positive electrode of the memristor. The positive voltage signal V _{out of the} memristor is as Output pulse signal.

The resistance sensor will sense external environmental signals and its resistance will change with the sensed environmental signals. Therefore, the resistance sensor can act as an adjustable resistor, allowing the memristor to work in the local active area and perform normal pulse emission to sense the environment. Information and voltage dividing function, it is an adjustable resistor with sensing ability; in addition, when facing the integration of multiple sensing circuits, the resistance sensor can reduce the encoding of the sensing encoding circuit caused by the differences between the memristor devices. error. In addition, the parasitic capacitance of the memristor itself acts as a necessary capacitance in the memristive neuron circuit, playing an integral role to simulate changes in neuron membrane potential, without the need for additional external capacitance; through the above design, the circuit structure can be simplified, making the neuron The circuit is more compact and the integration density is increased. In addition, the area reduction and hardware saving can also improve energy efficiency.

Specifically, the input voltage V _in first charges the parasitic capacitance C through the charging loop composed of the resistive sensor and the parasitic capacitance C of the memristor itself. As the capacitor voltage increases, the voltage across the memristor also continues to rise. High, when the threshold voltage V _th of the memristor is reached, the memristor undergoes a threshold transition and switches to a low resistance state. At this time, the current flowing through the memristor surges, and then the parasitic capacitance C passes through the loop formed by the memristor rapidly. Discharge, the voltage across the memristor continues to decrease until it is lower than the holding voltage V _h , and the memristor returns to the high resistance state. At this time, the current through the memristor returns to a state close to zero, and the parasitic capacitance C is discharged. At the end, another cycle of charging and discharging process begins under the stimulation of the input voltage V _in , so a series of voltage pulses V _out can be generated at the output end of the neuron circuit.

Figure 4 shows the pulse firing characteristics of sensory neurons under different resistive sensor resistance values R. It can be seen that, with other conditions remaining unchanged, as the adjustable resistance R increases, the pulse firing frequency of memristive neurons significantly decreases. The increase in R will directly lead to a decrease in charging current, so it will take longer to charge to the same threshold voltage and accumulate the same charge, thus increasing the integration time and resulting in a decrease in neuron firing frequency. This result reflects the influence of circuit parameters on the spike firing characteristics of memristive neurons, and also demonstrates the influence of the sensors in each sensory neuron of the sensory neuron array on pulse frequency encoding.

In summary, the present invention connects the sensors in the neuron array part according to the preset number of rows and the preset number of columns, and configures the threshold conversion memristor for the sensors in each row and connects them to form a cross array structure. It saves hardware usage, has good scalability, and has a simple circuit structure, which is conducive to high-density and large-scale array integration, and greatly improves integrability.

In an optional implementation, the above-mentioned resistance sensor is a pressure sensor, which is used to sense pressure changes in the external environment. Its resistance value decreases as the pressure in the external environment increases, thereby causing the pulse frequency output by the sensing neuron to increase. , while the amplitude remains basically unchanged.

In an optional implementation, the above-mentioned resistance sensor is a photosensitive sensor, which is used to sense changes in the light intensity of the external environment. Its resistance value decreases with the increase of the external light intensity, thereby causing the pulse output by the sensing neuron to The frequency increases while the amplitude remains essentially the same.

In an optional implementation, the above-mentioned resistance sensor is a curvature sensor, used to sense changes in the curvature of the object being measured. When the surface is curved, ductile deformation occurs, causing the resistance of the diaphragm to change, and then the resistance of the object being measured changes. Curvature changes are converted into electrical signal output.

In an optional implementation, the material of the above-mentioned memristor is a metal-insulator transition phase change material (for example: _VOX , _NbO , IGZO, perovskite materials, etc.), metal-insulator transition phase change materials or photosensitive resistive oxide materials as functional layers, the resistive switching characteristics of the device change with changes in ambient temperature; at this time, the above-mentioned memristor also It is used to sense temperature changes in the external environment. Its threshold voltage decreases as the temperature of the external environment increases, which in turn causes the amplitude of the pulses output by the sensing neurons to decrease and the frequency to increase. If a photosensitive oxide material is used, in addition to sensing temperature, it can also respond to light signals in the environment. The resistive switching characteristics of the device composed of the photosensitive resistive oxide material as a functional layer change with changes in ambient light intensity; At this time, when the material of the above-mentioned memristor is a photosensitive resistive variable oxide material, the above-mentioned memristor is also used to sense changes in the intensity of light in the external environment, and its resistance value decreases as the intensity of light in the external environment increases, thereby making the perception The pulse frequency output by the neuron increases, but the amplitude remains basically unchanged. Multi-modal sensing can be achieved through the cooperation of the above-mentioned memristors and sensors. In addition, the above-mentioned memristor material has fast switching speed, low device energy consumption, and can convert the external information sensed by the sensor into pulse signals without the need for an analog-to-digital converter, which is beneficial to reducing the energy of the sensing neuron array part. consumption and improve calculation accuracy.

In order to further illustrate the above-mentioned multi-modal sensing process, a detailed description is given below with reference to a specific embodiment:

The sensor in this embodiment is a pressure sensor, and the memristor is a phase change material whose resistive switching characteristics change with changes in ambient temperature. The resistive switching layer material is a metal-insulator transition such as VO _x or NbO _x . The material phase change Mainly a thermally induced phase change. Figure 5 is a schematic diagram of the IV characteristic curve of the memristive device at different temperatures and a schematic diagram of the sensing neuron circuit structure in the present invention; Figure (a) is the IV characteristic curve of the memristive device at different temperatures; Figure (b) It is a schematic diagram of the specific circuit structure of the sensory neuron unit. It can be seen from the graph (a) in Figure 5 that as the temperature increases, the threshold voltage V _th and the holding voltage V _h of the device both decrease, and V _th has a stronger temperature dependence than V _h . The decrease becomes more obvious as the temperature increases. Therefore, the hysteresis window of the memristor IV characteristic curve also decreases as the temperature increases. The resistive switching process of the memristive device in this embodiment is a complex process of electrothermal coupling. The threshold transition phenomenon is related to the positive feedback of current and temperature in the device. Therefore, when the ambient temperature increases, the resistive switching dynamic process of the device will accelerate. , the thermal feedback effect is strengthened, which reduces the voltage V _th required for threshold transition to occur. In addition, (b) in Figure 5 also shows the specific circuit structure of the sensory neuron unit in this embodiment. Specifically, the circuit is composed of the above-mentioned pressure sensor and a temperature-sensitive memristor with threshold transition characteristics connected in series, in which the positive voltage signal V _out of the memristor is used as the output pulse signal. The pressure sensor acts as an adjustable resistor to sense pressure and divide pressure; while the temperature-sensing memristor acts to sense temperature and process pressure signals and couple multi-modal information. In addition, the parasitic capacitance of the memristor itself acts as a memristor. The necessary capacitance in the neuron circuit plays an integral role in simulating changes in neuron membrane potential.

As the core processing unit, the memristor in the present invention can not only process the pressure signal of the external sensor, but also sense the environmental temperature information to achieve the coupling of its temperature information with other modal information. Figure 6 is a diagram showing the output pulse characteristics of sensing neurons under the same pressure and different temperatures in this embodiment. It can be seen that under the same pressure, as the temperature increases, the amplitude of the output pulse of the multimodal sensing neuron circuit decreases and the frequency increases. This is because the decrease in the device threshold voltage V _th caused by the increase in temperature will cause the integration process of the multi-modal sensing neuron circuit to become faster, thereby increasing the frequency of neuron pulses and at the same time reducing the amplitude of the output pulse. Therefore, the amplitude and emission frequency of the output pulse encode the environmental temperature information, which is also the key to realizing temperature sensing in the present invention. In order to more clearly understand the impact of temperature and pressure on the output pulses of neuron units in the multi-modal sensing neuron array of the present invention, Figure 7 further shows that the pulse amplitude and pulse frequency output by the sensing neurons in this embodiment change with temperature and pressure. changing characteristics diagram. It can be seen that at the same temperature, as the external pressure sensed on the sensor increases, the amplitude of the output pulses of the sensing neurons remains basically unchanged, while the firing frequency increases significantly. When the external pressure increases, the pressure sensor resistance R decreases, which causes the charging current to increase, so that the capacitor requires less time to accumulate the same charge. When the device threshold voltage is constant, the output voltage pulse of the sensing neuron is The amplitude remained essentially unchanged, while the frequency of firing increased significantly. Therefore, the frequency of pulses can encode the external pressure information sensed on the pressure sensor. To sum up, the pulse amplitude only decreases with the increase of temperature and remains basically unchanged with the increase of pressure. Therefore, the pulse amplitude cannot encode the multi-modal environment (pressure and temperature) in this embodiment; while the increase of temperature or pressure will This causes an increase in pulse emission frequency. Therefore, because the memristor can encode complex analog signals (temperature and pressure) in the environment into pulse signals of different frequencies, the present invention can realize pulse encoding of external multi-modal signals.

Based on this, the sensory neuron array in this embodiment can realize pulse encoding of external multi-modal signals, and encode complex analog signals in the environment into pulse signals of different frequencies, thereby realizing multi-modal fusion perception and calculation. Specifically, the sensors in the array can be changed according to the required sensing information, that is, different sensors can be used. As a core processing unit, memristor can not only process signals from external sensors, but also sense information such as ambient light/electricity/heat to achieve coupling of its perceived information with the modal information of the sensor. Compared with the existing technology, the present invention makes full use of the perceptual coding and compact structure characteristics of memristive neural circuits to design multi-modal perceptual arrays, effectively solving the problems of complex circuit integration, high delay, high energy consumption and other problems of traditional perceptual processing integrated systems. It saves hardware usage and has the advantages of parallel processing, high density, small area, and low energy consumption. It can be applied to edge intelligent sensing computing scenarios.

In a second aspect, the present invention provides a signal recognition circuit, including: a perceptual neuron circuit and a memristive neural network circuit; the perceptual neuron circuit is the perceptual neuron circuit provided in the first aspect of the present invention; the relevant technical solution is the same as this The sensory neuron circuit provided by the first aspect of the invention will not be described in detail here;

Among them, the memristive neural network circuit includes: a memristive input neuron circuit, a memristive synaptic array and a memristive output neuron circuit connected in sequence;

The memristive input neuron circuit includes a plurality of memristive input neurons, all of which are non-volatile memristors; each memristive input neuron and the positive electrode of the memristor of each sensing neuron in the sensing neuron array are one by one. Correspondingly connected;

The memristive synapses in the memristive synapse array are non-volatile memristors;

The memristive output neuron circuit includes multiple memristive output neurons, all of which are non-volatile memristors;

In the training mode, multiple sets of external environment information (such as set pressure and/or temperature) are preset, and the sensing neuron circuit is used to obtain the pulse signal under each set of external environment information and input it to the memristive neural network circuit. Training is performed (the training label is the corresponding external environment information);

In the recognition mode, the sensing neuron circuit is used to convert the external environment information it perceives into pulse signals, and output them to the memristive neural network circuit to identify and classify the external environment information.

It should be noted that the memristors used in the memristive neural network circuit are all non-volatile devices with multi-resistive state characteristics.

In an optional implementation, when the resistance sensor in the sensing neuron circuit is a pressure sensor, the above-mentioned external environment information includes external pressure information.

In an optional implementation, when the material of the memristor is a metal-insulator transition phase change material or a photosensitive resistive oxide material, the external environment information also includes external temperature information.

If a photosensitive oxide material is used, in addition to sensing temperature, it can also respond to light signals in the environment. At this time, the above-mentioned external environment information also includes external light signal information.

In a specific embodiment, FIG. 8 is a diagram illustrating the application of the signal recognition circuit provided by the present invention to multi-modal signal sensing of temperature and pressure on the palm. This embodiment applies the sensory neuron array to the perception of multi-modal signals of temperature and pressure on the palm and uses a memristive neural network circuit to calculate and identify them. By encoding the simultaneous multi-modal signals (temperature and pressure signals) on the palm into different voltage signals as inputs to the sensory neuron array, the sensory neuron array can sense and calculate these input signals and encode them into Pulse signals of different frequencies are output and used as input to the memristive neural network circuit for neuromorphic calculation and recognition. The color somewhere on the palm represents its corresponding normalized frequency intensity under the coupling condition of temperature and pressure, which in turn reflects the pressure and temperature on the palm. At this point, its perceptual encoding function is completed.

Then there is the neuromorphic computing part after perception, which is composed of a memristive neuron circuit (memristive input neuron circuit and memristive output neuron circuit) and a memristive synapse array. Each memristor in the memristive input neuron circuit Input neurons are connected to each row and column of the sensory neuron array. Figure 9 is a diagram showing the process of neuromorphic calculation of pulse signals by the memristive neural network circuit after sensing in the present invention and its results. This part is used to process the pulse signals with different frequencies output by the sensory neuron array and related to the multi-modal information on the palm, and perform identification and classification. A memristive neuromorphic computing system uses a memristor-based neural network. The morphological chip can receive the processed pulse signal and perform calculation and identification through the pulse neural network. In the memristive synaptic array in the memristive neuromorphic chip, the memristive device is a non-volatile device with multi-resistive state characteristics. It can be seen from the result display that the memristive neural network in this embodiment reaches a recognition accuracy of 97% after multiple training cycles, that is, it has a high-precision recognition effect on multi-modal information on the palm.

In summary, the present invention proposes a memristor-based sensing neuron circuit and a corresponding signal recognition circuit, which can encode the modal signals in the perceived environment into pulse signals of different frequencies and input them into the memristor neural network for processing. Identify calculations. Compared with traditional CMOS perception processing hardware, this invention cleverly utilizes the threshold transition characteristics of volatile memristors and the resistance switching characteristics affected by temperature to achieve multi-modal information processing on the palm without the need for high-cost analog-to-digital conversion. devices and complex control circuits, which greatly simplifies circuit design, reduces energy consumption, is conducive to high-density scale integration, has good application prospects, and promotes the development of neuromorphic sensing and computing. In addition, it is in line with modern Compared with some single memristor neuron circuits, the multi-modal sensing neuron array in the present invention has multiple sensing points and can sense large-scale information at the same time, further improving energy efficiency and reducing hardware area.

It is easy for those skilled in the art to 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 substitutions and improvements, etc., made within the spirit and principles of the present invention, All should be included in the protection scope of the present invention.

Claims (10)

1. A memristor-based sensing neuron circuit, characterized by comprising: a sensing neuron array; each sensing neuron in the array converts the external environment information it perceives into a pulse signal for output in parallel ; Each sensing neuron in the array includes a resistance sensor and a threshold transition memristor with an anode connected to the resistance sensor; the other end of the resistance sensor in the same row is connected to the same word line of the array is used to connect the power supply; the negative electrode of the memristor in the same column is connected to the same bit line of the array and is grounded; the initial state of the memristor is a high resistance state; Among them, in the process of sensing external environment information, the resistance value of the resistance sensor changes with changes in the external environment, which in turn causes the voltage at both ends of the memristor connected to it to change; when the voltage at both ends of the memristor reaches its threshold, When the voltage is applied, the memristor undergoes a threshold transition to a low-resistance state and begins to discharge, outputting pulses through the positive electrode until the voltage across the memristor is lower than its holding voltage, and the memristor returns to a high-resistance state. 2. The sensing neuron circuit according to claim 1, characterized in that the resistance sensor is a pressure sensor, used to sense pressure changes in the external environment, and its resistance decreases as the pressure in the external environment increases, thereby causing The frequency of pulses output by the sensory neurons increases. 3. The sensory neuron circuit according to claim 1 or 2, characterized in that the material of the memristor is a phase change material of metal-insulator transition, or a photosensitive resistive oxide material. 4. The sensory neuron circuit according to claim 3, characterized in that when the material of the memristor is a phase change material of metal-insulator transition or a photosensitive resistive oxide material, the memristor also It is used to sense temperature changes in the external environment. Its threshold voltage decreases as the temperature of the external environment increases, thereby causing the pulse amplitude output by the sensing neuron to decrease and the frequency to increase. 5. The sensory neuron circuit according to claim 3, characterized in that when the material of the memristor is a photosensitive resistive oxide material, the memristor is also used to sense changes in the light intensity of the external environment, Its resistance decreases as the light intensity of the external environment increases, thereby causing the pulse frequency output by the sensing neuron to increase and the amplitude to remain unchanged. 6. A signal recognition circuit, characterized in that it includes: a perceptual neuron circuit and a memristive neural network circuit; the perceptual neuron circuit is the perceptual neuron circuit according to any one of claims 1 to 5; Wherein, the memristive neural network circuit includes: a memristive input neuron circuit, a memristive synaptic array and a memristive output neuron circuit connected in sequence; The memristive input neuron circuit includes a plurality of memristive input neurons, and each memristive input neuron is connected to the positive electrode of the memristor of each sensing neuron in the sensing neuron array in a one-to-one correspondence; The memristive synapses in the memristive synapse array are non-volatile memristors; The memristive output neuron circuit includes a plurality of memristive output neurons; In the training mode, multiple sets of external environment information are preset, and the sensing neuron circuit is used to obtain the pulse signal under each set of external environment information and input it into the memristive neural network circuit for training; In the identification mode, the sensing neuron circuit is used to convert the external environment information it perceives into pulse signals, and output them to the memristive neural network circuit to identify the external environment information. 7. The signal recognition circuit according to claim 6, wherein when the resistance sensor in the sensing neuron circuit is a pressure sensor, the external environment information includes external pressure information. 8. The signal recognition circuit according to claim 6 or 7, characterized in that when the material of the memristor is a phase change material of metal-insulator transition or a photosensitive resistive oxide material, the external environment information Also includes outside temperature information. 9. The signal recognition circuit according to claim 6 or 7, characterized in that when the material of the memristor is a photosensitive resistive oxide material, the external environment information also includes external light signal information. 10. An electronic chip, characterized by comprising the sensing neuron circuit described in any one of claims 1-5, or the signal recognition circuit described in any one of claims 6-9.