CN117669676A - Memristor-based sensing neuron circuit and application - Google Patents
Memristor-based sensing neuron circuit and application Download PDFInfo
- Publication number
- CN117669676A CN117669676A CN202311555274.2A CN202311555274A CN117669676A CN 117669676 A CN117669676 A CN 117669676A CN 202311555274 A CN202311555274 A CN 202311555274A CN 117669676 A CN117669676 A CN 117669676A
- Authority
- CN
- China
- Prior art keywords
- memristor
- sensing
- neuron
- circuit
- memristive
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 210000002569 neuron Anatomy 0.000 title claims abstract description 130
- 239000000463 material Substances 0.000 claims description 34
- 230000008859 change Effects 0.000 claims description 31
- 238000013528 artificial neural network Methods 0.000 claims description 23
- 230000007704 transition Effects 0.000 claims description 23
- 210000002364 input neuron Anatomy 0.000 claims description 18
- 238000000034 method Methods 0.000 claims description 14
- 230000008569 process Effects 0.000 claims description 12
- 230000007423 decrease Effects 0.000 claims description 11
- 239000012212 insulator Substances 0.000 claims description 11
- 239000012782 phase change material Substances 0.000 claims description 11
- 210000004205 output neuron Anatomy 0.000 claims description 10
- 238000005286 illumination Methods 0.000 claims description 9
- 238000012549 training Methods 0.000 claims description 8
- 210000000225 synapse Anatomy 0.000 claims description 7
- 210000001044 sensory neuron Anatomy 0.000 claims description 6
- 230000007613 environmental effect Effects 0.000 claims description 4
- 230000000946 synaptic effect Effects 0.000 claims description 4
- 230000009466 transformation Effects 0.000 claims description 3
- 230000003287 optical effect Effects 0.000 claims description 2
- 230000010354 integration Effects 0.000 abstract description 11
- 230000003071 parasitic effect Effects 0.000 abstract description 11
- 238000005265 energy consumption Methods 0.000 abstract description 10
- 238000013461 design Methods 0.000 abstract description 4
- 238000010586 diagram Methods 0.000 description 25
- 238000012545 processing Methods 0.000 description 14
- 238000004364 calculation method Methods 0.000 description 13
- 239000003990 capacitor Substances 0.000 description 10
- 230000008447 perception Effects 0.000 description 10
- 230000008878 coupling Effects 0.000 description 6
- 238000010168 coupling process Methods 0.000 description 6
- 238000005859 coupling reaction Methods 0.000 description 6
- 230000008901 benefit Effects 0.000 description 5
- 210000005036 nerve Anatomy 0.000 description 5
- 230000009286 beneficial effect Effects 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 238000010304 firing Methods 0.000 description 4
- 239000012528 membrane Substances 0.000 description 4
- 238000011161 development Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 230000009467 reduction Effects 0.000 description 3
- 230000001953 sensory effect Effects 0.000 description 3
- 230000005611 electricity Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 239000002346 layers by function Substances 0.000 description 2
- 230000001537 neural effect Effects 0.000 description 2
- 238000010276 construction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000005284 excitation Effects 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 239000010410 layer Substances 0.000 description 1
- 238000010297 mechanical methods and process Methods 0.000 description 1
- 230000005226 mechanical processes and functions Effects 0.000 description 1
- 239000002184 metal Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 230000000630 rising effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
- G06N3/063—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using electronic means
- G06N3/065—Analogue means
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Theoretical Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Neurology (AREA)
- Testing Or Calibration Of Command Recording Devices (AREA)
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
Technical Field
The invention belongs to the technical field of intelligent perception, and particularly relates to a perception neuron circuit based on a memristor and application thereof.
Background
With the continuous development of intelligent perception calculation, the construction of an artificial intelligent robot puts higher demands on a hardware system. For example, when robots are operated in some complex environments (autopilot, intelligent fire, deep sea exploration), it is necessary to be able to sense and handle efficiently in real time.
However, in the conventional architecture, analog data collected by the sensor is first converted into a digital signal by an analog-to-digital converter, then stored in a memory, and transmitted to a computing unit, resulting in high power consumption and low efficiency. In contrast, neuromorphic sensory-inspired neuromorphic sensory computing systems have the advantages of high energy efficiency, strong robustness, high flexibility, and low fault tolerance in processing sensory information. However, the current neuromorphic sensing processing hardware technology is mainly implemented based on a traditional CMOS circuit, and with the development of intelligent sensing, a large amount of environmental information needs to be collected and converted into a digital signal which can be used for processing, which tends to make the circuit more complex. The CMOS circuit has a complex structure and poor scalability, and is not beneficial to high-density scale integration; the hardware based on the CMOS circuit needs extremely large operation times when processing complex multi-mode signals, so that extremely high power consumption is caused; in addition, in many operating scenarios, the large and multiple data transfers required between sensing, storage and processing limit their system response time.
Disclosure of Invention
Aiming at the defects or improvement demands of the prior art, the invention provides a sensing neuron circuit based on a memristor and application thereof, which are used for solving the technical problems that the existing neuromorphic sensing processing hardware technology cannot be relatively poor in integration level and relatively high in energy consumption.
To achieve the above object, in a first aspect, the present invention provides a memristor-based sensing neuron circuit, including: an array of sensory neurons; each sensing neuron in the array converts the sensed external environment information into a pulse signal in parallel for output;
each sensing neuron in the array comprises a resistance sensor and a threshold value transformation type memristor with a positive electrode connected with the resistance sensor; the other ends of the sensors in the same row are connected to the same word line of the array and are used for connecting a power supply; the cathodes of memristors on 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;
in the process of sensing external environment information, the resistance value of the resistance sensor changes along with the change of the external environment, so that the voltage at two ends of a memristor connected with the resistance sensor changes; when the voltage at the two ends of the memristor reaches the threshold voltage, the memristor is subjected to threshold transition to be in a low-resistance state, and discharge is started, pulses are output through the positive electrode until the voltage at the two ends of the memristor is lower than the holding voltage of the memristor, and the memristor returns to be in a high-resistance state.
Further preferably, the resistance sensor is a pressure sensor, and is configured to sense a pressure change of an external environment, and the resistance value of the resistance sensor decreases with an increase of the external environment pressure, so that the pulse frequency output by the sensing neuron increases.
Further preferably, the material of the memristor is a phase change material of metal-insulator transition, or a photoresistance-variable oxide material.
Further preferably, when the material of the memristor is a phase change material of metal-insulator transition or a photoresistance-variable oxide material, the memristor is further used for sensing temperature change of an external environment, and the threshold voltage of the memristor is reduced along with the rise of the temperature of the external environment, so that the pulse amplitude of the output of the sensing neuron is reduced and the frequency is increased.
Further preferably, when the memristor is made of a photoresistor oxide material, the memristor is further used for sensing the change of the illumination intensity of the external environment, and the resistance value of the memristor is reduced along with the enhancement of the illumination intensity of the external environment, so that the pulse frequency output by the sensing neuron is increased and the amplitude is unchanged.
In a second aspect, the present invention provides a signal recognition circuit comprising: a sensing neuron circuit and a memristive neural network circuit; the sensing neuron circuit is provided in the first aspect of the invention;
wherein, memristive neural network circuit includes: the memristive input neuron circuit, the memristive synaptic array and the memristive output neuron circuit are sequentially connected;
the memristive input neuron circuit comprises a plurality of memristive input neurons, and each memristive input neuron is correspondingly connected with the positive electrode of the memristor of each sensing neuron in the sensing neuron array one by one;
memristive synapses in the memristive synapse array are non-volatile memristors;
the memristive output neuron circuit includes a plurality of memristive output neurons;
in a training mode, presetting a plurality of groups of external environment information, wherein a sensing neuron circuit is used for acquiring pulse signals under each group of external environment information and inputting the pulse signals into a memristor neural network circuit for training;
in the recognition mode, the sensing neuron circuit is used for converting the sensed external environment information into a pulse signal and outputting the pulse signal to the memristor neural network circuit so as to recognize the external environment information.
Further preferably, when the resistance sensor in the sensory neuron circuit is a pressure sensor, the 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 photoresistance oxide material, the external environment information further includes external temperature information.
Further preferably, when the material of the memristor is a photoresistance oxide material, the external environment information further includes external optical signal information.
In a third aspect, the present invention provides an electronic chip comprising the sensing neuron circuit provided in the first aspect of the present invention, or the signal recognition circuit provided in the second aspect of the present invention.
In general, through the above technical solutions conceived by the present invention, the following beneficial effects can be obtained:
1. the invention provides a sense neuron circuit based on a memristor, which constructs an integrated sense neuron array through word lines and bit lines, and each sense neuron in the array converts the sensed external environment information into a pulse signal in parallel for output; each sensing neuron in the array comprises a resistance sensor and a threshold-switching memristor connected in series; the resistance sensor senses external environment signals and the resistance of the resistance sensor changes along with the sensed environment signals, so that the resistance sensor can serve as an adjustable resistance, the memristor works in a local active area to perform normal pulse emission, and the resistance sensor plays roles of sensing environment information and voltage division and is an adjustable resistance with sensing capability; in addition, the resistance sensor also reduces the coding error of the sensing coding circuit caused by the device-to-device variability of the memristor when a plurality of sensing circuits are integrated. In addition, the parasitic capacitance of the memristor serves as a necessary capacitance in a memristor neuron circuit, and plays an integral role in simulating the change of neuron membrane potential without additionally connecting with a capacitor; the design can simplify the circuit structure, so that 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.
2. Furthermore, the sensing neuron circuit provided by the invention has the advantages that the memristor is made of a phase-change material converted from a metal insulator or a light-sensitive resistive oxide material, signals of an external sensor can be processed, and information such as ambient light/electricity/heat and the like can be sensed, so that the coupling of sensing information and modal information of the sensor is realized, and multi-modal sensing is realized. Besides, the memristor based on the materials is high in switching speed and low in device energy consumption, external information perceived by a sensor can be converted into pulse signals without an analog-to-digital converter, and the method is beneficial to reducing the energy consumption of a multi-mode perception neuron array part and improving the calculation accuracy.
3. The invention also provides a signal identification circuit which comprises the sensing neuron circuit and the memristive neural network circuit provided by the first aspect of the invention; the sensing neuron circuit is provided in the first aspect of the invention; the memristor input neuron circuit comprises a plurality of memristor input neurons, each memristor input neuron is connected with the positive electrode of the memristor of each sensing neuron in the sensing neuron array in a one-to-one correspondence manner, information of a plurality of signal points can be sensed and processed in parallel, meanwhile, signals output by the cross array can be directly transmitted to the memristor neural network circuit for identification calculation, and the mode of parallel sensing and calculation greatly improves the efficiency of processing large-scale information, further improves the energy efficiency and reduces the hardware area.
Drawings
FIG. 1 is a schematic diagram of a sensing neuron circuit based on a memristor according to an embodiment of the present disclosure;
FIG. 2 is a graph of memristor I-V characteristics for pulse encoding provided by an embodiment of the present disclosure;
FIG. 3 is a schematic diagram of a specific structure of a sensing neuron according to an embodiment of the present invention;
FIG. 4 is a diagram showing the pulse emission characteristics of a sensing neuron according to an embodiment of the present invention under different resistance values R of a resistive sensor; wherein, (a) - (e) are respectively schematic diagrams of pulse signals emitted by sensing neurons under different resistance values R of the resistance sensor; (f) A frequency statistical diagram of pulse signals emitted by sensing neurons under different resistance values R of the resistance sensors;
FIG. 5 is a schematic diagram of I-V characteristic curves and sensing neuron circuit structures of a memristor device provided by the embodiments of the present disclosure at different temperatures; wherein, (a) is an I-V characteristic curve diagram of the memristor at different temperatures; (b) A specific circuit structure diagram of the sensing neuron unit;
FIG. 6 is a graph of output pulse characteristics of sensing neurons at different temperatures at the same pressure in an embodiment of the invention; wherein, (a) - (d) are respectively output pulse signal diagrams of sensing neurons under the same pressure and different temperatures;
FIG. 7 is a graph of pulse amplitude and pulse frequency versus temperature and pressure for a sensory neuron output provided by an embodiment of the present invention; wherein, (a) is a characteristic diagram of the change of the pulse amplitude of the sensing neuron output with the temperature and the pressure provided by the embodiment of the invention; (b) The change characteristic diagram of the pulse frequency output by the sensing neuron according to the temperature and the pressure is provided for the embodiment of the invention;
FIG. 8 is a schematic diagram showing the application of the signal recognition circuit provided by the embodiment of the invention to multi-mode signal sensing of temperature and pressure on the palm;
FIG. 9 is a diagram showing the process of performing neuromorphic computation on a pulse signal by a perceived memristive neural network circuit and a resulting representation thereof, provided by an embodiment of the present invention; the method comprises the steps of (a) performing nerve morphology calculation on pulse signals by a perceived memristive neural network circuit; (b) is a structural schematic diagram of a memristive neural network circuit; (c) The signal identification circuit provided by the embodiment of the invention has the characteristic that the identification accuracy rate of the signal identification circuit changes along with the period.
Detailed Description
The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention. In addition, the technical features of the embodiments of the present invention described below may be combined with each other as long as they do not collide with each other.
To achieve the above object, in a first aspect, as shown in fig. 1, the present invention provides a memristor-based sensing neuron circuit, including: an array of sensory neurons; each sensing neuron in the array converts the sensed external environment information into a pulse signal in parallel for output;
each sensing neuron in the array comprises a resistance sensor and a threshold value transformation type memristor with a positive electrode connected with the resistance sensor; the other ends of the sensors in the same row are connected to the same word line of the array and are used for connecting a power supply; the cathodes of memristors on 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;
in the process of sensing external environment information, the resistance value of the resistance sensor changes along with the change of the external environment, so that the voltage at two ends of a memristor connected with the resistance sensor changes; when the voltage at the two ends of the memristor reaches the threshold voltage, the memristor is subjected to threshold transition to be in a low-resistance state, and discharge is started, pulses are output through the positive electrode until the voltage at the two ends of the memristor is lower than the holding voltage of the memristor, and the memristor returns to be in a high-resistance state.
The above-mentioned sensing neuron array is an integration of pulse-coded compact circuit, and the processed output signal V out Is a pulse signal.
The sensing neuron array is a multi-mode sensing neuron cross array and comprises memristors connected according to preset rows and columns, and sensors respectively connected with the memristors of each row and each column. The sensor is used for sensing the change of the external multi-mode environment and converting the environment information into an electrical change quantity. The sensor herein can be changed according to the desired sensing information, i.e. different sensors can be used. The unit of each specific row and column position (sensing neuron) in the sensing neuron array is a 1T1S neuron unit formed by connecting a sensor and a memristor in series. The integrated sensing neuron array connects the memristors of each column in the array together by bit lines and connects the sensors of each row by word lines, the word lines and bit lines constituting the connection portions of the row-column devices of the sensing neuron array. The sensing neuron array provided by the invention can sense and process information of a plurality of signal points simultaneously in parallel.
Each neuron unit in the sensing neuron array is composed of a memristor and a sensor; wherein, one end of the sensor is connected with a power supply V in The other end of the memristor is connected with the positive electrode of the memristor; the positive electrode end of the memristor is connected to the sensor, and the negative electrode of the memristor is grounded. In the memristor neuron circuit, a capacitor is often needed to be used as an integrator to change the membrane potential, an external capacitor is generally adopted, one end of the capacitor is connected with the positive electrode of the memristor, and the other end of the capacitor is grounded. The memristor I-V characteristic diagram for pulse coding in the present invention is shown in fig. 2. The memristive device has good cycle uniformity. As can be seen from fig. 2, the memristors used should possess the following characteristics:
(1) Has a threshold voltage V th And a holding voltage V h ；
(2) Has a volatile threshold transition characteristic.
Specifically, as the input voltage increases gradually, the threshold voltage V of the device is exceeded th When the device is in the low resistance state, the device is switched to the low resistance state; at this time, the input voltage is gradually reduced when the voltage across the device is lower than the holding voltage V hold When the device is in the high resistance state, the device is automatically restored.
Fig. 3 is a schematic diagram of a specific structure of a sensing neuron according to the present invention (where C is a parasitic resistance of a memristor, which is located in the memristor, and is only illustrated herein). The circuit is formed by connecting the resistance sensor and the memristor with the threshold transition characteristic in series, wherein the resistance sensor is arranged above the threshold transition memristor and is connected with the positive electrode of the memristor, and the positive electrode voltage signal V of the memristor out As an output pulse signal.
The resistance sensor senses external environment signals and the resistance of the resistance sensor changes along with the sensed environment signals, so that the resistance sensor can serve as an adjustable resistance, the memristor works in a local active area to perform normal pulse emission, the effect of sensing environment information and voltage division is achieved, and the resistance sensor is an adjustable resistance with sensing capability; in addition, the resistance sensor can reduce coding errors of the sensing coding circuit caused by device-to-device variability of the memristor when being integrated in the face of a plurality of sensing circuits. In addition, the parasitic capacitance of the memristor serves as a necessary capacitance in a memristor neuron circuit, and plays an integral role in simulating the change of neuron membrane potential without additionally connecting with a capacitor; through the design, the circuit structure can be simplified, so that the neuron circuit is more compact, the integration density is improved, and in addition, the reduction of the area and the saving of hardware can also improve the energy efficiency.
Specifically, the input voltage V in Firstly, a parasitic capacitor C is charged through a charging loop formed by a resistance sensor and the parasitic capacitor C of the memristor, along with the rising of the capacitor voltage, the voltage at two ends of the memristor is also continuously increased, and when the threshold voltage V of the memristor is reached th When the memristor is switched into a low-resistance state by threshold transition, current flowing through the memristor at the momentThe parasitic capacitance C is rapidly discharged through a loop formed by the parasitic capacitance C and the memristor, and the voltage at two ends of the memristor is continuously reduced until the parasitic capacitance C is lower than the holding voltage V h The memristor returns to a high-resistance state, the current passing through the memristor returns to a state close to zero, the discharge of the parasitic capacitor C is ended, and the voltage V is input in A charge-discharge process is started for another period under the excitation of (a) so as to generate a series of voltage pulses V at the output end of the neuron circuit out 。
The sensing neuron pulse emission characteristic diagram under different resistance values R of the resistance sensor is shown in fig. 4. It can be seen that the pulsing frequency of memristive neurons decreases significantly as the adjustable resistance R increases, with other conditions unchanged. An increase in R directly results in a decrease in charging current, so that it takes longer to charge to the same threshold voltage and accumulate the same charge, thus increasing the integration time, resulting in a decrease in the firing frequency of the neuron. This result demonstrates the impact of circuit parameters on the memristive neuron pulsing characteristics, while demonstrating the impact of the sensor in each of the sensing neurons of the sensing neuron array on the pulse frequency encoding.
In summary, the invention connects the sensors in the neuron array part according to the preset number of rows and the preset number of columns, configures the threshold transition memristors for the sensors in each row and connects the threshold transition memristors to form a cross array structure, saves the use amount of hardware, has good scalability and simple circuit structure, is beneficial to high-density and large-scale array integration, and greatly improves the integrability.
In an optional implementation manner, the resistance sensor is a pressure sensor, and is used for sensing pressure change of an external environment, and the resistance value of the resistance sensor is reduced along with the increase of the pressure of the external environment, so that the pulse frequency output by the sensing neuron is increased, and the amplitude of the pulse frequency is kept unchanged basically.
In an optional implementation manner, the resistance sensor is a photosensitive sensor, and is used for sensing the change of the illumination intensity of the external environment, and the resistance value of the resistance sensor is reduced along with the enhancement of the external illumination intensity, so that the pulse frequency output by the sensing neuron is increased, and the amplitude of the pulse frequency is basically kept unchanged.
In an alternative embodiment, the resistance sensor is a curvature sensor, and is used for sensing curvature change of the measured object, when the surface is bent, toughness deformation is generated, so that the resistance of the diaphragm is changed, and the curvature change of the measured object is converted into an electrical signal to be output.
In an alternative embodiment, the memristor material is a metal-insulator transition phase change material (e.g., VO X 、NbO X Etc.), or a light-sensitive photoresistive oxide material (e.g.: ITO, IGZO, perovskite materials, etc.), the resistance characteristics of a device composed of a metal-insulator transition phase change material or a photoresistance oxide material as a functional layer change with changes in ambient temperature; at this time, the memristor is further used for sensing the temperature change of the external environment, and the threshold voltage of the memristor is reduced along with the rise of the temperature of the external environment, so that the pulse amplitude output by the sensing neuron is reduced and the frequency is increased. If a photosensitive oxide material is used, the photosensitive oxide material can also respond to light signals in the environment besides sensing temperature, and the resistance change characteristic of a device formed by taking the photosensitive resistance change oxide material as a functional layer changes along with the change of the illumination intensity of the environment; at this time, when the memristor is made of a photoresistance oxide material, the memristor is further used for sensing the change of the illumination intensity of the external environment, and the resistance value of the memristor is reduced along with the enhancement of the illumination intensity of the external environment, so that the pulse frequency output by the sensing neuron is increased, and the amplitude of the pulse frequency is basically kept unchanged. The multi-mode sensing can be realized through the cooperation of the memristor and the sensor. In addition, the memristor material has high switching speed and low device energy consumption, and external information perceived by a sensor can be converted into a pulse signal without an analog-to-digital converter, so that the energy consumption of a perception neuron array part is reduced, and the calculation accuracy is improved.
To further illustrate the above multi-modal sensing process, the following detailed description is provided in connection with one embodiment:
the sensor in the embodiment selects a pressure sensor, and the memristor selects a device with resistance change characteristics along with the change of the ambient temperatureThe material of the variable resistance layer is VO x Or NbO x Phase change materials with metal-insulator transition, and the phase change of the materials is mainly a thermally induced phase change. FIG. 5 is a schematic diagram of the I-V characteristic diagram and the sensing neuron circuit structure of the memristor device of the present disclosure at different temperatures; wherein, figure (a) is an I-V characteristic diagram of a memristive device at different temperatures; fig. (b) is a schematic diagram of a specific circuit structure of the sensing neuron unit. As can be seen from the graph (a) in FIG. 5, the threshold voltage V of the device increases with increasing temperature th And a holding voltage V h Are all reduced, and V th Ratio V h The temperature dependence is stronger, and the reduction along with the temperature rise is more obvious, so that the hysteresis window of the memristor I-V characteristic curve is also continuously reduced along with the temperature rise. The resistive switching process of the memristor device in this embodiment is a complex process of electrothermal coupling, and the threshold transition phenomenon is related to positive feedback of current and temperature in the device, so that when the ambient temperature increases, the resistive switching mechanical process of the device will accelerate, the thermal feedback effect will be enhanced, and the voltage V required for threshold transition will be reduced th . In addition, the specific circuit configuration of the sensing neuron unit in this embodiment is also shown in the (b) diagram in fig. 5. Specifically, the circuit is formed by connecting the pressure sensor and the temperature sensing memristor with threshold transition characteristics in series, wherein the positive voltage signal V of the memristor out As an output pulse signal. The pressure sensor is used as an adjustable resistor to sense pressure and divide pressure; the temperature sensing memristor has the functions of sensing temperature and processing pressure signals and coupling multi-mode information, and in addition, parasitic capacitance of the memristor serves as necessary capacitance in a memristor neuron circuit, and the temperature sensing memristor has the function of integrating to simulate change of neuron membrane potential.
The memristor is used as a core processing unit, so that the pressure signal of the external sensor can be processed, and the environmental temperature information can be sensed, so that the coupling of the temperature information and other modal information can be realized. Fig. 6 is a graph showing the output pulse characteristics of the sensing neurons at different temperatures under the same pressure in the present embodiment. It can be seen that the multi-modal sensory nerves are activated with increasing temperature at the same pressureThe amplitude of the output pulse of the meta-circuit decreases and the frequency increases. This is because of the device threshold voltage V caused by the temperature rise th The reduction of the frequency can lead to the fast integration process of the multi-mode sensing neuron circuit, further increase the frequency of the neuron issuing pulse, and simultaneously reduce the amplitude of the output pulse, so that the amplitude and the issuing frequency of the output pulse encode the environmental temperature information, which is also the key point of realizing temperature sensing. In order to more clearly recognize the effect of temperature and pressure on the output pulses of the neuron units in the multi-modal sensing neuron array of the present invention, fig. 7 further shows a graph of the pulse amplitude and pulse frequency of the sensing neuron output versus temperature and pressure in this embodiment. It can be seen that at the same temperature, the amplitude of the output pulses of the sensing neurons is substantially unchanged, while the firing frequency increases significantly, with an increase in the external pressure perceived on the sensor. When the external pressure increases, the pressure sensor resistance R decreases, which results in an increase in the charging current, so that the capacitance requires less time to accumulate the same charge, and in the case of a certain device threshold voltage, the amplitude of the output voltage pulse of the sensing neuron remains substantially unchanged, while the firing frequency increases significantly. Thus, the firing frequency of the pulses may encode ambient pressure information sensed at the pressure sensor. In summary, the pulse amplitude only decreases with increasing temperature and remains substantially unchanged with increasing pressure, so the pulse amplitude cannot encode the multi-modal environment (pressure and temperature) in this embodiment; the increase of temperature or pressure can cause the increase of the pulse release frequency, so that the memristor can realize the pulse coding of external multi-mode signals by coding complex analog signals (temperature and pressure) in the environment into pulse signals with different frequencies.
Based on the above, the sensing neuron array in the embodiment can realize pulse coding of external multi-mode signals, and encode complex analog signals in the environment into pulse signals with different frequencies, so as to realize multi-mode fusion sensing and calculation. In particular, the sensors in the array can be varied according to the desired sensing information, i.e. different sensors can be used. The memristor is used as a core processing unit, not only can process signals of an external sensor, but also can sense information such as ambient light/electricity/heat and the like so as to realize the coupling of sensing information and modal information of the sensor. Compared with the prior art, the multi-mode sensing array is designed by fully utilizing the sensing coding and compact structure characteristics of the memristor neural circuit, the problems of complex circuit integration, high delay, high energy consumption and the like of the traditional sensing processing integrated system are effectively solved, the hardware usage amount is saved, and the multi-mode sensing array has the advantages of parallel processing, high density, small area, low energy consumption and the like, and can be applied to scenes of intelligent sensing calculation of edges.
In a second aspect, the present invention provides a signal recognition circuit comprising: a sensing neuron circuit and a memristive neural network circuit; the sensing neuron circuit is provided in the first aspect of the invention; the related technical solution is the same as the sensing neuron circuit provided in the first aspect of the present invention, and will not be described herein;
wherein, memristive neural network circuit includes: the memristive input neuron circuit, the memristive synaptic array and the memristive output neuron circuit are sequentially connected;
the memristive input neuron circuit comprises a plurality of memristive input neurons which are all nonvolatile memristors; each memristor input neuron is correspondingly connected with the positive electrode of each memristor of each sensing neuron in the sensing neuron array one by one;
memristive synapses in the memristive synapse array are non-volatile memristors;
the memristive output neuron circuit comprises a plurality of memristive output neurons which are all nonvolatile memristors;
in a training mode, presetting a plurality of groups of external environment information (such as set pressure and/or temperature), wherein a sensing neuron circuit is used for acquiring pulse signals under each group of external environment information and inputting the pulse signals into a memristor neural network circuit for training (a training label is corresponding external environment information);
in the recognition mode, the sensing neuron circuit is used for converting the sensed external environment information into a pulse signal and outputting the pulse signal to the memristor neural network circuit so as to recognize and classify the external environment information.
It should be noted that, memristors used in the memristive neural network circuit are all nonvolatile devices with multi-resistance state characteristics.
In an alternative embodiment, when the resistive sensor in the sensing neuron circuit is a pressure sensor, the external environment information includes external pressure information.
In an alternative embodiment, when the material of the memristor is a phase change material of metal-insulator transition or a photoresistance-variable oxide material, the external environment information further includes external temperature information.
If a photosensitive oxide material is used, the sensor can respond to the light signal in the environment besides sensing the temperature, and the external environment information also comprises external light signal information.
In a specific embodiment, fig. 8 is a schematic diagram showing the application of the signal recognition circuit provided by the invention to multi-mode signal sensing of temperature and pressure on the palm. The embodiment applies the sensing neuron array to sensing of temperature and pressure multi-mode signals on the palm and calculates and identifies the signals by using the memristive neural network circuit. By encoding the multi-modal signals (temperature and pressure signals) that are present simultaneously on the palm as different voltage signals as inputs to the sensing neuron array, the sensing neuron array can sense and calculate these input signals and encode them as pulse signals with different frequencies for output and as inputs to the memristive neural network circuit for neuromorphic calculation and recognition. The color of a certain position on the palm represents the corresponding normalized frequency intensity under the condition of temperature and pressure coupling, and then the pressure and the temperature of the certain position on the palm are reflected, so that the sensing coding function of the palm is completed.
The perceived nerve morphology calculation part is composed of a memristor neuron circuit (a memristor input neuron circuit and a memristor output neuron circuit) and a memristor synapse array, wherein each memristor input neuron in the memristor input neuron circuit is connected with each row and each column in the perceived neuron array. FIG. 9 is a diagram showing the process of neural morphology calculation of pulse signals by the perceived memristive neural network circuit and the result thereof. The part is used for processing pulse signals with different frequencies, which are output by the sensing neuron array and are related to multi-mode information on the palm, and carrying out recognition and classification. The memristive synaptic array in the memristive nerve morphology chip is characterized in that the memristive device is a nonvolatile device with multi-resistance state characteristics. As can be seen from the result display diagram, the memristive neural network in the embodiment has the recognition accuracy reaching 97% after a plurality of training periods, namely has a high-accuracy recognition effect on the multi-mode information on the palm.
In summary, the invention provides a memristor-based sensing neuron circuit and a corresponding signal recognition circuit, which can encode modal signals in a sensed environment into pulse signals with different frequencies and input the pulse signals into a memristor neural network for recognition calculation. Compared with the traditional CMOS perception processing hardware, the multi-mode perception neuron array has the advantages that the threshold transition characteristic of the volatile memristor and the resistance change characteristic influenced by temperature are skillfully utilized to realize multi-mode information processing on the palm, an analog-digital converter and a complex control circuit with high cost are not needed, the circuit design is greatly simplified, the energy consumption is reduced, the high-density scale integration is facilitated, the application prospect is good, the development of nerve morphology perception and calculation is promoted, and besides, compared with the traditional single memristor neuron circuit, the multi-mode perception neuron array has a plurality of perception points, can simultaneously perceive large-scale information, further improves the energy efficiency and reduces the hardware area.
It will be readily appreciated by those skilled in the art that the foregoing description is merely a preferred embodiment of the invention and is not intended to limit the invention, but any modifications, equivalents, improvements or alternatives falling within the spirit and principles of the invention are intended to be included within the scope of the invention.
Claims (10)
1. A memristor-based sensing neuron circuit, comprising: an array of sensory neurons; each sensing neuron in the array converts the sensed external environment information into a pulse signal in parallel for output;
each sensing neuron in the array comprises a resistance sensor and a threshold value transformation type memristor with a positive electrode connected with the resistance sensor; the other ends of the resistance sensors in the same row are connected to the same word line of the array and are used for connecting a power supply; the cathodes of memristors on 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;
in the process of sensing external environment information, the resistance value of the resistance sensor changes along with the change of the external environment, so that the voltage at two ends of a memristor connected with the resistance sensor changes; when the voltage at the two ends of the memristor reaches the threshold voltage, the memristor is subjected to threshold transition to be in a low-resistance state, and discharge is started, pulses are output through the positive electrode until the voltage at the two ends of the memristor is lower than the holding voltage of the memristor, and the memristor returns to be in a high-resistance state.
2. The sensing neuron circuit according to claim 1, wherein the resistance sensor is a pressure sensor for sensing a pressure change of the external environment, and the resistance value thereof decreases with an increase of the external environment pressure, so that the pulse frequency of the sensing neuron output increases.
3. The sensing neuron circuit according to claim 1 or 2, wherein the memristor material is a metal-insulator transition phase change material, or a photoresistance-change oxide material.
4. A sensing neuron circuit according to claim 3, wherein when the memristor material is a metal-insulator transition phase change material or a photoresistive oxide material, the memristor is further configured to sense a change in temperature of the external environment, and the threshold voltage thereof decreases with an increase in temperature of the external environment, so that the pulse amplitude of the sensing neuron output decreases and the frequency increases.
5. The sensing neuron circuit according to claim 3, wherein when the memristor is made of a photoresistive oxide material, the memristor is further used for sensing changes in external environment illumination intensity, and the resistance value of the memristor decreases with the increase of the external environment illumination intensity, so that the pulse frequency output by the sensing neuron is increased and the amplitude of the pulse frequency output by the sensing neuron is unchanged.
6. A signal recognition circuit, comprising: a sensing neuron circuit and a memristive neural network circuit; the sensing neuron circuit is the sensing neuron circuit according to any one of claims 1 to 5;
wherein, the memristive neural network circuit includes: the memristive input neuron circuit, the memristive synaptic array and the memristive output neuron circuit are sequentially connected;
the memristive input neuron circuit comprises a plurality of memristive input neurons, and each memristive input neuron is correspondingly connected with the positive electrode of the memristor of each sensing neuron in the sensing neuron array one by one;
memristive synapses in the array of memristive synapses are non-volatile memristors;
the memristive output neuron circuit includes a plurality of memristive output neurons;
in a training mode, presetting a plurality of groups of external environment information, wherein the sensing neuron circuit is used for acquiring pulse signals under each group of external environment information and inputting the pulse signals into the memristor neural network circuit for training;
in the recognition mode, the sensing neuron circuit is used for converting the sensed external environment information into a pulse signal and outputting the pulse signal to the memristor neural network circuit so as to recognize the external environment information.
7. The signal recognition circuit of claim 6, wherein the ambient information includes ambient pressure information when the resistive sensor in the sensory neuron circuit is a pressure sensor.
8. The signal recognition circuit of claim 6 or 7, wherein the ambient information further includes ambient temperature information when the material of the memristor is a metal-insulator transition phase change material or a photoresistance oxide material.
9. The signal recognition circuit of claim 6 or 7, wherein the external environmental information further includes external optical signal information when the material of the memristor is a photoresistance oxide material.
10. An electronic chip comprising a sensing neuron circuit according to any one of claims 1 to 5, or a signal recognition circuit according to any one of claims 6 to 9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311555274.2A CN117669676A (en) | 2023-11-21 | 2023-11-21 | Memristor-based sensing neuron circuit and application |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311555274.2A CN117669676A (en) | 2023-11-21 | 2023-11-21 | Memristor-based sensing neuron circuit and application |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117669676A true CN117669676A (en) | 2024-03-08 |
Family
ID=90076197
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311555274.2A Pending CN117669676A (en) | 2023-11-21 | 2023-11-21 | Memristor-based sensing neuron circuit and application |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117669676A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117692003A (en) * | 2023-12-11 | 2024-03-12 | 广东工业大学 | Neuron-based analog-to-digital converter |
| CN118313424A (en) * | 2024-04-09 | 2024-07-09 | 湖南大学重庆研究院 | Construction method of intelligent sensing circuit and pulse camera adopting intelligent sensing circuit |
-
2023
- 2023-11-21 CN CN202311555274.2A patent/CN117669676A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117692003A (en) * | 2023-12-11 | 2024-03-12 | 广东工业大学 | Neuron-based analog-to-digital converter |
| CN118313424A (en) * | 2024-04-09 | 2024-07-09 | 湖南大学重庆研究院 | Construction method of intelligent sensing circuit and pulse camera adopting intelligent sensing circuit |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN117669676A (en) | Memristor-based sensing neuron circuit and application | |
| US11361216B2 (en) | Neural network circuits having non-volatile synapse arrays | |
| CN112183739B (en) | Hardware architecture of memristor-based low-power-consumption pulse convolution neural network | |
| Liu et al. | Reduction and IR-drop compensations techniques for reliable neuromorphic computing systems | |
| Dong et al. | Multimodal neuromorphic sensory-processing system with memristor circuits for smart home applications | |
| US20190122105A1 (en) | Training of artificial neural networks | |
| Kadetotad et al. | Parallel architecture with resistive crosspoint array for dictionary learning acceleration | |
| US11157803B2 (en) | Neuromorphic device including a synapse having a variable resistor and a transistor connected in parallel with each other | |
| Dong et al. | Convolutional neural networks based on RRAM devices for image recognition and online learning tasks | |
| Jiang et al. | Design and hardware implementation of neuromorphic systems with RRAM synapses and threshold-controlled neurons for pattern recognition | |
| US11620505B2 (en) | Neuromorphic package devices and neuromorphic computing systems | |
| WO2018228424A1 (en) | Method and device for training neural network | |
| Chen et al. | Optimal tuning of memristor conductance variation in spiking neural networks for online unsupervised learning | |
| Fang et al. | Neuromorphic algorithm-hardware codesign for temporal pattern learning | |
| Wei et al. | Emerging Memory-Based Chip Development for Neuromorphic Computing: Status, Challenges, and Perspectives | |
| CN112734022B (en) | Four-character memristor neural network circuit with recognition and sequencing functions | |
| Milo et al. | Brain-inspired recurrent neural network with plastic RRAM synapses | |
| CN109359734B (en) | Neural network synaptic structure based on memristor unit and adjusting method thereof | |
| Haensch | Analog computing for deep learning: Algorithms, materials & architectures | |
| CN116863296A (en) | Self-adaptive optimization sensing and storage integrated target detection and recognition system and method | |
| CN114861902A (en) | Processing unit, operation method thereof and computing chip | |
| Zhao et al. | A current-feedback method for programming memristor array in bidirectional associative memory | |
| CN114186667B (en) | Mapping method of cyclic neural network weight matrix to memristor array | |
| US11106971B2 (en) | Neuromorphic device and a synapse network including a post-synaptic neuron having a subtracting circuit | |
| KR20210158698A (en) | Apparatus and method for performing predetermined operation using circuit |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |