Receptive field module, pulse coupled electronic retina and implementation method thereof

Abstract

The application discloses a receptive field module, a pulse coupling electronic retina and an implementation method thereof. The pulse coupling electronic retina comprises a pulse type electronic photoreceptor array and an electronic visual neuron. The pulse type electronic photoreceptor comprises a photosensitive Mott memory resistor and a series resistor. The positive terminal of the photosensitive Mott memory resistor is connected with a power supply, the negative terminal of the photosensitive Mott memory resistor is connected with one end of the series resistor, and the other end of the series resistor is grounded. The number of the pulse type electronic photoreceptors in the pulse type electronic photoreceptor array is n. The electronic visual neuron comprises an MIFG transistor and a Mott memory resistor. The n+1 gate input ends of the MIFG transistor are respectively connected with the output signal ends of the n photosensitive Mott memory resistors and a common control signal end, and the source of the MIFG transistor is grounded. The positive electrode of the Mott memory resistor is connected with the power supply, and the negative electrode of the Mott memory resistor is connected with the drain of the MIFG transistor. The retina and the implementation method thereof have the characteristics of fast calculation speed and low power consumption.

Description

Technical Field

[0001] This invention belongs to the field of brain-like intelligence technology, and relates to a receptive field module, a pulse-coupled electronic retina, and a method for implementing them. Background Technology

[0002] With the arrival of the post-Moore's Law era, the von Neumann architecture faces increasingly prominent challenges, such as the memory wall, power consumption wall, and difficulties in improving intelligence. Therefore, single-function computing chips are gradually evolving into intelligent sensing chips. With the continuous development of artificial intelligence technology, neuromorphic intelligent sensing chips will become the general-purpose computing power and powerful hardware foundation of the post-Moore's Law era. Currently, artificial intelligence has become the core driving force of a new round of technological revolution and industrial transformation. Neuromorphic intelligent robots based on neuromorphic intelligent sensing chips will undoubtedly lead the development direction of artificial intelligence.

[0003] In terms of perception, brain-like intelligent robots require keen "eyes" to acquire external information. Vision is the most important mode of human perception, with over 80% of external information acquired through the visual system. Similarly, machine vision systems play a crucial role in intelligent robots. However, current machine vision systems are primarily based on digital image sensors, such as charge-coupled devices (CCDs) and active pixel image sensors (APS), targeting the traditional von Neumann architecture's central computing unit (CPU), rather than brain-like spiking neural network (SNN) computing units. The transmission path of image information involves a photosensitive unit → amplifier → analog-to-digital converter (ADC) → memory, transmitting it to the CPU in digital encoded form. Then, an artificial neural network (ANN) based on the von Neumann architecture performs extensive computations, completing image preprocessing (such as scaling, rotation, and normalization), image segmentation, and image description processes, ultimately enabling the recognition of external visual image information. Among these processes, image segmentation technology plays a vital role in image recognition, directly impacting the target recognition result. Based on the phenomenon of synchronized pulse firing in the cerebral cortex of animals such as cats and monkeys, Pulse Coupled Neural Networks (PCNNs) are powerful image segmentation tools. They possess functions such as information integration, thresholding, and pulse generation. They can extract effective information from complex backgrounds without learning or training, and their signal form and processing mechanism are more consistent with the physiological basis of the human visual nervous system. Therefore, PCNNs are widely used in image segmentation, feature extraction, and edge detection. However, current PCNNs require digital image sensors for image acquisition and von Neumann architecture CPUs for image processing, resulting in slow computation speed and high power consumption. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a receptive field module, a pulse-coupled electronic retina and its implementation method, which have the characteristics of fast calculation speed and low power consumption.

[0005] To achieve the above objectives, the present invention discloses a receptive field module comprising a pulsed electronic photoreceptor array and an electronic visual neuron;

[0006] The pulsed electronic photoreceptor includes a photosensitive Mott memristor and a series resistor. The positive terminal of the photosensitive Mott memristor is connected to the power supply, the negative terminal of the photosensitive Mott memristor is connected to one end of the series resistor, and the other end of the series resistor is grounded.

[0007] The pulsed photoreceptor array contains n pulsed photoreceptors. The electronic visual neuron includes a MIFG transistor and a Mott memristor. The n+1 gate input terminals of the MIFG transistor are connected to the output signal terminals and the common control signal terminal of the n photosensitive Mott memristors, respectively. The source of the MIFG transistor is grounded.

[0008] The positive terminal of the MOT memristor is connected to the power supply, and the negative terminal of the MOT memristor is connected to the drain of the MIFG transistor.

[0009] The photosensitive Mott memristor uses VO x As a resistive switching layer, a planar structure with two coplanar electrodes is adopted, and VOx material is deposited on the coplanar electrodes.

[0010] The MIFG transistor uses a stack of high-k gate + floating gate + high-k gate as the gate stack.

[0011] High-k gates are made of HfLaO, TaLaO, Al2O3 or HfO2 materials.

[0012] The floating gate is made of Ti, Al or Ni material.

[0013] The Mott memristor uses VO x or NbO x The material serves as the resistive switching layer of the MOT memristor.

[0014] MIFG transistors (T0) use Si, IGZO, ITZO, or MoS2 as the channel material.

[0015] The photoreceptor array and the photoreceptor array are integrated in a three-dimensional monolithic manner, with the photoreceptor array located above the photoreceptor array.

[0016] This invention discloses a pulse-coupled electronic retina, comprising an array of receptive field modules composed of receptive field modules.

[0017] This invention discloses a method for implementing a pulse-coupled electronic retina, comprising the following steps:

[0018] The pulsed photoreceptors in the pulsed photoreceptor array generate electrical pulse signals under illumination. The electronic visual neuron performs a weighted summation of the electrical pulse signals generated by each pulsed photoreceptor with the control signal input from the common control signal terminal, and determines whether to output an electrical pulse based on the calculation result.

[0019] When the calculated result is greater than the threshold voltage V of the MIFG transistor T When this occurs, an electrical pulse is output.

[0020] The present invention has the following beneficial effects:

[0021] In specific operation, the receptive field module and the pulse-coupled electronic retina based on the receptive field module described in this invention generate electrical pulse signals under illumination by pulsed photoreceptors. The electronic visual neurons perform weighted summation calculations on the electrical pulse signals generated by each pulsed photoreceptor and the control signal input from the common control signal terminal, and determine whether to output an electrical pulse based on the calculation result, so as to realize the integration of photosensing-encoding-computation and achieve rapid processing of visual information. It has the advantages of simple structure, multiple functions and low power consumption, and is more suitable for brain-like spiking neural networks. It solves the problems of slow visual information processing speed, high power consumption and incompatibility of information encoding method with brain-like SNN faced by traditional CMOSFET-based visual information processing systems. Attached Figure Description

[0022] Figure 1a A schematic diagram of the 3D structure of a pulse-coupled electronic retina;

[0023] Figure 1b A 3D structural diagram of a single receptive field module;

[0024] Figure 1c This is a functional schematic diagram of a pulse-coupled electronic retina.

[0025] Figure 1d This is a schematic diagram of the pulse-coupled electronic retina.

[0026] Figure 2a A schematic diagram illustrating the principle of pulse-coupled electronic retina segmentation of graphic information;

[0027] Figure 2b A grayscale image of the red channel information extracted from the original color image;

[0028] Figure 2c Image segmentation results for pulse-coupled electronic retina;

[0029] Figure 3a This is the equivalent circuit diagram of an electronic visual neuron;

[0030] Figure 3b For Mott memristor R M A typical IV curve. Detailed Implementation

[0031] To enable those skilled in the art to better understand the present invention, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of the present invention, not all embodiments, and are not intended to limit the scope of the present invention. Furthermore, in the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessary confusion regarding the concepts disclosed in the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort should fall within the scope of protection of the present invention.

[0032] The accompanying drawings show structural schematic diagrams according to embodiments disclosed in this invention. These drawings are not drawn to scale, and some details have been enlarged for clarity, and some details may have been omitted. The shapes of the various regions and layers shown in the drawings, as well as their relative sizes and positional relationships, are merely exemplary and may deviate from reality due to manufacturing tolerances or technical limitations. Furthermore, those skilled in the art can design regions / layers with different shapes, sizes, and relative positions as needed.

[0033] refer to Figure 1a , Figure 1b , Figure 1c , Figure 1d and Figure 2a As shown, the pulse-coupled electronic retina of the present invention includes a receptive field module array;

[0034] The receptive field module includes an array of m*m pulsed photoreceptors that integrate perception and coding, and an electronic visual neuron that integrates computation and coding, where m is an odd number greater than or equal to 3.

[0035] In this embodiment, the pulsed photoreceptor includes a photosensitive Mott memristor R. P and series resistor R C Among them, the photosensitive Mott memristor R P The positive terminal is connected to the power supply V. DD Photosensitive Mott memristor R P The negative terminal is connected in series with resistor R C One end is connected to the other end, and a series resistor R is connected. C The other end is grounded.

[0036] The electronic optic neuron includes a MIFG (Multi-Input Floating Gate) transistor T0 and a Mott memristor R. M In this case, the n+1 gate input terminals of the MIFG transistor T0 are respectively connected to n photosensitive Mott memristors R P Output signal terminals V1~V n It is connected to the common control signal terminal Vc, the source of MIFG transistor T0 is grounded, and n is a positive integer greater than or equal to 1.

[0037] The Mott memristor R M The positive terminal is connected to the power supply V. DD Connected, Mott memristor R M The negative terminal is connected to the drain of the MIFG transistor T0.

[0038] In this embodiment, the photosensitive Mott memristor R P Adopting VO x As a resistive switching layer, a planar structure with two coplanar electrodes is adopted, VO x Material is deposited on coplanar electrodes to ensure the photosensitive Mott memristor R P It has high photosensitivity.

[0039] In this embodiment, the MIFG transistor T0 uses a stack of high-k gate + floating gate + high-k gate as the gate stack. The high-k gate is made of HfLaO, TaLaO, Al2O3 or HfO2 material, and the floating gate is made of Ti, Al or Ni material, so that the MIFG transistor T0 has good leakage current performance and high stability. The MIFG transistor T0 uses Si, IGZO, ITZO or MoS2 as the channel material.

[0040] In this embodiment, to obtain better frequency characteristics, the Mott memristor R M Adopting VO x or NbO x Material as a Mott memristor R M The resistive switching layer.

[0041] In this embodiment, for the sake of simplicity, the photoreceptor array and the electronic visual neuron array are integrated in a three-dimensional monolithic manner, with the photoreceptor array located above the electronic visual neuron array.

[0042] In this embodiment, the common control terminal of all electronic visual neurons is connected to the common control signal terminal Vc, so that the control signal Vc with the same access time can realize the threshold segmentation function of visual information.

[0043] The method for implementing pulse-coupled electronic retina according to the present invention includes the following steps:

[0044] The pulsed photoreceptors in the pulsed photoreceptor array generate electrical pulse signals under illumination. The electronic visual neuron performs a weighted summation of the electrical pulse signals generated by each pulsed photoreceptor with the control signal input at the common control signal terminal Vc, and determines whether to output an electrical pulse based on the calculation result.

[0045] When the calculated result is greater than the threshold voltage V of the MIFG transistor T0 T When the time is right, an electrical pulse is output.

[0046] Example 1

[0047] In this embodiment, the pulse-coupled electronic retina is composed of a receptive field module array. Each receptive field module comprises a 3×3 pulsed photoreceptor array and a pulse-coupled electronic optic neuron. The pulsed photoreceptor generates oscillating pulses only under illumination and encodes the illumination information into a pulse frequency, thus simultaneously possessing event triggering and photosensitive-pulse encoding integrated functions. The pulse-coupled electronic optic neuron includes an IGZO MIFG transistor TO and a Mott memristor R. M It can perform weighted calculations on the electrical pulse signals generated by nine pulse-type photoreceptors within the sensing field and the control signal input at a common control signal terminal Vc, and determine whether to output an electrical pulse based on the calculation results. It also has functions such as information integration, threshold output, and pulse generation, i.e., calculation-pulse output integrated function.

[0048] The specific working principle of the electronic retina described in this embodiment is as follows:

[0049] The pulse signals (V1, V2, ..., V9) generated by the electronic visual neuron pairs within a receptive field module and the pulsed electronic photoreceptor array connected to them, and the control signal (V) applied to the common control signal terminal Vc, are all related to the control signals (V1, V2, ..., V9) generated by these pairs of electronic visual neurons and the pulsed electronic photoreceptor array connected to them. c Perform a weighted summation operation, and then determine the calculation result (i.e., the floating gate voltage V of MIFG transistor T0). F Is it greater than the threshold voltage V of the MIFG transistor T0? T When it is greater than the threshold voltage V T When the time is right, the MIFG transistor T0 turns on, driving the Mott memristor R. M When an electrical pulse is generated, the state of the electronic visual neuron is recorded as 1; otherwise, the MIFG transistor T0 outputs no pulse, and the state of the electronic visual neuron is recorded as 0. Simultaneously, the common control terminal of all electronic visual neurons receives the same control signal V. cThe threshold for the output electrical pulses of the electronic optic neurons is adjusted so that the receptive field module only outputs electrical pulses when the received light intensity is greater than a certain threshold; otherwise, no electrical pulse is output. By detecting the electrical pulse output of all electronic optic neurons within a fixed time period, the visual information perceived by the pulse-type electronic photoreceptor array can be efficiently binarized, thereby completing threshold segmentation of visual image information. In addition, the receptive field modules work in parallel and simultaneously, thus possessing parallel computing capabilities.

[0050] The segmentation effect of electronic retina on graphic information is as follows: Figure 2b and Figure 2c As stated above.

[0051] The working principle of electronic visual neurons is as follows: Figure 3a and Figure 3b As shown, where, Figure 3a The equivalent circuit diagram of an electronic visual neuron, C p For Mott memristor R M Parasitic capacitance; Figure 3b For Mott memristor R M A typical IV curve, R H R L V TH and V Hold These are respectively Mott memristors R M The high-resistance state resistance, low-resistance state resistance, threshold resistive voltage, and holding voltage.

[0052] The multiple input terminals of the MIFG transistor T0 are capacitively coupled to the floating gate (C1, C2, ..., C9), and the floating gate voltage V F The value is determined by the weighted sum of the input voltage signals (V1, V2, ..., V9) and the amount of charge Q stored in the floating gate, i.e.

[0053]

[0054] in, C0 is the capacitance between the floating gate and the channel of the MIFG transistor T0, w i (=C i / C TOT ) represents the weight of the corresponding input voltage signal, w i =C i / C TOT When no high voltage is used to perform charge erase / write operations on the floating gate transistor, Q = 0.

[0055]

[0056] When V F The threshold voltage V of the MIFG transistor T0 is greater than tWhen the time is right, the channel of the MIFG transistor T0 is turned on, driving the sandwich-structured Mott memristor R. M The output frequency-encoded electrical pulse, at point D, according to Kirchhoff's laws, yields:

[0057]

[0058] Where τ=R M C p I D R is the drain current of the MIFG transistor T0. M For Mott memristor R M The resistance value, C P For Mott memristor R M Parasitic capacitance.

[0059] Mott memristor R M From the IV curve, we can see that when u D (t)≥V TH At that time, R M =R L ; when u D (t)≤V Hold At that time, R M =R H .

[0060] When the MIFG transistor T0 is operating in the saturation region, then

[0061] I D =B(V) F -V t ) 2 (5)

[0062] in, μ, W, L and C ox These represent the channel carrier mobility, channel width, channel length, and capacitance per unit area of ​​the gate dielectric between the floating gate and the channel for the MIFG transistor T0.

[0063] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A receptive field module, characterized by, Including pulsed photoreceptor arrays and electronic visual neurons; Pulsed photosensitive sensors include photosensitive Mott memristors (R... P ) and series resistance (R) C ), among which, photosensitive Mott memristor (R P The positive terminal of the power supply (V) is connected to the power source. DD ), photosensitive Mott memristor (R P The negative terminal of ) is connected in series with the resistor (R) C Connect one end of the resistor (R) to the series resistor (R). C The other end of the device is grounded; The number of the pulse type electronic light receptors in the pulse type electronic light receptor array is n, and the electronic optic nerve cell comprises an MIFG transistor (T0) and a Mott memory resistor (R M ), wherein n+1 gate input ends of the MIFG transistor (T0) are connected with output signal ends of the n light-sensitive Mott memory resistors (R P ) and a common control signal end (Vc) respectively, and a source of the MIFG transistor (T0) is grounded. The positive pole of the Mott memory resistor (R M ) is connected with the power supply (V DD ), and the negative pole of the Mott memory resistor (R M ) is connected with the drain of the MIFG transistor (T0). A photo-sensitive Mott memory resistor (R P ) uses VO x as a resistance change layer, adopts a planar structure with two electrodes in the same plane, and deposits VO x material on the electrodes in the same plane. The MIFG transistor (T0) uses a stack of high-k gate + floating gate + high-k gate as the gate stack.

2. The field of view module of claim 1, wherein, High-k gates are made of HfLaO, TaLaO, Al2O3, or HfO2 materials; The floating gate is made of Ti, Al or Ni material.

3. The field of view module of claim 1, wherein, MIFG transistors (T0) use Si, IGZO, ITZO, or MoS2 as the channel material.

4. The field of view module of claim 1, wherein, The Mott memory resistor (R M ) uses VO x or NbO x material as the resistance change layer of the Mott memory resistor (R M ).

5. The field of view module of claim 1, wherein, The photoreceptor array and the photoreceptor array are integrated in a three-dimensional monolithic manner, with the photoreceptor array located above the photoreceptor array.

6. An impulse-coupled electronic retina, characterized by It includes an array of receptive field modules comprising the receptive field modules described in any one of claims 1-5.

7. A method of implementing the pulse coupled electronic retina of claim 6, wherein, Includes the following steps: The pulsed photoreceptors in the pulsed photoreceptor array generate electrical pulse signals under illumination. The electronic visual neuron performs a weighted summation of the electrical pulse signals generated by each pulsed photoreceptor with the control signal input from the common control signal terminal (Vc), and determines whether to output an electrical pulse based on the calculation result.

8. The method of implementing a pulse coupled electronic retina according to claim 7, wherein, When the calculation result is greater than a threshold voltage of the MIFG transistor (T0) V T then an electrical pulse is output.

Images