WO2021184407A1 - 一种视网膜形态光电传感阵列及其图片卷积处理方法 - Google Patents

一种视网膜形态光电传感阵列及其图片卷积处理方法 Download PDF

Info

Publication number
WO2021184407A1
WO2021184407A1 PCT/CN2020/081313 CN2020081313W WO2021184407A1 WO 2021184407 A1 WO2021184407 A1 WO 2021184407A1 CN 2020081313 W CN2020081313 W CN 2020081313W WO 2021184407 A1 WO2021184407 A1 WO 2021184407A1
Authority
WO
WIPO (PCT)
Prior art keywords
photoelectric sensor
substrate
layer
dielectric layer
photoelectric
Prior art date
Application number
PCT/CN2020/081313
Other languages
English (en)
French (fr)
Inventor
缪峰
梁世军
王晨宇
Original Assignee
南京大学
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 南京大学 filed Critical 南京大学
Priority to US17/769,609 priority Critical patent/US20240234617A9/en
Publication of WO2021184407A1 publication Critical patent/WO2021184407A1/zh

Links

Images

Classifications

    • H—ELECTRICITY
    • H01—ELECTRIC ELEMENTS
    • H01L—SEMICONDUCTOR DEVICES NOT COVERED BY CLASS H10
    • H01L31/00—Semiconductor devices sensitive to infrared radiation, light, electromagnetic radiation of shorter wavelength or corpuscular radiation and specially adapted either for the conversion of the energy of such radiation into electrical energy or for the control of electrical energy by such radiation; Processes or apparatus specially adapted for the manufacture or treatment thereof or of parts thereof; Details thereof
    • H01L31/08—Semiconductor devices sensitive to infrared radiation, light, electromagnetic radiation of shorter wavelength or corpuscular radiation and specially adapted either for the conversion of the energy of such radiation into electrical energy or for the control of electrical energy by such radiation; Processes or apparatus specially adapted for the manufacture or treatment thereof or of parts thereof; Details thereof in which radiation controls flow of current through the device, e.g. photoresistors
    • H01L31/10—Semiconductor devices sensitive to infrared radiation, light, electromagnetic radiation of shorter wavelength or corpuscular radiation and specially adapted either for the conversion of the energy of such radiation into electrical energy or for the control of electrical energy by such radiation; Processes or apparatus specially adapted for the manufacture or treatment thereof or of parts thereof; Details thereof in which radiation controls flow of current through the device, e.g. photoresistors characterised by potential barriers, e.g. phototransistors
    • H01L31/101—Devices sensitive to infrared, visible or ultraviolet radiation
    • H01L31/112—Devices sensitive to infrared, visible or ultraviolet radiation characterised by field-effect operation, e.g. junction field-effect phototransistor
    • H01L31/113—Devices sensitive to infrared, visible or ultraviolet radiation characterised by field-effect operation, e.g. junction field-effect phototransistor being of the conductor-insulator-semiconductor type, e.g. metal-insulator-semiconductor field-effect transistor
    • H01L31/1136—Devices sensitive to infrared, visible or ultraviolet radiation characterised by field-effect operation, e.g. junction field-effect phototransistor being of the conductor-insulator-semiconductor type, e.g. metal-insulator-semiconductor field-effect transistor the device being a metal-insulator-semiconductor field-effect transistor
    • 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • 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
    • G06N3/08—Learning methods
    • H—ELECTRICITY
    • H01—ELECTRIC ELEMENTS
    • H01L—SEMICONDUCTOR DEVICES NOT COVERED BY CLASS H10
    • H01L27/00—Devices consisting of a plurality of semiconductor or other solid-state components formed in or on a common substrate
    • H01L27/14—Devices consisting of a plurality of semiconductor or other solid-state components formed in or on a common substrate including semiconductor components sensitive to infrared radiation, light, electromagnetic radiation of shorter wavelength or corpuscular radiation and specially adapted either for the conversion of the energy of such radiation into electrical energy or for the control of electrical energy by such radiation
    • H01L27/144—Devices controlled by radiation
    • H01L27/146—Imager structures
    • H01L27/14601—Structural or functional details thereof
    • H01L27/14609—Pixel-elements with integrated switching, control, storage or amplification elements
    • H01L27/1461—Pixel-elements with integrated switching, control, storage or amplification elements characterised by the photosensitive area
    • H—ELECTRICITY
    • H01—ELECTRIC ELEMENTS
    • H01L—SEMICONDUCTOR DEVICES NOT COVERED BY CLASS H10
    • H01L27/00—Devices consisting of a plurality of semiconductor or other solid-state components formed in or on a common substrate
    • H01L27/14—Devices consisting of a plurality of semiconductor or other solid-state components formed in or on a common substrate including semiconductor components sensitive to infrared radiation, light, electromagnetic radiation of shorter wavelength or corpuscular radiation and specially adapted either for the conversion of the energy of such radiation into electrical energy or for the control of electrical energy by such radiation
    • H01L27/144—Devices controlled by radiation
    • H01L27/146—Imager structures
    • H01L27/14601—Structural or functional details thereof
    • H01L27/14609—Pixel-elements with integrated switching, control, storage or amplification elements
    • H01L27/14612—Pixel-elements with integrated switching, control, storage or amplification elements involving a transistor
    • H01L27/14616—Pixel-elements with integrated switching, control, storage or amplification elements involving a transistor characterised by the channel of the transistor, e.g. channel having a doping gradient
    • H—ELECTRICITY
    • H01—ELECTRIC ELEMENTS
    • H01L—SEMICONDUCTOR DEVICES NOT COVERED BY CLASS H10
    • H01L31/00—Semiconductor devices sensitive to infrared radiation, light, electromagnetic radiation of shorter wavelength or corpuscular radiation and specially adapted either for the conversion of the energy of such radiation into electrical energy or for the control of electrical energy by such radiation; Processes or apparatus specially adapted for the manufacture or treatment thereof or of parts thereof; Details thereof
    • H01L31/02—Details
    • H01L31/0224—Electrodes
    • H01L31/022408—Electrodes for devices characterised by at least one potential jump barrier or surface barrier
    • H—ELECTRICITY
    • H01—ELECTRIC ELEMENTS
    • H01L—SEMICONDUCTOR DEVICES NOT COVERED BY CLASS H10
    • H01L31/00—Semiconductor devices sensitive to infrared radiation, light, electromagnetic radiation of shorter wavelength or corpuscular radiation and specially adapted either for the conversion of the energy of such radiation into electrical energy or for the control of electrical energy by such radiation; Processes or apparatus specially adapted for the manufacture or treatment thereof or of parts thereof; Details thereof
    • H01L31/0248—Semiconductor devices sensitive to infrared radiation, light, electromagnetic radiation of shorter wavelength or corpuscular radiation and specially adapted either for the conversion of the energy of such radiation into electrical energy or for the control of electrical energy by such radiation; Processes or apparatus specially adapted for the manufacture or treatment thereof or of parts thereof; Details thereof characterised by their semiconductor bodies
    • H01L31/0256—Semiconductor devices sensitive to infrared radiation, light, electromagnetic radiation of shorter wavelength or corpuscular radiation and specially adapted either for the conversion of the energy of such radiation into electrical energy or for the control of electrical energy by such radiation; Processes or apparatus specially adapted for the manufacture or treatment thereof or of parts thereof; Details thereof characterised by their semiconductor bodies characterised by the material
    • H01L31/0264—Inorganic materials
    • H01L31/032—Inorganic materials including, apart from doping materials or other impurities, only compounds not provided for in groups H01L31/0272 - H01L31/0312
    • H—ELECTRICITY
    • H01—ELECTRIC ELEMENTS
    • H01L—SEMICONDUCTOR DEVICES NOT COVERED BY CLASS H10
    • H01L31/00—Semiconductor devices sensitive to infrared radiation, light, electromagnetic radiation of shorter wavelength or corpuscular radiation and specially adapted either for the conversion of the energy of such radiation into electrical energy or for the control of electrical energy by such radiation; Processes or apparatus specially adapted for the manufacture or treatment thereof or of parts thereof; Details thereof
    • H01L31/18—Processes or apparatus specially adapted for the manufacture or treatment of these devices or of parts thereof
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N25/00—Circuitry of solid-state image sensors [SSIS]; Control thereof
    • H04N25/47—Image sensors with pixel address output; Event-driven image sensors; Selection of pixels to be read out based on image data
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P70/00—Climate change mitigation technologies in the production process for final industrial or consumer products
    • Y02P70/50—Manufacturing or production processes characterised by the final manufactured product

Definitions

  • the invention relates to the cross-fields of photoelectric sensors, vision chips, artificial neural networks, and semiconductor manufacturing processes, and in particular to a photoelectric device with a retinal shape, a manufacturing method thereof, and a photoelectric sensor array with the photoelectric device and a photoelectric sensor array for matching
  • the input visual information is processed and recognized.
  • the visual system is the main way for the human body to perceive information from the outside world. Studies have shown that the amount of information transmitted through the vision system can account for more than 80% of the total amount of perceptual information. Therefore, by modeling the vision system and using computers to further exert its functions, researchers have opened up the research direction of computer vision. Since the 1960s, computer vision has developed rapidly and has been widely used in various fields, such as manufacturing, medical treatment, military, and security. In some specific aspects, computer vision has achieved results comparable to, or even better than, humans.
  • the natural biological vision system still has huge advantages in function and efficiency.
  • the human visual system has the characteristics of automatically adapting to different brightness environments, and its detection brightness range spans 9 orders of magnitude (10-3-106cd/m2); through eyeball control movement, the human eye can quickly and automatically track what is happening Offset target; using the sparse coding function of optic ganglion cells, visual information (equivalent to a high-definition picture with more than 100 million pixels) generated at the retina can be transmitted to the visual cortex with low latency and decoded correctly; small sample learning The ability allows the vision system to quickly learn sample features and realize recognition after inputting a few image samples; while achieving the above functions, the power consumption of the human brain is 20W, the operating frequency is as low as ⁇ 10Hz, and the power density (Power Density) ) Only one-thousandth of the existing general-purpose processors.
  • the researchers put forward the concept of "neuromorphic vision chip", the purpose is to simulate the structure and function of the human visual system, Apply its huge advantages in function and power consumption to the field of computer vision.
  • CMOS integrated circuits can be designed to simulate the early processing of visual information in the retina and improve the efficiency of information processing.
  • Professor Mead’s research team used the honeycomb circuit structure to successfully bionic the photoreceptor cells and bipolar cells in the retina, achieving functions such as edge recognition and wide-range brightness response; in addition, the visual system combined the signal collection and
  • the close integration of early processing and the hierarchical features of later processing have also been used for reference and formed a computing architecture integrating signal acquisition, storage, and processing; and since the beginning of the 21st century, reference has been made to the information processing mode of the vision system.
  • the researchers also proposed two information processing drive modes, frame-driven and event-driven, to achieve a more efficient neuromorphic vision processing system.
  • CMOS-based integrated circuits to simulate cells in the retina will increase the complexity of the circuit and reduce the fill factor of the vision chip; fast-developing 3D integrated circuits are still unable to be fabricated A vertical hierarchical structure similar to the retina is developed; the existing computing architecture is still unable to compare with the biological vision system in terms of power consumption.
  • the vision detector and the processor in the current neuromorphic vision processing system are separated, and the recognition of the target ultimately needs to be completed on a dedicated accelerator.
  • there is no photodetector which can not only detect and process visual information at the same time, but also has the ability of learning and reasoning, and can identify the target in real time.
  • the present invention provides a retina-shaped optoelectronic device, its manufacturing method, and a photoelectric sensor array with the optoelectronic device, which solves the problem of complex circuits and high power consumption caused by simulating cells in the retina.
  • the slow processing speed caused by the separation of the visual detection and the processor, the separation of the photoelectric sensor and the visual information processor, and the artificial neural network.
  • the photoelectric device with a retinal morphology is a vertically stacked heterojunction structure with a bottom electrode, a dielectric layer, a channel layer, a source electrode, and a drain electrode on a substrate.
  • the source electrode and The drain electrodes face each other and are placed at both ends of the channel layer.
  • the bottom electrode, the source electrode and the drain electrode are made of materials used for flexible electrodes, inert metals or semi-metals, and the dielectric layer is made of insulating materials.
  • the channel layer material is a bipolar material, and the base includes a substrate and an insulating material layer grown on the surface of the substrate.
  • the substrate material includes silicon, polyimide or polydimethylsiloxane, and the material used for the insulating material layer grown on the surface of the substrate is silicon oxide, aluminum oxide, hafnium zirconium oxide or boron nitride.
  • the bipolar material of the channel layer is graphene, tungsten selenide, molybdenum telluride, black phosphorus or palladium selenide.
  • the material of the dielectric layer is any one or more of boron nitride, silicon oxide, aluminum oxide and hafnium zirconium oxide.
  • a method for manufacturing a photoelectric device with a retinal shape comprising:
  • S2 obtains a dielectric layer directly on the bottom electrode or first obtains a dielectric layer material on the bottom electrode, and then uses a material transfer method to vertically stack different kinds of dielectric layer materials to prepare a dielectric layer with a multilayer structure;
  • S3 directly grows the bipolar material of the channel layer on the dielectric layer; or first obtains the bipolar material of the channel layer, and then uses the material transfer method to transfer the bipolar material to the dielectric layer to form the channel layer;
  • S4 prepares source and drain electrodes on the surface of the channel layer.
  • the method for directly obtaining the dielectric layer on the bottom electrode is a chemical vapor deposition method, a chemical vapor transmission method, a molecular beam epitaxy method, an atomic layer deposition method, or a hydrothermal method.
  • the method for directly growing the bipolar material of the channel layer on the dielectric layer is a chemical vapor deposition method, a chemical vapor transport method, a molecular beam epitaxy method, an atomic layer deposition method, or a hydrothermal method.
  • the method for preparing the bottom electrode on the surface of the substrate is as follows:
  • S11 adopts ultraviolet lithography, electron beam lithography or mask method to prepare and design the bottom electrode shape on the substrate;
  • a photoelectric sensor array for retinal morphology which has the above-mentioned photoelectric sensor for retinal morphology.
  • a visual image convolution processing method based on a photoelectric sensor array of retinal morphology comprising:
  • bit line corresponding to the photoelectric device in each row is connected in series
  • the signal line is set on each photoelectric device, and the signal line corresponding to the photoelectric device in each column is connected in series.
  • the bit line and the signal line apply source-drain voltage to the photoelectric sensor device at a specific position in the array;
  • a word line is arranged on each photoelectric device of the photoelectric sensor array, which is used to apply a back gate voltage to the photoelectric sensor device of a specific row in the array;
  • step (3) an m ⁇ m convolution kernel is used to complete the convolution operation of the entire photoelectric sensor array.
  • Step 1 Input the information to be identified into the photoelectric sensor array, and set the back gate voltage of all photoelectric devices to 0V;
  • Step 2 collect the output current I of the photoelectric sensor array and input it into the following Sigmoid activation function:
  • I is the output current of the photoelectric sensor array
  • ⁇ is the normalization coefficient
  • Step 3 After calculating the value of the activation function f, compare it with the target value, then make a judgment and perform the error back propagation operation according to the following formula:
  • ⁇ k is the error used in the kth training
  • f k is the output value of the Sigmoid activation function during the kth training
  • Step 4 After the error is subsequently transferred to the first layer, the initial back gate value in the photoelectric sensor array is updated through the following functional relationship:
  • n is the step size
  • ⁇ is the step size of the grid voltage change
  • P is the input of visual image information
  • round is the rounding function
  • conv is the convolution function to complete a training process
  • Step 5 Repeat steps 1-4 until the error calculated in step 3 is close to or equal to 0, that is, the target picture is successfully identified from all input pictures.
  • the present invention utilizes the physical properties of the new material to design a brand-new photoelectric sensor device with retinal morphology, which better simulates the cellular functions of the human visual system, optimizes device performance and reduces circuit complexity;
  • the photoelectric sensor of the present invention The array can realize the convolutional neural network of visual information detection, processing and recognition at the same time. By adjusting the gate voltage of each pixel device in the optoelectronic device, this optoelectronic device can be reconstructed into different convolutional neural networks for realization The operation and recognition of different processing and processing of visual information; the recognition method is simple; 3.
  • the reconfigurable artificial retina morphology sensor of the present invention for simultaneous detection, information processing and recognition of image information can be further integrated into a retinal morphology optical sensor chip, It is used for real-time application scenarios of edge computing such as intelligent security and health care.
  • FIG. 1 is a schematic diagram of the structure of the optoelectronic device in Embodiment 2 of the present invention.
  • Figure 2 is an optical picture of the optoelectronic device in embodiment 2 of the present invention.
  • Fig. 3 is the photoelectric response characteristics of the optoelectronic device according to the second embodiment of the present invention.
  • Fig. 3a is a graph of the photocurrent response of the device with the change of the back gate voltage when the light intensity is constant;
  • Fig. 3b is when the back gate is constant, The photoelectric current of the device varies with the light intensity;
  • Figure 3c shows the photoelectric response characteristics of the device with different light wavelengths under the same back gate;
  • FIG. 4 is a schematic diagram of the photoelectric sensor array and peripheral control circuit of the photoelectric device according to the present invention.
  • Figure 5 shows the different operations of the photoelectric sensor array based on the photoelectric device of the retinal morphology for image processing under different back grid adjustments according to the present invention, from top to bottom: inversion, edge enhancement, and brightness correction;
  • Fig. 6 is a schematic diagram of the process of training a convolutional neural network to realize image recognition.
  • the present invention first discloses a photoelectric device with a retinal shape and a manufacturing method thereof.
  • the bionic photoelectric device is a vertically stacked heterojunction structure with a bottom electrode, a dielectric layer, a channel layer, a source electrode, and a drain electrode on a substrate.
  • the source electrode It faces each other with the drain electrode and is placed on both ends of the channel layer.
  • the bottom electrode, source electrode and drain electrode are made of the material used for flexible electrodes, inert metal or semi-metal.
  • the material of the dielectric layer is insulating material, and the material of the channel layer
  • the base includes a substrate and an insulating material layer grown on the surface of the substrate.
  • the substrate material can be silicon, polyimide or polydimethylsiloxane
  • the insulating material layer grown on the surface of the substrate adopts silicon oxide, aluminum oxide, hafnium zirconium oxide or boron nitride.
  • It can be composed of one of boron nitride, silicon oxide, aluminum oxide, hafnium zirconium oxide, etc., or it can be composed of multiple types to form a vertical heterostructure.
  • the bipolar material of the channel layer has bipolar electrical characteristics, that is, through the adjustment of the field effect, this type of material can be either an electronic type doping material or a hole type doping material, including graphene, Tungsten selenide, molybdenum telluride, black phosphorus or palladium selenide.
  • the bottom electrode and the source and drain electrodes are composed of inert metals, flexible electrodes or semi-metal materials, such as platinum, gold, palladium, indium tin oxide, graphene, etc.
  • the present invention also discloses a method for manufacturing a photoelectric device with retinal morphology, which includes:
  • the electrode material is a conductive material that can be prepared by physical vapor deposition or magnetron sputtering, such as inert metals, titanium nitride, etc., or a semi-metal two-dimensional atomic crystal material such as graphene, etc.
  • the preparation method of the electrode of the two materials Slightly different.
  • the specific steps include:
  • the materials used are inert metals such as gold, platinum, palladium, etc., or conductive compounds such as Titanium nitride, etc.;
  • the prepared semi-metal two-dimensional atomic crystal film electrode is transferred to the desired position on the sample.
  • S2 directly obtain a dielectric layer on the bottom electrode or first obtain a dielectric layer material on the bottom electrode, and then use a material transfer method to vertically stack different kinds of dielectric layer materials to prepare a dielectric layer with a multilayer structure;
  • the dielectric layer is composed of insulating materials, including boron nitride, silicon oxide, aluminum oxide, hafnium zirconium oxide, and the like. It can be directly obtained by chemical vapor deposition (CVD), chemical vapor transport (CVT), molecular beam epitaxy (MBE), atomic layer deposition (ALD), magnetron sputtering, hydrothermal method, etc.;
  • CVD chemical vapor deposition
  • CVD chemical vapor transport
  • MBE molecular beam epitaxy
  • ALD atomic layer deposition
  • magnetron sputtering hydrothermal method, etc.
  • CVT chemical vapor transport
  • the dielectric layer may be composed of only one material, or after the corresponding material is obtained, a heterojunction dielectric layer having a multilayer structure in the vertical direction may be obtained by a transfer method.
  • the dielectric layer can be grown directly on the surface of the bottom electrode, or transferred from other substrates to the upper surface of the bottom electrode by a transfer method.
  • S3 directly grows the bipolar material of the channel layer on the dielectric layer; or first obtains the bipolar material of the channel layer, and then uses the material transfer method to transfer the bipolar material to the dielectric layer to form the channel layer.
  • the channel layer is composed of bipolar materials, including tungsten selenide, molybdenum telluride, black phosphorus, graphene, palladium selenide, and the like. It can be directly obtained by chemical vapor deposition (CVD), molecular beam epitaxy (MBE), atomic layer deposition (ALD), magnetron sputtering, hydrothermal method, etc.; it can also be obtained by chemical vapor transmission (CVT) method.
  • CVD chemical vapor deposition
  • MBE molecular beam epitaxy
  • ALD atomic layer deposition
  • CVT chemical vapor transmission
  • the bulk material is then obtained by mechanical peeling or ultrasonic spin coating methods to obtain a layered material; the dielectric layer can be grown directly on the surface of the bottom electrode, or transferred from other substrates to the upper surface of the dielectric layer by a transfer method.
  • any type of electrode material, dielectric material, and channel material can be used in the present invention.
  • the "conditional permission" mentioned here refers to the need to meet some limited conditions, such as: electrode materials, dielectric materials and channel materials cannot be dissolved in water and acetone, nor can they be dissolved in photoresist; electrode materials, dielectric materials and channel materials The channel material cannot react chemically with water, acetone and photoresist.
  • Those skilled in the art can use other metallic or semi-metallic materials for the electrode material in the present invention, other insulating materials for the dielectric material, and other bipolar materials for the channel material as long as the limited conditions are met.
  • S4 prepares source and drain electrodes on the surface of the channel layer.
  • the materials and preparation methods of the source electrode and the drain electrode are the same as those of the bottom electrode, and will not be repeated here.
  • the present invention provides multiple embodiments for manufacturing optoelectronic devices, which specifically include:
  • Preparation of the metal bottom electrode First, spin-coating a layer of PMMA (polymethyl methacrylate) on the substrate with a homogenizer, expose the specified electrode pattern on the PMMA by electron beam exposure, and use the developer to make the electrode The pattern is exposed to expose the underlying substrate; then an electron beam evaporation is used to grow a metal film with a thickness of about 40nm on the PMMA and the substrate; finally, the PMMA and the substrate are put into an acetone solution to dissolve the PMMA and take away the excess Metal film. What remains on the substrate is the metal film bottom electrode of the previously specified shape.
  • PMMA polymethyl methacrylate
  • the preparation of the dielectric layer firstly use atomic layer deposition to grow an aluminum oxide film with a thickness of about 10nm on a substrate with a bottom electrode; then use a mechanical lift-off method to obtain a thin layer of hexagonal boron nitride with a thickness between 5-50nm; The thin layer of hexagonal boron nitride is transferred to the surface of the aluminum oxide film by a transfer method, thereby forming an aluminum oxide/hexagonal boron nitride heterostructure dielectric layer.
  • a thin layer of tungsten selenide is directly prepared on the surface of the dielectric by the method of chemical vapor deposition (CDV).
  • Preparation of source and drain electrodes The preparation of source and drain electrodes is the same as the preparation of bottom electrodes mentioned above, but the source and drain electrodes must have an overlapping area with the bottom electrode in the horizontal direction, and ensure that there is a channel between the source and drain electrodes The material of the layer.
  • the preparation method of the metal bottom electrode is as follows: First, a layer of PMMA is spin-coated on the substrate with a homogenizer, and the specified electrode pattern is exposed on the PMMA by electron beam exposure. , And use a developer to expose the electrode pattern to expose the underlying substrate; then use electron beam evaporation to grow a metal film with a thickness of about 40nm on the PMMA and the substrate; finally, put the PMMA and the substrate into the acetone solution to make the PMMA Dissolve, thereby taking away excess metal film. What remains on the substrate is the metal film bottom electrode of the previously specified shape.
  • the preparation of the dielectric layer firstly use atomic layer deposition to grow an aluminum oxide film (Al 2 o 3 ) with a thickness of about 10 nm on a substrate with a bottom electrode; then use a mechanical lift-off method to obtain a thin layer of hexagonal boron nitride (h- BN), the thickness is between 5-50 nm; the thin layer of hexagonal boron nitride is transferred to the surface of the aluminum oxide film by a transfer method, thereby forming an aluminum oxide/hexagonal boron nitride heterostructure dielectric layer.
  • Al 2 o 3 aluminum oxide film
  • h- BN hexagonal boron nitride
  • Preparation of the channel layer first obtain a thin layer of tungsten selenide (WSe 2 ) by means of mechanical peeling, and then transfer to the surface of the dielectric layer by means of transfer.
  • WSe 2 tungsten selenide
  • source electrode is Source and drain electrode is Drain.
  • the preparation of source and drain electrodes is the same as the preparation of bottom electrode mentioned above, but the source and drain electrodes must have an overlapping area with the bottom electrode in the horizontal direction, and ensure The material of the channel layer exists between the source and drain electrodes.
  • Preparation of metal bottom electrode first spin-coating a layer of PMMA on the substrate with a homogenizer, expose the specified electrode pattern on the PMMA by electron beam exposure, and expose the electrode pattern with a developer to expose the underlying substrate ; Then use electron beam evaporation to grow a metal film with a thickness of about 40nm on the PMMA and the substrate; finally, put PMMA and the substrate into an acetone solution to dissolve the PMMA, thereby taking away the excess metal film. What remains on the substrate is the metal film bottom electrode of the previously specified shape.
  • Preparation of the dielectric layer firstly, an aluminum oxide film with a thickness of about 5-50 nm is grown on a substrate with a bottom electrode by atomic layer deposition.
  • a thin layer of tungsten selenide is directly prepared on the surface of the dielectric by the method of chemical vapor deposition (CDV).
  • Preparation of source and drain electrodes The preparation of source and drain electrodes is the same as the preparation of bottom electrodes mentioned above, but the source and drain electrodes must have an overlapping area with the bottom electrode in the horizontal direction, and ensure that there is a channel between the source and drain electrodes The material of the layer.
  • Preparation of metal bottom electrode first spin-coating a layer of PMMA on the substrate with a homogenizer, expose the specified electrode pattern on the PMMA by electron beam exposure, and expose the electrode pattern with a developer to expose the underlying substrate ; Then use electron beam evaporation to grow a metal film with a thickness of about 40nm on the PMMA and the substrate; finally, put PMMA and the substrate into an acetone solution to dissolve the PMMA, thereby taking away the excess metal film. What remains on the substrate is the metal film bottom electrode of the previously specified shape.
  • Preparation of the dielectric layer firstly, an aluminum oxide film with a thickness of about 5-50 nm is grown on a substrate with a bottom electrode by atomic layer deposition.
  • Preparation of the channel layer first obtain a thin layer of tungsten selenide by means of mechanical stripping, and then transfer to the surface of the dielectric layer by means of transfer.
  • Preparation of source and drain electrodes The preparation of source and drain electrodes is the same as the preparation of bottom electrodes mentioned above, but the source and drain electrodes must have an overlapping area with the bottom electrode in the horizontal direction, and ensure that there is a channel between the source and drain electrodes The material of the layer.
  • Preparation of metal bottom electrode first spin-coating a layer of PMMA on the substrate with a homogenizer, expose the specified electrode pattern on the PMMA by electron beam exposure, and expose the electrode pattern with a developer to expose the underlying substrate ; Then use electron beam evaporation to grow a metal film with a thickness of about 40nm on the PMMA and the substrate; finally, put PMMA and the substrate into an acetone solution to dissolve the PMMA, thereby taking away the excess metal film. What remains on the substrate is the metal film bottom electrode of the previously specified shape.
  • a thin layer of hexagonal boron nitride is obtained by a mechanical peeling method, the thickness is between 5-50 nm; the thin layer of hexagonal boron nitride is transferred to the surface of the aluminum oxide film by a transfer method.
  • a thin layer of tungsten selenide is directly prepared on the surface of the dielectric by the method of chemical vapor deposition (CDV).
  • Preparation of source and drain electrodes The preparation of source and drain electrodes is the same as the preparation of bottom electrodes mentioned above, but the source and drain electrodes must have an overlapping area with the bottom electrode in the horizontal direction, and ensure that there is a channel between the source and drain electrodes The material of the layer.
  • Preparation of metal bottom electrode first spin-coating a layer of PMMA on the substrate with a homogenizer, expose the specified electrode pattern on the PMMA by electron beam exposure, and expose the electrode pattern with a developer to expose the underlying substrate ; Then use electron beam evaporation to grow a metal film with a thickness of about 40nm on the PMMA and the substrate; finally, put PMMA and the substrate into an acetone solution to dissolve the PMMA, thereby taking away the excess metal film. What remains on the substrate is the metal film bottom electrode of the previously specified shape.
  • a thin layer of hexagonal boron nitride is obtained by a mechanical peeling method, the thickness is between 5-50 nm; the thin layer of hexagonal boron nitride is transferred to the surface of the aluminum oxide film by a transfer method.
  • Preparation of the channel layer first obtain a thin layer of tungsten selenide by means of mechanical stripping, and then transfer to the surface of the dielectric layer by means of transfer.
  • Preparation of source and drain electrodes The preparation of source and drain electrodes is the same as the preparation of bottom electrodes mentioned above, but the source and drain electrodes must have an overlapping area with the bottom electrode in the horizontal direction, and ensure that there is a channel between the source and drain electrodes The material of the layer.
  • the photoelectric device exhibits different photoelectric response characteristics under different back gate voltage, light intensity and wavelength, and the result is shown in Figure 3.
  • Figure 3a shows the response characteristics of optoelectronic devices under different back gate voltages.
  • Figure 3b shows the response characteristics of optoelectronic devices under different light intensities.
  • Figures 3c and 3d show that optoelectronic devices can work in the entire visible light spectrum.
  • a photoelectric sensor array of retinal morphology which has the photoelectric device of retinal morphology manufactured in the above-mentioned embodiment. Combining edge circuits on the basis of arrays can make photoelectric sensor chips for retinal morphology.
  • the above-mentioned photoelectric sensor array integrates the above-mentioned retina-shaped photoelectric devices on a plane to form a photoelectric sensor array. Through the design of supporting peripheral circuits, the source and drain voltages and gate voltages of any devices in the array can be controlled. This photoelectric sensor array can simultaneously realize the detection and processing of visual information, and is used to realize the convolutional neural network. By adjusting the gate voltage of each pixel device in the optoelectronic device, the optoelectronic device can be reconstructed into different convolutional neural networks for different processing and recognition of visual information.
  • the information processing chip that is, the reconfigurable artificial retina morphology sensor that integrates simultaneous detection, information processing and recognition is further integrated into the retinal morphology optical sensor chip, and further information processing is performed. It can be used in real-time application scenarios of edge computing such as smart security and health care.
  • a visual image convolution processing method based on the photoelectric sensor array of the retinal morphology is composed of the photoelectric devices manufactured in the embodiment, and its schematic diagram is shown in FIG. 4.
  • This array can complete the convolution operation in image processing.
  • the convolution operation of a 3 ⁇ 3 convolution kernel is specifically shown in Figure 4a above: the existing convolution kernel A:
  • P ij is the pixel value of the pixel located in row i and column j in the picture
  • Outputs is the final convolution operation result.
  • the convolution operation includes nine multiplications and eight additions.
  • the convolution kernel A starts from the upper left corner of the picture P and continuously multiplies the 3 ⁇ 3 part of the picture P along the dashed arrow.
  • the final calculation result is reassembled in the order of multiplication, and a new one is obtained.
  • picture. This picture is the result of picture P obtained under the convolution operation of the convolution kernel A.
  • the light response coefficient (G ij (V g ij )) of the devices in the array is adjusted by adjusting the back gate voltage V g ij applied to each device in the array.
  • the mode corresponds to the matrix element (a ij ) in the convolution kernel.
  • P ij represents the light intensity value of the picture pixel. Since the photocurrent (I photocurrent ) of the device is the product of the light intensity and the photoresponse matrix, the process of the device receiving light and generating the photocurrent can be regarded as a multiplication operation. Connecting multiple devices in series and using Kirchhoff's law to add up the photocurrent generated by different devices can be regarded as an addition operation. Therefore, the whole process can be regarded as the convolution operation of the array composed of light intensity and the corresponding light responsivity matrix:
  • Figure 4c shows how a large-scale photoelectric sensor array of a bionic photoelectric device with retinal morphology realizes the convolution operation of the image.
  • a bit line is arranged on each optoelectronic device of the photosensor array, the bit line corresponding to each row of optoelectronic devices is connected in series, and a signal line is arranged on each optoelectronic device, and the signal line corresponding to each photoelectric device in each column is connected in series. Lines and signal lines apply source and drain voltages to optoelectronic devices at specific positions in the array;
  • a word line is arranged on each photoelectric device of the photosensor array for applying back gate voltage to the photoelectric device of a specific row in the array.
  • the bit line and the word line are used to input the corresponding voltages for the optoelectronic devices in a specific column that intersect.
  • the back gate voltage is input to the corresponding optoelectronic devices through the word lines to complete part of the convolution operation and output the photocurrent I.
  • Figure 5 shows three different convolution operations performed on the input Lena image using the convolution kernel implemented by the 3 ⁇ 3 photoelectric sensor array in the principle verification stage using the method described in the above steps, which are inversion and edge. Enhancement and brightness correction. From the results, the results of the convolution operation performed by the photoelectric sensor array on the Lena graph are similar to the simulation results using the same convolution kernel, indicating the reconfigurable characteristics of our prototype photoelectric sensor array in terms of visual information processing.
  • a visual image recognition method based on a bionic photoelectric sensor array based on the morphology of the retina comprising:
  • Step 1 Input the information to be identified into the photoelectric sensor array, and set the back gate voltage of all photoelectric devices to 0V;
  • Step 2 collect the output current I of the photoelectric sensor array and input it into the following Sigmoid activation function:
  • I is the output current of the photoelectric sensor array
  • ⁇ is the normalization coefficient
  • Step 3 After calculating the value of the activation function f, compare it with the target value, then make a judgment and perform the error back propagation operation according to the following formula:
  • ⁇ k is the error used in the kth training
  • f k is the output value of the Sigmoid activation function during the kth training
  • Step 4 After the error is subsequently transferred to the first layer, the initial back gate value in the photoelectric sensor array is updated through the following functional relationship:
  • n is the step size
  • ⁇ is the step size of the grid voltage change
  • P is the input of visual image information
  • round is the rounding function
  • conv is the convolution function to complete a training process
  • Step 5 Repeat steps 1-4 until the error calculated in step 3 is close to or equal to 0, that is, the target picture is successfully identified from all input pictures.
  • the photoelectric sensor array is composed of Example 1 in the device, and its schematic diagram is shown in FIG. 6.
  • the "n”, “j", and “u” letters that the array needs to recognize are shown in Figure 6a.
  • the light response of each device in the photoelectric sensor array is trained as shown in Figure 6b to form three different convolution kernels. Recognize the letters given in Figure 6a.
  • Figure 6c shows the relationship between the average recognition rate for these three letters and the number of training times, and the illustration in Figure 6c shows the distribution of the back gate value of the 3 ⁇ 3 device array at the initial and after training.
  • Figure 6d shows the classification of three different writings for each of the three different types of letters "n", "j", and "u”.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Power Engineering (AREA)
  • Microelectronics & Electronic Packaging (AREA)
  • Computer Hardware Design (AREA)
  • Electromagnetism (AREA)
  • Condensed Matter Physics & Semiconductors (AREA)
  • Theoretical Computer Science (AREA)
  • Manufacturing & Machinery (AREA)
  • Molecular Biology (AREA)
  • Data Mining & Analysis (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Biomedical Technology (AREA)
  • Computational Linguistics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Biophysics (AREA)
  • Multimedia (AREA)
  • Signal Processing (AREA)
  • Inorganic Chemistry (AREA)
  • Chemical & Material Sciences (AREA)
  • Solid State Image Pick-Up Elements (AREA)

Abstract

本发明公开了一种视网膜形态光电传感阵列及其图片卷积处理方法，该光电传感器件其是在基底上具有底电极、电介质层、沟道层、源电极和漏电极的垂直堆叠的异质结结构，所述源电极和漏电极相互对峙，并置于所述沟道层的两端，所述底电极、源电极和漏电极的材料为柔性电极所用材料、惰性金属或者半金属，所述电介质层的材料为绝缘材料，所述沟道层材料为双极性材料，所述基底包括衬底以及生长于衬底表面的绝缘材料层。本发明利用新材料的物理特性，设计全新的类脑光电传感器件，更好模拟人类视觉系统的细胞功能，实现了同时传感、信息处理和识别的可重构人工神经网络。

Description

一种视网膜形态光电传感阵列及其图片卷积处理方法 技术领域
本发明涉及光电传感器、视觉芯片、人工神经网络、半导体制备工艺的交叉领域，具体涉及一种视网膜形态光电器件、其制造方法及具有该光电器件的光电传感阵列以及光电传感阵列用于对输入的视觉信息进行加工和识别的应用。
背景技术
视觉系统是人体感知外界信息的主要途径。有研究表明，通过视觉系统传输的信息量可以占到感知信息总量的80％以上。因此，通过对视觉系统建模，并利用计算机进一步发挥其功能，研究人员开辟出了计算机视觉这一研究方向。自上世纪60年代以来，计算机视觉迅猛发展，被广泛应用于各个领域，如制造、医疗、军事、安防等。在某些特定方面，计算机视觉已经取得了和人类相当，甚至优于人类的成果。
但是，天然的生物视觉系统仍然在功能和效率上具备巨大的优势。例如，人体视觉系统具有对于不同亮度的环境自动适应的特性，其探测的亮度范围横跨9个数量级(10-3-106cd/m2)；通过眼球控制运动，人眼可以快速自动的追踪正在发生偏移的目标；利用视神经节细胞的稀疏编码功能，视网膜处产生的视觉信息(等效于超过1亿像素的高清图片)可以低延时的传输到视觉皮层并被正确解码；小样本学习的能力可以使得视觉系统在输入少数图像样本之后，便可以快速学习样本特征并实现识别；在实现以上功能的同时，人脑的功耗是20W，工作频率低至～10Hz，功耗密度(Power Density)只有现有通用处理器的千分之一。因此，为了进一步的将人类视觉系统的优势与现有的计算机视觉成果相结合，研究人员提出了“神经形态视觉芯片”这一概念，目的就是通过对人体视觉系统的结构和功能上的模拟，将其在功能和功耗上的巨大优势应用在计算机视觉领域。
该领域诞生于超大规模集成电路发展早期，加州理工大学的卡尔-米德教授率先提出，可以通过设计特殊的CMOS集成电路，来模拟视网膜中发生的视觉信息早期处理过程，提升信息的处理效率。米德教授的研究团队利用蜂窝状电路结构，成功对视网膜中的感光细胞和双极性细胞进行了仿生，实现了诸如边缘识别，大范围亮度响应等功能；另外，视觉系统将信号的采集和早期处理紧密结合，并对后期处理进行分层的特点，也被借鉴并形成了信号采集、存储、处理一体化的计算架构；而进入二十一世纪以来，参 考了视觉系统的信息处理模式，研究人员又提出了帧驱动和事件驱动两种信息处理驱动模式，实现了更为高效的神经形态视觉处理系统。
然而，这一领域还面临着很多问题：例如，利用基于CMOS的集成电路来模拟视网膜中的细胞，会导致电路复杂程度的增加，降低视觉芯片的填充因子；快速发展的3D集成电路依然无法制备出类似视网膜的垂直分层结构；现有的计算架构在功耗上还是无法与生物视觉系统相提并论。此外，更重要的是，目前的神经形态视觉处理系统中的视觉探测器和处理器是分离的，对于目标物的识别最终需要在专有的加速器上完成。目前尚未存在一种光电探测器，其本身不仅可以同时进行视觉信息探测和处理，而且还具有学习和推理的能力，能够实时识别出目标物。
发明内容
发明目的：为了克服现有技术的不足，本发明提供一种网膜形态光电器件、其制造方法及具有该光电器件的光电传感阵列，其解决模拟视网膜中的细胞导致电路复杂，功耗高以及由于视觉探测与处理器的分离造成的处理速度慢、光电传感器和视觉信息处理器以及人工神经网络分离的问题。
技术方案：本发明所述的一种视网膜形态光电器件，其是在基底上具有底电极、电介质层、沟道层、源电极和漏电极的垂直堆叠的异质结结构，所述源电极和漏电极相互对峙，并置于所述沟道层的两端，所述底电极、源电极和漏电极的材料为柔性电极所用材料、惰性金属或者半金属，所述电介质层的材料为绝缘材料，所述沟道层材料为双极性材料，所述基底包括衬底以及生长于衬底表面的绝缘材料层。
进一步地，包括：
所述衬底材料包括硅、聚酰亚胺或聚二甲基硅氧烷，所述生长于衬底表面的绝缘材料层采用的材料为氧化硅、氧化铝、铪锆氧或氮化硼。
进一步地，包括：
所述沟道层的双极性材料为石墨烯、硒化钨、碲化钼、黑磷或硒化钯。
进一步地，包括：
所述电介质层的材料为氮化硼、氧化硅、氧化铝和铪锆氧中任一种或多种组成。
一种视网膜形态光电器件的制造方法，该制造方法包括：
S1在基底表面制备出底电极；
S2在底电极上直接获得电介质层或在底电极上首先获得电介质层材料，再利用材 料转移方法将不同种的电介质层材料垂直堆叠，制备具有多层结构的电介质层；
S3在电介质层上直接生长沟道层的双极性材料；或者首先获得沟道层的双极型材料，然后利用材料转移的方法将双极型材料转移到电介质层上，形成沟道层；
S4在沟道层表面制备源电极和漏电极。
进一步地，包括：
所述S2中，底电极上直接获得电介质层的方法为化学气相沉积法、化学气相传输法、分子束外延法、原子层沉积法或水热法。
进一步地，包括：
所述S3中，在电介质层上直接生长沟道层的双极性材料的方法为化学气相沉积法、化学气相传输法、分子束外延法、原子层沉积法或水热法。
进一步地，包括：
所述S1中，在基底表面制备出底电极的方法为：
S11采用紫外光刻法、电子束光刻法或掩膜版法在衬底上制备设计好底电极形状；
S12制备出底电极。
一种视网膜形态光电传感阵列，其具有上述的视网膜形态光电传感器件。
一种根据视网膜形态光电传感阵列实现的视觉图片卷积处理方法，该方法包括：
(1)在光电传感器阵列的每个光电器件上中设置位线，每一行的光电器件对应的位线串联，在每个光电器件上设置信号线，每一列的光电器件对应的信号线串联，所述位线和信号线为阵列中特定位置的光电传感器件施加源漏电压；
(2)在光电传感器阵列的每个光电器件上中设置字线，用于为阵列中特定一行的光电传感器件施加背栅电压；
(3)利用位线和字线为交叉的特定一列中的光电传感器件输入对应电压，同时将背栅电压通过字线输入给对应的光电器件，完成部分卷积运算，并输出结果，即：
Figure PCTCN2020081313-appb-000001
其中，
Figure PCTCN2020081313-appb-000002
为第m行第1列的背栅电压值，P m1为第m行第1列上光电器件的视觉图片信息输入，
Figure PCTCN2020081313-appb-000003
为第m行第1列的光响应度；m为卷积核的总行数；
(4)根据步骤(3)的方法采用m×m的卷积核完成整个光电传感器阵列的卷积运算。
一种根据视网膜形态仿生光电传感器阵列实现的视觉图片识别方法，其特征在于，该方法包括：
步骤1、将待识别信息输入到光电传感器阵列中，并将所有光电器件的背栅电压设置为0V；
步骤2、随后采集光电传感器阵列的输出电流I，并将它输入到下面的Sigmoid激活函数：
f＝(1+e -αI) -1
其中，I即为光电传感器阵列的输出电流，α为归一化系数；
步骤3、计算出激活函数f的值后，与目标值进行比较后，然后作出判断并根据下面的公式执行误差反向传播操作：
Figure PCTCN2020081313-appb-000004
其中，δ k为第k次训练时所用的误差，
Figure PCTCN2020081313-appb-000005
为第k次训练时的理论输出值，f k为第k次训练时的Sigmoid激活函数的输出值；
步骤4、误差随后传递到第一层后，通过下面的函数关系实现对光电传感器阵列中初始的背栅值进行更新：
Figure PCTCN2020081313-appb-000006
其中，n为步长，β为栅压变化的步长，P为视觉图片信息的输入，round为四舍五入求整函数，conv为卷积函数，以此完成一个训练流程；
步骤5、循环步骤1-4，直到步骤3中计算出的误差接近或者等于0，即，从所有输入图片中成功识别出目标图片。
有益效果：1、本发明利用新材料的物理特性，设计全新的视网膜形态光电传感器件，更好地模拟人类视觉系统的细胞功能，优化器件性能并降低电路复杂程度；2、本发明的光电传感器阵列可以同时实现视觉信息的探测、处理和识别的卷积神经网络，通过调节光电器件中每一个像素器件的栅压，这种光电器件可以被重构成为不同的卷积神经网络，用于实现视觉信息的不同处理加工的操作和识别；识别方法简单；3、本发明中对图片信息的同时探测、信息处理和识别的可重构人工视网膜形态传感器能够进一步集成为视网膜形态光学传感芯片，用于智能安防、健康医疗等边缘计算的实时应用场景。
附图说明
图1是本发明实施例2中的光电器件结构示意图；
图2是本发明实施例2中的光电器件光学图片；
图3是本发明实施例2所述的光电器件的光电响应特性；其中，图3a为光强不变时，器件光电流响应随背栅电压的变化图；图3b为背栅不变时，器件光电流随光强的变化图；图3c为相同背栅下，不同光波长的器件光电响应特性；
图4是本发明所述的光电器件的光电传感器阵列与外围控制电路示意图；
图5是本发明所述的不同背栅调节下，基于视网膜形态光电器件的光电传感器阵列对图片处理的不同操作，从上到下依次是：反相、边缘增强、亮度校正；
图6是训练卷积神经网络以实现对图像的识别的过程示意图。
具体实施方式
本发明首先公开一种视网膜形态光电器件及其制造方法，该仿生光电器件是在基底上具有底电极、电介质层、沟道层、源电极和漏电极的垂直堆叠的异质结结构，源电极和漏电极相互对峙，并置于沟道层的两端，底电极、源电极和漏电极的材料为柔性电极所用材料、惰性金属或者半金属，电介质层的材料为绝缘材料，沟道层材料为双极性材料，基底包括衬底以及生长于衬底表面的绝缘材料层。
其中，衬底材料可为硅、聚酰亚胺或聚二甲基硅氧烷，生长于衬底表面的绝缘材料层采用的材料为氧化硅、氧化铝、铪锆氧或氮化硼。
可以由氮化硼、氧化硅、氧化铝、铪锆氧等其中的一种组成，也可以由多种组成，构成垂直方向的异质结构。
沟道层的双极性材料具有双极性的电学特点，即通过场效应的调节方式，该类材料既可以成为电子型掺杂材料，也可以成为空穴型掺杂材料，包括石墨烯、硒化钨、碲化 钼、黑磷或硒化钯。
底电极与源漏电极由惰性金属、柔性电极或者半金属材料组成，如铂、金、钯、氧化铟锡、石墨烯等
其次，本发明还公开一种视网膜形态光电器件的制造方法，该制造方法包括：
S1在基底表面制备出底电极；
电极材料为可用物理气相沉积或磁控溅射等方法制备的导电材料如惰性金属、氮化钛等，或为半金属型二维原子晶体材料如石墨烯等，两种材料的电极的制备方法略有不同。
首先，对于可用物理气相沉积或磁控溅射等方法制备的导电材料来说，具体步骤包括：
(1)先在衬底上旋涂一层光刻胶，用电子束光刻或紫外光刻的手段在衬底上的指定位置曝光出自定义的电极图案(所需电极图案暴露，其余部分被光刻胶遮盖)，用显影液显影出图案；
(2)用电子束蒸发，热蒸发或磁控溅射的手段在制备好图案的衬底上生长一层导电材料薄膜，所用材料为惰性金属如金、铂、钯等，或为导电化合物如氮化钛等；
(3)将该衬底的放入丙酮溶液中，利用丙酮溶解多余的光刻胶，多余的光刻胶表面的导电材料薄膜也会随之从衬底上脱落，衬底上留下的即是设计好图案的导电薄膜电极。
其次，对于半金属型二维原子晶体材料电极来说，
首先选用在衬底上制备好的单层，多层或厚层半金属型二维原子晶体薄膜，旋涂一层光刻胶，用电子束光刻或紫外光刻的手段在石墨烯上的指定位置曝光出自定义的电极图案(所需电极图案被光刻胶遮盖，其余部分暴露)，用显影液显影出图案；
接着用等离子体刻蚀的手段将多余的半金属型二维原子晶体刻蚀掉，留下被光刻胶遮盖的所需电极图案的半金属型二维原子晶体材料；
将该衬底放入丙酮溶液中，利用丙酮溶解残余的光刻胶，衬底上留下的即是设计好图案的半金属型二维原子晶体薄膜电极；
最后将制备好的半金属型二维原子晶体薄膜电极转移至样品上的所需位置。
S2在底电极上直接获得电介质层或在底电极上首先获得电介质层材料，再利用材料转移方法将不同种的电介质层材料垂直堆叠，制备具有多层结构的电介质层；
具体的，电介质层由绝缘材料组成，包括氮化硼、氧化硅、氧化铝、铪锆氧等。可以通过化学气相沉积(CVD)、化学气相传输(CVT)、分子束外延(MBE)、原子层沉积(ALD)、磁控溅射、水热法等方法直接获得；
也可以通过化学气相传输(CVT)的方法先获得体块材料，然后采取机械剥离或超声旋涂等方法得到层状材料；
电介质层可以是只有一种材料构成，也可以在得到对应的材料之后，利用转移的方法获得在垂直方向具有多层结构的异质结电介质层。电介质层可以直接生长在底电极的表面，也可以通过转移的方法从其他衬底转移到底电极上表面。
S3在电介质层上直接生长沟道层的双极性材料；或者首先获得沟道层的双极型材料，然后利用材料转移的方法将双极型材料转移到电介质层上，形成沟道层。
具体的，沟道层由双极型材料组成，包括硒化钨、碲化钼、黑磷、石墨烯、硒化钯等。可以通过化学气相沉积(CVD)、分子束外延(MBE)、原子层沉积(ALD)、磁控溅射、水热法等方法直接获得；也可以通过化学气相传输(CVT)的方法先获得体块材料，然后采取机械剥离或超声旋涂等方法得到层状材料；电介质层可以直接生长在底电极的表面，也可以通过转移的方法从其他衬底转移到电介质层上表面。
只要条件允许，本发明可以选用任何类型的电极材料、电介质材料和沟道材料。这里所述的“条件允许”是指需要满足一些限定性条件，比如：电极材料、电介质材料和沟道材料不能溶于水和丙酮，也不能溶于光刻胶；电极材料、电介质材料和沟道材料不能与水，丙酮和光刻胶发生化学反应。本领域技术人员，只要满足限定性条件，本发明中的电极材料可以用其他金属性或者半金属性材料，电介质材料可以用其他绝缘材料，沟道材料也可以用其他双极型材料。
S4在沟道层表面制备源电极和漏电极。
源电极和漏电极的材料以及制备方法均与底电极相同，在此不再赘述。
本发明提供多个制造光电器件的实施例，具体包括：
实施例1
金属底电极的制备：首先用匀胶机在衬底上旋涂一层PMMA(聚甲基丙烯酸甲酯)，用电子束曝光的方法在PMMA上曝出指定的电极图案，并用显影液使电极图案暴露从而露出下面的衬底；接着用电子束蒸发在PMMA和衬底上生长厚度为40nm左右的金属薄膜；最后，将PMMA与衬底一起放入丙酮溶液，使得PMMA溶解，从而带走多余 的金属薄膜。衬底上留下的便是前面指定形状的金属膜底电极。
电介质层的制备：首先利用原子层沉积在有底电极的衬底上生长厚度为10nm左右的氧化铝薄膜；然后利用机械剥离法获得薄层的六方氮化硼，厚度为5-50nm之间；利用转移的方法将薄层六方氮化硼转移到氧化铝薄膜表面，从而形成氧化铝/六方氮化硼的异质结构电介质层。
沟道层的制备：利用化学气相沉积(CDV)的方法在电介质表面直接制备薄层硒化钨。
源漏电极的制备：源漏电极的制备与前面提到的底电极的制备相同，但是源漏电极必须要与底电极在水平方向有交叠的区域，并保证源漏电极之间存在沟道层的材料。
实施例2
如图1和2所示，金属底电极(Bottom Electrode)的制备方法为：首先用匀胶机在衬底上旋涂一层PMMA，用电子束曝光的方法在PMMA上曝出指定的电极图案，并用显影液使电极图案暴露从而露出下面的衬底；接着用电子束蒸发在PMMA和衬底上生长厚度为40nm左右的金属薄膜；最后，将PMMA与衬底一起放入丙酮溶液，使得PMMA溶解，从而带走多余的金属薄膜。衬底上留下的便是前面指定形状的金属膜底电极。
电介质层的制备：首先利用原子层沉积在有底电极的衬底上生长厚度为10nm左右的氧化铝薄膜(Al 2o 3)；然后利用机械剥离法获得薄层的六方氮化硼(h-BN)，厚度为5-50nm之间；利用转移的方法将薄层六方氮化硼转移到氧化铝薄膜表面，从而形成氧化铝/六方氮化硼的异质结构电介质层。
沟道层的制备：利用机械剥离的方法先获得薄层硒化钨(WSe 2)，再利用转移的方法转移到电介质层表面。
源漏电极的制备：源极为Source，漏极为Drain，源漏电极的制备与前面提到的底电极的制备相同，但是源漏电极必须要与底电极在水平方向有交叠的区域，并保证源漏电极之间存在沟道层的材料。
实施例3
金属底电极的制备：首先用匀胶机在衬底上旋涂一层PMMA，用电子束曝光的方法在PMMA上曝出指定的电极图案，并用显影液使电极图案暴露从而露出下面的衬底；接着用电子束蒸发在PMMA和衬底上生长厚度为40nm左右的金属薄膜；最后，将PMMA与衬底一起放入丙酮溶液，使得PMMA溶解，从而带走多余的金属薄膜。衬底 上留下的便是前面指定形状的金属膜底电极。
电介质层的制备：首先利用原子层沉积在有底电极的衬底上生长厚度为5-50nm左右的氧化铝薄膜。
沟道层的制备：利用化学气相沉积(CDV)的方法在电介质表面直接制备薄层硒化钨。
源漏电极的制备：源漏电极的制备与前面提到的底电极的制备相同，但是源漏电极必须要与底电极在水平方向有交叠的区域，并保证源漏电极之间存在沟道层的材料。
实施例4
金属底电极的制备：首先用匀胶机在衬底上旋涂一层PMMA，用电子束曝光的方法在PMMA上曝出指定的电极图案，并用显影液使电极图案暴露从而露出下面的衬底；接着用电子束蒸发在PMMA和衬底上生长厚度为40nm左右的金属薄膜；最后，将PMMA与衬底一起放入丙酮溶液，使得PMMA溶解，从而带走多余的金属薄膜。衬底上留下的便是前面指定形状的金属膜底电极。
电介质层的制备：首先利用原子层沉积在有底电极的衬底上生长厚度为5-50nm左右的氧化铝薄膜。
沟道层的制备：利用机械剥离的方法先获得薄层硒化钨，再利用转移的方法转移到电介质层表面。
源漏电极的制备：源漏电极的制备与前面提到的底电极的制备相同，但是源漏电极必须要与底电极在水平方向有交叠的区域，并保证源漏电极之间存在沟道层的材料。
实施例5
金属底电极的制备：首先用匀胶机在衬底上旋涂一层PMMA，用电子束曝光的方法在PMMA上曝出指定的电极图案，并用显影液使电极图案暴露从而露出下面的衬底；接着用电子束蒸发在PMMA和衬底上生长厚度为40nm左右的金属薄膜；最后，将PMMA与衬底一起放入丙酮溶液，使得PMMA溶解，从而带走多余的金属薄膜。衬底上留下的便是前面指定形状的金属膜底电极。
电介质层的制备：利用机械剥离法获得薄层的六方氮化硼，厚度为5-50nm之间；利用转移的方法将薄层六方氮化硼转移到氧化铝薄膜表面。
沟道层的制备：利用化学气相沉积(CDV)的方法在电介质表面直接制备薄层硒化钨。
源漏电极的制备：源漏电极的制备与前面提到的底电极的制备相同，但是源漏电极必须要与底电极在水平方向有交叠的区域，并保证源漏电极之间存在沟道层的材料。
实施例6
金属底电极的制备：首先用匀胶机在衬底上旋涂一层PMMA，用电子束曝光的方法在PMMA上曝出指定的电极图案，并用显影液使电极图案暴露从而露出下面的衬底；接着用电子束蒸发在PMMA和衬底上生长厚度为40nm左右的金属薄膜；最后，将PMMA与衬底一起放入丙酮溶液，使得PMMA溶解，从而带走多余的金属薄膜。衬底上留下的便是前面指定形状的金属膜底电极。
电介质层的制备：利用机械剥离法获得薄层的六方氮化硼，厚度为5-50nm之间；利用转移的方法将薄层六方氮化硼转移到氧化铝薄膜表面。
沟道层的制备：利用机械剥离的方法先获得薄层硒化钨，再利用转移的方法转移到电介质层表面。
源漏电极的制备：源漏电极的制备与前面提到的底电极的制备相同，但是源漏电极必须要与底电极在水平方向有交叠的区域，并保证源漏电极之间存在沟道层的材料。
实施例1，3-6制造的光电器件结构与实施例中的相同，但材料有所不同，附图没有一一列举。
本光电器件对不同背栅电压、光强和波长下展现出不同的光电响应特性，结果如图3所示。图3a展示了光电器件在不同背栅电压下的响应特性。图3b展示了不同光强的光照射下光电器件的响应特性。图3c和3d展示了光电器件能够工作在整个可见光光谱范围内。
一种视网膜形态光电传感阵列，其具有上述实施例制造的视网膜形态光电器件。在阵列的基础上结合边缘电路，可以制作视网膜形态光电传感芯片。
上述的光电传感器阵列是将上述的视网膜形态光电器件在平面上集成，形成光电传感器阵列，通过配套的外围电路设计，可以实现对于阵列中任意器件的源漏电压与门电压的控制。这种光电传感器阵列可以同时实现视觉信息的探测和处理，用于实现卷积神经网络。通过调节光电器件中每一个像素器件的栅压，这种光电器件可以被重构成为不同的卷积神经网络，用于实现视觉信息的不同处理和识别。
信息处理芯片，即将集同时探测、信息处理和识别的可重构人工视网膜形态传感器进一步集成为视网膜形态光学传感芯片，并进行更进一步的信息处理。可以用于智能安 防、健康医疗等边缘计算的实时应用场景。
一种根据视网膜形态光电传感阵列实现的视觉图片卷积处理方法，首先本光电传感器阵列由实施例制造的光电器件组成，其示意图如图4所示。
本阵列可以完成图像处理中的卷积运算。如一个3×3卷积核的卷积运算具体如上图4a所示：现有卷积核A：
Figure PCTCN2020081313-appb-000007
和图片P。做卷积时，从图片P的最左上角选取3×3的阵列，将该阵列的像素值与卷积核的数值相乘，其运算规则如下：
Figure PCTCN2020081313-appb-000008
P ij是图片中位于i行j列的像素的像素值，Outputs是最后的卷积运算成果。
该卷积运算包含了九个乘法与八个加法。按照这个方法，卷积核A从图片P的左上角开始，沿着虚线箭头不断与图片P的3×3部分相乘，最后的运算结果按照相乘的顺序重新组装，可以获得一张新的图片。该图片便是图片P在卷积核A的卷积运算下得到的结果。
为了利用我们的阵列实现这一功能，如图4b所示，将阵列中器件的光响应系数(G ij(V g ij))通过调控施加在阵列中每个器件的背栅电压V g ij的方式与卷积核中的矩阵元(a ij)对应。P ij是代表图片像素的光强值。由于器件的光电流(I photocurrent)是光强与光响应度矩阵的乘积，因此器件接收光照并产生光电流的过程可以被看成是乘法运算。将多个器件进行串联，利用基尔霍夫定律将不同器件产生的光电流相加可以被看成是加法运算。因此，整个过程便可以看作将光强组成的阵列与对应的光响应度矩阵组成的阵列做了卷积运算：
Figure PCTCN2020081313-appb-000009
图4c展示了一个较大规模的视网膜形态仿生光电器件的光电传感器阵列如何实现对图片的卷积运算。
在光电传感器阵列的每个光电器件上中设置位线，每一行的光电器件对应的位线串联，在每个光电器件上设置信号线，每一列的光电器件对应的信号线串联，所述位线和 信号线为阵列中特定位置的光电器件施加源漏电压；
在光电传感器阵列的每个光电器件上中设置字线，用于为阵列中特定一行的光电器件施加背栅电压。
利用位线和字线为交叉的特定一列中的光电器件输入对应电压，同时将背栅电压通过字线输入给对应的光电器件，完成部分卷积运算，并输出光电流I。
图5展示了利用上述步骤所述的方法在原理验证阶段采用了3×3光电器传感器阵列实施的卷积核对输入的Lena图片进行的三种不同的卷积运算操作，分别是反相、边缘增强、亮度校正。从结果看，光电传感器阵列对Lena图进行的卷积运算结果和采用相同卷积核的模拟结果类似，表明我们原型光电传感器阵列在视觉信息加工处理方面的可重构特性。
需要说明的是，上述各实施例是实验室实验处理过程，实际工业生产中，根据其相应的生产条件得做出相应的调整，只要这种调整不破坏本发明的精神，属于本发明的保护范围。
一种根据视网膜形态仿生光电传感阵列实现的视觉图片识别方法，该方法包括：
步骤1、将待识别信息输入到光电传感器阵列中，并将所有光电器件的背栅电压设置为0V；
步骤2、随后采集光电传感器阵列的输出电流I，并将它输入到下面的Sigmoid激活函数：
f＝(1+e -αI) -1
其中，I即为光电传感器阵列的输出电流，α为归一化系数；
步骤3、计算出激活函数f的值后，与目标值进行比较后，然后作出判断并根据下面的公式执行误差反向传播操作：
Figure PCTCN2020081313-appb-000010
其中，δ k为第k次训练时所用的误差，
Figure PCTCN2020081313-appb-000011
为第k次训练时的理论输出值，f k为第k次训练时的Sigmoid激活函数的输出值；
步骤4、误差随后传递到第一层后，通过下面的函数关系实现对光电传感器阵列中初始的背栅值进行更新：
Figure PCTCN2020081313-appb-000012
其中，n为步长，β为栅压变化的步长，P为视觉图片信息的输入，round为四舍五入求整函数，conv为卷积函数，以此完成一个训练流程；
步骤5、循环步骤1-4，直到步骤3中计算出的误差接近或者等于0，即，从所有输入图片中成功识别出目标图片。
具体的，本光电传感器阵列由器件中的实例1组成，其示意图如图6所示。本阵列需要识别的“n”、“j”、“u”字母如图6a所示，通过如图6b的方式训练光电传感器阵列中每一个器件的光响应，形成三个不同的卷积核，识别图6a中的给出的字母。图6c展示了针对这三个字母的平均识别率和训练次数的关系，图6c的插图展示了初始和训练完成后3×3器件阵列的背栅值分布情况。图6d给出了分别针对三类不同字母“n”、“j”、“u”中每一类三种不同写法的分类情况。

Claims (11)

  1. 一种视网膜形态光电传感器件，其特征在于，其是在基底上具有底电极、电介质层、沟道层、源电极和漏电极的垂直堆叠的异质结结构，所述源电极和漏电极相互对峙，并置于所述沟道层的两端，所述底电极、源电极和漏电极的材料为柔性电极所用材料、惰性金属或者半金属，所述电介质层的材料为绝缘材料，所述沟道层材料为双极性材料，所述基底包括衬底以及生长于衬底表面的绝缘材料层。
  2. 根据权利要求1所述的视网膜形态光电传感器件，其特征在于，所述衬底材料包括硅、聚酰亚胺或聚二甲基硅氧烷，所述生长于衬底表面的绝缘材料层采用的材料为氧化硅、氧化铝、铪锆氧或氮化硼。
  3. 根据权利要求1所述的视网膜形态光电传感器件，其特征在于，所述沟道层的双极性材料为石墨烯、硒化钨、碲化钼、黑磷或硒化钯。
  4. 根据权利要求1所述的视网膜形态光电传感器件，其特征在于，所述电介质层的材料为氮化硼、氧化硅、氧化铝和铪锆氧中任一种或多种组成。
  5. 一种权利要求1-4任一项所述的视网膜形态光电传感器件的制造方法，其特征在于，该制造方法包括：
    S1在基底表面制备出底电极；
    S2在底电极上直接获得电介质层或在底电极上首先获得电介质层材料，再利用材料转移方法将电介质层材料垂直堆叠，制备具有多层结构的电介质层；
    S3在电介质层上直接生长沟道层的双极性材料；或者首先获得沟道层的双极型材料，然后利用材料转移的方法将双极型材料转移到电介质层上，形成沟道层；
    S4在沟道层表面制备源电极和漏电极。
  6. 根据权利要求5所述的视网膜形态光电传感器件的制造方法，其特征在于，所述S2中，底电极上直接获得电介质层的方法为化学气相沉积法、化学气相传输法、分子束外延法、原子层沉积法或水热法。
  7. 根据权利要求5所述的视网膜形态光电传感器件的制造方法，其特征在于，所述S3中，在电介质层上直接生长沟道层的双极性材料的方法为化学气相沉积法、化学气相传输法、分子束外延法、原子层沉积法或水热法。
  8. 根据权利要求5所述的视网膜形态光电传感器件的制造方法，其特征在于，所述S1中，在基底表面制备出底电极的方法为：
    S11采用紫外光刻法、电子束光刻法或掩膜版法在衬底上制备设计好底电极形状；
    S12制备出底电极。
  9. 一种视网膜形态光电传感器阵列，其特征在于，其具有权利要求1-4任一项所述的视网膜形态光电传感器件。
  10. 一种根据权利要求9所述的视网膜形态光电传感器阵列实现的视觉图片卷积处理方法，其特征在于，该方法包括：
    (1)在光电传感器阵列的每个光电器件上中设置位线，每一行的光电器件对应的位线串联，在每个光电器件上设置信号线，每一列的光电器件对应的信号线串联，所述位线和信号线为阵列中特定位置的光电传感器件施加源漏电压；
    (2)在光电传感器阵列的每个光电器件上中设置字线，用于为阵列中特定一行的光电传感器件施加背栅电压；
    (3)利用位线和字线为交叉的特定一列中的光电传感器件输入对应电压，同时将背栅电压通过字线输入给对应的光电器件，完成部分卷积运算，并输出结果，即：
    Figure PCTCN2020081313-appb-100001
    其中，
    Figure PCTCN2020081313-appb-100002
    为第m行第1列的背栅电压值，P m1为第m行第1列上光电器件的视觉图片信息输入，
    Figure PCTCN2020081313-appb-100003
    为第m行第1列的光响应度；m为卷积核的总行数；
    (4)根据步骤(3)的方法采用m×m的卷积核完成整个光电传感器阵列的卷积运算。
  11. 一种根据权利要求9所述的视网膜形态仿生光电传感器阵列实现的视觉图片识别方法，其特征在于，该方法包括：
    步骤1、将待识别信息输入到光电传感器阵列中，并将所有光电器件的背栅电压设置为0V；
    步骤2、随后采集光电传感器阵列的输出电流I，并将它输入到下面的Sigmoid激活函数：
    f＝(1+e -αI) -1
    其中，I即为光电传感器阵列的输出电流，α为归一化系数；
    步骤3、计算出激活函数f的值后，与目标值进行比较后，然后作出判断并根据下 面的公式执行误差反向传播操作：
    Figure PCTCN2020081313-appb-100004
    其中，δ k为第k次训练时所用的误差，
    Figure PCTCN2020081313-appb-100005
    为第k次训练时的理论输出值，f k为第k次训练时的Sigmoid激活函数的输出值；
    步骤4、误差随后传递到第一层后，通过下面的函数关系实现对光电传感器阵列中初始的背栅值进行更新：
    Figure PCTCN2020081313-appb-100006
    其中，n为步长，β为栅压变化的步长，P为视觉图片信息的输入，round为四舍五入求整函数，conv为卷积函数，以此完成一个训练流程；
    步骤5、循环步骤1-4，直到步骤3中计算出的误差接近或者等于0，即，从所有输入图片中成功识别出目标图片。
PCT/CN2020/081313 2020-03-17 2020-03-26 一种视网膜形态光电传感阵列及其图片卷积处理方法 WO2021184407A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US17/769,609 US20240234617A9 (en) 2020-03-17 2020-03-26 Retinomorphic sensor array and image convolution method therefor

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010189068.4 2020-03-17
CN202010189068.4A CN111370526B (zh) 2020-03-17 2020-03-17 一种视网膜形态光电传感阵列及其图片卷积处理方法

Publications (1)

Publication Number Publication Date
WO2021184407A1 true WO2021184407A1 (zh) 2021-09-23

Family

ID=71211807

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/081313 WO2021184407A1 (zh) 2020-03-17 2020-03-26 一种视网膜形态光电传感阵列及其图片卷积处理方法

Country Status (2)

Country Link
CN (1) CN111370526B (zh)
WO (1) WO2021184407A1 (zh)

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111770245B (zh) * 2020-07-29 2021-05-25 中国科学院长春光学精密机械与物理研究所 一种类视网膜图像传感器的像素结构
CN111950720A (zh) * 2020-08-26 2020-11-17 南京大学 一种新型类脑视觉系统
CN113451423B (zh) * 2021-07-27 2023-03-24 湖南大学 一种基于等离激元效应的异质结光电突触器件
CN114141884A (zh) * 2021-12-14 2022-03-04 上海集成电路制造创新中心有限公司 可重构肖特基二极管
CN114784050B (zh) * 2022-06-17 2022-09-27 湖南大学 一种神经形态视觉传感器及其应用和制备方法

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106653850A (zh) * 2016-09-23 2017-05-10 南京大学 基于石墨烯/碳纳米管复合吸收层的人工神经突触晶体管
US20180287056A1 (en) * 2017-04-04 2018-10-04 SK Hynix Inc. Synapse and synapse array
CN110215606A (zh) * 2019-06-25 2019-09-10 武汉华星光电技术有限公司 一种人造视网膜神经柔性微电极装置及其制作方法
CN110276440A (zh) * 2019-05-19 2019-09-24 南京惟心光电系统有限公司 一种基于光电计算阵列的卷积运算加速器及其方法

Family Cites Families (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104252622A (zh) * 2014-10-15 2014-12-31 倪蔚民 移动终端前置和虹膜识别一体化光电成像系统及方法
CN105810750B (zh) * 2014-12-29 2019-02-01 中芯国际集成电路制造(上海)有限公司 一种碳纳米管神经元器件及其制作方法
CN105070781B (zh) * 2015-05-21 2017-03-01 清华大学 一种碳纳米管柔性光敏器件及其制备方法
CN106847817B (zh) * 2015-12-04 2019-07-23 中国科学院宁波材料技术与工程研究所 具有多端结构的人造突触器件及其制法和应用
WO2018193333A1 (ja) * 2017-04-21 2018-10-25 株式会社半導体エネルギー研究所 画像処理方法および受像装置
CN107527968A (zh) * 2017-07-16 2017-12-29 北京工业大学 一种石墨烯‑二硫化钼侧向异质结光电探测器结构
CN108281554B (zh) * 2018-01-26 2020-03-17 电子科技大学 一种量子点结构光电探测器及其制备方法
CN208507681U (zh) * 2018-04-02 2019-02-15 云南师范大学 一种基于二硫化钼与二硒化钨垂直异质结的铁电存储器
CN109817756A (zh) * 2019-01-16 2019-05-28 复旦大学 基于二维异质结光波长诱导的光电存储器及其制备方法

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106653850A (zh) * 2016-09-23 2017-05-10 南京大学 基于石墨烯/碳纳米管复合吸收层的人工神经突触晶体管
US20180287056A1 (en) * 2017-04-04 2018-10-04 SK Hynix Inc. Synapse and synapse array
CN110276440A (zh) * 2019-05-19 2019-09-24 南京惟心光电系统有限公司 一种基于光电计算阵列的卷积运算加速器及其方法
CN110215606A (zh) * 2019-06-25 2019-09-10 武汉华星光电技术有限公司 一种人造视网膜神经柔性微电极装置及其制作方法

Also Published As

Publication number Publication date
CN111370526A (zh) 2020-07-03
US20240136461A1 (en) 2024-04-25
CN111370526B (zh) 2023-12-26

Similar Documents

Publication Publication Date Title
WO2021184407A1 (zh) 一种视网膜形态光电传感阵列及其图片卷积处理方法
US11861489B2 (en) Convolutional neural network on-chip learning system based on non-volatile memory
Wang et al. Networking retinomorphic sensor with memristive crossbar for brain-inspired visual perception
Choi et al. Curved neuromorphic image sensor array using a MoS2-organic heterostructure inspired by the human visual recognition system
Liao et al. Neuromorphic vision sensors: Principle, progress and perspectives
Seo et al. Artificial optic-neural synapse for colored and color-mixed pattern recognition
Feng et al. 2D photonic memristor beyond graphene: progress and prospects
WO2022041091A1 (zh) 一种新型类脑视觉系统
CN112542515A (zh) 一种光电调控神经突触晶体管及其制备方法
CN109460819B (zh) 一种用于模拟生物体光突触的方法及器件
Lee et al. Adaptive precision cnn accelerator using radix-x parallel connected memristor crossbars
Guo et al. Retina-inspired in-sensor broadband image preprocessing for accurate recognition via the flexophototronic effect
Lee et al. Nanomaterial-based synaptic optoelectronic devices for in-sensor preprocessing of image data
CN113497063A (zh) 一种基于二维铁电半导体的异源突触电子器件及其制备方法
Yang et al. Research and Application of Visual Object Recognition System Based on Deep Learning and Neural Morphological Computation
Fan et al. ReS 2 charge trapping synaptic device for face recognition application
Kim et al. Tactile neuromorphic system: convergence of triboelectric polymer sensor and ferroelectric polymer synapse
Jiang et al. Self-Powered Organic Optoelectronic Synapses with Binarized Weights for Noise-Suppressed Visual Perception and High-Robustness Inference
US20240234617A9 (en) Retinomorphic sensor array and image convolution method therefor
Wang et al. Digital image processing realized by memristor-based technologies
Lv et al. Low-dimensional optoelectronic synaptic devices for neuromorphic vision sensors
Mohta et al. Optically excited artificial synapse based on α-In2Se3 FETs on Ta2O5
Ahmed et al. Introduction to Neuromorphic Computing Systems
Kumar et al. Neuromorphic Spatiotemporal Information Processing Using Neuro-Photodetector Systems
Zhang et al. High performance artificial visual system with plasmon-enhanced 2d material neural network

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20925522

Country of ref document: EP

Kind code of ref document: A1

WWE Wipo information: entry into national phase

Ref document number: 17769609

Country of ref document: US

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20925522

Country of ref document: EP

Kind code of ref document: A1