An artificial vision system based on carbon nanotube / all-inorganic perovskite quantum dots and a manufacturing method thereof

Abstract

The application relates to the research and application field of a new type of nanometer semiconductor material neuromorphic visual sensor, in particular to an artificial vision system based on carbon nanotubes / inorganic perovskite quantum dots and a manufacturing method. The inorganic perovskite quantum dots have excellent photoelectric response performance and adjustable spectral response range, and the carbon nanotubes have excellent carrier mobility and current switch ratio. In the application, the inorganic perovskite quantum dots are used as a photosensitive layer and a photo-generated charge trapping layer, and the high-purity semiconductor carbon nanotube film is used as a charge transport layer, a photoelectric transistor is formed based on the new material system, the advantages of the semiconductor carbon nanotube and the inorganic perovskite quantum dot are fully utilized, super-high response, super-high detection and super-high signal-to-noise ratio are obtained, and the photoelectric detection capacity is obtained, and the biological synapse behavior can be simulated to realize the functions of the artificial vision system such as array-level image detection, image memory and image learning.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of research and application of a new type of nanometer semiconductor material neuromorphic vision sensor, in particular to an artificial vision system based on carbon nanotubes / inorganic perovskite quantum dots and a manufacturing method thereof. BACKGROUND

[0002] The human vision system is essential for survival and learning. The vision system composed of retina and visual cortex realizes efficient image processing, in which the retina can sense light stimulation and perform parallel preprocessing of image information, and then transmit the information to the visual cortex of the brain for further processing [1] . In recent years, artificial vision systems have used conventional complementary metal oxide silicon image sensors or charge coupled device cameras [2] connected with digital systems performing machine vision algorithms [3] . However, these conventional digital artificial vision systems often suffer from high power consumption, large size and high cost in practical applications [4] . Inspired by living organisms, neuromorphic vision sensors that integrate image sensing, storage and processing functions in one are expected to solve these problems [5][6][7] .

[0003] In the development of neuromorphic vision sensor field, high-performance devices with ultra-high response, ultra-high detection and ultra-high signal-to-noise ratio are the key to solving the problems of poor imaging in extreme dark conditions, non-flexibility, low integration and poor stability [8] , and are an important factor to realize flexible wearable devices and intelligent robot vision systems. In terms of material selection, inorganic perovskite quantum dots have excellent light absorption efficiency and high stability [9] ; carbon nanotubes significantly improve the signal-to-noise ratio of sensor detection due to their excellent carrier mobility and high current on-off ratio

[10] . The combination of the two has good flexibility and can be uniformly and large-area film formed and has long-term stability, providing a new strategy for the design and construction of high-performance neuromorphic vision sensors and large-scale array integration.

[0004] [1] Kolb, H. How the retina works: much of the construction of an image takes place in the retina itself through the use of specialized neural circuits. Am. Sci. 91, 28-35 (2003).

[0005] [2]Zhang,K.et al.Origami silicon optoelectronics for hemisphericalelectronic eye systems.Nat.Commun.8,1782(2017).

[0006] [3]LeCun,Y.,Bengio,Y.&Hinton,G.Deep learning.Nature 521,436-444(2015).

[0007] [4]Giacomo Indiveri,R.D.Neuromorphic vision sensors.Science 288,1189-1190(2000).

[0008] [5]Wang,Y.et al.Photonic synapses based on inorganic perovskitequantum dots for neuromorphic computing.Adv.Mater.30,1802883(2018).

[0009] [6]Zhou,F.et al.Optoelectronic resistive random access memory forneuromorphic vision sensors.Nat.Nanotechnol.14,776-782(2019).

[0010] [7]Mennel,L.et al.Ultrafast machine vision with 2D material neuralnetwork image sensors.Nature 579, 62-66(2020).

[0011] [8]Chai,Y.In-sensor computing for machinevision.Nature579,32-33(2020).

[0012] [9]Yang, D.etal.CsPbBr3quantumdots2.0:benzenesulfonicacid equivalentligand awakens complete purification.Adv.Mater.31,1900767(2019).

[0013]

[10] Qu, T. et al. A flexible carbon nanotube sen-memory device. Adv. Mater. 32, 1907288 (2020). Summary of the Invention

[0014] The purpose of this invention is to provide an artificial vision system and its fabrication method based on carbon nanotubes / all-inorganic perovskite quantum dots. It utilizes a uniform, large-area-capable carbon nanotube / all-inorganic perovskite quantum dot composite film as the channel material, overcoming the problems of poor imaging, lack of flexibility, low integration, and poor stability of previous vision sensors under extremely low light conditions. This results in ultra-high photoelectric performance (responsivity, light-to-dark current ratio, external quantum efficiency, and detectivity) and high storage performance, and it can simulate biological synaptic behavior (long / short-term plasticity, double-pulse facilitation, etc.) for data processing. Furthermore, by training a highly integrated sensor array under low-light conditions, it achieves image reinforcement learning that simulates biological processes, opening up new applications for physical devices in the field of artificial intelligence.

[0015] The technical solution of the present invention:

[0016] An artificial vision system based on carbon nanotubes / all-inorganic perovskite quantum dots is disclosed. The artificial vision system is a highly integrated neuromorphic vision sensor array, with the following specific structure: The substrate is a rigid substrate or a flexible substrate. The top of the substrate has a gate and leads, which are used to connect the gate and source to an external test circuit, respectively. The top and sides of the gate and leads are covered with a gate insulating layer. The top of the gate insulating layer is arranged with a source, a semiconductor channel, a drain, and leads in sequence. The source and drain are connected through the semiconductor channel. The leads are used to connect the drain to an external test circuit. The semiconductor channel is a composite thin film material of a lower layer of semiconducting carbon nanotubes and an upper layer of all-inorganic perovskite quantum dots. The all-inorganic perovskite quantum dots serve as a photosensitive layer and a photogenerated charge trapping layer, while the semiconducting carbon nanotube thin film serves as a charge transport layer.

[0017] The artificial vision system based on carbon nanotubes / all-inorganic perovskite quantum dots uses a phototransistor as the neuromorphic vision sensor based on semiconductor channel material. The neuromorphic vision sensor has the characteristics of a phototransistor, a memory, and an artificial neural synapse. Under illumination, the all-inorganic perovskite quantum dots generate photogenerated carriers and can capture and release photogenerated electrons, enabling the vision sensor to store information and process information in a manner similar to that of a biological neural synapse.

[0018] The artificial vision system based on carbon nanotubes / all-inorganic perovskite quantum dots uses CsPbX3 or a combination of PbS and CsPbX3 as the inorganic perovskite quantum dots, where X = Cl, Br or I, and the quantum yield is 80-90%; the carbon nanotubes are semiconducting carbon nanotubes.

[0019] The method for fabricating the artificial vision system based on carbon nanotubes / all-inorganic perovskite quantum dots includes the following steps:

[0020] 1) Fabricate the gate and leads on the substrate;

[0021] 2) A gate insulating layer was prepared using atomic layer deposition.

[0022] 3) Etch lead connection windows on the insulating layer;

[0023] 4) Fabricate the source and drain electrodes and leads on the insulating layer;

[0024] 5) Deposition and channel patterning of semiconducting carbon nanotube thin films;

[0025] 6) Spin-coating all-inorganic perovskite quantum dots to obtain a uniform all-inorganic perovskite quantum dot film.

[0026] In the fabrication method of the artificial vision system based on carbon nanotubes / all-inorganic perovskite quantum dots, step 1) involves fabricating a gate lead and a source lead. First, a titanium layer with a thickness ranging from 4 to 6 nm is deposited on the substrate, and then a gold layer with a thickness ranging from 50 to 60 nm is deposited.

[0027] In the fabrication method of the artificial vision system based on carbon nanotubes / all-inorganic perovskite quantum dots, step 2) involves using atomic layer deposition to prepare an alumina insulating layer with a thickness greater than 60 nm.

[0028] In the fabrication method of the artificial vision system based on carbon nanotubes / all-inorganic perovskite quantum dots, step 3) involves etching the alumina insulating layer using an 85wt% phosphoric acid solution, controlling the etching time to be 4–8 min and the etching temperature to be 60–70 °C.

[0029] In the method for fabricating an artificial vision system based on carbon nanotubes / all-inorganic perovskite quantum dots, step 5) involves preparing a semiconducting carbon nanotube film by mixing and diluting semiconducting carbon nanotubes with toluene at a volume ratio of 1:5 to 10, and then sonicating the mixture for 20 to 40 minutes to form a semiconducting carbon nanotube solution. Simultaneously, the product from step 4) is spin-coated with hexamethyldisilazane and immersed in the prepared semiconducting carbon nanotube solution to obtain a large-area uniform semiconducting carbon nanotube film on the product.

[0030] In the method for fabricating the artificial vision system based on carbon nanotubes / all-inorganic perovskite quantum dots, step 6) involves spin-coating all-inorganic perovskite quantum dots dispersed in n-hexane using a spin coater, controlling the spin-coating speed to be 2000–3000 rpm and the time to be 50–60 s; wherein the concentration of the n-hexane solution containing the all-inorganic perovskite quantum dots is 5–15 mg / mL.

[0031] The design concept of this invention:

[0032] This invention proposes an artificial vision system and its fabrication method based on carbon nanotubes / all-inorganic perovskite quantum dots. All-inorganic perovskite quantum dots serve as the photosensitive layer and photogenerated charge trapping layer, while a high-purity semiconducting carbon nanotube film serves as the charge transport layer. The excellent light absorption efficiency and high stability of all-inorganic perovskite quantum dots, along with the superior carrier mobility and high current on / off ratio of carbon nanotubes, improve the sensor's detection signal-to-noise ratio. The composite of these two materials is used as the channel material to address problems such as poor imaging, lack of flexibility, low integration, and poor stability under extremely low-light conditions. Based on the grating effect, the sensor possesses ultra-high photoelectric performance (responsivity, light-to-dark current ratio, external quantum efficiency, and detectivity) and excellent storage performance. Furthermore, it can simulate biological synaptic behavior (long / short-term plasticity, double-pulse facilitation, etc.) for data processing, realizing the image reinforcement learning process that simulates biological organisms, thus opening up new applications for carbon-based electronic devices in the field of artificial intelligence.

[0033] The advantages and beneficial effects of this invention are:

[0034] 1. The composite thin film of carbon nanotubes and all-inorganic perovskite quantum dots involved in this invention can overcome the shortcomings of single materials in the optoelectronic field, thereby achieving high performance. In addition, both materials have good flexibility and long-term stability and can be formed into films on a large area on a substrate, which is conducive to the large-scale fabrication of neuromorphic vision sensors.

[0035] 2. The neuromorphic visual sensor constructed in this invention simultaneously possesses the characteristics of a photoelectric sensor, a memory, and an artificial neural synapse. Under 405nm laser irradiation, its responsivity reaches 5.1 × 10⁻⁶. 7 A / W, light-to-dark current ratio exceeds 10 6The detectivity reaches 2×10 16 Jones, with an information storage time exceeding 10,000 seconds, can successfully simulate synaptic behaviors such as short-term plasticity, long-term plasticity, and bipulse facilitation in organisms.

[0036] 3. This invention enables the large-scale fabrication of highly integrated neuromorphic visual sensor arrays, which are trained using weak light pulse signals to simulate the image reinforcement learning process of biological organisms, thus opening up new applications of physical devices in the field of artificial intelligence.

[0037] 4. All-inorganic perovskite quantum dots possess excellent photoelectric response performance and tunable spectral response range, while carbon nanotubes exhibit superior carrier mobility and current on / off ratio. Combining these two materials fully leverages their photon trapping, photogenerated carrier generation, and transport capabilities, significantly improving the photoelectric response and image acquisition capabilities of artificial vision systems. This invention employs all-inorganic perovskite quantum dots as the photosensitive layer and photogenerated charge trapping layer, and high-purity semiconducting carbon nanotube films as the charge transport layer. Based on this novel material system, the phototransistor fully utilizes the advantages of the composite of semiconducting carbon nanotubes and inorganic perovskite quantum dots to achieve ultra-high responsivity, ultra-high detectivity, and ultra-high signal-to-noise ratio photoelectric detection capabilities. Furthermore, it can simulate biological synaptic behavior, realizing array-level image detection, image memory, and image learning functions of artificial vision systems, providing a prototype device reference for the development of future human-eye-inspired intelligent systems. Attached Figure Description

[0038] Figure 1 A flowchart illustrating the manufacturing process of a neuromorphic visual sensor based on carbon nanotubes / all-inorganic perovskite quantum dots.

[0039] Figure 2 The images show the transmission electron microscope (TEM) image, size statistical distribution, and X-ray diffraction pattern of all-inorganic perovskite quantum dots (CsPbBr3-QD). (a) TEM image of CsPbBr3-QD; (b) size statistical distribution of CsPbBr3-QD, with Size (nm) on the x-axis and Counts on the y-axis; (c) X-ray diffraction pattern of CsPbBr3-QD, with 2Theta (degree) on the x-axis and Intensity (au) on the y-axis.

[0040] Figure 3Absorption spectra of CNT, CsPbBr3-QD, and CNT / CsPbBr3-QD composite films on quartz substrates are characterized. (a) Absorption spectra of CNT, CsPbBr3-QD, and CNT / CsPbBr3-QD composite films on quartz substrates, with the horizontal axis representing wavelength (nm) and the vertical axis representing absorbance (au); (b) Magnified absorption spectrum of the CNT film.

[0041] Figure 4 The optoelectronic properties of CNT and CsPbBr3-QD thin films as channel materials alone are shown. (a) CNT transistors under dark and light conditions (173 μW / cm²). 2 Transfer characteristic curve, horizontal axis V GS Represents gate voltage (V), with the vertical axis I. DS (a) Source-drain current (A); (b) CsPbBr3-QD transistor under dark and light conditions (173 μW / cm²). 2 Transfer characteristic curve, horizontal axis V GS Represents gate voltage (V), with the vertical axis I. DS Represents the source-drain current (A).

[0042] Figure 5 Design and characterization of a CNT / CsPbBr3-QD neuromorphic visual sensor. (a) Schematic diagram of the visual sensor structure, where Source represents the source electrode, CsPbBr3 represents inorganic perovskite quantum dots, Light represents laser irradiation, Drain represents the drain electrode, CNT represents carbon nanotubes, Gate represents the gate electrode, and Substrate represents the substrate; (b) Scanning electron image (SEM) of the CNT thin film, with the inset showing a light microscope image of the visual sensor, and Channel representing the channel; (c) Atomic force image (AFM) of the CsPbBr3-QD thin film; (d) V DS When I = 1V, under incident light irradiation with different power densities DS -V GS Curve, x-axis V GS Represents gate voltage (V), with the vertical axis I. DS (a) Represents the source-drain current (A), and the laser wavelength used is 405nm; (e) Schematic diagram of the energy band and working principle of the vision sensor.

[0043] Figure 6 The photoelectric performance of the CNT / CsPbBr3-QD neuromorphic visual sensor was characterized using a wavelength of 405 nm. Among the data, (a) shows the responsivity and external quantum efficiency (EQE) as a function of incident light power density, with a bias condition of V... DS =1V, VGS =5V, the horizontal axis represents power density (W / cm²) 2 (a) The vertical axis Responsivity represents the responsivity (A / W); (b) The detectivity curve as a function of incident light power density, with the bias condition being V. DS =1V, V GS =5V, the horizontal axis represents power density (W / cm²) 2 (c) Photoresponse cycling performance, under illumination conditions I DS Apply a pulsed gate voltage I under rising, dark conditions DS The descent occurs under the bias condition V. DS =1V, V GS =5V, illumination conditions are 405nm laser, power density is 0.78W / cm² 2 The horizontal axis, Time, represents time (s), and the vertical axis, I... DS The source-drain current (A) is represented by "Optical", which represents the optical pulse signal, specifically, the incident light power density is 0.78 W / cm². 2 The erase gate voltage pulse is 0 V with a pulse width of 100 ms. "Electrical" represents the gate voltage pulse signal, specifically, a gate voltage pulse of 0 V with a pulse width of 100 ms; (d) Response time: rise time is 3.3 ms, fall time is 1.1 ms, and the bias condition is V. DS =1V, V GS =5V, illumination conditions are 405nm laser, power density is 0.78W / cm² 2 The horizontal axis, Time, represents time (ms), and the vertical axis, Current, represents relative intensity (au).

[0044] Figure 7 Characterization of the flexible properties of neuromorphic visual sensing. Among them, (a) dark conditions and 1.43 μW / cm². 2 Under illumination, under different bending strain conditions, I DS -V GS Curve, bias condition V DS =1V, V GS =5V, illumination condition is 516nm laser, V GS Represents grid voltage (V), ε represents bending strain (%), I DS (a) Source-to-leakage current (A); (b) Photocurrent stability test under different bending strains and illumination power densities, with bias condition V. DS =1V, V GS=5V, illumination condition is 516nm laser; the horizontal axis ε represents bending strain (%), and the vertical axis Photocurrent represents photocurrent (μA).

[0045] Figure 8 To characterize the long-term stability of the neuromorphic vision sensor, the bias condition is V. DS =1V, V GS =5V, illumination condition is 516nm laser; in the figure, the horizontal axis is V GS Represents gate voltage (V), with the vertical axis I. DS Represents the source-drain current (A).

[0046] Figure 9 To optimize the storage performance of the neuromorphic vision sensor, the wavelength used for signal writing is 516 nm, and the power density is 0.78 W / cm². 2 The electrical erase information 0V pulse duration is 100ms. Among them, (a) the transfer characteristic curves of optical signal writing information and electrical signal erasure information, optical signal irradiation for 20s to write information, the horizontal axis V GS (a) Represents gate voltage (V), and the vertical axis P represents source-drain current (A); (b) Information storage time exceeds 10000s, and information is written under the condition of 10s of optical signal illumination, where P represents power density (W / cm²). 2 V GS (c) Information write / erase cycle characteristics: information is written after 500ms of light signal illumination. The horizontal axis Time represents time (s), and the vertical axis I represents the gate voltage (V). DS Represents the source-drain current (A).

[0047] Figure 10 To simulate the behavior of light-tunable neural synapses in a neuromorphic vision sensor, a wavelength of 516 nm was used, with a bias condition of V. DS =1V, V GS =5V. Where (a)-(d) are at a light power density of 6.8μW / cm². 2 Under these conditions, the device simulated the transition of synapses from short-term to long-term plasticity as the pulse irradiation time increased. The irradiation times were 10 ms, 100 ms, 4000 ms, and 10000 ms, respectively. The horizontal axis represents time (s), and the vertical axis represents time (seconds). DS (e) The device successfully simulated double-pulse facilitation, with the double-pulse facilitation coefficient gradually decreasing to 100% as the pulse interval time increased, and the illumination power density being 48 W / cm². 2The horizontal axis represents the pulse interval (s), and the vertical axis represents the double-pulse facilitation factor (%); (f) Long-term enhancement effect, with an optical pulse width of 20ms and an optical pulse interval of 500ms, the horizontal axis represents the number of pulses, and the vertical axis represents the double-pulse facilitation factor (%); DS Represents the source-drain current (μA).

[0048] Figure 11 This is a neuromorphic visual sensor array. (a) The visual sensor array is mounted on an external test printed circuit board (scale bar, 40 mm); (b) The visual sensor leads are bonded to the external test printed circuit board (scale bar, 5 mm); (c) A photograph of the flexible visual sensor array (5 mm); (d) A light microscope photograph of the 32×32 visual sensor array (scale bar, 500 μm); (e) A magnified light microscope photograph of a portion of the 32×32 visual sensor array (scale bar, 50 μm); (f) A light microscope photograph of a single device in the 32×32 visual sensor array (scale bar, 20 μm); (g) Under low light power density of 1 W / cm² 2 Under illumination conditions, the process of training a neuromorphic visual sensor to simulate the image reinforcement learning of the digit "8" in a biological organism was demonstrated using initial pulses and different numbers of light pulses. As the number of training light pulses gradually increased, the image of the digit "8" became increasingly clear. The laser wavelength used was 405nm, the pulse width was 250ms, the pulse interval was 250ms, and the bias condition was V. DS =1V, V GS =5V. Detailed Implementation

[0049] In its specific implementation, this invention relates to an artificial vision system and its fabrication method based on carbon nanotubes (CNTs) / all-inorganic perovskite quantum dots (CsPbBr3-QDs). CsPbBr3-QDs are used as the photosensitive layer and the photogenerated charge trapping layer, while a high-purity semiconducting carbon nanotube film serves as the charge transport layer. This allows the sensor to simultaneously possess sensing performance, storage performance, and data processing capabilities that simulate biological synaptic behavior. Furthermore, a highly integrated sensor array is trained using weak light pulse signals, achieving image reinforcement learning that simulates biological processes, thus opening up new applications for physical devices in the field of artificial intelligence. The integration density is 1024 pixels.

[0050] The feasibility of the present invention will be further demonstrated below through examples.

[0051] Example

[0052] In this embodiment, a method for fabricating an artificial vision system based on carbon nanotubes / all-inorganic perovskite quantum dots is as follows:

[0053] The present invention describes a method for fabricating a carbon nanotube / all-inorganic perovskite quantum dot artificial vision system using techniques such as ultraviolet exposure, electron beam evaporation (EVE), plasma etching, wet etching, atomic layer deposition (ALD), and spin coating.

[0054] (1) For the fabrication of visual sensors on flexible substrates, the flexible substrate polyethylene naphthalate (PEN) will deform during the heating process. By optimizing the pretreatment of the flexible substrate and the low-temperature heating process of the photoresist, including temperature and time, the successful fabrication of flexible devices can be ensured. Finally, when the flexible substrate is pretreated at 190℃ for 3 hours, and then immersed in Remover PG (a photoresist produced by Microchem Corporation) and isopropanol (IPA) for 20 minutes respectively to remove the particulate matter generated during the pretreatment process; high-precision patterns can be fabricated on the flexible substrate by heating LOR 3A at 150℃ for 20 minutes and S-1813 at 120℃ for 2 minutes.

[0055] (2) For the large-scale fabrication of highly integrated sensor arrays, it is necessary to optimize the size of the electrodes and leads, as well as the spacing between electrodes, between electrodes and leads, and between leads. Due to the small size of highly integrated sensor arrays, incomplete metal stripping after electron beam evaporation of metal electrodes and leads can lead to array fabrication failure. Therefore, this invention optimizes the shape and size of the electrodes and the distribution and connection method of the leads to ensure the accuracy of the sensor array.

[0056] (3) For the large-scale fabrication of highly integrated sensor arrays, gates and leads are fabricated on the substrate. The leads and electrodes are composites of titanium (Ti) layer and gold (Au) layer. First, a Ti layer with a thickness range of 4 to 6 nm is deposited on the substrate, and then an Au layer with a thickness range of 50 to 60 nm is deposited.

[0057] (4) For the large-scale fabrication of highly integrated sensor arrays, the absence of leakage current devices is required. This necessitates optimizing the deposition conditions of the atomic layer deposition of the aluminum oxide (Al2O3) insulating layer and reducing the overlap area between the upper and lower metal layers of the insulating layer. Different temperatures (250℃, 200℃, 150℃) were tested to deposit aluminum oxide films of varying thicknesses (30nm, 40nm, 50nm, 60nm, 70nm, 80nm, 90nm, 100nm). Narrow gate electrodes were designed to avoid overlap with the source and drain electrodes. Furthermore, the lead size was reduced to decrease the overlap area between the upper and lower leads of the insulating layer. Ultimately, a 150℃ deposition of an 80nm Al2O3 insulating layer ensured that the array did not leak current under gate voltage conditions of -5V to 5V.

[0058] (5) The lead connection window was etched on the insulating layer. For opening the Al2O3 thin film, the wet etching conditions were optimized using an 85wt% phosphoric acid solution. Different temperatures (50℃, 60℃, 70℃, 80℃) and etching times (4min, 5min, 6min, 7min, 8min) were tried. Finally, etching at 70℃ for 6min ensured that the Al2O3 insulating layer in the metal connection window area was fully removed. Then, the source and drain electrodes and leads were fabricated on the insulating layer.

[0059] (6) The deposition of carbon nanotube (CNT) films was optimized by trying different CNT to toluene volume ratio solutions; different temperatures (20℃, 40℃, 60℃, 80℃, etc.); different heating methods (hot plate and water bath heating); different deposition times (30min, 1h, 1.5h, 2h, 3h, 4h, etc.); and whether to spin-coat the modifier hexamethyldisilazane (HMDS). Under the condition of CNT to toluene volume ratio of 1:10, a semiconducting carbon nanotube solution was formed by sonication for 30min; the product obtained in step (5) was spin-coated with HMDS and then immersed in the prepared semiconducting carbon nanotube solution and heated in a water bath at 60℃ for 2h to obtain a large-area uniform semiconducting carbon nanotube film with optimal performance, thus realizing the deposition and channel patterning of the semiconducting carbon nanotube film.

[0060] (7) All-inorganic perovskite quantum dots dispersed in n-hexane were spin-coated using a spin coater. The concentration of the n-hexane solution containing all-inorganic perovskite quantum dots was 10 mg / mL. For spin-coating CsPbBr3-QD films, different spin speeds of 1000 r / min, 1500 r / min, 3000 r / min, 4000 r / min, and 2000 r / min were tried. The optimal spin-coating speed was finally found to be 3000 r / min, and the spin-coating time was 60 s, resulting in a uniform all-inorganic perovskite quantum dot film.

[0061] like Figure 1 As shown, the neuromorphic visual sensor and array are constructed through the following six steps: (a) Selecting a suitable substrate, including rigid and pretreated PEN flexible substrates → (b) Depositing gate electrodes and metal leads using ultraviolet lithography and electron beam evaporation (EVE) → (c) Depositing an 80 nm thick Al2O3 film using ALD and etching the Al2O3 film with phosphoric acid solution to create windows → (d) Depositing source and drain electrodes and metal leads using ultraviolet lithography and EVE → (e) Depositing and patterning CNT films → (f) Spin-coating CsPbBr3-QD films.

[0062] like Figure 2As shown, CsPbBr3-QD was characterized as follows: (a) Transmission electron microscopy (TEM) image of CsPbBr3-QD shows that the quantum dot size is below 10 nm; (b) Statistical distribution of CsPbBr3-QD size shows that the average quantum dot size is 8.5 nm; (c) X-ray diffraction pattern of CsPbBr3-QD. Figure 2 It can be seen that the sharp X-ray diffraction peaks indicate that the prepared quantum dot material has good crystallinity.

[0063] like Figure 3 As shown, the absorption spectra of CNT, CsPbBr3-QD, and CNT / CsPbBr3-QD composite films were characterized. (a) The absorption spectra of the three films show that CsPbBr3-QD is a light-absorbing material with an absorption edge of 525 nm; (b) The absorption spectrum of the CNT film in Figure (a) is magnified, which further shows that CsPbBr3-QD is a light-absorbing material in the CNT / CsPbBr3-QD composite film.

[0064] like Figure 4 As shown, the photoelectric properties of CNT and CsPbBr3-QD films were characterized individually as channel materials at an incident wavelength of 405 nm. (a) CNT transistor under dark and light conditions (173 μW / cm²) 2 (a) The transfer characteristic curve shows that CNTs have virtually no photoelectric properties under these conditions; (b) CsPbBr3-QD transistors under dark and light conditions (173 μW / cm²) 2 The transfer characteristic curves show that the CsPbBr3-QD film is not conductive in the horizontal direction.

[0065] like Figure 5 As shown, the CNT / CsPbBr3-QD neuromorphic visual sensor was characterized by material properties and photoelectric performance testing. (a) Schematic diagram of the visual sensor structure, the device is a buried gate transistor structure, using CNT / CsPbBr3-QD thin film as the channel material; (b) Scanning electron microscopy (SEM) image of the CNT thin film shows that the CNT thin film is very dense, and the inset is a light microscope image of the visual sensor; (c) Atomic force imaging (AFM) image of the CsPbBr3-QD thin film shows that the CsPbBr3-QD size is very small and the film is very uniform; (d) When V DS At 1V, as the incident light power density increases from 0 to 1.7μW / cm², 2 The device's I DS -V GS The curves show a clear photoelectric response, with the maximum ratio of illumination current to dark current approaching 10. 7(e) Schematic diagram of the energy band and working principle of the vision sensor. The working principle is mainly as follows: The upper figure shows that under dark conditions, due to the mismatch of the energy band structures of CNT and CsPbBr3-QD, energy band bending and built-in power generation are formed at the interface. The lower figure shows that under illumination conditions, the photogenerated electrons are captured by CsPbBr3-QD to produce a grating effect, thereby obtaining excellent photoelectric performance.

[0066] like Figure 6 As shown, the optoelectronic performance of the device was characterized with an incident wavelength of 405 nm. (a) When V DS =1V, V GS At 5V, the responsivity and external quantum efficiency (EQE) decrease with increasing incident light power density, and the maximum responsivity and EQE of the device are 5.1 × 10⁻⁶. 7 A / W and 1.6×10 16 %; (b) The detectivity curve as a function of incident light power density is consistent with the trend in Figure (a), and the maximum detectivity of the device is 2 × 10⁻⁶. 16 Jones.(c) under bias condition V DS =1V, V GS =5V, incident light power density is 0.78W / cm² 2 When the erase gate voltage pulse is 0V and the pulse width is 100ms, under illumination conditions, I DS After applying a pulsed gate voltage under rising, dark conditions, I DS The decrease indicates that the device has good cycling performance in terms of photoresponsivity; (d) under bias condition V DS =1V, V GS =5V, incident light power density is 0.78W / cm² 2 When the erase gate voltage pulse is 0V and the pulse width is 100ms, it indicates a fast response time with a rise time of 3.3ms and a fall time of 1.1ms.

[0067] like Figure 7 As shown, the flexibility of the device is characterized with an incident light wavelength of 516 nm and a bias condition of V. DS =1V, V GS =5V. (a) At incident light power densities of 0 and 1.43 μW / cm² 2 At that time, I under different bending strain conditions DS -V GS The curves are almost identical, indicating that the device has excellent flexibility. (b) Photocurrent stability tests under different bending strains and illumination power densities show that the device is very stable under bending conditions.

[0068] likeFigure 8 As shown, under the bias condition V DS =1V, V GS =5V, incident light wavelength of 516nm, long-term stability characterization of visual sensor, I before and after being placed in air for 8 months. DS -V GS The fact that the curves are almost identical indicates that the device has long-term stability.

[0069] like Figure 9 As shown, the storage performance of the device was characterized with an incident wavelength of 516 nm and a power density of 0.78 W / cm². 2 The electrical erase information 0V pulse duration is 100ms. (a) Optical signal writing information and electrical signal erasure information I DS -V GS (a) The curve shows that the device can optically write and electrically erase information; (b) The information storage time variation curve shows that the information storage time exceeds 1000s; (c) The information write / erase cycle curve shows that the device can repeatedly write and erase information.

[0070] like Figure 10 As shown, the device was tested to simulate the optically tunable neural synaptic behavior using a wavelength of 516 nm and a bias condition of V. DS =1V, V GS =5V. (a)-(d) at a power density of 6.8μW / cm² 2 At that time, as the illumination pulse time increased, the device simulated the transition from short-term plasticity to long-term plasticity; (e) by applying two consecutive power densities of 48 μW / cm². 2 The light pulses simulated the two-pulse facilitation characteristics of organisms. The two-pulse facilitation coefficient is defined as A2 divided by A1, where A1 and A2 are the current amplitudes when the first and second light pulses are applied, respectively. When the device is stimulated by the first light pulse, the current in the CNT channel increases, and the photogenerated electrons trapped in the CsPbBr3-QDs require a longer decay time. When the device is stimulated by the second light pulse, the CsPbBr3-QDs trap more electrons, resulting in a higher current value in the CNT channel. In addition, the PPF coefficient gradually decreases as the pulse interval increases, eventually approaching 100%. (f) Under different power density conditions, 500 light pulses were continuously applied to the device to simulate the long-term enhancement effect of organisms, indicating that the light signal can be gradually learned and memorized by the device. The current increases with the number of light pulse stimuli, and the number of light pulses required to obtain the same target synaptic weight decreases with the increase of incident power density, indicating that the learning speed gradually accelerates with the increase of incident power density.

[0071] likeFigure 11 As shown, the neuromorphic visual sensor array was characterized and tested. (a) The visual sensor array was mounted on an external test printed circuit board for testing; (b) The visual sensor was connected to the external test printed circuit board via wire bonding; (c) Photograph of the flexible visual sensor array; (d)-(f) Enlarged views of the overall and partial aspects of the visual sensor array; (g) Under low light power density of 1 W / cm² 2 Under illumination conditions, the process of training a neuromorphic visual sensor to simulate the image reinforcement learning of the digit "8" in a biological organism using original state and different numbers of light pulses is demonstrated. As the number of training light pulses gradually increases, the image of the digit "8" becomes clearer. The laser wavelength used is 405nm, the pulse width is 250ms, the pulse interval is 250ms, and the bias condition is V. DS =1V, V GS =5V.

[0072] The results of the embodiments demonstrate that this invention proposes an artificial vision system and its fabrication method based on carbon nanotubes / all-inorganic perovskite quantum dots, wherein CsPbBr3-QD serves as the photosensitive layer and photogenerated charge trapping layer, and a high-purity semiconducting carbon nanotube film serves as the charge transport layer. Due to the photogating effect, the sensor possesses ultra-high photoelectric performance (responsivity, brightness-to-current ratio, external quantum efficiency, and detectivity) and high storage performance, and can simulate biological synaptic behavior (long / short-term plasticity, double-pulse facilitation, etc.) for data processing. Simultaneously, using weak light pulse signals to train a highly integrated sensor array realizes the process of image reinforcement learning that simulates biological processes, opening up new applications of physical devices in the field of artificial intelligence.

Claims

1. An artificial vision system based on carbon nanotubes / all-inorganic perovskite quantum dots, characterized in that, The artificial vision system is a highly integrated neuromorphic vision sensor array, with the following specific structure: the substrate is a rigid substrate or a flexible substrate, and the top of the substrate is a gate and gate leads and source leads, which are used to connect the gate and source to the external test circuit, respectively. The gate and gate leads are covered with a gate insulating layer on the top and both sides. The source, semiconductor channel, drain, and drain leads are arranged sequentially on the top of the gate insulating layer. The source and drain are connected through the semiconductor channel. The drain leads are used to connect the drain to an external test circuit. The semiconductor channel is a composite thin film material of a lower layer of semiconducting carbon nanotubes and an upper layer of all-inorganic perovskite quantum dots. The all-inorganic perovskite quantum dots serve as a photosensitive layer and a photogenerated charge trapping layer, while the semiconducting carbon nanotube film serves as a charge transport layer. The inorganic perovskite quantum dots are CsPbX3, or a combination of PbS and CsPbX3, where X = Cl, Br, or I, and the quantum yield is 80-90%. Neuromorphic vision sensors based on semiconductor channel materials are phototransistors. They simultaneously possess the characteristics of photoelectric sensors, memory, and artificial neural synapses. Under illumination, all-inorganic perovskite quantum dots generate photogenerated carriers and can capture and release photogenerated electrons, enabling the vision sensor to store information and process information in a manner similar to that of biological neural synapses.

2. A method for fabricating an artificial vision system based on carbon nanotubes / all-inorganic perovskite quantum dots as described in claim 1, characterized in that, Includes the following steps: 1) Fabricate the gate and leads on the substrate; 2) A gate insulating layer is prepared using atomic layer deposition. 3) Etch the connection window between the source and drain leads on the gate insulating layer; 4) Fabricate the source and drain electrodes and drain leads on the gate insulating layer; 5) Deposition and channel patterning of semiconducting carbon nanotube thin films; 6) Spin-coating all-inorganic perovskite quantum dots to obtain a uniform all-inorganic perovskite quantum dot film.

3. The method for fabricating an artificial vision system based on carbon nanotubes / all-inorganic perovskite quantum dots according to claim 2, characterized in that, In step 1), the fabricated leads include gate leads and source leads. First, a titanium layer with a thickness ranging from 4 to 6 nm is deposited on the substrate, and then a gold layer with a thickness ranging from 50 to 60 nm is deposited.

4. The method for fabricating an artificial vision system based on carbon nanotubes / all-inorganic perovskite quantum dots according to claim 2, characterized in that, In step 2), an alumina gate insulating layer is fabricated using atomic layer deposition, and the thickness of the gate insulating layer is greater than 60 nm.

5. The method for fabricating an artificial vision system based on carbon nanotubes / all-inorganic perovskite quantum dots according to claim 2, characterized in that, In step 3), an 85wt% phosphoric acid solution is used to etch the aluminum oxide gate insulating layer, and the etching time is controlled within the range of 4~8 min, and the etching temperature is 60~70℃.

6. The method for fabricating an artificial vision system based on carbon nanotubes / all-inorganic perovskite quantum dots according to claim 2, characterized in that, In step 5), the method for preparing the semiconducting carbon nanotube film is as follows: the semiconducting carbon nanotubes and toluene are mixed and diluted at a volume ratio of 1:5~10, and ultrasonicated for 20~40 min to form a semiconducting carbon nanotube solution; at the same time, the product after step 4) is spin-coated with hexamethyldisilazane and immersed in the prepared semiconducting carbon nanotube solution to obtain a large-area uniform semiconducting carbon nanotube film on the product.

7. The method for fabricating an artificial vision system based on carbon nanotubes / all-inorganic perovskite quantum dots according to claim 2, characterized in that, In step 6), an inorganic perovskite quantum dots dispersed in n-hexane are spin-coated using a spin coater, with the spin-coating speed controlled at 2000~3000 rpm and the time at 50~60 s; wherein the concentration of the hexane solution of the inorganic perovskite quantum dots is 5~15 mg / mL.

Images