Self-rectification memristor based on hafnium zirconium oxide / tellurium oxide heterojunction and preparation and application thereof

Abstract

The invention discloses a self-rectification memristor based on a hafnium zirconium oxide / tellurium oxide heterojunction and preparation and application of the self-rectification memristor, tellurium dioxide is used as an evaporation source, a TiN / HZO substrate is placed in vacuum evaporation equipment for evaporation of a tellurium dioxide film, finally, a metal electrode is evaporated on the tellurium dioxide film, and air annealing is performed to obtain the self-rectification memristor. According to the present invention, the material has characteristics of good switching characteristic, good cycle stability, excellent self-rectification characteristic and broad spectrum response, and can be used in storage devices, coded locks, light response detectors, light synapses, multi-mode neuromorphic calculation units and other related aspects by using the characteristics.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of semiconductor device preparation and application, and particularly relates to a self-rectifying memristor based on hafnium-zirconium oxide / tellurium oxide heterojunction and preparation and application thereof. BACKGROUND

[0002] In the era of big data and artificial intelligence, the "memory wall" and "power wall" problems of the separation of computing and storage under the von Neumann architecture are increasingly prominent, and have become a bottleneck restricting the further development of computing efficiency. Memristor is widely regarded as the core electronic device for building the next generation of high-density non-volatile storage and neuromorphic computing systems due to its natural simple structure, high speed, low power consumption and potential for integration of storage and computing. However, when we place the traditional memristor in a high-density cross array, a fatal problem emerges - "sneak current". In the traditional cross array, when a specific device (such as the device located at the intersection of the row line and the column line) is read or written, the current does not only flow through the selected target device, but also forms a parallel parasitic path through other unselected cells. This is like a crisscross city road network, when you only want a car to go to a specific destination, but many "shortcut" vehicles block other roads, causing the signal of the target path to be unclear, ultimately causing read operation errors, write energy consumption surge, and array size limitation, and a series of serious challenges.

[0003] It is in this background that the heterojunction self-rectifying memristor emerges as the times require, which represents an innovative approach to "root cause" from the physical level of the device. Compared with ordinary memristor, its core advantage lies in the integration of the memristor function layer and the heterojunction with rectifying characteristics, realizing "one device, two functions". Using the inherent unidirectional conduction characteristics of the heterojunction (such as p-n junction, Schottky junction, etc.), it naturally has self-rectifying ability. When a reverse bias is applied, the device is in a high-resistance off state, and the current is effectively cut off. This means that in array operation, the sneak current flowing through non-target cells is greatly suppressed, so that transistor-free integration of large scale and high density can be achieved, which is its most fundamental advantage. At the same time, its self-rectifying characteristic ensures that each "synapse" cell can be independently and accurately regulated. At the same time, it usually has good multi-resistance state characteristics, which can more reliably realize continuous adjustment of synaptic weights, directly simulate learning rules such as long-term potentiation (LTP) and long-term depression (LTD), and pave the way for building large-scale, high-energy-efficiency brain-like computing chips. SUMMARY

[0004] In order to overcome the shortcomings of the prior art, the purpose of the present application is to provide a self-rectifying memristor based on hafnium-zirconium oxide / tellurium oxide (HZO / TeO2) heterojunction and preparation and application thereof, In order to achieve the above object, the technical scheme adopted by the present application is: In a first aspect, the present application provides a self-rectifying memristor based on hafnium-zirconium oxide / tellurium oxide heterojunction, wherein the heterojunction of the self-rectifying memristor is formed by hafnium-zirconium oxide and tellurium oxide.

[0005] Further, the structure of the self-rectifying memristor is TiN / HZO / TeO2 / A, wherein A is a metal electrode, which can be a Ni or Au electrode, and preferably a Ni electrode.

[0006] In a second aspect, the present application provides a preparation method of the self-rectifying memristor based on hafnium-zirconium oxide / tellurium oxide heterojunction as described in the first aspect, wherein a tellurium dioxide is used as an evaporation source, a TiN / HZO substrate is placed in a vacuum evaporation device to evaporate a tellurium dioxide film, and finally a metal electrode is evaporated on the tellurium dioxide film.

[0007] In the present application, the Te-O source is evaporated on the TiN / HZO substrate by heating in a vacuum coating device to prepare a HZO / Te-O heterojunction film. Through X-ray photoelectron spectroscopy (XPS) analysis of the sample, the atomic proportions of Te and O are 33.41% and 66.59%, respectively. Therefore, the ratio of Te to O in the Te-O film is close to 1:2. Thus, the present application selects TeO2 as the evaporation source.

[0008] Further, the HZO thickness of the TiN / HZO substrate is 2-10 nm, the thickness of the tellurium dioxide film is 2-10 nm, and the device performance is best when the thickness is 6 nm, and the film is deposited on the surface of the HZO. After the metal electrode is evaporated, the device is subjected to air annealing at a temperature of 100-500°C for 10-60 min.

[0009] In a third aspect, the present application provides an application of the self-rectifying memristor based on hafnium-zirconium oxide / tellurium oxide heterojunction as described in the first aspect to a memory, wherein information is written by different wavelength light pulses, and the post-pulse current is used to represent information. The self-rectifying memristor exhibits good switching characteristics and excellent self-rectifying characteristics, wherein the maximum switching ratio is 107, and the maximum rectification ratio is 490. The self-rectifying characteristics can reduce or eliminate the leakage current problem in traditional memories and reduce power consumption.

[0010] In a fourth aspect, the present application provides an application of the self-rectifying memristor based on hafnium-zirconium oxide / tellurium oxide heterojunction as described in the first aspect to a light response detector, wherein the self-rectifying memristor has a wide spectral response and can capture multiple waveband light signals simultaneously to realize high-precision and multifunctional detection and analysis. The typical light response waveband range is 254 nm-520 nm. When the incident light is incident on the tellurium oxide at an oblique angle, the current increases, i.e., the post-pulse current is higher than the pre-pulse current.

[0011] Furthermore, the input signal is incident on the memristor of the present invention at an oblique incidence angle of approximately 45°.

[0012] In one embodiment, three different wavelengths of light were selected for in-depth investigation. The results showed that the three types of light exhibited similar trends: the current of the device changed accordingly when the number, intensity, frequency, and pulse width of the input light source pulses changed. Because the device responds differently to different lights, the magnitude of the change in post-pulse current for different lights also varies, but the overall trend shows that the post-pulse current increases with the increase of light intensity.

[0013] Furthermore, the three wavelengths of light were set to 254 nm, 365 nm, and 405 nm because the device responds strongly to these three wavelengths.

[0014] A fifth aspect of the present invention provides the application of the self-rectified memristor based on the hafnium zirconium oxide / tellurium oxide heterojunction described in the first aspect for photosynapses, wherein applying a light pulse to the self-rectified memristor is defined as a learning process, and not applying a light pulse is defined as a forgetting process, a forgetting time is set, and the current after each pulse is monitored to simulate the process of memory deepening in the human brain during repeated learning.

[0015] Furthermore, the light pulse was set to a 365 nm light source with a duration of 50 ms.

[0016] The sixth aspect of the present invention provides another application of the self-rectified memristor based on the hafnium zirconium oxide / tellurium oxide heterojunction described in the first aspect for photosynapses, wherein a first optical pulse is applied to the self-rectified memristor as a conditioned stimulus, a second optical pulse with a different wavelength from the first optical pulse is applied as an unconditioned stimulus, a threshold current is set, and the current after the pulse is monitored to simulate the associative learning process of the human brain.

[0017] Furthermore, the first optical pulse has a wavelength of 470 nm, the second optical pulse has a wavelength of 365 nm, and the pulse width is 1 s.

[0018] A seventh aspect of the present invention provides an application of the self-rectified memristor based on the hafnium zirconium oxide / tellurium oxide heterojunction described in the first aspect for a combination lock. For an M-bit input password, there are N digit selections (N>M). Each digit is assigned one of the self-rectified memristors. For any digit of the password, it is designed to be the current after a light pulse corresponding to the rectified memristor at a certain moment. Unlocking is completed only when the monitoring current of all M self-rectified memristors after light pulse input meets the requirements.

[0019] Further, N=10, representing the numbers 0-9, is a 10-bit password gate, and the 10-bit password corresponds to 10 different devices. The current of the device at a certain time (for example, the 20th second) is used as the basis for judgment (for example, each device applies different conditions of light or no light), and 10 different inputs are obtained. Only when the 10-bit input is consistent with the unlocking signal of the password lock, the password lock is unlocked.

[0020] Further, the light pulse is set to a 365 nm light source with a pulse width of 1 s.

[0021] In an eighth aspect of the present application, the hafnium-zirconium oxide / tellurium oxide heterojunction-based self-rectifying memristor of the first aspect is applied to a multi-modal neuromorphic computing unit. The self-rectifying memristor is first subjected to a light pulse and then subjected to an electrical pulse, so that a single device can complete weight increase and decrease, corresponding to multiple conductance states. The multiple conductance states are combined with a neural network to perform target recognition.

[0022] In one embodiment, when the input is a negative voltage pulse, the post-pulse current increases with the increase of the number, frequency and pulse width of the electrical pulse. When the input is a positive voltage pulse, the post-pulse current decreases with the increase of the number, frequency and pulse width of the electrical pulse.

[0023] At the same time, when the input signal is a light source, the post-pulse current of the device increases with the increase of the light. When the input signal is a positive voltage pulse, the post-pulse current of the device decreases with the increase of the light.

[0024] Using the above properties, the device is first subjected to 100 light pulses, and then subjected to 100 positive voltage pulses after the light pulses end. The conductance of the device will first increase and then decrease. Further, the light pulse is set to a 365 nm light source with a pulse width of 200 ms and a pulse interval of 100 ms. The electrical pulse is set to a 0.2 V voltage with a pulse width of 200 ms and a pulse interval of 100 ms.

[0025] Compared with the prior art, the present application has the following advantages: 1. The present application realizes the use of twice thermal evaporation to manufacture a hafnium-zirconium oxide / tellurium oxide heterojunction-based self-rectifying memristor device. The device is simple to operate, the process is simple and easy to operate, and the cost is low.

[0026] 2. The heterojunction-based memristor not only ensures good memristive performance and excellent self-rectifying characteristics, but also has excellent optoelectronic synaptic performance, realizing the application of password decryption. BRIEF DESCRIPTION OF DRAWINGS

[0027] Figure 1 is a schematic diagram of the structure of the self-rectifying memristor of the present application and the application of light and electrical pulses.

[0028] Figure 2 is a learning-forgotten curve of the HZO / TeO2 heterojunction memristor of the embodiment of the application with the thickness of TeO2 being 6 nm I - V curve.

[0029] Figure 3 is a learning-forgotten curve of the HZO / TeO2 heterojunction memristor of the embodiment of the application with the thickness of TeO2 being 0 nm, the electrode being Ni and the HZO / TeO2 heterojunction memristor being annealed I - V curve.

[0030] Figure 4 is a learning-forgotten curve of the HZO / TeO2 heterojunction memristor of the embodiment of the application with the thickness of TeO2 being 2 nm, the electrode being Ni and the HZO / TeO2 heterojunction memristor being annealed I - V curve.

[0031] Figure 5 is a learning-forgotten curve of the HZO / TeO2 heterojunction memristor of the embodiment of the application with the thickness of TeO2 being 4 nm, the electrode being Ni and the HZO / TeO2 heterojunction memristor being annealed I - V curve.

[0032] Figure 6 is a learning-forgotten curve of the HZO / TeO2 heterojunction memristor of the embodiment of the application with the thickness of TeO2 being 6 nm, the electrode being Au and the HZO / TeO2 heterojunction memristor being annealed I - V curve.

[0033] Figure 7 is a learning-forgotten curve of the HZO / TeO2 heterojunction memristor of the embodiment of the application with the thickness of TeO2 being 6 nm, the electrode being Ni and the HZO / TeO2 heterojunction memristor being unannealed I - V curve.

[0034] Figure 8 is a change of ΔI of the embodiment of the application under different numbers of light pulses. PSC

[0035] Figure 9 is a change of ΔI of the embodiment of the application under different intensities of light pulses. PSC

[0036] Figure 10 is a change of ΔI of the embodiment of the application under different frequencies of light pulses. PSC

[0037] Figure 11 is a change of ΔI of the embodiment of the application under different widths of light pulses. PSC

[0038] Figure 12 is a simulation of a learning-forgotten curve of the embodiment of the application. ​​​​

[0039] Figure 13 is the mouse appears in the embodiment of the application.

[0040] Figure 14 is the hammer appears in the embodiment of the application.

[0041] Figure 15 is the mouse and hammer appears in the embodiment of the application.

[0042] Figure 16 is the change of the baby when the mouse appears after the mouse and hammer appears in the embodiment of the application.

[0043] Figure 17 is the password lock schematic diagram in the embodiment of the application.

[0044] Figure 18 is the password unlocking in the embodiment of the application.

[0045] Figure 19 is the current change under different electric pulse numbers in the embodiment of the application.

[0046] Figure 20 is the current change under different electric pulse widths in the embodiment of the application.

[0047] Figure 21 is the current change under different electric pulse frequencies in the embodiment of the application.

[0048] Figure 22 is the light enhancement and electric inhibition in the embodiment of the application. DETAILED DESCRIPTION

[0049] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings of the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments of the application, all other embodiments obtained by those skilled in the art without creative work fall within the protection scope of the application.

[0050] The application is a kind of HZO / TeO2 heterojunction self rectification memristor and its preparation method. Tellurium oxide (TeO2) is a kind of semiconductor material with unique photoelectric properties. Its most prominent feature is high carrier mobility and controllable conductance characteristics, which is derived from the multivalent state of tellurium element and the low formation energy of oxygen vacancy. After combining with HZO, the heterojunction memristor formed shows not only self rectification characteristics, but also wide spectrum response. Using these characteristics, the application can realize the application in the fields of memory, password lock, light response detector, optical synapse, multi-modal neuromorphic computing unit and the like.

[0051] The present application is based on a hafnium-zirconium oxide / tellurium oxide heterojunction self-rectifying memristor, which adopts a heterojunction formed by hafnium-zirconium oxide and tellurium oxide, and the structure thereof can be represented as TiN / HZO / TeO2 / A, as shown in Figure 1 Figure, wherein A is a metal electrode, typically a Ni electrode, TeO2 is a thin film structure, and TiN / HZO is a substrate, wherein the thickness of HZO is preferably 2-10 nm. Taking the Ni electrode as an example, the preparation process of the memristor is specifically described as follows: Step 1: washing the TiN / HZO substrate with deionized water, placing the washed and dried substrate on a substrate in a vacuum evaporation device, and placing the substrate in the vacuum evaporation device.

[0052] Step 2: placing a TeO2 evaporation source in the vacuum plating device, evaporating it onto the surface of the TiN / HZO substrate by electric heating. The evaporation rate can be selected to be 0.3-1 Å / s, typically 0.4 Å / s or 0.5 Å / s, and the film thickness is 2-10 nm, typically 6 nm, deposited on the surface of HZO.

[0053] Step 3: adding a Ni target to the vacuum plating device, and again using the method of thermal evaporation to deposit a Ni electrode on the TeO2 film. The evaporation rate is 0.3-2 Å / s, typically 1 Å / s, and the electrode thickness is 0-200 nm (not including 0), typically 100 nm, and the obtained Ni electrode is in the form of a cylinder.

[0054] Step 4: placing the device in a vacuum annealing table, and performing air annealing at 100-500℃ for 10-60 min, typically annealing parameters such as 100℃ for 20 min, and setting the temperature rising time to 10 min, and after completing the annealing, the TiN / HZO / TeO2 / Ni memristor of the present application is successfully prepared.

[0055] The HZO / TeO2 heterojunction self-rectifying memristor of the present application exhibits good switching characteristics and cycle stability, and also has excellent self-rectifying characteristics due to the HZO / TeO2 heterojunction.

[0056] To further illustrate the beneficial effects of the present application, a series of performance tests and applications of the heterojunction memristor of the present embodiment are performed, and the specific implementation is as follows: Example 1: thickness of tellurium oxide and data analysis.

[0057] This embodiment employs a heterojunction memristor with a TiN / HZO / TeO2 / Ni structure, where the HZO thickness is 10 nm, the TeO2 thickness is 6 nm, Ni is the top electrode, and TiN is the bottom electrode. To study its electrical characteristics, a bias voltage is applied between the top electrode (TE) and the bottom electrode (BE), with the voltage set to a flyback voltage ranging from -5 V to 5 V. When the voltage increases from 0 V to 5 V, the device current is relatively large; when the voltage flies back from 5 V to 0 V, the device current is relatively small. Figure 2 As shown, when the voltage scans from -5 V to 0 V, electrons begin to be injected from Ni into the TeOx layer and captured by deep-level traps in the amorphous TeOx. The trap levels filled with electrons provide a new conductive path, making it easier for electrons to pass through the TeOx layer. Therefore, as the voltage increases from 0 V, the current increases. However, with the continued action of the external electric field, electrons flow from the TeOx layer to Ni under the influence of the electric field force, and electrons in the defects are released, disrupting the conductive path. Therefore, when the voltage scans back from 5 V to 0 V, the current decreases. The on / off ratio is defined as I when the voltage is 2 V. LRS / I HRS The maximum switching ratio is 107.

[0058] Devices I-V The results show that, in addition to its basic resistive switching characteristics, this memristor also exhibits excellent rectification characteristics. Due to the different properties of HZO and TeO2, the charge redistributes after bonding, forming a built-in potential barrier at their interface. When a forward bias is applied, the barrier lowers and conduction occurs; when a reverse bias is applied, the barrier rises and turn-off occurs. This results in a larger current at positive voltage and a very small current at negative voltage. The rectification ratio is set at Ic for 3V and -3V voltages. LRS The ratio is 490, and its maximum rectification ratio is 490.

[0059] Further investigation was conducted on the device performance with different TeO2 thicknesses, specifically when the thickness was 0 nm, i.e., the intermediate layer was only HZO. I - V The curve reversed counterclockwise at 5V, creating a "window" as shown. Figure 3 As shown, this is because pure HZO contains oxygen vacancies (XPS results show an oxygen vacancy content of 14.80%). When a sufficiently large positive voltage is applied, the electric field drives these oxygen vacancies to move directionally and aggregate in the thin film, forming a conductive filament. This filament provides a low-resistance channel for electrons, enabling the device to switch from a high-resistance state to a low-resistance state, as represented by the "window" in the curve. When the thickness increases to 2 nm, the window of the curve becomes smaller, and the on / off ratio becomes lower, as shown in the figure. Figure 4As shown. Because the movement of oxygen vacancies after adding the TeO2 layer is affected not only by the external electric field but also by the built-in electric field between HZO and TeO2, but the influence of the built-in electric field is weaker at this point, and the inversion direction of the curve is consistent with the former. When the thickness increases to 4 nm, I - V The curve reverses clockwise, as... Figure 5 As shown, its mechanism is the same as that at 6 nm, but its uniformity and electrical properties are not as good as those of the device at 6 nm. With the increase of film thickness, the surface roughness decreases and the uniformity of device performance is improved.

[0060] Based on the above experiments, since the self-rectification characteristic can reduce or eliminate leakage current problems in traditional memory and reduce power consumption, the self-rectified memristor of this invention can be used to develop high-density, low-power memory, and can also simulate synaptic behavior in biological neurons.

[0061] Example 2, Electrode Comparison.

[0062] In this embodiment, the effects of different electrodes on device performance were investigated. Figure 2 , Figure 6 The devices under Ni and Au electrodes were shown respectively. I - V The curves show that when the electrode is changed from Ni to Au, the on / off ratio, rectification ratio, and uniformity of the memristor all decrease. This is because Ni has a higher work function, which forms a higher Schottky barrier when it comes into contact with TeO2, resulting in better rectification performance.

[0063] Example 3: Annealing comparison.

[0064] In this embodiment, the main focus is on the impact of annealing on device performance, such as... Figure 2 and Figure 7 As shown in the figure, the curves before and after annealing of the device can be seen intuitively. Annealing improves the uniformity of the device, thereby increasing the switching ratio and rectification ratio of the device. The annealing conditions are 100 °C in air for 20 min.

[0065] Four different wavelengths of optical synapses were achieved.

[0066] This invention, based on a self-rectified memristor of HZO / TeO2, not only possesses basic electrical characteristics but also exhibits a broad-spectrum (254 nm-520 nm) optical response. It can simultaneously capture multi-band optical signals, achieving high-precision, multi-functional detection and analysis, and can be used as a photoresponse detector. It is suggested that the incident light be obliquely incident on the tellurium oxide surface, such as... Figure 1The typical response wavelengths are 254 nm, 365 nm, 405 nm, 470 nm, and 520 nm. In this embodiment, three light sources with strong responses (254 nm, 365 nm, and 405 nm) are studied. When the light source is incident on the memristor at an angle of about 45°, the current of the device increases sharply. Therefore, by changing the parameters of the light source, the change in current after the pulse is observed. It is found that the current after the pulse increases with the increase in the number, intensity, frequency, and pulse width of the light pulse. In order to better show the change in current, the column chart of the current difference (ΔEPSC) before and after the pulse of different light sources with the change in the number, intensity, frequency, and pulse width of the light pulse is as shown in FIG. 6. Figure 8 、 Figure 9 、 Figure 10 and Figure 11 , which provides a basis for the application of subsequent devices.

[0067] Example 5, simulation of biological memory behavior.

[0068] In this embodiment, the self-rectifying memristor is used to simulate the biological learning and memory process. Specifically, the application of a light pulse to the self-rectifying memristor is defined as a learning process, and the non-application of a light pulse is defined as a forgetting process. The forgetting time is set, and the current after each pulse is monitored to simulate the process of deepening memory in repeated learning of the human brain.

[0069] In a specific embodiment, the application of a 365 nm light pulse with a pulse width of 50 ms to the self-rectifying memristor is regarded as a learning process, and the non-application of light is regarded as a forgetting process. First, a light pulse is applied to the self-rectifying memristor for 30 times, and no light pulse is applied within 10 s after the end of the pulse. It is observed that the current at this time is much higher than the current before the pulse. At the same time, in order to better simulate the biological learning behavior, the number of learning times is reduced in the subsequent study, and 30, 20, and 10 learning times are performed in turn. It can be seen that under the same forgetting time, even if the number of learning times is reduced each time, the current after the pulse will still increase, as shown in FIG. 5. Figure 12 This process is consistent with the learning-forgetting mechanism of the human brain. Even if the number of repeated learning times in the learning is less than that of the previous time, the memory will be deepened after each learning due to the increase in the cumulative number of times. Therefore, the repeated learning and memory behavior of the biological organism is more closely simulated.

[0070] Example 6, simulation of conditional stimulus.

[0071] In the human body, there are two special stimuli, unconditioned stimuli (such as food, pain, cold, etc.) directly act on the target organs through biological channels to trigger basic physiological responses. Conditioned stimuli (such as traffic lights, language) establish a connection with unconditioned reflex through learning, rely on the participation of the cerebral cortex, adapt to environmental changes, and the two work together to ensure basic survival and enhance environmental adaptability. In this embodiment, the self-rectifying memristor is still used as a phototransmitter to simulate John Watson's Little Albert experiment. That is, a first light pulse is applied to the self-rectifying memristor as a conditioned stimulus, a second light pulse with a wavelength different from the first light pulse is applied as an unconditioned stimulus, a threshold current is set, and the post-pulse current is monitored to simulate the human brain's associative learning process.

[0072] In a specific embodiment, a 254 nm light pulse is used as a conditioned stimulus (CS) to simulate the appearance of a white mouse, and a 365 nm light pulse is used as an unconditioned stimulus (US) to simulate a loud noise made by a hammer. At the same time, 0.3 μA is set as the threshold current, and the post-pulse current at 60 s is observed as the response indicator. When it exceeds the threshold value, it means that the stimulus triggers the instinctive response of the baby Albert - crying. As shown in Figure 13 , the US is applied first, and it can be seen that the current does not reach the threshold, indicating that Albert does not cry; as shown in Figure 14 , after the application of CS, the current exceeds the threshold, which is consistent with the results of conditioned and unconditioned stimuli. However, after applying US and CS for a period of time, CS is applied again, and the response indicator exceeds the threshold current at this moment, indicating that baby Albert at this moment is afraid of the appearance of the white mouse, that is, the instinctive response is triggered by the conditioned stimulus after training, as shown in Figure 15 and Figure 16 . This process successfully replicates John Watson's Little Albert experiment, reproduces the process of human brain associative learning, simulates the rules of human conditioned emission formation, and demonstrates the application prospects of the device.

[0073] Example 7, decryption of the password gate.

[0074] In this embodiment, a four-digit password lock is designed, as shown in Figure 17 , only when the number of input digits and the order are all consistent with the decryption conditions, the password lock will be unlocked. In order to achieve such results, first, 10 different self-rectifying memristors (for example, 10 different memristors made in the same batch under the same conditions) are selected to represent the digits "0" to "9". Then, 0.9 uA is selected as the threshold current, and the current of each device at 20 s is taken, and when the value exceeds 0.9 uA, it is defined as valid input, otherwise it is invalid input and is not paired with the password lock. The password is set to 1345, and then any 4 of the 10 devices are applied with a 1 s pulse width and a 365 nm wavelength light source, and the current is observed. As shown inFigure 18 As shown, when the input number is "1345", the password lock is unlocked, and when the input is other cases, such as "6789", the unlocking fails.

[0075] Example 8, light enhancement and electrical inhibition.

[0076] In this embodiment, in addition to studying the response of the device to the light signal, the electrical response of the device is also studied, and the results show that when a negative voltage is applied to the device, the observation of the electrical response shows that the post-pulse current increases with the increase of the number, intensity, frequency and pulse width of the electrical pulse, which is a typical electrical pulse characteristic. When a positive voltage is applied to the device, the post-pulse current of the device is also higher than the pre-pulse, but with the increase of the number, frequency and pulse width of the electrical pulse, the post-pulse current decreases, as shown in Figure 19 , Figure 20 and Figure 21 , which provides a basis for the combination of light pulse enhancement and electrical pulse inhibition.

[0077] When the input signal is a light source, the post-pulse current of the memristor increases with the increase of the light pulse, and when the input signal is a positive voltage, the post-pulse current decreases with the increase of the voltage. By using the above properties, the two are combined, the device is first subjected to a light pulse, and then the device is subjected to an electrical pulse, and finally the current curve of the device shows a trend of first increasing and then weakening, which will provide a good foundation for the subsequent application of neuromorphic computing.

[0078] In a specific embodiment, 100 365 nm light pulses with a pulse width of 100 ms are first applied to the self-rectifying memristor, and then 100 0.2 V pulse voltages are applied after the light pulse ends. The device current increases with the increase of the number of light pulses and decreases with the increase of the number of electrical pulses, as shown in Figure 22 That is, the light pulse produces excitatory post-synaptic current, and the electrical pulse produces inhibitory post-synaptic current, thereby achieving the use of a single device to complete the increase and decrease of weights, corresponding to multiple conductance states, and combining the multiple conductance states of the device with the neural network, that is, the application of image classification, recognition and the like can be realized.

[0079] For example, in the scenario of identifying a specific action (such as a gesture, gait) in a video, the light pulse can quickly and in parallel process the visual information of the video (excitation) and extract features. The electrical pulse can be responsible for "trimming" irrelevant information or suppressing background noise (inhibition). This balance between excitation and inhibition allows the network to more accurately identify dynamic behaviors that have specific patterns in time and space, rather than just static images. For another example, the electrical stimulation signal is used as the inhibition mechanism of the neural network. When the light excitation signal causes some neurons or pathways to be overactive, the electrical inhibition can intervene in time to balance the network state and ensure that it works within a stable and controllable range.

[0080] For another example, a light signal is directly received by a light sensing unit (such as a photodiode) and an excitatory response is generated (analogous to cone / rod cells in the retina). At the same time, a local electrical inhibition signal can be applied by a microcircuit integrated inside the sensor. This inhibition can enhance the contrast of image edges, suppress information in uniform areas, thus achieving feature extraction and noise reduction at the source of data generation, greatly reducing the amount of data that needs to be uploaded. Based on this, preliminary processing can be performed while using light perception information.

[0081] For another example, in autonomous driving, the visual system can quickly identify a feasible path and moving obstacles, and generate an excitatory signal for "moving forward" or "turning". At the same time, when a collision is detected by its tactile or close-range sensor (such as laser radar, ultrasonic wave), an electrical inhibition signal will be sent out to instantly inhibit the motor command that leads to the collision. This hybrid optical-electrical system can achieve millisecond-level reflex behavior driven directly by perception, which is faster and safer than the traditional serial architecture of "perception-computation-execution".

[0082] In summary, the present application realizes the self-rectifying characteristics of the device by constructing the HZO / TeO2 heterojunction through a low-cost and simple operation method. At the same time, the device in the present application shows a wide spectral response to deep ultraviolet to visible light, and the decryption function of the password is realized by using the light response. In addition, due to the rectifying characteristics of the heterojunction, the heterojunction shows different response characteristics to positive and negative voltages, and the combination of the optical synapse and the electrical synapse successfully realizes the enhancement of the optical pulse and the inhibition of the electrical pulse, which lays a foundation for neuromorphic computing applications.

Claims

1. A self-rectifying memristor based on hafnium-zirconium oxide / tellurium oxide heterojunction, characterized in that, The heterojunction of the self-rectifying memristor is formed by hafnium-zirconium-oxygen and tellurium oxide.

2. The hafnium-zirconium oxide / tellurium oxide heterojunction-based self- rectifying memristor of claim 1, wherein, The structure of the self-rectifying memristor is TiN / HZO / TeO2 / A, wherein A is a metal electrode.

3. The method of making hafnium-zirconium oxide / tellurium oxide heterojunction based self rectifying memristor of claim 1, wherein, A TiN / HZO substrate is placed in a vacuum evaporation device to evaporate a tellurium dioxide film with tellurium dioxide as an evaporation source, and finally a metal electrode is evaporated on the tellurium dioxide film.

4. The production method according to claim 3, characterized by, The TiN / HZO substrate has a HZO thickness of 2-10 nm, and the tellurium dioxide film has a thickness of 2-10 nm and is deposited on the surface of the HZO; after the metal electrode is evaporated, the device is subjected to air annealing at a temperature of 100-500°C for 10-60 min.

5. Use of the hafnium-zirconium oxide / tellurium oxide heterojunction based self- rectifying memristor of claim 1 for memory, characterized in that, Information is written with light pulses of different wavelengths, and the post-pulse current is used as information representation.

6. Application of the hafnium-zirconium-oxygen / tellurium oxide heterojunction-based self-rectifying memristor of claim 1 to an optical response detector, wherein the optical response wavelength range of the self-rectifying memristor is 254 nm-520 nm, and the incident light is incident on the tellurium oxide at an oblique angle.

7. Use of the hafnium-zirconium oxide / tellurium oxide heterojunction based self- rectifying memristor of claim 1 for an optical synapse, characterized in that, Applying light pulses to the self-rectifying memristor is defined as a learning process, and not applying light pulses is defined as a forgetting process, a forgetting time is set, the post-pulse current of each time is monitored, and the process of deepening memory in repeated learning of the human brain is simulated.

8. Use of the hafnium-zirconium oxide / tellurium oxide heterojunction based self- rectifying memristor of claim 1 for an optical synapse, characterized in that, Applying a first light pulse to the self-rectifying memristor as a conditioned stimulus and applying a second light pulse with a wavelength different from the first light pulse as an unconditioned stimulus, a threshold current is set, the post-pulse current is monitored, and the process of human brain associative learning is simulated.

9. Use of the hafnium-zirconium oxide / tellurium oxide heterojunction based self- rectifying memristor of claim 1 for a password lock, characterized in that, For an M-bit input password, there are N digital choices, N>M, and one self-rectifying memristor is arranged for each bit, for any password, it is designed as the post-pulse current of the self-rectifying memristor at a certain time, and only when the monitored currents of all M self-rectifying memristors after light pulse input meet the requirements, the unlocking is completed.

10. Use of the hafnium-zirconium oxide / tellurium oxide heterojunction based self- rectifying memristor of claim 1 for multi-modal neuromorphic computing units, characterized in that, Applying a light pulse to the self-rectifying memristor first and then applying an electric pulse to realize weight increase and decrease of a single device, corresponding to multiple conductance states, combining the multiple conductance states with a neural network to perform target recognition.

Images