Reconfigurable gadolinium oxide memristor based on electrode engineering, preparation method thereof and reservoir computing system

Abstract

This invention relates to a reconfigurable gadolinium oxide memristor based on electrode engineering, its fabrication method, and a reservoir computing system. The reconfigurable gadolinium oxide memristor comprises a bottom electrode, a gadolinium oxide layer, and a top electrode stacked sequentially. In this invention, gadolinium oxide is used as the resistive switching material, which possesses both excellent ion migration capabilities and spontaneous relaxation characteristics. Therefore, a reconfigurable gadolinium oxide memristor with both volatile and non-volatile properties is developed based on electrode engineering to meet the functional requirements of the reservoir and readout layers in reservoir computing. In reservoir computing, the volatile Al / Gd₂O₃ / Pt memristor acts as the physical reservoir to achieve nonlinear mapping of the input signal; the non-volatile Ta / Gd₂O₃ / Pt memristor serves as the readout network, enabling parallel computation and identification, effectively solving the problems of slow data processing speed, poor real-time performance, and high integration difficulty in existing memristor-based reservoir computing systems.

Description

Technical Field

[0001] This invention relates to the field of semiconductor technology, specifically to a reconfigurable gadolinium oxide memristor based on electrode engineering, its fabrication method, and a reservoir computing system. Background Technology

[0002] With the rapid development of science and technology, artificial intelligence (AI) has been widely applied in fields such as intelligent driving, intelligent education, smart homes, and smart factories. However, with the popularization of applications, the demand for AI computing power and power consumption has also increased exponentially. When processing data, neural networks such as deep neural networks (DNNs) suffer from problems such as large training parameters and gradient explosion, hindering the speed of data processing and computation. Reservoir computing, as a three-layer network structure, consists of an input layer, a reservoir layer, and a readout layer. The connection between the input layer and the reservoir layer is fixed and does not require training. Only the weights of the reservoir layer and the readout layer need to be trained to obtain the final result, thus offering advantages such as fewer training parameters and fewer floating-point operations.

[0003] Reserve-pool computation nonlinearly maps the input to a high-dimensional space through the reservoir layer, enriching the feature information. The output layer only needs to learn the mapping relationship between the reservoir state and the target output through linear regression and optimize the weights to obtain the prediction result. To fully realize the potential of reserve-pool computation, utilizing the dynamic migration and spontaneous relaxation characteristics of ions within the physical device for nonlinear information processing is a current research hotspot. However, existing reserve-pool computation systems based on memristors suffer from slow computation speed, incompatibility between materials and device structures, and low integration density, which severely impact data processing speed and computational performance, and lead to a significant increase in process costs. Summary of the Invention

[0004] Based on the above description, the present invention provides a reconfigurable gadolinium oxide memristor based on electrode engineering, its preparation method and a reservoir computing system, aiming to achieve high-speed computing, improve system integration and process compatibility.

[0005] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: The present invention provides a reconfigurable gadolinium oxide memristor based on electrode engineering, comprising a substrate, an adhesive layer, a bottom electrode, a gadolinium oxide layer and a top electrode stacked sequentially.

[0006] Furthermore, the material of the top electrode includes aluminum or tantalum.

[0007] Furthermore, the thickness of the gadolinium oxide layer is 10 nm-20 nm.

[0008] Furthermore, the substrate is made of silicon, the adhesive layer is made of titanium, and the bottom electrode is made of platinum.

[0009] Furthermore, the substrate has a thickness of 500 μm, the adhesive layer has a thickness of 10 nm-20 nm, the bottom electrode has a thickness of 100 nm-200 nm, and the top electrode has a thickness of 100 nm-200 nm.

[0010] This invention also proposes a method for fabricating a reconfigurable gadolinium oxide memristor based on electrode engineering, as described above, comprising: S1. Obtain a substrate, and sequentially form an adhesive layer and a bottom electrode on the substrate; S2. A gadolinium oxide layer is formed on the bottom electrode; S3. Define a top electrode region on the gadolinium oxide layer, coat it with photoresist, bake, expose, and develop it. After removing the photoresist from the top electrode region, form a top electrode in the top electrode region, and strip it to obtain a reconfigurable gadolinium oxide memristor based on electrode engineering.

[0011] Furthermore, step S1 includes: Obtain the substrate and ultrasonically clean it sequentially with acetone, anhydrous ethanol and deionized water for 10 min-20 min, with an ultrasonic power of 60 W-80 W. After drying, the adhesive layer and the bottom electrode are sputtered sequentially onto the substrate in an argon atmosphere.

[0012] Furthermore, in step S1, the power during sputtering of the adhesive layer and the bottom electrode is 70 W-90 W, and the pressure of the argon gas is 0.4 Pa-0.5 Pa.

[0013] This invention also proposes a reservoir computing system based on electrode engineering for reconfigurable gadolinium oxide memristors, comprising: An input layer is used to convert the target to be identified into a timing signal and input it into the reservoir layer; A reservoir layer includes multiple reservoir units, each reservoir unit including a volatile memristor, and the reservoir layer is used to map an input signal into a high-dimensional feature vector. The readout layer includes multiple non-volatile memristors arranged in an array. The readout layer is used to perform a dot product operation between the output value of the reservoir layer and the weights of the conductance-mapped output neurons to obtain the final output result. The volatile memristor and the non-volatile memristor include the reconfigurable gadolinium oxide memristor based on electrode engineering as described above, wherein the top electrode material of the volatile memristor is aluminum, and the top electrode material of the non-volatile memristor is tantalum.

[0014] Compared with the prior art, the technical solution of this application has the following beneficial technical effects: 1. A reservoir computing system based on reconfigurable gadolinium oxide memristors, implemented through electrode engineering, utilizes the same substrate, bottom electrode, and resistive switching layer materials. Short-term memory and multi-level storage characteristics are achieved only by changing the top electrode, adapting to the functional requirements of the reservoir layer and readout layer in a fully connected reservoir computing system. Electrode engineering reduces manufacturing difficulty and improves system integration density and process compatibility, thus solving the problem of inconsistent material systems between the reservoir layer and readout layer. Simultaneously, the input signal is encoded into a 4-bit pulse sequence and input to the reservoir layer constructed from Al / Gd₂O₃ / Pt volatile memristors. The reservoir layer generates a current response, which is then encoded into a voltage and fed into the readout network composed of Ta / Gd₂O₃ / Pt non-volatile memristors. The network performs a dot product operation between the reservoir layer output value and the conductance-mapped output neuron weights to obtain the final prediction result. The fully connected readout layer supports direct parallel computation of data, effectively avoiding latency losses during software scheduling and instruction execution. In scenarios requiring rapid response (such as real-time signal processing and sensor data acquisition), the hardware-based readout layer can complete data reading and preliminary processing more quickly, reducing system latency and thus exhibiting excellent real-time characteristics.

[0015] 2. By utilizing the migration and spontaneous relaxation behavior of ions inside the memristor to perform nonlinear information processing, nonlinear decay behavior can be obtained without the need for additional auxiliary circuit design, such as Mackey-Glass nonlinear circuits and ReLU integral leakage circuits. Therefore, the system requires low power consumption. Attached Figure Description

[0016] Figure 1 A schematic diagram of an embodiment of the reconfigurable gadolinium oxide memristor based on electrode engineering provided by the present invention; Figure 2 This is a schematic diagram of the reservoir calculation system based on electrode engineering for a reconfigurable gadolinium oxide memristor provided in Embodiment 2 of the present invention; Figure 3 This is a flowchart of the fabrication method of a reconfigurable gadolinium oxide memristor based on electrode engineering provided in Embodiment 2 of the present invention; Figure 4 This is a current-voltage scanning diagram of the Al / Gd2O3 / Pt and Ta / Gd2O3 / Pt memristors provided in Embodiment 1 of the present invention; Figure 5 This is a schematic diagram of the 4-bit output state implemented by the Al / Gd2O3 / Pt memristor provided in Embodiment 1 of the present invention; Figure 6 This is a schematic diagram of the conductance enhancement and suppression achieved by the Ta / Gd2O3 / Pt memristor provided in Embodiment 1 of the present invention under pulse action; Figure 7This is a schematic diagram of the reservoir computing system provided in Embodiment 2 of the present invention, including the entire process of audio preprocessing and feature mapping of the reservoir layer to the readout network; Figure 8 This is a schematic diagram showing the change in audio recognition rate of the reservoir calculation system provided in Embodiment 2 of the present invention with the training cycle.

[0017] The attached diagram lists the components represented by each number as follows: 100. Reconfigurable gadolinium oxide memristor based on electrode engineering; 1. Substrate; 2. Adhesive layer; 3. Bottom electrode; 4. Gadolinium oxide layer; 5. Top electrode. Detailed Implementation

[0018] To facilitate understanding of this application, a more complete description will be provided below with reference to the accompanying drawings, which illustrate embodiments of the present application. However, the present application can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that the disclosure of this application will be thorough and complete.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application.

[0020] With the rapid development of science and technology, artificial intelligence (AI) has been widely applied in fields such as intelligent driving, intelligent education, smart homes, and smart factories. However, with the popularization of applications, the demand for AI computing power and power consumption has also increased exponentially. When processing data, neural networks such as deep neural networks (DNNs) suffer from problems such as large training parameters and gradient explosion, hindering the speed of data processing and computation. Reservoir computing, as a three-layer network structure, consists of an input layer, a reservoir layer, and a readout layer. The connection between the input layer and the reservoir layer is fixed and does not require training. Only the weights of the reservoir layer and the readout layer need to be trained to obtain the final result, thus offering advantages such as fewer training parameters and fewer floating-point operations.

[0021] Reserve-pool computation nonlinearly maps the input to a high-dimensional space through the reservoir layer, enriching the feature information. The output layer only needs to learn the mapping relationship between the reservoir state and the target output through linear regression and optimize the weights to obtain the prediction result. To fully realize the potential of reserve-pool computation, utilizing the dynamic migration and spontaneous relaxation characteristics of ions within the physical device for nonlinear information processing is a current research hotspot. However, reserve-pool computation systems based on memristors suffer from slow computation speed, incompatibility between materials and device structures, and low integration density, which severely impact data processing speed and computational performance, and lead to a significant increase in process costs.

[0022] In view of this, see Figure 1 The present invention provides a reconfigurable gadolinium oxide memristor 100 based on electrode engineering, comprising a substrate 1, an adhesive layer 2, a bottom electrode 3, a gadolinium oxide layer 4 and a top electrode 5 stacked sequentially.

[0023] In the technical solution of this invention, a gadolinium oxide layer 4 is used. Gadolinium oxide material has excellent ion migration and spontaneous relaxation characteristics, which can be used as a physical reservoir layer to effectively realize high-dimensional nonlinear mapping of input signals, enrich timing features, and thus improve the reservoir computing's ability to process complex dynamic data. The gadolinium oxide layer 4 has good contact ability with metal. Compared with traditional memristor-based reservoir computing systems, good contact makes the interface barrier uniform and the energy barrier threshold of ion migration consistent, thereby significantly reducing the fluctuation range of operating voltage and making the device's set / reset threshold voltage more stable. This is not only crucial for the large-scale integration and precise control of memristor arrays, but also helps to overcome the technical limitations of existing reservoir computing in terms of real-time performance and energy efficiency.

[0024] Furthermore, the material of the top electrode 5 includes aluminum or tantalum.

[0025] In the technical solution of this invention, when the top electrode is aluminum (Al), aluminum has a high oxygen affinity, thereby forming AlO. x The barrier layer makes it difficult for electrons to cross Gd2O3 / AlO x The interface barrier, bearing most of the voltage drop, makes it difficult for conductive filaments with oxygen vacancies to form. Current conduction mainly occurs through electron migration and tunneling, resulting in volatile memristor characteristics in the device. When the top electrode is tantalum (Ta), due to Ta's weak affinity and TaO... x The electron affinity of TaO is relatively large, therefore... x The effect of the layer on electrical performance is negligible. Under the same current limiting conditions, oxygen vacancy conductive filaments will be preferentially formed to connect the upper and lower electrodes. A negative Reset scan voltage needs to be applied to break the conductive filaments, thereby giving the device non-volatile memristor characteristics.

[0026] Meanwhile, in this invention, by simply changing the top electrode material, volatile or non-volatile operating modes can be flexibly implemented on the same gadolinium oxide device without changing the core material system or device structure, significantly improving the adaptability and versatility of the hardware architecture.

[0027] Furthermore, the thickness of the gadolinium oxide layer 4 is 10 nm-20 nm.

[0028] In the technical solution of the present invention, when the thickness is controlled at 10 nm-20 nm, a sufficiently strong electric field can be used to drive oxygen vacancy migration at a lower operating voltage, ensuring fast and repeatable resistance switching; at the same time, it avoids the increase in pinhole defects, breakdown risk or device inconsistency caused by excessively thin film (<10 nm), and also prevents problems such as excessively high operating voltage and slow response speed caused by excessively thick film (>20 nm), thereby achieving an optimized balance between performance and reliability.

[0029] Furthermore, the substrate 1 is made of silicon, the adhesive layer 2 is made of titanium, and the bottom electrode 3 is made of platinum.

[0030] In the technical solution of this invention, platinum (Pt) is used as the bottom electrode material, which has high conductivity, high work function, and strong chemical stability in oxidizing environments. During the operation of gadolinium oxide memristor devices, platinum hardly participates in oxygen ion exchange reactions, which can effectively prevent electrode oxidation or metal ion injection into the gadolinium oxide layer, thereby maintaining a clear interface, reducing parasitic effects, and ensuring that the memristor switching behavior is mainly dominated by the oxygen vacancy dynamics inside the gadolinium oxide layer, thus improving the repeatability and reliability of device performance. Titanium (Ti) is used as the bonding layer, which can form a stable silicide interface with silicon at one end and a good metal bond with platinum at the other end, significantly improving the adhesion of the bottom electrode in subsequent high-temperature or electrical stress processes, preventing platinum layer peeling or wrinkling, and ensuring the long-term structural integrity of the device. By using a silicon (Si) substrate, there is no need to introduce non-standard processes or new material systems, and it can be directly integrated into existing CMOS back-end processes, providing a process basis for realizing monolithic heterogeneous integration of gadolinium oxide-based reservoir computing units and logic circuits.

[0031] It should be noted that the substrate is silicon, specifically referring to the substrate of the reconfigurable gadolinium oxide memristor based on electrode engineering. The substrate can be an SOI substrate or a pure silicon substrate; the specific type is not limited here. Specifically, in some embodiments of the present invention, a Si / SiO2 substrate is used.

[0032] Furthermore, the substrate has a thickness of 500 μm, the adhesive layer has a thickness of 10 nm-20 nm, the bottom electrode has a thickness of 100 nm-200 nm, and the top electrode has a thickness of 100 nm-200 nm.

[0033] This invention also proposes a method for fabricating a reconfigurable gadolinium oxide memristor based on electrode engineering, as described above, comprising: S1. Obtain a substrate, and sequentially form an adhesive layer and a bottom electrode on the substrate; S2. A gadolinium oxide layer is formed on the bottom electrode; S3. Define a top electrode region on the gadolinium oxide layer, coat it with photoresist, bake, expose, and develop it. After removing the photoresist from the top electrode region, form a top electrode in the top electrode region, and strip it to obtain a reconfigurable gadolinium oxide memristor based on electrode engineering.

[0034] In the technical solution of this invention, by adopting a standard micro-nano fabrication process of deposition-photolithography-lift, complex etching steps are eliminated, avoiding plasma damage or chemical corrosion to the gadolinium oxide functional layer and effectively protecting its internal oxygen vacancy distribution and interface integrity. Simultaneously, the process modules used (such as sputtering and ultraviolet lithography) are highly compatible with existing CMOS production lines, facilitating large-scale, high-yield manufacturing.

[0035] Furthermore, step S1 includes: Obtain the substrate and ultrasonically clean it sequentially with acetone, anhydrous ethanol and deionized water for 10 min-20 min, with an ultrasonic power of 60 W-80 W. After drying, the adhesive layer and the bottom electrode are sputtered sequentially onto the substrate in an argon atmosphere.

[0036] In the technical solution of this invention, acetone, anhydrous ethanol and deionized water are used for sequential ultrasonic cleaning, which can efficiently remove organic contaminants, particulate impurities and residual moisture from the substrate surface. By adjusting the ultrasonic cleaning parameters, the substrate can be thoroughly cleaned while avoiding excessive ultrasonic damage. By performing magnetron sputtering in an argon atmosphere, the reaction of reactive gases such as oxygen and water vapor with the metal during the deposition process can be avoided to generate oxide impurities, thereby ensuring the high purity, density and good conductivity of the titanium adhesive layer and the platinum base electrode.

[0037] Furthermore, in step S1, the power during sputtering of the adhesive layer and the bottom electrode is 70 W-90 W, and the pressure of the argon atmosphere is 0.4 Pa-0.5 Pa.

[0038] In the technical solution of this invention, under a relatively low argon pressure of 0.4 Pa to 0.5 Pa, the mean free path of the sputtered particles is relatively long, and the atoms experience fewer collisions before reaching the substrate, resulting in higher kinetic energy. This is beneficial for forming dense, continuous, and fine-grained titanium / platinum thin films. Simultaneously, the moderate sputtering power of 70 W to 90 W avoids target overheating, particle splashing, or excessive film stress caused by excessive energy, effectively suppressing defects such as pinholes, voids, and columnar crystals, and improving the overall quality of the film.

[0039] This invention also proposes a reservoir computing system based on electrode engineering for reconfigurable gadolinium oxide memristors, see reference. Figure 2 ,include: An input layer is used to convert the target to be identified into a timing signal and input it into the reservoir layer; A reservoir layer includes multiple reservoir units, each reservoir unit including a volatile memristor, and the reservoir layer is used to map an input signal into a high-dimensional feature vector. The readout layer includes multiple non-volatile memristors arranged in an array. The readout layer is used to perform a dot product operation between the output value of the reservoir layer and the weights of the conductance-mapped output neurons to obtain the final output result. The volatile memristor and the non-volatile memristor include the reconfigurable gadolinium oxide memristor based on electrode engineering as described above, wherein the top electrode material of the volatile memristor is aluminum, and the top electrode material of the non-volatile memristor is tantalum.

[0040] In the technical solution of this invention, both volatile and non-volatile memristors are based on the same gadolinium oxide material system and sandwich structure. Dual functions can be achieved simply by switching the top electrode material. This "homogeneous heterogeneous" design avoids process conflicts and interface incompatibility problems caused by multiple material systems. It can be integrated on the same wafer through standard photolithography and stripping processes to achieve a high-density, high-consistency hybrid memristor array, which greatly improves system integration and manufacturing yield.

[0041] The technical solution of the present invention will be further described in detail below with reference to specific embodiments. It should be understood that the following embodiments are only used to explain the present invention and are not intended to limit the present invention.

[0042] Unless otherwise specified, all materials and reagents used in the following examples are commercially available.

[0043] Example 1 This embodiment provides a reconfigurable gadolinium oxide memristor based on electrode engineering, such as... Figure 1 As shown, Al / Gd₂O₃ / Pt devices and Ta / Gd₂O₃ / Pt devices are prepared by the following methods: Step 1: Use a diamond cutter to cut the wafer into 1×1 cm pieces. 2 Experimental silicon wafers; Step 2: Transfer several cut silicon wafers to a beaker and place them in an ultrasonic cleaner. Clean them step by step with acetone, anhydrous ethanol and deionized water in sequence. The ultrasonic cleaning time of the silicon wafers in each solution is 15 min and the ultrasonic power is 70 W. Step 3: Place the platinum target and titanium target with a purity of 99.99% into target slots No. 2 and No. 3 respectively, fix them with iron pressure rings and cover them with anode covers; Step four: Fix the cleaned silicon wafer onto the rotating substrate, place it in the vacuum chamber, and wait for the vacuum chamber to be pumped to 5×10⁻⁶. -4 When the pressure reaches Pa, close the main valve, open the inflation valve, introduce 30 sccm of Ar, adjust the chamber pressure to 0.45 Pa, set the target power to 80W, and sputter the Ti adhesive layer and Pt bottom electrode in sequence. Step 5: Disassemble the Pt and Ti targets, replace them with gadolinium oxide targets of 99.99% purity, place them in target cell No. 3, secure them with iron clamping rings, and cover with the anode cover; close the chamber door and evacuate to 5×10⁻⁶. -4 When the pressure reaches Pa, close the main valve, open the gas filling valve, introduce 30 sccm of Ar, adjust the chamber pressure to 0.45 Pa, set the target power to 60 W, and start sputtering a 16 nm Gd2O3 thin film. Step six: Define the top electrode region through operations such as spin coating of photoresist, baking, exposure, and development; Step seven: Following steps three and four, two different metals, aluminum and tantalum, are sputtered as top electrodes, and then the Al / Gd2O3 / Pt volatile memristor and Ta / Gd2O3 / Pt non-volatile memristor are fabricated.

[0044] Example 2 This embodiment provides a reservoir computing system based on electrode engineering for reconfigurable gadolinium oxide memristors. (See reference...) Figure 2 and Figure 3 Its preparation method includes: Achieving reconfigurable characteristics: Volatile and non-volatile memristors that meet the functional requirements of the reservoir layer and readout layer in reservoir calculation are fabricated by electrode engineering methods, namely the Al / Gd2O3 / Pt volatile memristor and Ta / Gd2O3 / Pt non-volatile memristor provided in Example 1. Constructing a reservoir computing network: Al / Gd2O3 / Pt volatile memristors exhibit nonlinear current decay characteristics, enabling high-dimensional nonlinear mapping of features, thus serving as the physical reservoir layer; Ta / Gd2O3 / Pt nonvolatile memristors demonstrate stable multi-order conductance storage capabilities, meeting the requirements of weight mapping and in-situ computation, thus serving as the readout network for parallel computation; thereby completing the construction of a reservoir computing system comprising an input layer, a reservoir layer, and a readout layer.

[0045] Performance testing The electrical performance of the reconfigurable memristor was characterized using a B1500A semiconductor parameter analyzer. The tests included IV scanning, 4-bit output state testing, and multi-order conductance modulation and storage testing. The specific steps are as follows: Step 1, Connect the B1500A semiconductor parameter analyzer: Connect the two probes of the B1500A to the bottom and top electrodes of the memristor respectively. The voltage is applied by the top electrode and the bottom electrode is grounded. Step two: Using the DC module of the B1500A tester, voltage-current scanning tests were performed on the Al / Gd₂O₃ / Pt and Ta / Gd₂O₃ / Pt devices provided in Example 1, respectively. The results are as follows: Figure 4 As shown, Al electrode devices exhibit volatility, while Ta electrode devices exhibit non-volatility.

[0046] Step 3, test the nonlinear decay characteristics: Apply continuous action pulses to the Al / Gd2O3 / Pt device, and then monitor the current change under the action of the read pulses. The decay behavior conforms to the double exponential function by fitting. Step 4, Test the 4-bit output state: Encode the finally extracted features 0 and 1 into a 4-bit pulse sequence (0000 ~ 1111) and apply it to the Al / Gd2O3 / Pt memristor, then read the corresponding conductance state. The result is as follows: Figure 5 As shown, "0" indicates no pulse action, and "1" indicates a large write pulse; Step 5: Multi-order conductance modulation. By applying continuous positive and negative pulses, the Ta / Gd₂O₃ / Pt memristor achieves continuous conductance enhancement and suppression, as shown in the following figure. Figure 6 As shown, the obtained conductivity value can be used as the weight of the readout network.

[0047] Audio Recognition: To verify the feasibility of the constructed reservoir computing system, Fourier transform was used to convert the audio information into a signal in the frequency domain. (See [reference needed]). Figure 7 Resampling fixes the frequency of each audio sample at 8000 Hz. Then, the samples are divided into 32 frames, each with 16 channel values. These values ​​are binarized into 0 and 1, and then cut and spliced ​​into a 128×4 black and white image to extract information features. The feature vectors are then encoded into pulse sequences and fed into the reservoir layer. The output of the reservoir layer is used as the input to the readout network, and matrix multiplication is performed with the weights mapped to the Ta / Gd2O3 / Pt memristor. The fully connected readout network is iteratively trained to output the final recognition result.

[0048] The results are as follows Figure 8 As shown. According to Figure 8 It can be seen that the audio recognition accuracy can reach 98% within 100 training cycles, which verifies the feasibility of the constructed reservoir computing system.

[0049] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0050] In summary, the technical solution of this application has the following beneficial technical effects: 1. A reservoir computing system based on reconfigurable gadolinium oxide memristors, implemented through electrode engineering, utilizes the same substrate, bottom electrode, and resistive switching layer materials. Short-term memory and multi-level storage characteristics are achieved only by changing the top electrode, adapting to the functional requirements of the reservoir layer and readout layer in a fully connected reservoir computing system. Electrode engineering reduces manufacturing difficulty and improves system integration density and process compatibility, thus solving the problem of inconsistent material systems between the reservoir layer and readout layer. Simultaneously, the input signal is encoded into a 4-bit pulse sequence and input to the reservoir layer constructed from Al / Gd₂O₃ / Pt volatile memristors. The reservoir layer generates a current response, which is then encoded into a voltage and fed into the readout network composed of Ta / Gd₂O₃ / Pt non-volatile memristors. The network performs a dot product operation between the reservoir layer output value and the conductance-mapped output neuron weights to obtain the final prediction result. The fully connected readout layer supports direct parallel computation of data, effectively avoiding latency losses during software scheduling and instruction execution. In scenarios requiring rapid response (such as real-time signal processing and sensor data acquisition), the hardware-based readout layer can complete data reading and preliminary processing more quickly, reducing system latency and thus exhibiting excellent real-time characteristics.

[0051] 2. By utilizing the migration and spontaneous relaxation behavior of ions inside the memristor to perform nonlinear information processing, nonlinear decay behavior can be obtained without the need for additional auxiliary circuit design, such as Mackey-Glass nonlinear circuits and ReLU integral leakage circuits. Therefore, the system requires low power consumption.

Claims

1. A reconfigurable gadolinium oxide memristor based on electrode engineering, characterized in that, It includes a substrate, an adhesive layer, a bottom electrode, a gadolinium oxide layer, and a top electrode stacked in sequence.

2. The reconfigurable gadolinium oxide memristor based on electrode engineering according to claim 1, characterized in that, The material of the top electrode includes aluminum or tantalum.

3. The reconfigurable gadolinium oxide memristor based on electrode engineering according to claim 1, characterized in that, The thickness of the gadolinium oxide layer is 10 nm-20 nm.

4. The reconfigurable gadolinium oxide memristor based on electrode engineering according to claim 1, characterized in that, The substrate is made of silicon, the adhesive layer is made of titanium, and the bottom electrode is made of platinum.

5. The reconfigurable gadolinium oxide memristor based on electrode engineering according to claim 1, characterized in that, The substrate has a thickness of 500 μm, the adhesive layer has a thickness of 10 nm-20 nm, the bottom electrode has a thickness of 100 nm-200 nm, and the top electrode has a thickness of 100 nm-200 nm.

6. A method for fabricating a reconfigurable gadolinium oxide memristor based on electrode engineering as described in any one of claims 1 to 5, characterized in that, include: S1. Obtain a substrate, and sequentially form an adhesive layer and a bottom electrode on the substrate; S2. A gadolinium oxide layer is formed on the bottom electrode; S3. Define a top electrode region on the gadolinium oxide layer, coat it with photoresist, bake, expose, and develop it. After removing the photoresist from the top electrode region, form a top electrode in the top electrode region. Use acetone solution to strip it to obtain a reconfigurable gadolinium oxide memristor based on electrode engineering.

7. The method for fabricating a reconfigurable gadolinium oxide memristor based on electrode engineering according to claim 6, characterized in that, Step S1 includes: Obtain the substrate and ultrasonically clean it sequentially with acetone, anhydrous ethanol and deionized water for 10 min-20 min, with an ultrasonic power of 60 W-80 W. After drying, the adhesive layer and the bottom electrode are sputtered sequentially onto the substrate in an argon atmosphere.

8. The method for fabricating a reconfigurable gadolinium oxide memristor based on electrode engineering according to claim 7, characterized in that, In step S1, the power during sputtering of the adhesive layer and the bottom electrode is 70 W-90 W, and the pressure of the argon atmosphere is 0.4 Pa-0.5 Pa.

9. A reservoir computing system based on a reconfigurable gadolinium oxide memristor using electrode engineering, characterized in that, include: An input layer is used to convert the target to be identified into a timing signal and input it into the reservoir layer; A reservoir layer includes multiple reservoir units, each reservoir unit including a volatile memristor, and the reservoir layer is used to map an input signal into a high-dimensional feature vector. The readout layer includes multiple non-volatile memristors arranged in an array. The readout layer is used to perform a dot product operation between the output value of the reservoir layer and the weights of the conductance-mapped output neurons to obtain the final output result. The volatile memristor and the non-volatile memristor include the reconfigurable gadolinium oxide memristor based on electrode engineering as described in any one of claims 1 to 5, wherein the top electrode material of the volatile memristor is aluminum, and the top electrode material of the non-volatile memristor is tantalum.

Images