Preparation method and application of carbon nanotube heterojunction device

Abstract

The invention discloses a preparation method of a carbon nano tube heterojunction device, which comprises the following steps of: preparing a pre-patterned Au / Ni thin film metal electrode on a silicon wafer substrate containing silicon dioxide by adopting sputtering and stripping processes, wherein the distance between the electrodes is 1.5-2 microns, and the ratio of the width of the electrodes to the distance is 15-30; the electrodes are arranged on the silicon wafer, carbon nanotubes are assembled between the electrodes by adopting a dielectrophoresis technology, a heterogeneous nodule point array is formed between the carbon nanotubes and the electrodes, the silicon wafer is connected with an external adapter plate through a lead bonding technology, and the external adapter plate is used for inputting electric signals to the heterogeneous nodule point array and collecting the electric signals. The carbon nanotube heterojunction device prepared by the method is used for constructing a reserve pool, providing efficient reserve pool calculation, and realizing topology reconfigurable and physically realizing the reserve pool with extremely low complexity by performing mask expansion on a single node of the carbon nanotube heterojunction, so that the requirement of a high-complexity task is met.

Description

Technical Field

[0001] The present invention relates to the technical fields of brain-inspired computing and artificial intelligence, and particularly relates to a method for preparing a carbon nanotube heterojunction device, constructing a carbon nanotube reservoir, and processing data based on the constructed reservoir. Background Art

[0002] Reservoir Computing (RC) is an efficient machine learning method, especially suitable for processing time series signals. Traditional RC relies on a large-scale randomly connected node network, and has problems such as complex hardware implementation and high power consumption. Carbon nanotube heterojunctions have excellent electrical properties, providing the possibility for constructing miniaturized and low-power reservoirs.

[0003] Reservoir computing is a lightweight recurrent neural network. Its core idea is to use a fixed and randomly initialized reservoir to process input data and output through a simple linear readout layer. Since the input layer weights and reservoir weights are pre-generated and fixed, the reservoir computing network has few parameters and small computational complexity. Therefore, it can achieve low-cost, high-efficiency, and low-power time series task processing. At the same time, the concept of reservoir computing can be directly mapped to a physical system, and the natural dynamic characteristics of the physical system are used to construct a physical reservoir, which has a complex nonlinearity that software cannot achieve and a higher upper limit of the ability to process time series data.

[0004] Existing physical reservoir nodes often use electronic, optical, or mechanical nodes to implement, lacking the exploration of the implementation of new nonlinear physical nodes. Structurally, they often use a physically randomly interconnected topological structure to implement, and once generated, they are fixed, without flexibility, and the physical implementation is difficult.

[0005] Carbon Nanotube (CNT) is a new type of nanomaterial, which has received extensive attention due to its excellent electrical conductivity, mechanical properties, and nonlinear characteristics. The reservoir based on carbon nanotube heterojunctions provides new possibilities for the efficient implementation of reservoir computing due to its natural nonlinearity and easy hardware integration characteristics. By masking and expanding a single node of a strongly nonlinear carbon nanotube heterojunction, a reservoir with reconfigurable topology and extremely low physical implementation complexity can be realized. Summary of the Invention

[0006] To achieve the above object, the present invention provides a method for preparing a carbon nanotube heterojunction device, including:

[0007] S1: Preparing a pre-patterned Au / Ni thin film metal electrode on a silicon wafer substrate containing silicon dioxide by sputtering and stripping processes, where the electrode spacing is 1.5 - 2 μm, and the ratio of electrode width to spacing is 15 - 30; and

[0008] S2: Assemble carbon nanotubes between the electrodes by dielectrophoresis technology, wherein a heterojunction node array is formed between the carbon nanotubes and the electrodes, and the silicon wafer is connected to an external adapter board through wire bonding technology, and the external adapter board is used to input and collect electrical signals to / from the heterojunction node array.

[0009] Preferably, the raw semiconductor carbon nanotubes use toluene as a solvent and PCz as a dispersant. The step S2 further includes:

[0010] Extract the raw semiconductor carbon nanotube solution and dry it. Take the residual solid, mix it with ethanol and ultrasonicate to obtain a carbon nanotube solution;

[0011] Mix phosphomolybdic acid particles with ethanol and ultrasonicate to prepare a phosphomolybdic acid solution with a preset concentration, wherein the preset concentration of the phosphomolybdic acid solution is 250 - 400 μg / ml;

[0012] Take the carbon nanotube solution and the phosphomolybdic acid solution, mix and ultrasonicate them to obtain a phosphomolybdic acid - modified carbon nanotube solution; and

[0013] Place the probes at both ends of the Au / Ni thin - film electrode, and drip the phosphomolybdic acid - modified carbon nanotube solution. The signal generator generates a sinusoidal AC voltage signal to perform dielectrophoresis operation on the carbon nanotubes through the probes.

[0014] The present invention also discloses a method for constructing a reservoir based on a carbon nanotube heterojunction device, including:

[0015] Select a node from the heterojunction node array of the above - mentioned carbon nanotube heterojunction device, and collect the output signal of the node; and

[0016] Generate a mask matrix to perform mask expansion on the collected output signal, including:

[0017] Perform mask expansion on the output signal at each time step. Specifically, it includes multiplying the output signal with each column of the mask matrix one by one, and splicing them column - by - column to obtain a mask - expanded matrix;

[0018] Re - organize the mask - expanded matrix according to time steps, map the output signal to a high - dimensional state space, and construct a reservoir state matrix to complete the construction of the reservoir.

[0019] Preferably, selecting the node further includes: analyzing the volt - ampere characteristics of multiple heterojunction nodes, and selecting a strongly non - linear node as the collection node from the multiple nodes. When the non - linearity index 0 < λ ≤ 0.25, the node is a strongly non - linear node; when 0.25 < λ < 0.5, the node is a weakly non - linear node, and the value range of λ is 0 - 0.5.

[0020] Preferably, the mask matrix is a matrix randomly generated in a uniform distribution between [-1, 1] or the mask matrix is composed of 1 and -1, where the number of 1s and -1s in each row of the mask matrix is a preset ratio, and the mask matrix is randomly generated in a hypergeometric distribution manner.

[0021] Preferably, constructing the reservoir state matrix includes: arranging the signals expanded from the output signals of the same time step in the mask expansion in columns, and then arranging and splicing them in the order of time steps.

[0022] Preferably, each row in the mask matrix represents an independent mask sequence, and the dimension of the mask matrix matches the dimension of the high-dimensional state space.

[0023] The present invention further discloses a method for processing data by a reservoir constructed based on a carbon nanotube heterojunction device, including:

[0024] Selecting a node in the carbon nanotube heterojunction node array on the reservoir constructed based on the carbon nanotube heterojunction device;

[0025] Setting the parameters for constructing the reservoir and loading the data set, where the data set includes a training set and a test set;

[0026] Inputting the training set to the selected node and collecting the voltage signal of the selected node;

[0027] Performing mask expansion on the signal of each time step of the voltage signal, mapping the collected voltage signal from a low-dimensional space to a high-dimensional space, and completing the construction of the reservoir;

[0028] Inputting the reservoir state constructed by the training set into the readout layer and training the readout layer; and

[0029] Mapping the test set to a high-dimensional space through reservoir construction, and inputting the reservoir state of the test set into the trained readout layer to obtain the target data.

[0030] Preferably, the data set is one of time series data, speech sequence data, and motion state sequence data. When the data set is speech sequence data, it further includes obtaining the target data in a cross-validation manner, and the target data is the speech classification result.

[0031] Preferably, the interface of the reservoir constructed by the carbon nanotube heterojunction device is built on the MATLAB platform of the upper computer, where the interface includes a parameter configuration module, a result display module, and a saving module, and the interface is used to set the parameters for constructing the reservoir, display the reservoir state, reservoir evaluation indicators, and the results of the target data.

[0032] Generally speaking, through the above technical solutions conceived by the present invention, the following beneficial effects can be achieved:

[0033] In the carbon nanotube heterojunction node array of the present invention, it is a multi-node architecture. Different nodes are selected according to task requirements, which improves the flexibility of selection. And the node redundancy provides a large fault tolerance for sample production. Even if individual nodes fail, it does not affect the effectiveness of the overall reservoir computing. The multi-node design of the carbon nanotube heterojunction node array also facilitates the expansion of the number of nodes, supports a larger-scale reservoir state space, and meets the requirements of high-complexity tasks.

[0034] Compared with the existing physical reservoirs, it has low hardware complexity. By using a mask to expand a single node to construct the state of the reservoir, it supports the flexible adjustment of the reservoir topology structure to adapt to different computing task requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is a flowchart for the preparation of carbon nanotube heterojunctions according to an embodiment of the present invention;

[0036] Figure 2 It is a physical diagram of a carbon nanotube heterojunction node array according to an embodiment of the present invention;

[0037] Figure 3 It is a schematic diagram of a typical non-linear I-V curve of a carbon nanotube heterojunction node according to an embodiment of the present invention;

[0038] Figure 4 It is a flowchart for constructing a single-node reservoir based on mask expansion according to an embodiment of the present invention;

[0039] Figure 5 It is a topology diagram of a reservoir constructed based on a nanotube heterostructure according to an embodiment of the present invention;

[0040] Figure 6 It is an overall operation flowchart of a reservoir constructed according to an embodiment of the present invention for completing a digital voice classification task;

[0041] Figure 7 It is a result diagram of a reservoir constructed according to an embodiment of the present invention for completing a digital voice classification task;

[0042] Figure 8 It is a result diagram of a reservoir constructed according to an embodiment of the present invention for completing the NARMA10 time series prediction task. DETAILED DESCRIPTION OF THE INVENTION

[0043] To make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0044] In the present invention, terms such as "first" and "second" in the present invention and the accompanying drawings (if any) are used to distinguish similar objects and do not necessarily describe a specific order or sequence.

[0045] The preparation of the carbon nanotube heterojunction uses a single-walled carbon nanotube (SWCNT) material modified with phosphomolybdic acid (POM). Combining its excellent non-linear electrical characteristics, a heterojunction structure is constructed to achieve the dynamic non-linear response of a single-node reservoir.

[0046] Embodiment 1:

[0047] Figure 1 FIG. is a flowchart for the preparation of a carbon nanotube heterojunction according to an embodiment of the present invention. The preparation of the carbon nanotube heterojunction includes preparing an Au / Ni thin film metal electrode, assembling carbon nanotubes between the electrodes, and connecting the electrodes to a signal transfer PCB board. The specific flowcharts for preparing the metal electrode pair and assembling the carbon nanotubes are as Figure 1 shown, including the following steps:

[0048] S1: Prepare a pre-patterned Au / Ni thin film metal electrode on a silicon wafer substrate containing silicon dioxide by sputtering and stripping processes, where the electrode spacing is 1.5 - 2 μm, and the ratio of the electrode width to the spacing is 15 - 30; and

[0049] S2: Assemble carbon nanotubes between the electrodes by dielectrophoresis technology, where a heterojunction node array is formed between the carbon nanotubes and the electrodes, and the silicon wafer is connected to an external transfer board through wire bonding technology, where the external transfer board is used to input and collect electrical signals for the heterojunction node array.

[0050] Specifically, the stripping process is to immerse the silicon wafer substrate in a photoresist remover to dissolve the photoresist and obtain the thin film pattern deposited on the exposed substrate area.

[0051] After completion of the stripping, the silicon wafer containing the Au / Ni thin film metal electrode is annealed, for example, the silicon wafer is annealed in a vacuum environment at 400 °C for 4 hours to enhance the adhesion between the metal thin film and the substrate, improve the crystal structure of the thin film, and improve the conductivity and stability of the electrode. The annealing time here is not limited to 4 hours.

[0052] Further, in step S2, dielectrophoresis technology is used to assemble carbon nanotubes between electrodes. Before assembling the carbon nanotubes, it also includes processing the carbon nanotubes to obtain a phosphomolybdic acid-modified carbon nanotube solution. The processing procedure is as follows:

[0053] First, the original semiconductor carbon nanotubes are dissolved in toluene solvent with PCz (poly[9-(1-octylnonyl)-9H-carbazole]) as the dispersant; then the solution containing the original semiconductor carbon nanotubes is extracted, dried, and the residual solid is mixed with ethanol and ultrasonicated to obtain a carbon nanotube solution; secondly, phosphomolybdic acid particles are mixed with ethanol and ultrasonicated to prepare a phosphomolybdic acid solution with a preset concentration, where the preset concentration of the phosphomolybdic acid solution is 250 - 400 μg / ml; and then the carbon nanotube solution and the phosphomolybdic acid solution are mixed and ultrasonicated to obtain a phosphomolybdic acid-modified carbon nanotube solution.

[0054] In the embodiment of the present invention, the carbon nanotubes are single-walled carbon nanotubes with a length of 0.8 - 3.2 μm and a tube diameter of 1.2 - 1.8 nm. The above are only the embodiments of the present invention and do not limit the present invention.

[0055] Further, in step S2, dielectrophoresis treatment is also included for the carbon nanotubes assembled between the Au / Ni thin film metal electrodes. In the embodiment of the present invention, first, the probes are respectively placed at both ends of the AuNi thin film electrodes, and the phosphomolybdic acid-modified carbon nanotube solution is dripped towards the center of the electrodes. The signal generator generates a sinusoidal alternating voltage signal, and this voltage signal is transmitted to both ends of the AuNi thin film electrodes through the probes, thereby performing dielectrophoresis on the carbon nanotubes. For example, the peak-to-peak value of the sinusoidal alternating voltage signal is 16 Vpp and the frequency is 2 MHz. Here, the center of the electrodes contains the phosphomolybdic acid-modified carbon nanotube solution.

[0056] In an alternative embodiment, the silicon wafer prepared with the carbon nanotube heterojunction node array is connected to a designed and customized signal transfer PCB board through wire bonding technology, and the signal transfer PCB board is connected to the interface of the host computer to achieve signal transmission. Specifically, the PCB board has two columns of electrodes, namely the input end and the output end. Among them, the PCB electrodes are connected to the Au / Ni thin film metal electrodes through wire bonding. The input end electrodes are used to receive the timing signal from the host computer, and after the output end electrodes are connected in series with a grounding resistor, they are used to convert the electrical response of the carbon nanotube nodes into a voltage signal, and then the host computer collects the voltage signal. Specifically, the physical diagram of the PCB board connected to the carbon nanotube heterojunction node array is as Figure 2 shown.

[0057] Specifically, Figure 2The left input terminal electrode is used to receive the timing signal from the host computer, and a grounding resistor, such as a resistor with a resistance value of 100 Ω, is connected in series to the right output terminal electrode, which is used to convert the electrical response of the carbon nanotube node into a voltage signal, such as a voltage signal of 0 - 1 V, and then the host computer collects the above voltage signal. Figure 3 It is the typical non-linear I-V curve diagram of the carbon nanotube heterojunction node in Embodiment 1 of the present invention. Figure 3 It respectively shows the volt-ampere characteristic curves of nodes 12, 16, and 29 in Sample No. 2 of the present invention.

[0058] In an optional embodiment, the carbon nanotube heterojunction nodes are divided into strong non-linear nodes and weak non-linear nodes. The discrimination rule between strong non-linear nodes and weak non-linear nodes is as follows: First, normalize the volt-ampere characteristic curve of the node, and then calculate the integral area of the curve with respect to the x-axis, denoted as λ. λ is the non-linear index, and its value range is 0 to 0.5, which is used to measure the non-linearity degree of the node. When 0 < λ ≤ 0.25, the node is a strong non-linear node; when 0.25 < λ < 0.5, the node is a weak non-linear node. The method for distinguishing strong non-linear nodes and weak non-linear nodes also includes referring to the current-time curve and spectral characteristics of the node. The setting of the above non-linear index λ is only an embodiment of the present invention and cannot be used as a limitation of the present invention.

[0059] As Figure 3 shown, due to the dispersion of the nodes themselves, there are differences in the dynamic responses of each node. In order to analyze the non-linearity of each carbon nanotube heterojunction node, the electrical characteristics of the two-terminal device samples are tested and analyzed at room temperature and in the atmospheric environment. The semiconductor analyzer B1500 is used to cyclically test the device samples to obtain the I-V characteristics of the corresponding nodes. The input bias of each measurement cycle is scanned from 1 V to -1 V, and then returned from -1 V to 1 V. The pulse width of the bias voltage is about 124 ms. The I-V characteristics of nodes 12, 16, and 29 measured show obvious non-linear characteristics although the dynamic responses of each node are different. For example, in the positive and negative voltage ranges, the amplitude and growth trend of the current response are not completely symmetric, and the phenomenon of voltage pulses appears.

[0060] Embodiment 2:

[0061] The present invention discloses a method for constructing a reservoir based on the carbon nanotube heterojunction device in Embodiment 1, including the following steps:

[0062] Select a node from the heterojunction node array of the carbon nanotube heterojunction device in the above Embodiment 1, and collect the output signal of the node. In an optional implementation manner of the present invention, the selected node is a strong non-linear node. The specific judgment method of the strong non-linear node refers to the description in Embodiment 1. To avoid being cumbersome, it will not be elaborated here.

[0063] Next, a mask matrix is generated, and the collected output signal is subjected to mask expansion. In an alternative embodiment of the present invention, the mask matrix is a matrix randomly generated in a uniform distribution between [-1, 1] or composed of elements 1 and -1, where the number of 1s and -1s in each row of the mask matrix is a preset ratio, and the mask matrix is randomly generated in a hypergeometric distribution manner. For example, if the mask matrix size is set to [10, 8], composed of elements 1 and -1, and the positive-negative ratio is 4:1, then the mask matrix is as follows:

[0064]

[0065] The above is only one embodiment of generating the mask matrix in the present invention, and the idea of the present invention is not limited thereto, and other methods can also be selected to generate the mask matrix.

[0066] Furthermore, for the mask expansion of the collected output signal, first, the output signal of each time step is subjected to mask expansion, which specifically includes multiplying the output signal by each column of the mask matrix one by one and splicing them column by column to obtain a mask expansion matrix; secondly, the mask expansion matrix is reorganized according to the time step, mapping the output signal to a high-dimensional state space, and constructing a reservoir state matrix to complete the construction of the reservoir. Specifically, the signals expanded from the output signals of the same time step in the mask expansion are arranged column by column, and then arranged and spliced in the order of time steps. And each row in the mask matrix represents an independent mask sequence, and the dimension of the mask matrix matches the dimension of the high-dimensional state space. Furthermore, for a better understanding of the mask expansion process, a dataset with a sample number of I is used for illustration:

[0067] (1) The number of samples in the input dataset (including the training set or the test set) is I, and the step size of each sample is set to J, then the collected output signal is a sequence S with a length of I×J.

[0068] (2) Let the mask matrix size be [N, ML], where N is the number of expanded virtual neurons and ML is the number of virtual neuron states.

[0069] (3) Multiply each time step of the sequence S by each column of the mask matrix [N, ML] one by one and splice them column by column to obtain a mask expansion matrix E with a size of [N, ML×I×J].

[0070] (4) Reorganize the mask expansion matrix, extract the elements of each row of E at intervals of ML and splice them column by column to obtain sequences Pi (i = 1, 2, 3... ML), splice Pi row by row to obtain an extended state matrix Qj (j = 1, 2, 3... N), and the size of Qj is [ML, I×J].

[0071] (5) Concatenate Qj row by row to obtain a complete extended reservoir state matrix R with a size of [N×ML, I×J]. In Embodiment 2 of the present invention, the single-node state collected is extended, making full use of the dynamic characteristics of the strongly nonlinear carbon nanotube heterojunction single node, mapping the input signal into a high-dimensional state space for implementing complex computational tasks, such as Figure 4 The flowchart of constructing a reservoir in Embodiment 2 of the present invention is shown as follows.

[0072] In summary, through mask expansion, a high-dimensional non-linear mapping of the single-node state is performed to generate a high-dimensional reservoir state matrix to complete the construction of the reservoir; through operations such as concatenation and permutation recombination, the timing state of the original signal is maintained, enhancing the computational ability and dynamic characteristics of the system.

[0073] Embodiment 3:

[0074] The present invention discloses a method for processing data by a reservoir constructed based on carbon nanotube heterojunction devices, including the following steps:

[0075] First, select a node in the carbon nanotube heterojunction node array on the reservoir constructed based on the above carbon nanotube heterojunction devices; then set the parameters for reservoir construction and load the data set, where the data set includes a training set and a test set; further, input the training set to the selected node and collect the voltage signal of the selected node; perform mask expansion on the signal at each time step of the voltage signal to map the collected voltage signal from a low-dimensional space to a high-dimensional space, completing the construction of the reservoir.

[0076] Second, input the reservoir state constructed by the training set into the readout layer and train the readout layer; and map the test set to a high-dimensional space through reservoir construction, and input the reservoir state of the test set into the trained readout layer to obtain the target data result. The types of data processed by the reservoir constructed by the present invention include: time series data prediction, classification and recognition of voice data, and motion state control, etc.

[0077] For example, when using the reservoir constructed by the present invention to predict the trend of target data, the target data here refers to the prediction result of the target data; when the data set processed is voice sequence data, the target data here is the classification or recognition of the voice sequence data; when the data set processed is motion state sequence data, the target data is the future dynamic control data of the dynamic object. According to the type of the data set processed, the target data here is also different. The above are only some examples of the present invention and do not limit the present invention. Thus, according to Embodiment 3 of the present invention, by collecting the electrical signals of a single node in the carbon nanotube heterojunction node array and constructing a reservoir in the form of mask expansion, the processing of time series data is realized, including data prediction, voice classification and recognition, dynamic object control, etc.

[0078] In addition, to increase the convenience of actual operation, change the parameters for reservoir construction more conveniently, and view the results of reservoir calculation more intuitively, during the actual construction of the reservoir and the execution of tasks, the reservoir construction interface is implemented on the MATLAB platform of the host computer, including a parameter configuration module, a result display and saving module; when constructing the reservoir, various parameters for reservoir construction are set in the interface, and static two-dimensional diagrams, dynamic three-dimensional diagrams of the reservoir state, and calculation results of reservoir evaluation indicators MC (Memory Capacity), KR (Kernel Quality), and GR (Generalization Rank), as well as task calculation results are displayed in the interface. For the specific reservoir construction interface, the connection relationship between the carbon nanotube node array and the reservoir, refer to Figure 5 as shown.

[0079] Furthermore, to better explain the core idea of the present invention, in the present invention, the data of node 16 in sample 2 collected is taken as an example to illustrate Embodiment 3 of the present invention. Those skilled in the art should understand that the following examples are only for explaining the process of data processing implemented by the present invention and are not limitations of the present invention. First, select a node on the PCB signal transfer board of the carbon nanotube heterojunction node array with good wire bonding, for example, select Figure 3 node 16 in Figure 6 , then set the parameters for reservoir construction in the host computer and load the digital voice data set. In the present invention, the digital voice data set is taken as an example for illustration. Here, the data set can also be other types of data sets. For the processing of voice data, in addition to the methods disclosed in Embodiment 3 above, cross-validation of the voice data set is also included in the voice data processing process. Figure 6 Taking the ten-fold cross-validation of the voice data set as an example for illustration, here, choosing the ten-fold cross-validation is only one implementation method and is not limited thereto. The specific process of the ten-fold cross-validation can refer to

[0080] Specifically, the training set data is input into the node and the node voltage signal, i.e., the node state, is collected. The node voltage signal is transmitted to the host computer. After the signal acquisition is completed, the construction of the reservoir begins on the MATLAB platform of the host computer. The specific process of reservoir construction refers to the descriptions in Embodiment 2 and Embodiment 3 above and will not be elaborated here. After mask expansion, the reservoir state is input into the readout layer, and the readout layer is trained and optimized until the training is completed. The training algorithms for the readout layer here include: ridge regression algorithm, linear regression algorithm, etc. After the training of the readout layer is completed, the test set data is input into the readout layer. After the acquisition is also completed, it is transmitted to the host computer for mask expansion processing, and the classification result is calculated, such as the classification and recognition result of voice data. When the ten-fold cross-validation is completed, the average recognition rate of the voice data is further calculated and the confusion matrix is plotted.

[0081] Specifically, the confusion matrix of the digital voice classification results after ten-fold cross-validation is referred to as Figure 7 shown. The experimental results show that the classification recognition rate of the model of the reservoir constructed based on the single node of the carbon nanotube heterojunction in the TI 46-Word dataset of the present invention reaches 94%. This high recognition rate fully shows that the single-node reservoir based on the carbon nanotube heterojunction has significant advantages in processing voice data, and its strong non-linear characteristics and dynamic response ability can effectively improve the performance of the voice classification task.

[0082] Furthermore, in order to verify the excellent effect of the reservoir constructed in the present invention in time series prediction, in the present invention, the NARMA10 time series dataset is taken as an example for verification, and the normalized mean square error (NMSE) of the NARMA10 time series prediction is calculated. The line chart of the prediction result of the NARMA10 time series and the NMSE of the prediction result are referred to as Figure 8 shown. The experimental results show that the NMSE of the model of the reservoir constructed based on the single node of the carbon nanotube heterojunction in the present invention in the NARMA10 time series prediction task reaches 0.057732. This low error fully shows that the single-node reservoir based on the carbon nanotube heterojunction has significant advantages in processing time series signals, and its strong non-linear characteristics and dynamic response ability can effectively improve the performance of the time series prediction task.

[0083] In summary, in the alternative embodiment of the present invention, the carbon nanotube heterojunction node array has a 32-node architecture. Different nodes can be selected according to task requirements, which improves the flexibility of selection. And the node redundancy provides a large fault tolerance for sample production. Even if individual nodes fail, it does not affect the effectiveness of the overall reservoir computing. The multi-node design of the carbon nanotube heterojunction node array also facilitates the expansion of the number of nodes, supports a larger-scale reservoir state space, and meets the requirements of high-complexity tasks.

[0084] Those skilled in the art can easily understand that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for preparing a carbon nanotube heterojunction device, characterized in that: include: S1: Preparing a pre-patterned Au / Ni thin film metal electrode on a silicon wafer substrate containing silicon dioxide by sputtering and lift-off process, wherein the electrode spacing is 1.5 to 2 μm and the electrode width to spacing ratio is 15 to 30; and S2: Carbon nanotubes are assembled between the electrodes using dielectrophoresis technology, wherein a heterojunction node array is formed between the carbon nanotubes and the electrodes, and the silicon wafer is connected to an external adapter board through wire bonding technology, wherein the external adapter board is used to input and collect electrical signals for the heterojunction node array.

2. The method for preparing a carbon nanotube heterojunction device according to claim 1, characterized in that: The original semiconductor carbon nanotubes use toluene as solvent and PCz as dispersant, and the step S2 also includes: Extracting the original semiconductor carbon nanotube solution and drying it, taking the residual solid and mixing it with ethanol and ultrasonicating it to obtain a carbon nanotube solution; The phosphomolybdic acid particles are mixed with ethanol and subjected to ultrasonic treatment to prepare a phosphomolybdic acid solution of a preset concentration, wherein the preset concentration of the phosphomolybdic acid solution is 250-400 μg / ml; The carbon nanotube solution is mixed with the phosphomolybdic acid solution and subjected to ultrasonic treatment to obtain a phosphomolybdic acid-modified carbon nanotube solution; and The probes are placed at both ends of the Au / Ni thin film electrode respectively, the phosphomolybdic acid modified carbon nanotube solution is dripped, and a signal generator generates a sinusoidal AC voltage signal to perform the dielectrophoresis operation on the carbon nanotubes through the probes.

3. A method for constructing a storage pool based on a carbon nanotube heterojunction device, characterized in that: include: Selecting a node in the heterojunction node array of the carbon nanotube heterojunction device prepared by the method of claims 1-2, and collecting the output signal of the node; as well as Generating a mask matrix and performing mask expansion on the collected output signal, including: Performing mask expansion on the output signal of each time step, specifically comprising multiplying the output signal with each column of the mask matrix one by one, and splicing by column to obtain a mask expansion matrix; The mask expansion matrix is ​​reorganized according to the time step, the output signal is mapped to a high-dimensional state space, a reserve pool state matrix is ​​constructed, and the construction of the reserve pool is completed.

4. The method for constructing a storage pool based on a carbon nanotube heterojunction device according to claim 3, characterized in that: Selecting the node also includes: analyzing the volt-ampere characteristics of multiple nodes, selecting a strong nonlinear node from the multiple nodes as a collection node, when the nonlinear index is 0<λ≤0.25, the node is a strong nonlinear node, when 0.25<λ<0.5, the node is a weak nonlinear node, and the value range of λ is 0-0.

5.

5. The method for constructing a storage pool based on a carbon nanotube heterojunction device according to claim 3, characterized in that: The mask matrix is ​​a matrix randomly generated in a uniform distribution between [-1, 1] or the mask matrix consists of 1 and -1, wherein the number of 1 and -1 in each row of the mask matrix is ​​a preset ratio, and the mask matrix is ​​randomly generated in a hypergeometric distribution.

6. The method for constructing a storage pool based on a carbon nanotube heterojunction device according to claim 3, characterized in that: Constructing the reserve pool state matrix includes: arranging the signals expanded from the output signals of the same time step in the mask expansion in columns, and then arranging and splicing them in the order of time steps.

7. The method for constructing a storage pool based on a carbon nanotube heterojunction device according to claim 5, characterized in that: Each row in the mask matrix represents an independent mask sequence, and the dimension of the mask matrix matches the dimension of the high-dimensional state space.

8. A method for processing data in a storage pool constructed based on a carbon nanotube heterojunction device, characterized in that: include: Selecting a node in the carbon nanotube heterojunction node array on the reservoir constructed based on the carbon nanotube heterojunction device in claims 3-7; Set the parameters for building the reserve pool and load the data set, where the data set includes a training set and a test set; Inputting the training set to a selected node and collecting a voltage signal of the selected node; Performing mask expansion on the voltage signal at each time step, mapping the collected voltage signal from a low-dimensional space to a high-dimensional space, and completing the construction of the reserve pool; Inputting the reserve pool state constructed by the training set into the readout layer, and training the readout layer; as well as The test set is mapped to a high-dimensional space through a reserve pool, and the reserve pool state of the test set is input into the trained readout layer to obtain target data.

9. The method for processing data in a storage pool based on a carbon nanotube heterojunction device according to claim 8, characterized in that: The data set is one of time series data, speech sequence data and motion state sequence data. When the data set is speech sequence data, it also includes obtaining the target data in a cross-validation manner, and the target data is a speech classification result.

10. The method for processing data based on a reserve pool constructed based on a carbon nanotube heterojunction device according to claim 8, wherein the interface of the reserve pool constructed based on the carbon nanotube heterojunction device is built on a MATLAB platform of a host computer, wherein the interface comprises a parameter configuration module, a result display module and a saving module, wherein the interface is used to set the parameters for constructing the reserve pool, display the reserve pool status, the reserve pool evaluation index and the results of the target data.

Images