Reservoir computing system based on silver nanowire network and preparation method and application thereof

Abstract

The invention discloses a reserve pool computing system based on a silver nanowire network and a preparation method and application thereof, and belongs to the technical field of nanowires and neural networks. The reservoir computing system based on the silver nanowire network comprises a bottom multi-electrode array, wherein electrodes can serve as input ends or output ends; and the storage pool is a silver nanowire network, is prepared on the bottom multi-electrode array by instilling and spin-coating a silver nanowire solution, and is used for projecting input data to a high-dimensional nonlinear space. The computing system can be applied to data processing, a parallel input method of data can be combined with application methods such as virtual nodes and amplitude information at the same time, input feature information is greatly increased, the requirements for the complexity and the memory capacity of the reserve pool are reduced, and the feature extraction capacity of the reserve pool computing system is enhanced.

Description

Technical Field

[0001] The present invention belongs to the technical field of nanowires and neural networks, and in particular relates to a storage pool computing system based on a silver nanowire network and a preparation method and application thereof. Background Art

[0002] With the development of machine learning and artificial intelligence (AI), edge AI computing applications are gradually increasing. How to use limited computing resources and computing power to achieve richer AI functions has become a key challenge in the development of science and technology. Recurrent neural networks (RNNs) use hidden layers to share weights in the time dimension to obtain short-term memory capabilities, thereby adapting to ubiquitous time series data tasks, but their hardware implementation is still relatively difficult.

[0003] Neuromorphic computing aims to create energy-efficient computing systems by simulating the working mechanism of the brain. Studies have shown that the neuronal activity recorded in the brain can be decoded into behavior through linear readout. In combination with this biological mechanism, the hidden layer in the RNN is implemented through a physical system, and the network can be trained only by training the output layer weights. This is reservoir computing. Reservoir computing uses the laws of physics to build loop links within the system to achieve a lightweight architecture. It has the advantages of less training, fast speed, and low energy consumption. It is considered to be an ideal choice for adapting to a variety of edge computing scenarios.

[0004] At present, the reservoir computing system based on resistive random access memory (RRAM) has made great progress, but it is still limited by the challenges of device performance uniformity and stability. The nanowire network uses its self-organizing random structure to achieve time-domain changes in network resistance, better simulating the working state of brain neurons. In recent reports, the research on the reservoir computing system based on nanowire networks focuses more on the molecular dynamics within the network and application expansion, but how to better utilize the physical structure of the nanowire network reservoir, expand the network memory capacity, and enhance the network feature extraction capability still needs to be explored and studied. Summary of the invention

[0005] In view of the deficiencies in the prior art, the purpose of the present invention is to provide a storage pool computing system based on a silver nanowire network and a preparation method and application thereof, thereby solving the problems in the prior art.

[0006] The purpose of the present invention can be achieved through the following technical solutions:

[0007] The reservoir computing system based on the silver nanowire network includes:

[0008] Bottom multi-electrode array: the electrodes can be used as input or output;

[0009] Reservoir: It is a silver nanowire network, which is prepared by dripping and spin coating a silver nanowire solution on the bottom multi-electrode array, and is used to project input data into a high-dimensional nonlinear space.

[0010] Furthermore, the bottom multi-electrode array is made of Ag, the array size is n×n, and the shortest distance between electrodes is 200 μm.

[0011] Furthermore, the diameter of the silver nanowires is 120 nm and the average length is 45 μm.

[0012] Furthermore, the surface of the silver nanowires is covered with a 2-5 nm thick PVP shell layer.

[0013] The method for preparing the silver nanowire network-based reservoir computing system is characterized by comprising the following steps:

[0014] A silver multi-electrode array is fabricated on a silicon substrate using a photolithography machine;

[0015] Silver nanowires are dispersed in a solvent, and the silver nanowire solution is drop-casted and spin-coated on a silicon substrate with a bottom multi-electrode array to obtain a reservoir.

[0016] Furthermore, the specific steps of dripping and spin coating include:

[0017] Step 1, placing a silicon substrate on a temperature-controllable rotating table;

[0018] Step 2, drop the silver nanowire solution onto the bottom multi-electrode array;

[0019] Step 3, controlling the rotation of the rotating table to improve the distribution state of the silver nanowire solution;

[0020] Step 4: After the diffusion is completed, the temperature of the rotating table is increased, the solvent evaporates completely, and the drop casting is completed;

[0021] Step 5, repeating steps 2 to 4, and finally controlling the nanowire density to meet the required requirements.

[0022] Furthermore, in step 3, the rotation speed of the rotating table is 300 rpm.

[0023] Furthermore, in step 4, the temperature of the rotating table is raised to 60°C.

[0024] The application of the above-mentioned silver nanowire network-based reservoir computing system in data processing.

[0025] The applications include:

[0026] Asynchronous parallelism: The input data is split into several segments by row or column, and then input into several electrodes of the computing system in sequence. After each segment of data is input, the next segment of data is input. During the entire input process, except for the input electrode, the remaining electrodes act as output terminals to read out data at the same time, thereby realizing multiple network states using a single storage pool.

[0027] And, synchronous parallelism: the input data is split into several segments by row or column, and input into several electrodes of the computing system at the same time, and another electrode is selected as the output end to read out the data at the same time, so as to use the computing system to synchronously batch process multiple data features.

[0028] Beneficial effects of the present invention:

[0029] 1. The present invention uses a multi-electrode array as the bottom electrode to construct a reservoir computing system based on a silver nanowire network with multi-input and multi-output ports, and proposes an application method for synchronous or asynchronous parallel input. The reservoir is composed of a self-organized network of silver nanowires drop-cast on a silver bottom electrode array. When powered on, silver ions form conductive filaments in the PVP shell at the intersection of the nanowires, showing the resistive memory characteristics required for reservoir computing. Using the multi-input and multi-output structure of the system, graphic data can be processed in terms of rows, columns, colors, grayscale, etc., and input into the nanowire network reservoir in parallel.

[0030] 2. If the input is asynchronous, a single pool can be used to achieve multiple network states; if the input is synchronous, the network can batch process multiple data features at the same time. The parallel input method of data can also be combined with application methods such as virtual nodes and amplitude information to greatly increase the input feature information, reduce the demand for pool complexity and memory capacity, and enhance the feature extraction capability of the pool computing system. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0032] Figure 1 is a flow chart of the preparation of the storage pool computing system of the silver nanowire network of the present invention;

[0033] Figure 2 is a schematic diagram of the bottom multi-electrode array structure of the present invention;

[0034] Figure 3 is a graph of the measurement results of the computing system of the present invention;

[0035] Figure 4It is an application flow chart of the computing system of the present invention. DETAILED DESCRIPTION

[0036] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0037] Example 1

[0038] A reservoir computing system based on a silver nanowire network, comprising: a bottom multi-electrode array and a reservoir;

[0039] A bottom multi-electrode array, where the electrodes can function as either input or output;

[0040] Reservoir: It is a silver nanowire network, which is prepared by dripping and spin coating a silver nanowire solution on the bottom multi-electrode array, and is used to project input data into a high-dimensional nonlinear space.

[0041] The bottom multi-electrode array includes Figure 2 There are nine types (a)-(i) in the figure, which can be divided into internal electrodes and external electrodes. The internal and external electrodes use different network parameters, and the two are connected by wires. The internal electrodes are all circular, (a)-(g) are arranged in a square, among which the array scale of (a)-(c) is 8×8, and the array scale of (d)-(g) is 4×4; (h) and (i) are arranged in a "U" shape, both with 16 electrodes. The electrode diameters used are 60μm for (a)-(d) and (f), and 20μm for (e) and (g)-(i). The spacing between the centers of adjacent electrodes is: (a) is 400μm, (b) is 300μm, (c), (d) and (f) are 200μm, (e), (g) and (h) are 100μm, and (i) is 50μm. The external electrode arrays are all square, distributed around the periphery, and used to lead out external circuits. Among them, the side length of (a)-(c) is 200μm, the side length of (d) and (e) is 800μm, and the side length of (f)-(i) is 400μm.

[0042] The bottom multi-electrode array is made of Ag.

[0043] The diameter of the silver nanowires is 120 nm and the average length is 45 μm.

[0044] The surface of the silver nanowires is covered with a 2-5 nm thick PVP shell.

[0045] Example 2

[0046] In this embodiment, a method for preparing a storage pool computing system based on a silver nanowire network is disclosed, such as Figure 1 As shown in (a)-(f), the following steps are included:

[0047] S1, draw the layout of the bottom multi-electrode array (such as Figure 2 As shown), and based on the layout, a silver multi-electrode array is manufactured on a silicon substrate using a photolithography machine;

[0048] S2, dispersing silver nanowires in a solvent (isopropanol), drop-casting and spin-coating the silver nanowire solution on a silicon substrate with a bottom multi-electrode array to obtain a reservoir, thus completing the preparation of a reservoir computing system based on a silver nanowire network;

[0049] Among them, the specific steps of dripping and spin coating include:

[0050] S21, placing the silicon substrate on a temperature-controllable rotating table, which is preset to 30° C.;

[0051] S22, drop the silver nanowire solution onto the bottom multi-electrode array and slowly spread it by itself;

[0052] S23, controlling the rotation of the rotating table to improve the distribution state of the silver nanowire solution, the rotation speed of the rotating table is 300 rpm;

[0053] S24, after the diffusion is completed, the temperature of the rotating table is increased to 60°C to accelerate the solvent evaporation process and complete a drop casting;

[0054] S25, repeat S22-S24, and finally control the density of the nanowires to meet the required requirements.

[0055] The storage pool computing system based on the silver nanowire network prepared in this embodiment was measured, and its attenuation characteristic curve and its corresponding function fitting diagram are shown in FIG. Figure 3 As shown in (a), it can be seen that the curve is a current decay curve, and the exponential fitting degree is good. Figure 3 (b) shows the cumulative effect of the decay curve, which proves that the resulting silver nanowire network-based reservoir computing system has a short-term memory capability for historical inputs. Figure 3 (c) in the figure shows the differences between the attenuation curves of different input and output ends, which can verify the state richness of the reservoir computing system based on the silver nanowire network, increase the resolution of the system, and reduce the probability of overfitting of the readout layer. Figure 3 (d) and (e) in the figure show the relationship between the attenuation curve and the input pulse, and the input pulse related parameters in the subsequent parallel application method can be set based on this relationship. Figure 3 (f) in the figure shows the attenuation curve of the 4-bit sequence pulse input. Figure 3As can be seen from (g), the resulting silver nanowire network-based reservoir computing system has good resolution for the time series data.

[0056] Example 3

[0057] In this embodiment, the application of a storage pool computing system based on a silver nanowire network in data processing is introduced; Figure 4 As shown, the applications include asynchronous parallel and synchronous parallel;

[0058] 1) Asynchronous parallelism

[0059] like Figure 4 As shown in (a) in the figure, taking 5×5 input data as an example, the input data is divided into 5 segments by row or column, and input into the five electrodes of the computing system in sequence. After each segment of data is input, the next segment of data is input. During the entire input process, except for the input electrode, the remaining electrodes read out data at the same time as the output end, so as to realize multiple network states using a single reservoir. Peripheral gating circuits can be designed for simple control of both data input and output.

[0060] 2) Synchronous parallel

[0061] like Figure 4 As shown in (b), taking 5×5 input data as an example, the input data is split into 5 segments by row or column, and input into the five electrodes of the computing system at the same time, and another electrode is selected as the output end to read out the data at the same time, so that the computing system can synchronously batch process multiple data features.

[0062] The input data includes but is not limited to characteristic information such as rows, columns, colors, and grayscale of the graphic data.

[0063] The parallel data input method described in the embodiment of the present invention can also be combined with application methods such as virtual nodes and amplitude information, which greatly increases the input feature information, reduces the demand for reserve pool complexity and memory capacity, and enhances the feature extraction capability of the reserve pool computing system.

[0064] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0065] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements all fall within the scope of the present invention to be protected.

Claims

1. A storage pool computing system based on a silver nanowire network, characterized in that: include: Bottom multi-electrode array: the electrodes can be used as input or output; Reservoir: It is a silver nanowire network, which is prepared by dripping and spin coating a silver nanowire solution on the bottom multi-electrode array, and is used to project input data into a high-dimensional nonlinear space.

2. The silver nanowire network-based reservoir computing system according to claim 1, characterized in that: The bottom multi-electrode array is made of Ag, the array scale is n×n, and the shortest distance between electrodes is 200 μm.

3. The silver nanowire network-based reservoir computing system according to claim 1, characterized in that: The diameter of the silver nanowires is 120 nm and the average length is 45 μm.

4. The silver nanowire network-based reservoir computing system according to claim 1, characterized in that: The surface of the silver nanowire is covered with a 2-5 nm thick PVP shell layer.

5. The method for preparing a storage pool computing system based on a silver nanowire network according to any one of claims 1 to 4, characterized in that: The following steps are involved: A silver multi-electrode array is fabricated on a silicon substrate using a photolithography machine; Silver nanowires are dispersed in a solvent, and the silver nanowire solution is drop-casted and spin-coated on a silicon substrate with a bottom multi-electrode array to obtain a reservoir.

6. The method for preparing a storage pool computing system based on a silver nanowire network according to claim 5, characterized in that: The specific steps of dripping and spin coating include: Step 1, placing a silicon substrate on a temperature-controllable rotating table; Step 2, drop the silver nanowire solution onto the bottom multi-electrode array; Step 3, controlling the rotation of the rotating table to improve the distribution state of the silver nanowire solution; Step 4: After the diffusion is completed, the temperature of the rotating table is increased, the solvent evaporates completely, and the drop casting is completed; Step 5, repeating steps 2 to 4, and finally controlling the nanowire density to meet the required requirements.

7. The method for preparing a storage pool computing system based on a silver nanowire network according to claim 6, characterized in that: In step 3, the rotation speed of the rotating stage is 300 rpm.

8. The method for preparing a storage pool computing system based on a silver nanowire network according to claim 5, characterized in that: In step 4, the temperature of the rotating stage is raised to 60°C.

9. Application of the silver nanowire network-based reservoir computing system according to any one of claims 1 to 4 in data processing.

10. The use according to claim 9, characterized in that: The applications include: Asynchronous parallelism: The input data is split into several segments by row or column, and then input into several electrodes of the computing system in sequence. After each segment of data is input, the next segment of data is input. During the entire input process, except for the input electrode, the remaining electrodes act as output terminals to read out data at the same time, thereby realizing multiple network states using a single storage pool. And, synchronous parallelism: the input data is split into several segments by row or column, and input into several electrodes of the computing system at the same time, and another electrode is selected as the output end to read out the data at the same time, so as to use the computing system to synchronously batch process multiple data features.

Images