A hybrid topology micro-nano material physical reserve pool and a corresponding computing system

Abstract

The application belongs to the field of new computing hardware, and discloses a micro-nano material physical reserve pool with a hybrid topology structure and a corresponding computing system, wherein the micro-nano material physical reserve pool with the hybrid topology structure comprises a connecting plate, and a first micro-nano material network physical reserve pool and a second micro-nano material network physical reserve pool located on the front and back surfaces of the connecting plate respectively; the two micro-nano material network physical reserve pools are independently selected from ESN and LSM; the nodes of the two networks are stacked or connected in parallel. The first and second micro-nano material network physical reserve pools are respectively constructed on the front and back surfaces of the same connecting plate to form the physical reserve pool with the hybrid topology structure, so that the nonlinear mapping capability and parallel processing capability of the system can be improved, the complexity of training can be reduced, and the technical problems of low computing performance and limited scale of the existing carbon nanotube physical reserve pool can be solved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of new computing hardware, and more particularly, relates to a micro-nano material physical reservoir pool of hybrid topology and a corresponding computing system. BACKGROUND

[0002] Reservoir computing is a computing architecture derived from recurrent neural networks, which has the advantages of low training cost and fast learning. The internal connection weights of the reservoir are randomly generated and do not need to be adjusted, so they can be replaced by physical systems constructed by new materials and devices with nonlinear dynamics, such as micro-nano material-based network physical reservoir pool.

[0003] Among micro-nano materials, carbon nanotubes (CNT) are a new type of nanomaterial with unique structure, having outstanding advantages such as ultra-small size, high mobility, high saturation velocity, and long mean free path. Carbon nanotube-based reservoir pool system is also one of the research hotspots of physical reservoir pool.

[0004] Among them, the single-walled carbon nanotube (SWCNT) network physical reservoir pool of 2D structure is widely studied. In 2016-2017, Matthew Dale's team used SWCNT and polymers to construct a reservoir pool, using two polymers, polymethyl methacrylate (PMMA) and polybutyl methacrylate (PBMA), and using an evolutionary algorithm to pre-train the SWCNT / polymer device to configure it as a functional RC system. The device has completed tasks such as nonlinear sequence fitting, waveform generation, and MC calculation, outperforming silver nanowire network reservoir pool and demonstrating the advantages of carbon nanotubes as reservoir pool computing base materials; however, due to the use of conductive wire network dynamics by the SWCNT / polymer reservoir pool, the nonlinear response to input signals is insufficient, and the internal nonlinear dynamics of the network are not rich enough, resulting in general computing performance of the reservoir pool as a whole.

[0005] In 2022, Megumi Akai-Kasaya's team built a POM / SWCNT random reservoir pool network by using phosphomolybdic acid (POM) to adsorb carbon nanotubes and formed a composite structure, and implemented basic reservoir pool computing such as periodic sequence prediction tasks, NARMA2, and MC calculation. Among them, the functionalization of SWCNT by POM makes the composite structure have more complex and rich nonlinear electrical properties.

[0006] Due to the low interconnection probability of carbon nanotubes in the 2D structure SWCNT network physical reservoir pool, the 2D reservoir pool has the problems of limited scale and limited complexity of connection between nodes, resulting in limited nonlinear degree of the whole 2D reservoir pool and low overall computing performance of the reservoir pool.

[0007] Other micro-nano material constructed physical reserve pool, similar, also exist similar shortcomings. SUMMARY

[0008] In view of the above defects or improvement needs of the prior art, the purpose of the present application is to provide a micro-nano material physical reserve pool with a hybrid topology and a corresponding computing system, which can improve the nonlinear mapping capability and parallel processing capability of the system, and reduce the complexity of training, by constructing a first micro-nano material network physical reserve pool (such as a first carbon nanotube network physical reserve pool) and a second micro-nano material network physical reserve pool (such as a second carbon nanotube network physical reserve pool) on the front and back surfaces of the same connection board (such as a PCB board) to form a physical reserve pool with a hybrid topology. Taking a carbon nanotube network with carbon nanotubes as the micro-nano material as an example, based on the present application, two carbon nanotube networks (such as two 2D carbon nanotube networks) are assembled to form a physical reserve pool with a hybrid topology, and the resulting hybrid topology can solve the technical problems of low computing performance and limited scale of existing carbon nanotube physical reserve pools.

[0009] To achieve the above-mentioned purpose, according to one aspect of the present application, a micro-nano material physical reserve pool with a hybrid topology is provided, comprising a connection board, a first micro-nano material network physical reserve pool located on the front surface of the connection board, and a second micro-nano material network physical reserve pool located on the back surface of the connection board; wherein the first micro-nano material network physical reserve pool and the second micro-nano material network physical reserve pool are independently selected from: an echo state network (ESN) based micro-nano material network physical reserve pool, a liquid state machine (LSM) based micro-nano material network physical reserve pool;

[0010] The input nodes of the first micro-nano material network physical reserve pool are connected in parallel with the input nodes of the second micro-nano material network physical reserve pool, and the output nodes of the first micro-nano material network physical reserve pool are connected in parallel with the output nodes of the second micro-nano material network physical reserve pool; or the input nodes of the first micro-nano material network physical reserve pool are one-to-one short-circuited with the input nodes of the second micro-nano material network physical reserve pool, and the output nodes of the first micro-nano material network physical reserve pool are one-to-one short-circuited with the output nodes of the second micro-nano material network physical reserve pool.

[0011] As a further preferred embodiment of the present application, the micro-nano material is carbon nanotubes, and the first micro-nano material network physical reserve pool and the second micro-nano material network physical reserve pool are both carbon nanotube network physical reserve pools.

[0012] As a further preferred embodiment of the present application, the first micro-nano material network physical reserve pool and the second micro-nano material network physical reserve pool are both 2D carbon nanotube network physical reserve pools.

[0013] As a further preferred embodiment of the present application, the reservoir of the first micro-nano material network physical reservoir pool is prepared directly on the connection board, or is attached to the connection board after being prepared on a silicon substrate;

[0014] The reservoir of the second micro-nano material network physical reservoir pool is prepared directly on the connection board, or is attached to the connection board after being prepared on a silicon substrate.

[0015] As a further preferred embodiment of the present application, the connection board is a PCB board.

[0016] As a further preferred embodiment of the present application, the carbon nanotube is a semiconducting single-walled carbon nanotube.

[0017] As a further preferred embodiment of the present application, the echo state network (ESN) based micro-nano material network physical reservoir pool is an echo state network (ESN) based carbon nanotube network physical reservoir pool, specifically a POM / SWCNT / PBMA thin film device formed by components including phosphomolybdic acid, single-walled carbon nanotubes and polybutyl methacrylate.

[0018] The liquid state machine (LSM) based micro-nano material network physical reservoir pool is a liquid state machine (LSM) based carbon nanotube network physical reservoir pool, specifically a POM / SWCNT pulse device formed by components including phosphomolybdic acid and single-walled carbon nanotubes.

[0019] As a further preferred embodiment of the present application, one of the first micro-nano material network physical reservoir pool and the second micro-nano material network physical reservoir pool is an echo state network (ESN) based micro-nano material network physical reservoir pool, and the other is a liquid state machine (LSM) based micro-nano material network physical reservoir pool.

[0020] According to another aspect of the present application, the present application provides a computing system comprising the above-mentioned hybrid topology micro-nano material physical reservoir pool.

[0021] Compared with the prior art, the physical reservoir pool based on the hybrid topology of the present application stacks or connects the nodes of two networks to realize a composite node or a large-scale reservoir pool. At present, research on reservoir computing mainly focuses on theory and models, and device implementation of physical reservoirs also focuses on implementation and research of single networks. Based on implementation of single ESN and LSM network devices, the present application considers the high demand of computing performance of the reservoir pool for system nonlinearity, and enhances the nonlinear characteristics of the system and the computing performance of the reservoir pool by mixing multiple network devices in topology.

[0022] (a) Taking the ESN network and the mixed ESN network as examples, the capacity of the model can be expanded to enable it to capture and represent more complex dynamic relationships and features; in some complex prediction tasks, a single ESN may be difficult to achieve the desired accuracy, and by combining multiple ESNs, the prediction accuracy of complex systems can be improved by taking advantage of their synergy, and various application scenarios can also be more flexibly adapted.

[0023] Unlike simply increasing the number of nodes in the same ESN network, although increasing the number of nodes can improve the fitting ability of the network and better capture complex dynamic patterns in the data, the disadvantages brought by too many nodes cannot be ignored. For example, as the scale of the nodes expands, the computational complexity and storage requirements will also increase, resulting in low computational efficiency; too many nodes can make the network too complex and easily cause overfitting problems, reducing the generalization ability; increasing the number of nodes will increase the difficulty of adjusting parameters such as connection weights, output weights, and spectral radius. In addition, simply relying on increasing the number of nodes to improve the computational capacity of the physical reserve pool does not meet the development needs of electronic process integration. The present application uses the mixing of ESN networks and ESN networks to improve computational efficiency and enhance the model's ability to capture complex dynamic patterns, while optimizing the network structure to reduce storage requirements and overfitting risks. Similar advantages also apply to the mixing of LSM networks and LSM networks, and the mixing of ESN networks and LSM networks.

[0024] (b) Mixing LSM networks and LSM networks further improves the parallel processing capability of the system, enabling it to handle a large amount of spatiotemporal information simultaneously; at the same time, it can also expand the recognition range and depth of complex spatiotemporal patterns, making it perform better in tasks such as processing complex biological signals and speech recognition; the combination of multiple LSMs can form a more robust network structure, and even if some LSMs are disturbed or damaged, the entire system can still maintain good performance.

[0025] (c) Especially when one of the first and second micro-nano material network physical reserve pools is an echo state network (ESN) based micro-nano material network physical reserve pool, and the other is a liquid state machine (LSM) based micro-nano material network physical reserve pool:

[0026] (1) The present application parallelly combines ESN network structure and LSM network structure, ESN has good short-term memory ability, LSM is suitable for long-term dependence and processing complex dynamics, mixing the two can form a comprehensive response to different features in time series, enhancing the time response ability of the system;

[0027] (2) The liquid layer of the LSM topologically simulates the processing mode of the biological neural network, has strong nonlinearity and noise robustness, and after mixing, can make up for the deficiency of the ESN in a high-noise environment, and enhance the nonlinearity of the system;

[0028] (3) By mixing the two network structures, more diverse dynamic states can be formed in the reservoir, so that the output layer can make more accurate prediction based on richer state information, and the micro-nano material physical reservoir computing performance is improved. BRIEF DESCRIPTION OF DRAWINGS

[0029] Figure 1 It is a schematic diagram of the ESN network assembled on one side of the PCB provided by the application.

[0030] Figure 2 It is a schematic diagram of the LSM network assembled on one side of the PCB provided by the application.

[0031] Figure 3 It is a schematic diagram of the ESN network and the LSM network provided by the application.

[0032] Figure 4 It is a schematic diagram of the ESN network and the LSM network provided by the application.

[0033] Figure 5 It is a real photo of a silicon-based POM / CNT / PBMA thin film device as a physical reservoir of the ESN network provided by the application.

[0034] Figure 6 It is a real photo of a silicon-based POM / SWCNT pulse device as a physical reservoir of the LSM network provided by the application.

[0035] Figure 7 It is a schematic diagram of the reservoir computing test system provided by the application.

[0036] Figure 8 It is a schematic diagram of the results of the NARMA10 time series prediction task using a single silicon-based ESN network device in Example 1.

[0037] Figure 9 It is a schematic diagram of the results of the NARMA10 time series prediction task using a single silicon-based LSM network device in Example 1.

[0038] Figure 10 It is a schematic diagram of the results of the NARMA10 time series prediction task using a silicon-based ESN network device and a LSM network device in parallel provided by the application in Example 1.

[0039] Figure 11is a result schematic diagram of using two silicon-based ESN network devices stacks to perform a NARMA10 time series prediction task according to Embodiment 2 of the present application. DETAILED DESCRIPTION

[0040] In order to make the objects, technical solutions and advantages of the present application clearer, the present application 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 application and should not be used to limit the present application. In addition, the technical features involved in each embodiment of the present application described below can be combined with each other as long as they do not conflict with each other.

[0041] The carbon nanotube network in the 2D structure is taken as an example, and the PCB board is taken as an example of the connecting plate. In the following embodiments, the preparation methods of the ESN network and the LSM network are as follows (the carbon nanotube used is a semiconductor single-walled carbon nanotube purchased on the market):

[0042] The topological structure of the physical reserve pool of the ESN network is as shown in Figure 1 The microelectrode array as the connecting node can be directly designed on the PCB board or prepared by a graphic process to obtain a metal thin film electrode on a silicon substrate, and then the silicon wafer is attached to the PCB board. The main process of the graphic process is as follows: a 10*10 microelectrode array pattern is formed on the surface of the silicon wafer by photolithography, a nickel electrode thin film with a thickness of 50 nm and a gold electrode thin film with a thickness of 350 nm are obtained by sputtering and vacuum annealing at 400℃ for 4 hours, a POM / SWCNT / PBMA anisole solution is prepared, the SWCNT concentration is 500 μg / ml, the mass ratio of SWCNT to PBMA is 1%, and the concentration of POM is in the range of 200-800 μg / ml, and the concentration used here is 500 μg / ml, and finally a POM / SWCNT / PBMA network thin film is generated on the surface of the microelectrode array by dripping and drying at 100℃, and finally the electrodes on the silicon wafer are connected to the corresponding PCB board by wire bonding to obtain a microelectrode array that can be used for testing.

[0043] The CNT in the POM / SWCNT / PBMA network thin film randomly connects and fixes the nodes between the electrode arrays, constitutes the first layer network topology of the physical reserve pool of the ESN network, has the characteristics of random initialization and fixed connection weight, and has the characteristics of non-repeatability between the thin films prepared by the dripping process, which improves the complexity of the network; the nodes in the thin film and the input and output nodes on the PCB board constitute the second layer topology.

[0044] The topological structure of the physical reserve pool of the LSM network is as shown in Figure 2As shown, similar to the ESN network, the microelectrode array on the silicon substrate or the PCB board can be selected as needed; the main process of preparing the device on the silicon wafer is as follows: a 16*2 microelectrode array pattern is formed on the surface of the silicon wafer by photolithography, wherein the electrode spacing for building CNT is 2 μm, a 35 nm-thick nickel electrode film and a 500 nm-thick gold electrode film are obtained by sputtering and vacuum annealing at 400 ℃ for 4 hours, a POM / SWCNT ethanol solution is prepared by ultrasonic, wherein the concentrations of POM and SWCNT are 300 μg / ml and 20 μg / ml, CNT is randomly assembled between the electrode arrays on the microelectrode array surface by a dielectrophoresis (DEP) process to form a passage, and finally the electrodes on the silicon wafer are connected to the corresponding PCB board by wire bonding to obtain a microelectrode array that can be used for testing.

[0045] In the electrophoresis, the number of CNTs is related to the DEP parameters such as signal voltage amplitude, frequency, solution concentration and assembly time, and in the present application, the signal voltage amplitude is set to 16 V pp and the frequency is 2 MHZ. A sparse connection random network is generated between the planar electrodes by DEP, which constitutes the network topology inside the physical reserve pool of the LSM network, and has high nonlinearity and complexity.

[0046] Embodiment 1

[0047] This embodiment is to assemble the ESN network on the front surface of the same PCB board and assemble the LSM network on the back surface, as shown in Figure 1 、 Figure 2 .

[0048] The corresponding device can work in a stacked manner or in a parallel manner, specifically:

[0049] Figure 3 is one of the mixed topologies provided by the present application for stacking the ESN network and the LSM network, the physical reserve pool of the ESN network with M nodes and the physical reserve pool of the LSM network with M nodes are assembled on the two surfaces of the PCB board, respectively, the input and output nodes of the two physical reserve pools are short-circuited, that is, the M input nodes simultaneously provide input signals to the corresponding network nodes on the two surfaces, and after being processed in the reserve pools, the output nodes simultaneously collect the M output states of the two networks; since the connection nodes in the LSM network are one-to-one corresponding and the connection nodes in the ESN network are randomly distributed, the responses collected by the same output node may come from two or more input nodes, the output layer can collect more rich state information, when a certain node is out of order, other nodes can still work, thereby improving the complexity and fault tolerance of the system.

[0050] Figure 4The application provides another hybrid topology for connecting an ESN network and an LSM network in parallel, an ESN network physical reserve pool with M nodes and an LSM network physical reserve pool with N nodes are respectively assembled on two surfaces of a PCB, (M+N) input nodes simultaneously provide input signals, (M+N) output states are obtained after processing in the reserve pools, the two reserve pools are connected in parallel to expand the scale of the state space, enhance the processing capacity of the system for complex input and the storage capacity of the system for time sequence information, and thus improve the computing performance of the reservoir computing.

[0051] The silicon-based devices of the ESN and LSM physical reserve pools are as shown in Figure 5 and Figure 6 The silicon chip is attached to the PCB, the electrodes of the silicon-based devices are connected with the electrodes of the PCB by using a wire bonding process, the hardware system including an industrial computer and an NI data acquisition card is connected with the upper computer software system, signal input, acquisition and reserve pool computing task testing are performed, and a complete physical reserve pool computing system is formed, as shown in Figure 7 The reserve layer / liquid layer is the SWNT network device. Based on the computing system, independent ESN network devices and LSM network devices are respectively used to perform a nonlinear auto regressive moving average (NARMA) time sequence prediction task. Since NARMA10 has a higher order than NARMA2 (NARMA10 is a ten-order nonlinear model, and the model is more complex because more historical information needs to be considered; NARMA2 is a two-order nonlinear model, and the model is relatively simple and has a low computing complexity because the order is low), the performance and memory depth of the system in processing a complex nonlinear dynamic system are better tested. The application uses the NARMA10 task to perform calculation when verifying the performance of the network device, and the results are as shown in Figure 8 、 Figure 9 The physical reserve pool network device provided by the application can better complete the time sequence prediction task, and the normalized mean square error (NMSE) is 0.027647 and 0.061098 respectively.

[0052] Figure 10In order to calculate the result of the NARMA10 task after the ESN network device and the LSM network device are connected in parallel, the NMSE is 0.016605, which is more accurate than the prediction result of the independent ESN / LSM network, and it can be seen that the calculation performance of the hybrid network obtained based on the application is better than that of a single network. The structure of the carbon nanotube network physical reserve pool of the hybrid topology structure of the application can further improve the internal complexity of the network, enhance the nonlinear characteristics and dynamic characteristics of the whole network system, and improve the calculation performance of the carbon nanotube physical reserve pool. This is not only applicable to carbon nanotube networks, but also applicable to micro-nano material networks corresponding to other micro-nano materials.

[0053] Embodiment 2

[0054] In addition to the hybrid topology structure of the ESN network and the LSM network shown in Embodiment 1 Figure 3 and Figure 4 above, the hybrid network topology structure can also adopt other ways. Embodiment 2 is to assemble an ESN network on the front surface of the same PCB board and assemble an ESN network on the back surface (i.e., assemble two ESN networks on the same PCB board at the same time). Between the two networks, they can be stacked or connected in parallel.

[0055] Figure 11 In order to calculate the result of the NARMA10 task after the two ESN network devices are stacked, the NMSE is 0.016566, which is more accurate than the prediction result of the independent ESN network, and it can be seen that the calculation performance of the hybrid network obtained based on the application is better than that of a single network. The ESN network and the ESN network are mixed, which can expand the capacity of the model and enable it to capture and represent more complex dynamic relationships and characteristics; in some complex prediction tasks, a single ESN may not be able to achieve ideal accuracy, by combining multiple ESNs, the prediction accuracy of complex systems can be improved through their synergistic effect, and various different application scenarios can also be more flexibly adapted.

[0056] Embodiment 3

[0057] Embodiment 3 is to assemble an LSM network on the front surface of the same PCB board and assemble an LSM network on the back surface (i.e., assemble two LSM networks on the same PCB board at the same time). Between the two networks, they can be stacked or connected in parallel.

[0058] The combination of the LSM network and the LSM network can further improve the parallel processing capability of the system, so that a large amount of spatio-temporal information can be processed simultaneously; at the same time, the recognition range and depth of the system to the complex spatio-temporal pattern can be expanded, so that the system performs better in processing complex biological signals, speech recognition and other tasks; the combination of multiple LSMs can form a more robust network structure, and even if part of the LSM is disturbed or damaged, the entire system can still maintain good performance.

[0059] The above embodiments are only examples, and the same or different types of physical reserve pools can be assembled on the front and back surfaces of the PCB according to requirements, that is, whether to build a topology structure mixing the ESN network and the LSM network, or to build a topology structure mixing two ESN networks, or to build a topology structure mixing two LSM networks, can be selected according to actual computing requirements. In addition to the PCB, other connection boards can also be used. In addition, the above embodiments provide a structure design for assembling two 2D structure carbon nanotube networks, and in addition to the 2D structure carbon nanotube network, other micro-nano materials can also be used to form a network (of course, it can also be a 3D network device).

[0060] The parts not described in detail in the present application can be implemented by referring to the prior art, for example, the stacking or parallel connection of the nodes of the two networks (corresponding to the connection between the input layers of the two networks and the connection between the output layers of the two networks) is realized by a software system and a hardware system (for example, the hardware system mainly consists of an industrial computer, an NI data acquisition card and an acquisition interface input / output control circuit board; the software system mainly consists of LabVIEW and MATLAB platform host computer programs).

[0061] Those skilled in the art will readily understand that the above description is only a preferred embodiment of the present application and is not intended to limit the present application, and any modification, equivalent replacement and improvement made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A hybrid topological micro- and nanomaterial physical stock pool, characterized in that, The connecting plate, the first micro-nano material network physical reserve pool on the front of the connecting plate, and the second micro-nano material network physical reserve pool on the back of the connecting plate; wherein the first micro-nano material network physical reserve pool and the second micro-nano material network physical reserve pool are independently selected from an echo state network (ESN) based micro-nano material network physical reserve pool and a liquid state machine (LSM) based micro-nano material network physical reserve pool; The input nodes of the first micro-nano material network physical reserve pool and the input nodes of the second micro-nano material network physical reserve pool are connected in parallel to add the input nodes of the first micro-nano material network physical reserve pool and the input nodes of the second micro-nano material network physical reserve pool, and the output nodes of the first micro-nano material network physical reserve pool and the output nodes of the second micro-nano material network physical reserve pool are connected in parallel to add the output nodes of the first micro-nano material network physical reserve pool and the output nodes of the second micro-nano material network physical reserve pool; or the input nodes of the first micro-nano material network physical reserve pool and the input nodes of the second micro-nano material network physical reserve pool are connected in one-to-one short circuit, and the output nodes of the first micro-nano material network physical reserve pool and the output nodes of the second micro-nano material network physical reserve pool are connected in one-to-one short circuit.

2. The hybrid topological micro- and nanomaterial physical reserve pool of claim 1, wherein, The micro-nano material is a carbon nanotube, and the first micro-nano material network physical reserve pool and the second micro-nano material network physical reserve pool are both carbon nanotube network physical reserve pools.

3. The hybrid topography micro- and nanomaterial physical reserve pool according to claim 2, wherein, The first micro-nano material network physical reserve pool and the second micro-nano material network physical reserve pool are both 2D carbon nanotube network physical reserve pools.

4. The hybrid topography micro- and nanomaterial physical reserve pool of claim 1, wherein, The reservoir of the first micro-nano material network physical reserve pool is prepared directly on the connecting plate, or is prepared on a silicon substrate and then attached to the connecting plate; The reservoir of the second micro-nano material network physical reserve pool is prepared directly on the connecting plate, or is prepared on a silicon substrate and then attached to the connecting plate.

5. The hybrid topography micro- and nanomaterial physical reserve pool according to claim 4, wherein, The connecting plate is a PCB board.

6. The hybrid topography micro- and nanomaterial physical reserve pool according to claim 2, wherein, The carbon nanotube is a semiconductive single-walled carbon nanotube.

7. The hybrid topography micro- and nanomaterial physical reserve pool according to claim 1, wherein, The echo state network (ESN) based micro-nano material network physical reserve pool is an echo state network (ESN) based carbon nanotube network physical reserve pool, specifically a POM / SWCNT / PBMA thin film device formed by components including phosphomolybdic acid, single-walled carbon nanotubes, and polybutyl methacrylate; The liquid state machine (LSM) based micro-nano material network physical reserve pool is a liquid state machine (LSM) based carbon nanotube network physical reserve pool, specifically a POM / SWCNT pulse device formed by components including phosphomolybdic acid and single-walled carbon nanotubes.

8. The hybrid topography micro- and nanomaterial physical reserve pool of claim 1, wherein, One of the first micro-nano material network physical reserve pool and the second micro-nano material network physical reserve pool is an echo state network (ESN) based micro-nano material network physical reserve pool, and the other is a liquid state machine (LSM) based micro-nano material network physical reserve pool.

9. A computing system, comprising: The micro-nano material physical reserve pool with the mixed topology structure according to any one of claims 1-8.

Images