A plasma reservoir pre-computation system

Abstract

The application discloses a kind of plasma reservoir pool pre-computing systems, comprising: coherent light source, first objective, reservoir pool module, second objective, imaging plane are sequentially arranged in light path;The coherent light source emits the incident light beam carrying its own information through the first objective converges to the reservoir pool module;The reservoir pool module receives the incident light beam and converts it scattered into the imaging plane can capture transmission light beam;The second objective focuses the transmission light beam scattered by the reservoir pool module on the imaging plane;The imaging plane is used to capture the transmission light beam and image.Utilize plasma scattering to break through the diffraction limit, so that it can extract more abundant information from the input information carrying light beam or the holographic encoding light beam of the object to be measured.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of reservoir computing, in particular to a plasma reservoir pre-computing system. BACKGROUND

[0002] Reservoir computing, initially proposed in the early 21st century as echo state networks and liquid state machines, is a framework of neural computing, which includes a fixed computational reservoir that maps the input into a high-dimensional space and reads out the reservoir states from the high-dimensional space from a trainable output layer. One advantage of this computing framework is that the computational reservoir is fixed and does not require adaptive training, providing opportunities for implementation in various physical systems. Among these physical reservoir layers, reservoir computing based on photonic / optoelectronic technology has attracted much attention due to its good performance, and its architecture is generally divided into two categories: node array and time-delay system.

[0003] Traditional optical (or optoelectronic) reservoir computing systems generally include four parts: the first part is the optical signal input end, which is composed of one or more coherent light sources carrying information, the second part is the optical processing unit, i.e. the optical (or optoelectronic) reservoir, the third part is the optical signal intensity reading unit, and the fourth part is the algorithm processing of the read data signal. In order to achieve better computing effect, the overall hardware composition is usually complex, such as the input optical signal often needs to be modulated first and then injected into the reservoir through waveguide coupling; the reservoir itself needs to contain one or more optical (or optoelectronic) nonlinear devices, and at the same time needs to assemble complex components such as micro-ring resonators or design time-delay components and feedback systems; the optical signal output reading unit is generally composed of a single high-speed photodetector or a photodetector array such as an optical camera; the read data signal is generally processed using a linear regression model. The entire system needs to use specially designed optical devices and equipment, which has high cost, and at the same time has high restrictions on the form of input light beams, and cannot break through the diffraction limit to perform sub-micron optical computing. SUMMARY

[0004] Therefore, the purpose of the present application is to provide a plasma reservoir pre-computing system that can solve the above problems.

[0005] The present application provides a plasma reservoir pre-computing system, comprising: a coherent light source, a first objective lens, a reservoir module, a second objective lens and an imaging plane arranged in sequence along the light path.

[0006] The coherent light source emits an incident light beam carrying self-information to the reserve pool module through the first objective lens; the reserve pool module receives the incident light beam and converts it into a transmitted light beam that can be captured by the imaging plane; the second objective lens focuses the transmitted light beam scattered by the reserve pool module onto the imaging plane; and the imaging plane is used to capture the transmitted light beam and form an image.

[0007] Further, comprising: a data acquisition module and a machine learning model;

[0008] The data acquisition module is used to store the transmitted light beam after imaging.

[0009] The machine learning model is used to extract the information of the object to be measured in the transmitted light beam.

[0010] The machine learning module can be any one or combination of a neural network model, a linear regression model.

[0011] The coherent light source can adjust different incident angles to adjust the angle of the incident light beam to obtain different information.

[0012] The coherent light source can also emit the incident light beam to irradiate the object to be measured to produce an encoded light beam carrying object information through holographic encoding of the object to be measured.

[0013] The reserve pool module is a plasmonic nanometer multi-scatterer with a thickness of 200-5000nm.

[0014] The plasmonic nanometer multi-scatterer contains a plurality of nanometer metal particles corresponding to the nodes in the traditional reserve pool.

[0015] The diameter of the nanometer metal particles is 50-150nm, and the spacing between any adjacent nanometer metal particles is 0-300nm.

[0016] The nanometer metal particles are randomly distributed and solidified in a transparent medium.

[0017] The transparent medium can be any one of silica, glass, solid water, epoxy resin, polymer transparent film, and transparent plastic.

[0018] The beneficial effects of the present application are:

[0019] First, by using a nanometer metal multi-scatterer as a computational reserve pool, the diffraction limit is broken through by using plasmonic scattering, so that more abundant information can be extracted from the input light beam carrying information or the holographic encoded light beam of the object to be measured. The system cost is greatly reduced, the form of the input light beam is less limited, and sub-micron optical computing can be performed by breaking through the diffraction limit.

[0020] Second, the nanometer metal scatterer as a computing reserve pool has a certain degree of nonlinear capability. Compared with other complex photonic reserve pool calculations using specially designed nonlinear units, although the weak nonlinearity limits the computing capability of the optical part to a certain extent, the system building cost is greatly reduced, and the overall robustness of the system is stronger, which can well balance the hardware and software parts. By strengthening the software algorithm processing of the collected signal, a stronger machine learning model than the linear regression model used in traditional reserve pool calculation can be used to improve the overall computing power. BRIEF DESCRIPTION OF DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0022] Figure 1 is the system module diagram of the present application in which the light beam itself carries information.

[0023] Figure 2 is the system module diagram of the present application in which the light beam holographic encodes the object.

[0024] Figure 3 is the light intensity |E|2 diagram of the present application in which the light beam polarization is 0°.

[0025] Figure 4 is the light intensity |E|2 diagram of the present application in which the light beam polarization is 90°.

[0026] Figure 5 is the light intensity |E|2 diagram of the present application in which the light beam polarization parameters r1=0.87, r2=0.50, and φ=5.64.

[0027] Figure 6 is the prediction diagram of the polarization information parameter r1 in the present application.

[0028] Figure 7 is the prediction diagram of the polarization information parameter r2 in the present application.

[0029] Figure 8 is the prediction diagram of the polarization information parameter φ in the present application.

[0030] Figure 9 is the prediction diagram of the polarization information parameter r1 after offset in the present application.

[0031] Figure 10 is the prediction diagram of the polarization information parameter r2 after offset in the present application.

[0032] Figure 11 is a prediction map of the post-shift polarization information parameter φ in the present application. DETAILED DESCRIPTION

[0033] For the convenience of those skilled in the art to understand, the structure of the present application will be further described in detail in combination with the drawings: it should be understood that, in the present embodiment, the steps mentioned, except for the order specified, can be adjusted in actual need, even can be executed simultaneously or partially simultaneously.

[0034] As Figure 1 shown, the embodiment of the present application provides a kind of plasma reservoir pool pre-computing system, comprising:

[0035] Including: coherent light source, first objective lens, reservoir pool module, second objective lens, imaging plane are sequentially arranged along light path;

[0036] The coherent light source emits incident light beam carrying its own information to the reservoir pool module through the first objective lens;The reservoir pool module receives the incident light beam and converts it into a transmission light beam that can be captured by the imaging plane;The second objective lens focuses the transmission light beam scattered by the reservoir pool module onto the imaging plane;The imaging plane is used to capture the transmission light beam and image.

[0037] Wherein, the coherent light source can adjust different incident angles, for adjusting the incident light beam angle to obtain different information.

[0038] Wherein, the coherent light source can also emit the incident light beam to irradiate on the object to be measured, and carry out holographic encoding on the object to be measured to generate an encoded light beam carrying object information.

[0039] In the present embodiment, the hardware part of the whole system can be divided into two stages: the first stage is the beam holographic encoding stage, which is divided into two cases, 1) as Figure 1 shown, the light beam itself carries information, such as the polarization of the light beam;2) the light beam holographic encodes the object, as Figure 2 shown, the light beam carries part of the information of the object to be analyzed after irradiating on the object. The light source can use coherent input light source, and the method can adopt common coherent light preparation method, such as wavefront segmentation method or amplitude segmentation method. The second stage is the information extraction stage, which inputs the light beam carrying its own information or object information in the first stage into the plasma reservoir pool pre-computing module. The module uses plasma scattering to break through the diffraction limit, so as to extract more abundant information from the light beam. The light beam output from the module is output to the imaging plane through the focusing module.

[0040] Further comprising: a data acquisition module and a machine learning model;

[0041] The data acquisition module is configured to store the transmitted light beam after imaging.

[0042] The machine learning model is configured to extract information of the object to be measured in the transmitted light beam.

[0043] The machine learning module can be any one or combination of a neural network model and a linear regression model.

[0044] In this embodiment, the software part of the system is used to extract data in the imaging plane. Depending on the task, the data in the imaging plane is collected for data analysis of the object or further machine learning calculation. The data can be used as training samples to train the neural network model or the linear regression model. The trained neural network model or linear regression model realizes feature extraction, classification and prediction of data. The training method of the neural network model or the linear regression model is a prior art, which will not be described here.

[0045] The reservoir module is a plasmonic nanomultiscatterer with a thickness of 200-5000 nm.

[0046] The plasmonic nanomultiscatterer contains a plurality of nanometallic particles.

[0047] The diameter of the nanometallic particles is 50-150 nm, and the spacing between any adjacent nanometallic particles is 0-300 nm.

[0048] The nanometallic particles are randomly distributed and solidified in the transparent medium.

[0049] In this embodiment, the thickness of the reservoir module, the size of the nanometallic particles and the spacing can have certain flexibility. The nanometallic particles can have a diameter of 50-150 nm, which is within the conventional size range. The spacing of the nanometallic particles is close to the diameter of the nanometallic particles, and multiple layers of nanometallic particles are accommodated by selecting the thickness of the reservoir module, which ensures the occurrence of multiple scattering and enhances the dynamic complexity of the reservoir.

[0050] The transparent medium can be any one of silica, glass, solid water, epoxy resin, polymer transparent film and transparent plastic.

[0051] In the embodiment, the reservoir module is a plasmonic nanomulti-scatterer, metal nanoscatters are randomly distributed in the plasmonic nanomulti-scatterer, and the encoding information in the input light beam is converted into a transmission light pattern through multiple scattering of the metal nanoscatters, which can be acquired by a conventional imaging plane, and the nanometal particles can be selected from any one of nanogold particles, nanosilver particles and other nanometal particles.

[0052] The surface of the nanometal particles is excited to surface plasmons.

[0053] In the embodiment, when the incident information-carrying light beam converges on the plasmonic nanomulti-scatterer through the first objective lens, surface plasmons are excited on the surface of each nanometal particle, and multiple scattering occurs between the randomly distributed nanometal particles, so that the information in the incident light beam can be extracted in more detail.

[0054] The reservoir module does not contain a specially designed nonlinear device.

[0055] In the embodiment, the nonlinear capability of the plasmonic reservoir mainly comes from the quadratic nonlinearity of the light intensity signal when the output light beam is imaged, and the secondary part comes from the weak nonlinearity of each nanometal scatterer when the light is scattered.

[0056] The application is further described below in combination with FDTD (finite-difference time-domain) simulation.

[0057] In the FDTD simulation, a coherent light source with a wavelength of 550 nm and a Gaussian beam with a radius of 250 nm is selected, the reservoir module is made of a 500 nm thick silicon dioxide plate, the silicon dioxide plate solidifies three layers of nanogold particles with a diameter of 50 nm, the vertical spacing of each layer of nanogold particles is 120 nm, and the gap between two adjacent nanogold particles is between 20-50 nm. An imaging plane is selected by a camera system, which can be acquired after focusing through an objective lens. The machine module is selected by three trainable convolution layers and one trainable linear layer, and the loss function is MSE (mean square error). The polarization information encoded in the incident light beam has three parameters: r1 is the amplitude of the component along the x direction, r2 is the amplitude of the component along the y direction, and φ is the phase difference between the two components. As shown in Figure 3 、 Figure 4 、 Figure 5 Different light beam polarizations (0°: r1=1, r2=0, φ=0, 90°: r1=0, r2=1, φ=0, r1=0.87, r2=0.50, φ=5.64) produce different light intensity patterns in the imaging plane in the incident light beam, different light beam polarizations carry different light beam information, produce different light intensity patterns, and different images generated can be further analyzed.

[0058] After collecting the incident beam data, training and testing are performed, where the training dataset is 6000 data samples and the testing dataset is 1000 data samples. Assuming no shift between the incident beam positions (relative to the objective and reservoir module) in training and testing, the prediction of the three polarization information parameters is achieved, where Figure 6 the prediction of the polarization information parameter r1, Figure 7 the prediction of the polarization information parameter r2, Figure 8 the prediction of the polarization information parameter φ, the polarization information parameters r1, r2 and φ can be determined as 0.99898, 0.99909 and 0.94377 by linear regression prediction, and their MSE (Mean Squared Error) values are 4.68216e-05, 4.44993e-05 and 0.18881 respectively. It can be seen that the MSE values of r1, r2 and φ are extremely close to 0, indicating that the method has excellent prediction effect for the three polarization information parameters. It can be seen that the present application can well realize the collection of incident beam information and the prediction of beam polarization information according to the collected data.

[0059] In order to test the generalization ability of the system, mixed training data is used for training. The incident angle is adjusted, and data with a shift of ±50nm in the x, y and z directions of the incident beam is collected. Although the overall performance is affected by the shift of the incident position, the overall system can still achieve good prediction, where Figure 9 the prediction of the polarization information parameter r1 after the shift, Figure 10 the prediction of the polarization information parameter r2 after the shift, Figure 11 the prediction of the polarization information parameter φ after the shift, the polarization information parameters r1, r2 and φ can be determined as 0.95728, 0.95546 and 0.16909, and their MSE (Mean Squared Error) values are 0.00245, 0.00275 and 2.07569 respectively. It can be seen that the MSE values of r1, r2 and φ are extremely close to 0, indicating that the method still has good prediction effect for the three polarization information parameters. It can be seen that the present application can still well realize the prediction of the polarization information parameters of the light by using the polarization information data after the shift.

[0060] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer usable program code.

[0061] The present application is described in reference to the accompanying drawings, which use the legend: in which: like numerals represent like elements throughout the several figures, and in which: Figure 1 one or more of the illustrated flows or blocks Figure 1 an apparatus to perform the functions specified in one or more of the illustrated flows or blocks.

[0062] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the flow Figure 1 one or more of the illustrated flows or blocks Figure 1 an apparatus to perform the functions specified in one or more of the illustrated flows or blocks.

[0063] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the flow Figure 1 one or more of the illustrated flows or blocks Figure 1 an apparatus to perform the functions specified in one or more of the illustrated flows or blocks.

[0064] It should be noted that the use of any of the terms "first", "second" or the like does not connote any order, quantity, or importance, but rather are used to distinguish one element from another. It should also be noted that the terms "comprising", "including", "containing", and / or "having" are intended to be open-ended terms. Further, the singular forms "a", "an" and / or "the" are intended to include one or more, unless expressly specified otherwise.

[0065] While the preferred embodiments of the application have been described, additional variations and modifications can be made to the embodiments described and illustrated herein, without departing from the spirit and scope of the application. Accordingly, it is intended that all subject matter contained in the above description be interpreted as illustrative only and basis for claims.

[0066] Obviously, various modifications and changes can be made to the present application by those skilled in the art without departing from the spirit and scope of the present application. Accordingly, the present application intends to include all such modifications and changes as fall within the scope of the claims and their equivalents.

[0067] In the present application, unless otherwise explicitly specified and limited, the terms "mounting", "connection", "linking", "fixing" and the like should be understood in a broad sense, for example, can be fixed connection, can also be detachable connection, or integral; can be mechanical connection, can also be electrical connection; can be direct connection, can also be indirect connection through intermediate medium, can be internal communication of two elements or interaction relationship of two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0068] In the description of the present application, the description of the terms "one embodiment", "some embodiments", "example", "specific example" or "some examples" means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present application, the illustrative description of the above terms should not be understood as necessarily referring to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any appropriate manner in any one or more embodiments or examples. In addition, those skilled in the art can combine and combine the different embodiments or examples described in the present application and the features of the different embodiments or examples without contradiction.

Claims

1. A plasma reservoir precomputation system, characterized by, The application relates to a coherent light source, a first objective lens, a reservoir module, a second objective lens and an imaging plane arranged in sequence along a light path. The coherent light source emits an incident light beam carrying self-information, the incident light beam is converged to the reservoir module through the first objective lens, the reservoir module receives the incident light beam and converts the incident light beam into a transmission light beam which can be captured by the imaging plane through scattering, the second objective lens focuses the transmission light beam scattered by the reservoir module on the imaging plane, and the imaging plane is used for capturing the transmission light beam and imaging. The application further relates to a data acquisition module and a machine learning model. The data acquisition module is used for storing the transmission light beam after imaging. The machine learning model is used for extracting information of a to-be-detected object in the transmission light beam. The machine learning model is any one or combination of a neural network model and a linear regression model. The coherent light source can adjust different incident angles, and is used for adjusting the angle of the incident light beam to obtain different information. The coherent light source is obtained by emitting one or more coherent light beams to irradiate a to-be-detected object, and the to-be-detected object is holographically encoded through optical reflection, transmission or scattering to generate an encoded light beam carrying object information. The reservoir module is a plasmonic nanometer multi-scattering body with a thickness of 200-5000 nm. The plasmonic nanometer multi-scattering body contains a plurality of nanometer metal particles. The diameter of the nanometer metal particles is 50-150 nm, and any two adjacent nanometer metal particles are spaced apart by 0-300 nm. The nanometer metal particles are randomly distributed and solidified in a transparent medium. The transparent medium is any one of silica, glass, solid water, epoxy resin, a polymer transparent film and transparent plastic.

2. The plasma reservoir precomputation system of claim 1, wherein, ​

Images