Dual-modulation gradient reservoir computing system and method based on spin-polarized dynamics

Abstract

The application belongs to the technical field of reservoir computing, and relates to a double-modulation gradient reservoir computing system and method based on spin polarization dynamics. The system comprises a double-modulation input layer, a photonic reservoir layer and a double-modulation output layer. The double-modulation input layer modulates a to-be-processed signal to generate a control signal and adjusts the amplitude and polarization ellipticity of pump light, and obtains an input optical signal containing to-be-processed signal information based on the modulated pump light. The photonic reservoir layer comprises a spin vertical cavity surface laser emitter and a polarization beam splitter. The spin vertical cavity surface laser emitter outputs a gradient response optical signal containing time sequence characteristics of the to-be-processed signal under the excitation of the input optical signal. The polarization beam splitter performs polarization component matching and orthogonal polarization beam splitting on the gradient response optical signal, and outputs multiple polarization optical signals with mutually orthogonal polarization directions. The double-modulation output layer collects the polarization optical signals, and obtains a target output signal of the reservoir computing system based on the multiple polarization optical signals.

Description

Technical Field

[0001] This invention relates to the field of reservoir calculation technology, and in particular to a dual-modulation gradient reservoir calculation system and method based on spin polarization dynamics. Background Technology

[0002] As the scale and complexity of artificial intelligence models continue to rise, traditional electronic computing systems are increasingly approaching physical limits in terms of energy efficiency and bandwidth. The data migration overhead caused by the "separation of storage and computation" under the von Neumann architecture has become a core bottleneck for performance improvement. Photonic neuromorphic computing, on the other hand, directly "maps" the computation process to physical processes such as interference and diffraction of light fields by deeply coupling micro-nano photonic devices with neuromorphic computing models. This not only provides a feasible path for building low-latency, high-throughput intelligent computing systems, but also offers an architectural choice that achieves a new balance between energy efficiency, scalability, and system robustness for applications such as ultra-large-scale heuristic computing and edge intelligence.

[0003] The reservoir computing framework, with its inherent dynamic response and state memory characteristics, performs high-dimensional nonlinear mapping on time-varying input signals, thereby characterizing the temporal correlation and complex dynamic patterns in the input signals. By combining this framework with photonics technology, and utilizing the broadband transmission capabilities and multi-channel parallel characteristics of optical devices, the processing speed and bandwidth density of the system are improved while maintaining the dynamic richness of the reservoir, thus constructing a photonic reservoir computing architecture with neuromorphic information processing capabilities.

[0004] Neural network systems based on the aforementioned photonic reservoir computing architecture have been used for various pattern recognition tasks. For example, by encoding speech signals by light intensity or phase and then injecting them into a photonic reservoir, temporal features of the speech can be extracted in the optical domain, and the output layer can then determine the speech category. Related research results show that this type of photonic reservoir system can achieve high recognition accuracy in applications such as speech recognition and handwritten digit recognition, while maintaining low inference latency and high computational throughput.

[0005] Although computational architectures based on photonic reservoirs have shown potential in tasks such as speech recognition and channel equalization, most existing optical implementations only use single optical field parameters such as light intensity and phase to complete data weighting operations and nonlinear transformations. When faced with complex tasks with high-dimensional entanglement features or ultrafast dynamic characteristics (such as multi-person speech aliasing recognition in noisy environments, dynamic real-time voice command processing, and effective signal stripping in high-speed communication), such single-dimensional dynamic systems often face a trade-off bottleneck between "memory capacity" and "nonlinear mapping capability," making it difficult to achieve high-precision feature separation with a limited hardware node scale. Furthermore, existing technologies still face a dilemma in signal modulation and transmission strategies: using continuous analog optical signals for computation and transmission can fully leverage the large bandwidth of the optical path and achieve higher computation and transmission efficiency, but such signals are highly susceptible to line noise, device temperature changes, and hardware manufacturing deviations. Errors accumulate and amplify continuously during multi-stage computation and transmission, resulting in poor stability and low fault tolerance of the entire computing system. On the other hand, using discrete pulse signals similar to spiking neural networks for computation, while offering stronger anti-interference capabilities and more stable computation results, requires ultra-high sampling rates or extended data acquisition durations to accurately capture time-series computation information in high-speed optical paths. This significantly increases the design and control difficulty of back-end electrical control equipment and directly abandons the core advantages of photonic computing: low latency and high-speed real-time computation.

[0006] In summary, there is an urgent need to design a novel neural computing architecture that can integrate high-dimensional dynamic characteristics and take into account both high bandwidth and robustness, so as to break through the performance constraints caused by single-degree-of-freedom operations and significantly improve the data processing accuracy and system operation stability under complex dynamic tasks while ensuring low latency and high throughput. Summary of the Invention

[0007] Therefore, the technical problem to be solved by the present invention is to overcome the performance limitations of the photon reservoir computing architecture in the prior art due to the single degree of freedom operation, which makes it impossible to take into account the low computing latency, high throughput, data processing accuracy and operational stability under complex dynamic tasks.

[0008] To address the aforementioned technical problems, this invention provides a dual-modulation gradient reservoir computing system based on spin polarization dynamics, comprising: The dual-modulation input layer is used to modulate the signal to be processed to generate a control signal. The control signal is used to adjust the amplitude and polarization ellipticity of the pump light, and an input light signal containing the information of the signal to be processed is obtained based on the modulated pump light. The photon storage pool layer specifically includes: A spin vertical cavity surface laser emitter, connected to a dual modulation input layer, is used to change the total carrier concentration and the ratio of spin-up to spin-down carrier concentrations under the excitation of the input optical signal, forming a spin non-equilibrium carrier distribution. Thus, under the synergistic effect of the total carrier concentration and the spin non-equilibrium carrier distribution, a spin-related nonlinear polarization dynamic response is generated to the pulse combination at different times in the input optical signal, and the output is a gradient response optical signal that exhibits continuous layered differences in intensity level, polarization distribution, and transient trajectory. A polarization beam splitter, connected to a spin vertical cavity surface laser emitter, is used to perform polarization component matching and orthogonal polarization beam splitting on gradient response optical signals, outputting multiple polarized optical signals with mutually orthogonal polarization directions. A dual-modulation output layer, connected to the photon reservoir layer, is used to acquire polarized light signals and obtain the target output signal of the reservoir computing system based on the multi-path polarized light signals.

[0009] Preferably, the dual-modulation input layer includes: Pump lasers are used to generate pump light; A waveform generator is used to generate a mask signal. A signal coupling modulator, connected to a waveform generator, is used to couple and modulate the mask signal and the signal to be processed to generate a control signal. An electric amplifier, connected to a signal coupling modulator, is used to amplify control signals; A modulator, connected to the pump laser and an electrical amplifier, is used to adjust the amplitude of the pump light using a control signal; A polarization controller, connected to a modulator and an electrical amplifier, is used to adjust the polarization ellipticity of the pump light using a control signal, and to obtain an input optical signal containing information about the signal to be processed based on the modulated pump light.

[0010] Preferably, the dual-modulation input layer further includes: An optical isolator, connected to the pump laser, is used to suppress interference from the return light on the pump laser and improve the output stability of the pump light.

[0011] Preferably, the photon reservoir layer further includes: A beam splitter is used to split multiple polarized light signals with mutually orthogonal polarization directions and transmit the split polarized light signals to a dual-modulation output layer.

[0012] Preferably, the polarization beam splitter and the beam splitter prism are separate optical devices, with the beam splitter positioned in the optical path between the polarization beam splitter and the dual modulation output layer.

[0013] Preferably, the polarization beam splitter and the beam splitter prism are integrated on the same substrate to form a photonic integrated circuit.

[0014] Preferably, the dual-modulation output layer includes: A photodetector array, connected to a photon reservoir layer, is used to perform photoelectric conversion on various polarized light signals; A programmable gated array, connected to a photodetector array, is used to train the output of the photon reservoir layer, obtain the output weights of each polarized light signal, and thus obtain the target output signal of the reservoir computing system.

[0015] Preferably, the dual-modulation output layer further includes: The fiber collimator, connected to the photon reservoir layer, is used to collimate various polarized light signals. An optical amplifier, connected to an optical fiber collimator, is used to amplify the aligned polarized light signal and transmit the amplified polarized light signal to a photodetector array.

[0016] This invention also provides a method for calculating a dual-modulation gradient reservoir based on spin polarization dynamics. The method is implemented based on the aforementioned dual-modulation gradient reservoir calculation system based on spin polarization dynamics, and includes: A dual-modulation input layer is used to modulate the signal to be processed to generate a control signal. The amplitude and polarization ellipticity of the pump light are adjusted using the control signal. Based on the modulated pump light, an input light signal containing information about the signal to be processed is obtained. The input optical signal is injected into the spin vertical cavity surface laser emitter, so that the spin vertical cavity surface laser emitter changes the total carrier concentration and the ratio of spin-up carrier concentration to spin-down carrier concentration under the excitation of the input optical signal, forming a spin non-equilibrium carrier distribution. Thus, under the synergistic effect of the total carrier concentration and the spin non-equilibrium carrier distribution, a spin-related nonlinear polarization dynamic response is generated to the pulse combination of the input optical signal at different times, and the output gradient response optical signal exhibits continuous layered differences in intensity level, polarization distribution and transient trajectory. A polarization beam splitter is used to perform polarization component matching and orthogonal polarization beam splitting on the gradient response optical signal, outputting multiple polarized optical signals with mutually orthogonal polarization directions; Polarized light signals are acquired using a dual-modulation output layer, and the target output signal of the reservoir computing system is obtained based on the multi-channel polarized light signals.

[0017] Preferably, a dual-modulation input layer is used to modulate the signal to be processed to generate a control signal, and the amplitude and polarization ellipticity of the pump light are adjusted using the control signal, including: Pump light is generated using a pump laser; Use a waveform generator to generate a mask signal; A signal coupling modulator is used to couple and modulate the mask signal and the signal to be processed to generate a control signal for modulating the amplitude and polarization ellipticity of the pump light. The control signal is amplified using an electrical amplifier; The amplitude of the pump light is adjusted using a modulator based on a control signal; The polarization ellipticity of the pump light is adjusted by a polarization controller based on a control signal, and an input optical signal containing the information of the signal to be processed is obtained based on the modulated pump light.

[0018] The dual-modulation gradient reservoir computation system based on spin polarization dynamics provided in this application has the following advantages: This invention overcomes the inherent limitations of single-degree-of-freedom operations in existing technologies, and for the first time proposes to modulate the signal to be processed to generate a control signal. This control signal is then used to achieve two-dimensional coordinated control of the pump light's amplitude and polarization ellipticity. After the input light signal obtained from this two-dimensional coordinated modulation is injected into a spin-based vertical-cavity laser emitter (VCSEL), it can simultaneously change the total carrier concentration, the ratio of spin-up carriers to spin-down carriers within the VCSEL, thus breaking the original spin-carrier equilibrium state within the VCSEL and constructing a stable and controllable spin-nonequilibrium carrier distribution. Furthermore, under the synergistic coupling effect of the total carrier concentration and the spin-nonequilibrium carrier distribution, the VCSEL can excite a rich variety of spin-related nonlinear polarization dynamic responses to pulse combinations with different timing sequences and arrangements in the input light signal. This outputs a gradient response optical signal carrying complete temporal characteristics and exhibiting continuous hierarchical differences in intensity level, polarization distribution, and transient trajectory, giving the photon reservoir layer dual degrees of freedom in terms of light intensity and spin polarization. Then, a polarization beamsplitter performs polarization component matching and orthogonal polarization splitting on the gradient response optical signal, physically separating the nonlinear response features corresponding to different spin dynamics. After splitting, nonlinear features from different sources and with different response laws each occupy independent polarization transmission channels, no longer superimposing or interfering with each other. This achieves a physical expansion of the optical computation dimension, greatly enriching the feature mapping dimension and information storage capacity within the photon reservoir. Thus, without expanding the hardware node scale, it can efficiently complete the fine feature separation and data parsing of high-dimensional entangled features and dynamic signals, improving the data processing accuracy under complex dynamic tasks. Furthermore, relying on the dual modulation driving mechanism of amplitude and polarization, it can flexibly adapt to both continuous signals and time-series pulse signals as input forms. It can fully leverage the large bandwidth of the optical path to ensure high data throughput, while also using the multi-dimensional response characteristics of spin polarization to weaken signal errors caused by device temperature drift and link noise, thus significantly improving system stability. It can complete high-precision timing information encoding and extraction without relying on ultra-high sampling rates and long time-domain integration, preserving the advantage of ultra-low latency in photonic computing to the greatest extent. This effectively solves the problem that existing photonic reservoir computing architectures cannot simultaneously achieve low computational latency, high throughput, accuracy in handling complex tasks, and system stability. Attached Figure Description

[0019] To make the content of this invention easier to understand, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings, wherein: Figure 1 A schematic diagram of the dual-modulation gradient reservoir computing system based on spin polarization dynamics provided in this application; Figure 2 A flowchart of the dual-modulation gradient reservoir calculation method based on spin polarization dynamics provided in this application; Figure 3 A schematic diagram illustrating the dynamic response of the spin vertical-cavity surface-mount laser emitter provided in this application to input optical signals with different modulation schemes; wherein, Figure 3 (a) in the diagram is a schematic diagram of the dynamic response of a vertical cavity surface laser emitter to the input optical signal obtained after modulating the amplitude of the pump light. Figure 3 (b) in the figure is a schematic diagram of the dynamic response of the input light signal obtained by the vertical cavity surface laser emitter after modulating the polarization ellipticity of the pump light; Figure 4 A schematic diagram illustrating the prediction results of time series forecasting using a dual-modulation gradient reservoir computational system based on spin polarization dynamics, provided in this application; wherein, Figure 4 (a) in the diagram is a schematic representation of the target value. Figure 4 (b) in the diagram is a schematic of the predicted values; Explanation of reference numerals in the accompanying drawings: 1. Dual-modulation input layer; 11. Pump laser; 12. Waveform generator; 13. Signal coupling modulator; 14. Electrical amplifier; 15. Modulator; 16. Polarization controller; 17. Optical isolator; 2. Photon reservoir layer; 21. Spin vertical-cavity laser emitter; 22. Polarization beam splitter; 23. Beam splitter prism; 3. Dual-modulation output layer; 31. Photodetector array; 32. Programmable gate array; 33. Fiber collimator; 34. Optical amplifier. Detailed Implementation

[0020] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments described are not intended to limit the present invention.

[0021] Please see Figure 1 , Figure 1 The diagram shows the structure of the dual-modulation gradient reservoir computing system based on spin polarization dynamics provided in this application. The system specifically includes a dual-modulation input layer 1, a photon reservoir layer 2, and a dual-modulation output layer 3.

[0022] The dual-modulation input layer 1 is used to modulate the signal to be processed to generate a control signal. The control signal is used to adjust the amplitude and polarization ellipticity of the pump light, and an input optical signal containing the information of the signal to be processed is obtained based on the modulated pump light.

[0023] The photon reservoir layer 2 includes a spin vertical cavity surface laser emitter 21 and a polarization beam splitter 22.

[0024] The spin vertical cavity surface laser emitter 21 is connected to the dual modulation input layer 1. It is used to change the total carrier concentration and the ratio of spin-up carrier concentration to spin-down carrier concentration under the excitation of the input optical signal, forming a spin non-equilibrium carrier distribution. Thus, under the synergistic effect of the total carrier concentration and the spin non-equilibrium carrier distribution, it generates a spin-related nonlinear polarization dynamic response to the pulse combination at different times in the input optical signal, and outputs a gradient response optical signal that exhibits continuous layered differences in intensity level, polarization distribution and transient trajectory.

[0025] The polarization beam splitter 22 is connected to the spin vertical cavity surface laser emitter 21 and is used to perform polarization component matching and orthogonal polarization beam splitting on the gradient response optical signal, and output multiple polarized optical signals with mutually orthogonal polarization directions.

[0026] The dual-modulation output layer 3 is connected to the photon reservoir layer 2 and is used to collect polarized light signals. The target output signal of the reservoir computing system is obtained based on the multi-path polarized light signals.

[0027] Specifically, this application employs optical injection to perform dual-dimensional modulation of a spin vertical-cavity surface-mount laser emitter 21. The amplitude of the pump light is used to adjust the total carrier concentration within the spin vertical-cavity laser emitter 21, while the polarization ellipticity of the pump light is used to adjust the relative injection ratio of spin-up carriers to spin-down carriers, thus introducing spin degrees of freedom beyond traditional carrier total modulation. This dual modulation method not only induces distinguishable changes in the device's output state but also generates a hierarchical and gradual gradient response when input parameters change continuously: when the pump light power changes, the total carrier concentration, stimulated emission gain, and the competition between polarization modes within the device change accordingly; when the polarization ellipticity of the pump light changes, the relative injection ratio of spin-up carriers to spin-down carriers changes, thereby altering the degree of spin imbalance and the evolution path of the dual polarization modes within the device. Since the total carrier concentration modulation and spin imbalance modulation are coupled within the same device, they influence gain distribution, polarization competition, and relaxation. The process and time-domain evolution behavior mean that for input modulation with different amplitudes, different sequences and different combinations, the system can output a response result that exhibits continuous layered differences in intensity level, polarization distribution and transient trajectory, thus forming a gradient response characteristic oriented towards input data. That is, based on the synergistic effect of total carrier concentration and spin non-equilibrium carrier distribution, the spin vertical cavity surface laser emitter 21 can excite a spin-related nonlinear polarization dynamic response, thereby outputting a gradient response optical signal carrying the characteristic information of the signal to be processed and exhibiting continuous layered differences in intensity level, polarization distribution and transient trajectory under different pulse combination excitations.

[0028] Furthermore, based on the aforementioned gradient response mechanism, the photon reservoir layer 2 of this application does not merely perform simple on / off mapping or binary differentiation of input information. Instead, it projects the input data in the amplitude, polarization, and time domains onto multiple interconnected but mutually distinguishable dynamic state intervals, ensuring that similar inputs correspond to similar but distinct output states, and inputs with significant differences correspond to output states with higher separation. Thus, the system can achieve hierarchical expansion, nonlinear separation, and high-dimensional mapping of input data in the initial feature extraction stage, improving its ability to represent complex time-series signals, weakly differentiated patterns, and multi-class data. This provides richer and more stable feature information for subsequent readout training and classification decisions of the dual-modulation output layer.

[0029] It should be noted that this application does not merely perform conventional adjustment of the external polarization state of the laser output, but rather controls the spin carrier distribution inside the device by adjusting the polarization ellipticity of the pump light. This expands the input modulation from single carrier intensity modulation to two-dimensional synergistic modulation of total carrier concentration and spin imbalance. Compared with traditional modulation methods limited by carrier relaxation oscillations, this approach can fully leverage the spin-correlated response advantage of the spin vertical cavity laser emitter 21, improve the device's representation dimension and dynamic response capability of input information, and thus provide a physical basis for high-speed, multi-level information processing.

[0030] Furthermore, the dual-modulation input layer 1 includes a pump laser 11, a waveform generator 12, a signal coupling modulator 13, an electrical amplifier 14, a modulator 15, and a polarization controller 16.

[0031] Pump laser 11 is used to generate pump light. Optionally, such as Figure 1 As shown, the dual modulation input layer 1 also includes an optical isolator 17, which is connected to the pump laser 11 to suppress the interference of the return light on the pump laser 11 and improve the output stability of the pump light.

[0032] Waveform generator 12 is used to generate mask signals.

[0033] The signal coupling modulator 13 is connected to the waveform generator 12 and is used to couple and modulate the mask signal and the signal to be processed to generate a control signal.

[0034] The amplifier 14 is connected to the signal coupling modulator 13 and is used to amplify the control signal.

[0035] The modulator 15 is connected to the pump laser 11 and the electric amplifier 14 and is used to adjust the amplitude of the pump light using a control signal.

[0036] The polarization controller 16 is connected to the modulator 15 and the amplifier 14 and is used to adjust the polarization ellipticity of the pump light using a control signal, so as to obtain an input optical signal containing the information of the signal to be processed based on the modulated pump light.

[0037] Specifically, this application modulates the signal to be processed to generate a control signal for adjusting the amplitude and polarization ellipticity of the pump light. The modulator 15 and polarization controller 16 then adjust the amplitude and polarization ellipticity of the pump light based on this control signal to obtain an input optical signal containing information about the signal to be processed. This input optical signal is then used to excite the spin vertical-cavity laser emitter 21. Specifically, the amplitude of the pump light is used to adjust the total carrier concentration inside the spin vertical-cavity laser emitter 21, and the polarization ellipticity of the pump light is used to adjust the relative injection ratio of spin-up carriers to spin-down carriers, thereby breaking the carrier spin symmetry and introducing a spin degree-of-freedom control dimension to the system independent of the total carrier modulation. Under the combined effects of dual polarization mode competition and coupling, spin-flip relaxation, and intracavity nonlinear dynamics, the system can evolve distinguishable transient output trajectories and steady-state response levels for different input timing combinations, thus forming gradient response optical signal characteristics containing the feature information of the signal to be processed.

[0038] Furthermore, the photon reservoir layer 2 also includes a beam splitter 23, which is used to split multiple polarized light signals with mutually orthogonal polarization directions and transmit the split polarized light signals to the dual modulation output layer 3.

[0039] In some embodiments, the polarization beam splitter 22 and the beam splitter prism 23 are separate optical devices, and the beam splitter prism 23 is disposed in the optical path between the polarization beam splitter 22 and the dual modulation output layer 3.

[0040] In other embodiments, the polarization beam splitter 22 and the beam splitter prism 23 are integrated on the same substrate to form a photonic integrated circuit.

[0041] Specifically, the polarization component matching and transmission state adjustment of the output of the spin vertical cavity laser emitter 21 are performed by the polarization beam splitter 22 to output multiple polarized light signals. The beam splitter 23 is used to stably transmit each polarized light signal to the dual modulation output layer 3, so that the dual modulation output layer 3 can read out and fuse the feature information carried by each polarized light signal in parallel, thereby realizing further feature extraction and output mapping of the signal to be processed.

[0042] Furthermore, the dual-modulation output layer 3 includes a photodetector array 31 and a programmable gate array 32.

[0043] The photodetector array 31 is connected to the photon reservoir layer 2 and is used to perform photoelectric conversion on various polarized light signals.

[0044] The programmable gate array 32 is connected to the photodetector array 31 and is used to train the output of the photon reservoir layer 2 to obtain the output weights of each polarized light signal, thereby obtaining the target output signal of the reservoir computing system.

[0045] Furthermore, the dual-modulation output layer 3 also includes an optical fiber collimator 33 and an optical amplifier 34.

[0046] The fiber collimator 33 is connected to the photon reservoir layer 2 and is used to collimate the polarized light signals of each path.

[0047] The optical amplifier 34 is connected to the fiber collimator 33 to amplify the aligned polarized light signal and transmit the amplified polarized light signal to the photodetector array 31.

[0048] Specifically, since the spin vertical cavity surface laser emitter 21 can generate gradient response optical signals under the action of spin-related nonlinear polarization dynamics, these optical signals have differentiated and identifiable characteristics in terms of intensity distribution, temporal response, and dynamic evolution path. Therefore, after polarization beam splitting and channel division of the gradient response optical signals, this application reads out and processes each polarization signal in parallel in the dual modulation output layer 3. By assigning different output weights to each polarization optical signal and performing weighted superposition and output mapping, differentiated fusion of feature information of each path can be implemented, thereby enhancing the high-dimensional representation capability, parallel processing capability, and output discrimination capability of the photon reservoir computing system, and improving the accuracy and stability of the final output results.

[0049] In a specific example of this application, the output weights of the photon reservoir layer 2 can be trained using linear regression and ridge regression algorithms.

[0050] This application also provides a method for calculating a dual-modulation gradient reservoir based on spin polarization dynamics. This method is implemented based on the aforementioned dual-modulation gradient reservoir calculation system based on spin polarization dynamics. Figure 2 As shown, the method includes S10~S40: S10: The dual modulation input layer modulates the signal to be processed to generate a control signal. The amplitude and polarization ellipticity of the pump light are adjusted using the control signal. Based on the modulated pump light, an input light signal containing the information of the signal to be processed is obtained.

[0051] S20: The input optical signal is injected into the spin vertical cavity surface laser emitter, so that the spin vertical cavity surface laser emitter changes the total carrier concentration and the ratio of spin-up carrier concentration to spin-down carrier concentration under the excitation of the input optical signal, forming a spin non-equilibrium carrier distribution. Thus, under the synergistic effect of the total carrier concentration and the spin non-equilibrium carrier distribution, a spin-related nonlinear polarization dynamic response is generated to the pulse combination in the input optical signal at different times, and the output gradient response optical signal exhibits continuous layered differences in intensity level, polarization distribution and transient trajectory.

[0052] S30: Uses a polarization beam splitter to perform polarization component matching and orthogonal polarization beam splitting on the gradient response optical signal, and outputs multiple polarized optical signals with mutually orthogonal polarization directions.

[0053] S40: Polarized light signals are acquired using a dual-modulation output layer, and the target output signal of the reservoir computing system is obtained based on the multi-channel polarized light signals.

[0054] Furthermore, a dual-modulation input layer is used to modulate the signal to be processed to generate a control signal, which is then used to adjust the amplitude and polarization ellipticity of the pump light, including S100~S105: S100: Uses a pump laser to generate pump light.

[0055] S101: Use a waveform generator to generate a mask signal.

[0056] S102: The mask signal and the signal to be processed are coupled and modulated using a signal coupling modulator to generate a control signal for modulating the amplitude and polarization ellipticity of the pump light.

[0057] S103: Amplify the control signal using an electric amplifier.

[0058] S104: The amplitude of the pump light is adjusted by a modulator based on a control signal.

[0059] S105: The polarization ellipticity of the pump light is adjusted by the polarization controller based on the control signal, and the input light signal containing the information of the signal to be processed is obtained based on the modulated pump light.

[0060] The above solution will be further explained below with reference to a specific embodiment: In this embodiment, the aforementioned dual-modulation gradient reservoir computational system based on spin polarization dynamics is used to predict time series data. Specifically, firstly, a dual-modulation gradient reservoir computational system based on spin polarization dynamics is built and simulated through numerical simulation, such as... Figure 3 The diagram shows the dynamic response of the spin vertical cavity surface laser emitter provided in this application to input optical signals with different modulation schemes. Figure 3 (a) in the diagram is a schematic of the dynamic response of a vertical-cavity surface-mount laser (VCSEL) to the input optical signal after the amplitude of the modulated pump light is adjusted. The corresponding rate equation is expressed as: , , , Figure 3 Figure (b) shows the dynamic response of a vertical-cavity surface-mount laser (VCSEL) to the input light signal after modulating the polarization ellipticity of the pump light. The corresponding rate equation is expressed as: , , , in, This represents the total carrier density of a spin vertical cavity surface-mount laser emitter. Represents spin carrier density; The amplitude of the complex field representing polarized light; Indicates the light field attenuation rate; Indicates the linewidth enhancement factor; Represents linear dichroism; Indicates linear birefringence; This represents the nonradiative recombination rate of charge carriers; Indicates the bias pump current; Indicates the polarization ellipticity of the input charge carriers; Indicates the modulation depth of the input pulse; Represents a modulated periodic square wave pulse sequence; Indicates the spin relaxation rate; Represents an imaginary number.

[0061] Furthermore, using this system to predict time series data includes steps 1 through 4: Step 1: Use a waveform generator to generate a mask signal, and couple and modulate the time series data to be predicted to obtain control signals for modulating the amplitude and polarization ellipticity of the pump light, thereby modulating the pump light to generate the input light signal.

[0062] Step 2: Inject the input optical signal into the spin vertical cavity surface mount laser emitter (VCSPL), causing the VCSPL to change its total carrier concentration and the ratio of spin-up to spin-down carrier concentrations under the excitation of the input optical signal, forming a spin non-equilibrium carrier distribution. Thus, under the synergistic effect of the total carrier concentration and the spin non-equilibrium carrier distribution, the VCSPL generates a spin-related nonlinear polarization dynamic response to the pulse combinations at different times in the input optical signal, and outputs a gradient response optical signal.

[0063] Step 3: Use a polarization beam splitter to perform polarization component matching and orthogonal polarization beam splitting on the gradient response optical signal, and output multiple polarized optical signals with mutually orthogonal polarization directions.

[0064] Step 4: In the dual-modulation output layer, each polarized light signal is photoelectrically converted to obtain n electrical signals. The n electrical signals are trained to determine the output weights of each polarized light signal, and then the target time series data is predicted based on each polarized light signal and its corresponding output weights.

[0065] like Figure 4 The diagram shows the prediction results of time series forecasting using a dual-modulation gradient reservoir computing system based on spin polarization dynamics, as provided in this application. To highlight the prediction performance of the reservoir computing system under the dual-modulation scheme, this application only presents one set of representative results; among them, Figure 4 (a) in the diagram is a schematic representation of the target value. Figure 4 (b) in the diagram is a schematic of the predicted value, which is obtained by analyzing... Figure 4 Analysis revealed that the predicted time series obtained in this embodiment closely matches the actual time series. The normalized mean square error (NMSE) calculated under amplitude modulation is only 0.0007, and the normalized mean square error (NMSE) under polarization modulation is only 0.0001. This indicates that the dual-modulation gradient reservoir calculation system based on spin polarization dynamics proposed in this application significantly improves the accuracy of data processing by using a novel information modulation method, thereby exhibiting higher precision in tasks such as time series prediction and data recognition.

[0066] In summary, since the polarization ellipticity of the pump light not only changes the external polarization state of the light but also further determines the relative injection ratio of spin-up carriers to spin-down carriers, the modulation object of this application is not simply the output polarization characterization, but the spin non-equilibrium carrier distribution inside the spin vertical-cavity surface-mount laser emitter. This introduces spin degrees of freedom on top of traditional total carrier modulation, giving the reservoir computing system a higher information representation dimension and richer nonlinear dynamic response capability, thus improving the feature unfolding capability in complex temporal tasks and pattern recognition tasks. Specifically, this application adopts a dual-dimensional synergistic control method of pump light power and pump light polarization ellipticity, enabling the change in total carrier concentration and the spin non-equilibrium evolution process to occur coupled within the same device. This allows input information to be simultaneously mapped to the device's intensity response, polarization response, and temporal dynamic response. Compared to schemes relying solely on a single modulation parameter, this application can form richer state trajectories and a higher-dimensional dynamic mapping space within a single spin vertical-cavity surface-mount laser emitter, thus being more conducive to nonlinear projection, feature unfolding, and mode separation of complex input data.

[0067] It should be noted that in conventional electrically injected lasers, the modulation injection current mainly changes the total carrier concentration of the system. Its dynamic response is usually limited by the change in the number of carriers and the relaxation oscillation process. Even in existing spin VCSEL technology, the existing modulation methods focus on the current intensity or output polarization characterization level, without treating the total carrier injection and spin injection imbalance as two distinct and independent control objects. However, the control object of this application is expanded from the traditional single total carrier amount to two-dimensional control of the total carrier concentration and the difference between spin carriers. By utilizing the physical characteristic of Spin-VCSELs that are highly sensitive to spin selective injection, the polarization ellipticity of the pump light is transformed into an effective means of controlling the spin non-equilibrium state inside the device. This allows for the simultaneous control of the total carrier injection and spin injection imbalance, and under the combined action of dual polarization mode coupling, spin flipping, and nonlinear dynamics, a graded dynamic response corresponding to different timing combinations is generated.

[0068] It is also important to note that the dual modulation method in this application differs from the scheme of independently modulating two pump beams before applying them to the laser. This application does not use two independent pump beams carrying different modulation information; instead, it modulates the power and polarization ellipticity parameters of the same pump beam, so that the total carrier injection intensity and spin injection ratio synergistically act on the spin vertical-cavity surface-emitting laser under the same optical field path. Therefore, the two modulation components are not applied in parallel as two optical signals outside the device and then combined to form a composite effect, but are uniformly encapsulated in the same pump beam before entering the device, and together act on the same set of carrier evolution and polarization dynamics processes inside the device.

[0069] Compared to the method of separately modulating two pump beams, the single-pump dual-parameter modulation method of this application has at least the following technical differences: First, the modulation objects are different. The former corresponds to the separate control of two optical channels, while this application corresponds to the coordinated control of the intensity dimension and spin dimension in the same pump beam. Second, the action paths are different. In the former, different modulation information is usually injected through two optical fields separately and then forms a combined effect. In this application, the two types of modulation information coexist in the same pump beam from the input end and act on the internal state variables of the device. Third, the dynamic mechanisms are different. This application does not use the external combination relationship between the two pumps to form a response, but rather uses the coupled control of the total carrier concentration and spin non-equilibrium distribution by the same pump beam to directly form a stronger correlation dynamic mapping relationship inside the device. Therefore, this application has a more compact system structure and is more conducive to forming a continuous and controllable gradient response and high-dimensional feature unfolding capability in terms of state excitation mechanism.

[0070] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied 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.

[0071] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0072] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0073] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0074] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A dual-modulation gradient reservoir computational system based on spin polarization dynamics, characterized in that, include: The dual-modulation input layer is used to modulate the signal to be processed to generate a control signal. The control signal is used to adjust the amplitude and polarization ellipticity of the pump light, and an input light signal containing the information of the signal to be processed is obtained based on the modulated pump light. Specifically, it includes: Pump lasers are used to generate pump light; A waveform generator is used to generate a mask signal. A signal coupling modulator, connected to a waveform generator, is used to couple and modulate the mask signal and the signal to be processed to generate a control signal. An electric amplifier, connected to a signal coupling modulator, is used to amplify control signals; A modulator, connected to the pump laser and an electrical amplifier, is used to adjust the amplitude of the pump light using a control signal; A polarization controller, connected to a modulator and an electrical amplifier, is used to adjust the polarization ellipticity of the pump light using a control signal, and to obtain an input optical signal containing information about the signal to be processed based on the modulated pump light. The photon storage pool layer specifically includes: A spin vertical cavity surface laser emitter, connected to a dual modulation input layer, is used to change the total carrier concentration and the ratio of spin-up to spin-down carrier concentrations under the excitation of the input optical signal, forming a spin non-equilibrium carrier distribution. Thus, under the synergistic effect of the total carrier concentration and the spin non-equilibrium carrier distribution, a spin-related nonlinear polarization dynamic response is generated to the pulse combination at different times in the input optical signal, and the output is a gradient response optical signal that exhibits continuous layered differences in intensity level, polarization distribution, and transient trajectory. A polarization beam splitter, connected to a spin vertical cavity surface laser emitter, is used to perform polarization component matching and orthogonal polarization beam splitting on gradient response optical signals, outputting multiple polarized optical signals with mutually orthogonal polarization directions. A dual-modulation output layer, connected to the photon reservoir layer, is used to acquire polarized light signals and obtain the target output signal of the reservoir computing system based on the multi-path polarized light signals.

2. The dual-modulation gradient reservoir calculation system based on spin polarization dynamics according to claim 1, characterized in that, The dual-modulation input layer also includes: An optical isolator, connected to the pump laser, is used to suppress interference from the return light on the pump laser and improve the output stability of the pump light.

3. The dual-modulation gradient reservoir calculation system based on spin polarization dynamics according to claim 1, characterized in that, The photon storage pool layer also includes: A beam splitter is used to split multiple polarized light signals with mutually orthogonal polarization directions and transmit the split polarized light signals to a dual-modulation output layer.

4. The dual-modulation gradient reservoir calculation system based on spin polarization dynamics according to claim 3, characterized in that, The polarization beam splitter and the beam splitter prism are separate optical devices, with the beam splitter positioned in the optical path between the polarization beam splitter and the dual modulation output layer.

5. The dual-modulation gradient reservoir calculation system based on spin polarization dynamics according to claim 3, characterized in that, The polarization beam splitter and the beam splitter prism are integrated on the same substrate to form a photonic integrated circuit.

6. The dual-modulation gradient reservoir computing system based on spin polarization dynamics according to claim 1, characterized in that, The dual-modulation output layer includes: A photodetector array, connected to a photon reservoir layer, is used to perform photoelectric conversion on various polarized light signals; A programmable gated array, connected to a photodetector array, is used to train the output of the photon reservoir layer, obtain the output weights of each polarized light signal, and thus obtain the target output signal of the reservoir computing system.

7. The dual-modulation gradient reservoir computational system based on spin polarization dynamics according to claim 6, characterized in that, The dual-modulation output layer also includes: The fiber collimator, connected to the photon reservoir layer, is used to collimate various polarized light signals. An optical amplifier, connected to an optical fiber collimator, is used to amplify the aligned polarized light signal and transmit the amplified polarized light signal to a photodetector array.

8. A method for calculating a dual-modulation gradient reservoir based on spin polarization dynamics, characterized in that, The method is implemented based on the dual-modulation gradient reservoir computing system based on spin polarization dynamics as described in any one of claims 1 to 7, and includes: A dual-modulation input layer is used to modulate the signal to be processed to generate a control signal. This control signal is then used to adjust the amplitude and polarization ellipticity of the pump light. Based on the modulated pump light, an input optical signal containing information about the signal to be processed is obtained. Specifically, this includes: Pump light is generated using a pump laser; Use a waveform generator to generate a mask signal; A signal coupling modulator is used to couple and modulate the mask signal and the signal to be processed to generate a control signal for modulating the amplitude and polarization ellipticity of the pump light. The control signal is amplified using an electric amplifier; The amplitude of the pump light is adjusted using a modulator based on a control signal; The polarization ellipticity of the pump light is adjusted by a polarization controller based on a control signal, and an input optical signal containing information of the signal to be processed is obtained based on the modulated pump light. The input optical signal is injected into the spin vertical cavity surface laser emitter, so that the spin vertical cavity surface laser emitter changes the total carrier concentration and the ratio of spin-up carrier concentration to spin-down carrier concentration under the excitation of the input optical signal, forming a spin non-equilibrium carrier distribution. Thus, under the synergistic effect of the total carrier concentration and the spin non-equilibrium carrier distribution, a spin-related nonlinear polarization dynamic response is generated to the pulse combination of the input optical signal at different times, and the output gradient response optical signal exhibits continuous layered differences in intensity level, polarization distribution and transient trajectory. A polarization beam splitter is used to perform polarization component matching and orthogonal polarization beam splitting on the gradient response optical signal, outputting multiple polarized optical signals with mutually orthogonal polarization directions; Polarized light signals are acquired using a dual-modulation output layer, and the target output signal of the reservoir computing system is obtained based on the multi-channel polarized light signals.

Images