A reservoir computing device based on optically pumped spin VCSEL

Abstract

The present invention discloses a reservoir computing device based on optically pumped spin VCSEL, comprising an input layer, a reservoir and an output layer, wherein the input layer comprises a first driving laser, a second driving laser, a first arbitrary waveform generator, a second arbitrary waveform generator, a first modulator and a second modulator, the reservoir comprises an optically pumped spin VCSEL with an optical feedback loop for nonlinear mapping of input signals, and the output layer comprises two photoelectric converters and a dual-channel oscilloscope. The beneficial effects of the present invention are as follows: the polarization mode is stable and easy to control, and an optically pumped spin VCSEL with a highly stable and easy-to-control spin polarization mode is used as a reservoir, so that the polarization mode of the reservoir is not easily affected by the working environment without affecting the task performance of the reservoir device, and exhibits stable performance compared to an ordinary VCSEL reservoir.

Description

Technical Field

[0001] The invention relates to the technical fields of photon reservoir computing, optics, communication and spin electronics, and in particular to a reservoir computing device based on optically pumped spin VCSEL. Background Art

[0002] Artificial neural networks (ANNs) are brain-like computing structures that were created to cope with today's rapidly growing amount of information. After decades of continuous development and improvement, they have become a machine learning method that cannot be ignored and has great potential. Reservoir computing (RC) is derived from recurrent neural networks (RNNs) in ANNs, which helps to overcome the limitations of RNNs, such as training difficulties and complex structures. However, the large number of nonlinear nodes required for spatial RC still leads to high control and adjustment costs, which greatly limits its development and application.

[0003] In recent years, researchers have proposed using the dynamic response of a nonlinear system under a single delayed feedback as a virtual node to construct a time-delay RC, which greatly reduces the cost and difficulty of RC implementation. Semiconductor lasers that can exhibit rich nonlinearities under delayed feedback have also attracted extensive attention and research in the construction of RC due to their advantages such as fast speed, high bandwidth, and parallel processing. Among them, vertical-cavity surface-emitting semiconductor lasers (VCSELs) occupy an important position in the field of semiconductor laser RC due to their advantages such as high energy efficiency, low manufacturing cost, and rich polarization dynamics. They can effectively improve system processing speed, enhance performance, and achieve parallel processing. For example, the scheme of using a single VCSEL polarization dynamics to construct a reservoir using a dual-polarization mode RC formed by using a VCSEL with rotational feedback has been shown to have better performance than traditional single-mode RC (see [J.Vatin,D.Rontani,and M.Sciamanna,“Enhanced performance of areservoir computer using polarization dynamics in VCSELs,”Opt.Lett.43(18),4497–4500(2018).]); the scheme of using two mutually coupled VCSEL polarization dynamics using a four-channel RC (see [X.Guo,S.Xiang,Y.Zhang,et al.“Four-channels reservoir computing based onpolarization dynamics in mutually coupled VCSELs system,”Opt.Express.27(16),23293-23306(2019)]); the scheme of using VCSEL-based RC polarization dynamics to process two tasks in parallel (see [X.Guo,S.Xiang,Y.Zhang,et al.“Polarization multiplexing reservoir computingbased on a VCSEL with polarized optical feedback,” in IEEE Journal of SelectedTopics in Quantum Electronics, vol.26, no.1, pp.1-9, Jan.-Feb.2020, Artno.1700109]). However, the ordinary VCSELs used in these schemes have problems such as unstable polarization mode and difficult polarization control, which restricts the stability and performance of RC work. Summary of the invention

[0004] The object of the present invention is to provide a reservoir computing device based on optically pumped spin VCSEL to solve the problems raised in the above background technology.

[0005] To achieve the above-mentioned object, the present invention provides the following technical solutions: a reservoir computing device based on optically pumped spin VCSEL, comprising an input layer, a reservoir and an output layer;

[0006] The input layer includes a first driving laser, a second driving laser, a first arbitrary waveform generator, a second arbitrary waveform generator, a first modulator and a second modulator;

[0007] The reservoir includes an optically pumped spin VCSEL with an optical feedback loop for nonlinearly mapping the input signal;

[0008] The output layer includes two photoelectric converters and a dual-channel oscilloscope.

[0009] Preferably, the first arbitrary waveform generator and the second arbitrary waveform generator are used to respectively generate two pre-processed input signals S + (t) and S - (t).

[0010] Preferably, the two input signals S + (t) and S - (t) respectively modulating the polarized output light I of the first driving laser and the second driving laser by the first modulator and the second modulator + (t), I - (t), two paths of modulated light are formed and injected into the storage pool at the same time.

[0011] Preferably, the optically pumped spin VCSEL output has two modes, left-handed polarization and right-handed polarization, which are respectively affected by the injection of two polarization modulated lights generated by the input layer.

[0012] Preferably, the optically pumped spin VCSEL generates rich dynamic responses in the two polarization modes after passing through the optical feedback loop, thereby converting the input signal S + (t) and S - (t) Perform nonlinear mapping processing.

[0013] Preferably, the photoelectric converter is used to convert the optical signal into an electrical signal; the oscilloscope outputs E in the two spin polarization modes of the storage cell. + (t) and E - (t) Read separately to obtain the total response status and sample two groups of virtual nodes.

[0014] Preferably, the two groups of virtual nodes are multiplied by corresponding output weights to obtain the output result of the optically pumped spin VCSEL reserve pool, and the output weights are optimized by minimizing the mean square error between the target value and the RC output calculated by the reserve pool using a linear regression algorithm or a ridge regression algorithm.

[0015] Preferably, the two different spin polarization modes in the optically pumped spin VCSEL respectively handle two tasks simultaneously.

[0016] Preferably, the first modulator and the second modulator are phase modulators or intensity modulators, and modulation can be performed directly using the output of an arbitrary waveform generator instead of a modulator.

[0017] Preferably, the optical feedback loop is in the form of all-optical feedback or photoelectric feedback, and optical transmission adopts optical fiber or spatial optical transmission.

[0018] Beneficial Effects

[0019] The reservoir computing device based on optically pumped spin VCSEL provided by the present invention has a stable polarization mode and is easy to control. An optically pumped spin VCSEL with a highly stable and easily controllable spin polarization mode is used as a reservoir. Under the condition that the task performance of the reservoir device is not affected, the polarization mode of the reservoir is not easily affected by the working environment, and exhibits stable performance compared to ordinary VCSEL reservoirs. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a schematic diagram of the overall structure of the present invention;

[0021] Figure 2 A diagram of the prediction target and results for a single time series prediction task processed by the reserve pool system of the present invention;

[0022] Figure 3 This is a graph showing the results of the reserve pool system of the present invention processing the time series prediction and waveform recognition tasks in parallel. DETAILED DESCRIPTION

[0023] The following are specific embodiments of the present invention and the accompanying drawings to further describe the technical solution of the present invention, but the present invention is not limited to these embodiments.

[0024] Example

[0025] like Figure 1-3 As shown, a reservoir computing device based on optically pumped spin VCSEL includes an input layer, a reservoir and an output layer;

[0026] The input layer includes a first driving laser, a second driving laser, a first arbitrary waveform generator, a second arbitrary waveform generator, a first modulator and a second modulator;

[0027] The reservoir includes an optically pumped spin VCSEL with an optical feedback loop to process the input signal nonlinearly;

[0028] The output layer includes two optical-to-electrical converters and a two-channel oscilloscope.

[0029] Preferably, the first arbitrary waveform generator and the second arbitrary waveform generator are used to generate two pre-processed input signals S + (t) and S - (t).

[0030] Preferably, the two input signals S + (t) and S - (t) The polarized output light I of the first driving laser and the second driving laser modulated by the first modulator and the second modulator, respectively. + (t), I - (t), two paths of modulated light are formed and injected into the storage pool at the same time.

[0031] Preferably, the optically pumped spin VCSEL output has two modes, left-handed polarization and right-handed polarization, which are respectively affected by the injection of two polarization modulated lights generated by the input layer.

[0032] Preferably, the optically pumped spin VCSEL generates rich dynamic responses in the two polarization modes after passing through the optical feedback loop, thereby converting the input signal S + (t) and S - (t) Perform nonlinear mapping processing.

[0033] Preferably, the photoelectric converter is used to convert the optical signal into an electrical signal; the oscilloscope outputs E in the two spin polarization modes of the storage cell. + (t) and E - (t) Read separately to obtain the overall response status and sample two groups of virtual nodes.

[0034] Preferably, two groups of virtual nodes are multiplied by corresponding output weights to obtain the output result of the optically pumped spin VCSEL reservoir, and the output weights are optimized by minimizing the mean square error between the target value and the RC output calculated by the reservoir using a linear regression algorithm or a ridge regression algorithm.

[0035] Preferably, two different spin polarization modes in the optically pumped spin VCSEL respectively handle two tasks simultaneously.

[0036] Preferably, the first modulator and the second modulator are phase modulators or intensity modulators, and modulation can be performed directly using the output of an arbitrary waveform generator instead of a modulator.

[0037] Preferably, the optical feedback loop is in the form of all-optical feedback or photoelectric feedback, and the optical transmission adopts the form of optical fiber or spatial optical transmission.

[0038] In this specification, numerical simulation is used to simulate the performance of the optically pumped spin VCSEL reservoir computing device in processing the Santa Fe time series prediction task alone and the Santa Fe time series prediction and waveform recognition tasks simultaneously, verifying the single-task processing and parallel processing capabilities of the present invention.

[0039] For the optically pumped spin VCSEL reservoir calculation device, the rate equation is established as follows:

[0040]

[0041] Wherein, the subscripts + and - represent the right-hand polarization mode and the left-hand polarization mode, respectively. E represents the complex amplitude of the electric field of the optically pumped spin VCSEL, N is the total carrier density, m is the normalized carrier density difference, and E inj represents the injection effect of the driving laser, κ is the light field attenuation rate, α is the line width enhancement factor, and γ a and γ p They represent the linear dispersion effect and birefringence effect of the active region medium, γ is the carrier decay rate, and γ s is the spin reversal rate, η ± They represent the normalized optical pumping amount corresponding to the two polarizations, and the total optical pumping amount and the elliptic polarization degree are defined as η = η + +η - and P = (η + -η - ) / (η + +η - ), β is the spontaneous radiation noise factor, ξ ± is Gaussian white noise, k f and τ represent the feedback strength and delay time respectively, k inj represents the injection intensity, E0 is the injection intensity of the driving laser, and △f is the optical injection frequency detuning. ± (t) = u ± (n)×mask(t), where u ± (n) is the input corresponding to the task, and mask(t) is the mask signal. The values ​​of the parameters in the simulation are: κ = 230ns -1 ; α=4; γ=0.68ns -1 ; γ a =0;γ p =8.8*πns -1 ; P = 0.1; η = 1.5; γ s =30ns -1;Φ0=0; |E0|=1.5; △f=-20GHz; τ=1ns; k f =30ns -1 ;k inj =40ns -1 ;θ=10ps;β=1*10 -6 .

[0042] In the input layer, let the input u ± (n) are the same and correspond to the same task. The virtual node interval θ is set to 10ps. The response of the optical pump spin VCSEL in the two spin polarization modes in the feedback loop is sampled to obtain a total of 2τ / θ virtual nodes and use them for calculation, which can process a single task. Figure 2 The target, prediction result and error of a single time series prediction task are shown using the optically pumped spin VCSEL reservoir computing device. Figure 2 It can be seen that the prediction results are very close to the prediction targets, and the calculated normalized mean square error NMSE value is 0.002629, indicating that the optically pumped spin VCSEL reservoir computing device has good prediction performance.

[0043] In the input layer, let the input u ± (n) Corresponding to two different tasks, the responses of the optically pumped spin VCSEL in the two spin polarization modes are sampled separately and used for the calculation of the two tasks respectively, so that the two tasks can be processed in parallel. Figure 3 The figure shows the prediction target, prediction result, input waveform, classification target and result when the optically pumped spin VCSEL reservoir computing device is used to parallelly process the Santa Fe time series prediction and waveform recognition tasks. Figure 3 It can be seen that the time series prediction results are close to the target and the results of classifying different waveforms are consistent with the target. The normalized mean square error NMSE values ​​corresponding to the two tasks are 0.043326 and 0.040147 respectively, indicating that the optically pumped spin VCSEL reservoir computing device has good parallel computing capabilities.

[0044] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the invention content within the protection scope of the present invention.

Claims

1. A reservoir computing device based on optically pumped spin VCSEL, comprising an input layer, a reservoir and an output layer, characterized in that: The input layer includes a first driving laser, a second driving laser, a first arbitrary waveform generator, a second arbitrary waveform generator, a first modulator and a second modulator; The reservoir includes an optically pumped spin VCSEL with an optical feedback loop for nonlinearly mapping the input signal; The output layer includes two photoelectric converters and a dual-channel oscilloscope; The first arbitrary waveform generator and the second arbitrary waveform generator are used to generate two pre-processed input signals S + (t) and S - (t); The two input signals S + (t) and S - (t) respectively modulating the polarized output light I of the first driving laser and the second driving laser by the first modulator and the second modulator + (t), I - (t), forming two paths of modulated light and injecting them into the storage pool at the same time; The optically pumped spin VCSEL output has two modes, left-handed polarization and right-handed polarization, which are respectively affected by the two-path polarization modulated light injection generated by the input layer; The optically pumped spin VCSEL generates rich dynamic responses in two polarization modes after passing through the optical feedback loop, thereby converting the input signal S + (t) and S - (t) performing nonlinear mapping processing; The photoelectric converter is used to convert the optical signal into an electrical signal; the oscilloscope outputs E in the two spin polarization modes of the storage cell. + (t) and E - (t) Reading is performed separately to obtain the overall response status and sample two groups of virtual nodes; The two groups of virtual nodes are multiplied by the corresponding output weights to obtain the output result of the optically pumped spin VCSEL reservoir. The output weights are optimized by minimizing the mean square error between the target value and the RC output calculated by the reservoir using a linear regression algorithm or a ridge regression algorithm.

2. The reservoir computing device based on optically pumped spin VCSEL according to claim 1, characterized in that: The two different spin polarization modes in the optically pumped spin VCSEL respectively handle two tasks simultaneously.

3. The reservoir computing device based on optically pumped spin VCSEL according to claim 1, characterized in that: The first modulator and the second modulator are phase modulators or intensity modulators, or the arbitrary waveform generator output is directly used for modulation without using a modulator.

4. The reservoir computing device based on optically pumped spin VCSEL according to claim 1, characterized in that: The optical feedback loop is in the form of all-optical feedback or photoelectric feedback, and optical transmission adopts optical fiber or spatial optical transmission.

Images