Reconfigurable photonic reservoir system based on single micro-ring resonator

Abstract

The present application relates to a kind of reconfigurable photonic reservoir system based on single micro-ring resonator, belong to photonic computing and integrated optics field.For the problems of traditional reservoir node quantity is limited, delay feedback system performance and processing time are mutually restricted, the system is proposed by n signal transmission waveguide, n waveguide coupling region and a ring resonant cavity.The n≥1;Signal transmission waveguide input and output optical signal;Waveguide coupling region couples specific wavelength light to resonant cavity, and changes optical power;Ring resonant cavity transmits optical signal to different coupling region.By detecting the different wavelength optical power of each output port generates virtual node, combined with wavelength division multiplexing to realize reservoir computing.The structure breaks through the inherent contradiction between performance and time, significantly improves the integration potential and system flexibility, only needs single micro-ring to complete high-precision time series prediction and speech classification task, provides efficient solution for optical neural network chip.

Description

Technical Field

[0001] This invention belongs to the field of photonic computing and integrated optics technology, and relates to a reconfigurable photonic water storage system based on a single microring resonator. Background Technology

[0002] With the rapid development of machine learning technology, its progress is highly dependent on significant improvements in computing power, iterative upgrades of dedicated hardware, and the widespread availability of large-scale datasets. However, the explosive growth of data in modern society poses a severe challenge to building larger-scale and more complex machine learning models. Traditional electronic computing hardware faces fundamental constraints in terms of computing efficiency and energy consumption control, and these bottlenecks have become major obstacles to the development of artificial intelligence (AI).

[0003] Optical integration technology, with its inherent advantages in computing speed and bandwidth, offers a new path to overcome the bottlenecks of electronic computing. Currently, optically integrated devices have been applied to construct computing architectures such as artificial neural networks and spiking neural networks, demonstrating great potential in fields such as communication, data processing, and sensing. However, the integration density of optical devices is far lower than that of integrated circuits, and the difficulty of network control and calibration is significantly higher than that of electronic systems, making it difficult to realize large-scale optical neural networks (ONNs). The limited device size directly restricts the number of neurons, thus affecting computing performance.

[0004] Reservoir computing, as a special type of recurrent neural network, can replace physical nodes with virtual nodes, requiring only the training of output weights without updating internal weights, thus significantly reducing the requirements for the scale of physical devices. In recent years, optical reservoir computing has demonstrated high performance in benchmark tasks such as time series prediction and speech recognition. Existing optical reservoir systems mainly employ two implementation schemes:

[0005] Delay-feedback-based structures utilize devices such as asymmetric Mach-Zehnder interferometers to introduce delay feedback via waveguides. These schemes rely on time-division multiplexing to generate virtual nodes, but the number of nodes is positively correlated with the feedback delay, leading to an irreconcilable conflict between computational performance and processing time.

[0006] Spatial reuse schemes: physical nodes are constructed using optical node arrays or light scattering spots. The former is difficult to scale due to device cost, size, and control complexity; the latter requires special spot detection equipment, and faces alignment difficulties and integration challenges.

[0007] In contrast, wavelength division multiplexing (WDM) schemes can avoid the latency issues of time division multiplexing (TDM) and have better integration potential, enabling the implementation of a large number of virtual nodes within a limited size. However, existing WDM reservoir systems still rely on multi-device collaboration and have not yet resolved the core contradiction between structural complexity and scalability. Therefore, there is an urgent need for a novel photonic reservoir structure that can achieve device simplification and large-scale integration while ensuring high computing performance. Summary of the Invention

[0008] In view of this, the purpose of this invention is to provide a reconfigurable photonic reservoir system based on a single microring resonator. This reservoir structure does not use delays to generate virtual nodes, thus eliminating the constraint between computational performance and processing time. The structure itself has multiple spatial nodes, and by combining it with wavelength division multiplexing (WDM) methods, sufficient virtual nodes can be generated, thereby enabling the completion of high-precision, complex tasks.

[0009] To achieve the above objectives, the present invention provides the following technical solution:

[0010] A reconfigurable photonic water storage system based on a single microring resonator includes:

[0011] The system consists of n signal transmission waveguides, n waveguide coupling regions, and a ring resonant cavity; where n ≥ 1.

[0012] The signal transmission waveguide is used for inputting and outputting optical signals;

[0013] The waveguide coupling region connects the signal transmission waveguide and the ring resonant cavity, and is used to couple optical signals of a specific wavelength to the ring resonant cavity or the signal transmission waveguide, and change the signal optical power.

[0014] The ring resonant cavity is used to resonate optical signals of a specific wavelength and transmit the optical signals to different waveguide coupling regions;

[0015] Virtual nodes of the water storage tank are generated by detecting the optical signal power information of different wavelengths in the output ports of each signal transmission waveguide.

[0016] Furthermore, the signal transmission waveguide is a silicon-based single-mode waveguide or multi-mode waveguide, and its cross-section is ridge-shaped or strip-shaped.

[0017] Furthermore, the waveguide coupling region employs one of the following coupling methods:

[0018] Lateral or vertical coupling between silicon-based straight waveguides and ring resonators; lateral or vertical coupling between silicon-based bent waveguides and ring resonators.

[0019] Coupling forms include: coupling between two straight waveguides, single-point or multi-point coupling between a straight waveguide and a curved waveguide, and single-point or multi-point coupling between two curved waveguides.

[0020] Furthermore, the ring resonant cavity has a circular, racetrack-shaped, or irregularly shaped structure, and its waveguide is a silicon-based single-mode waveguide or multi-mode waveguide with a strip-shaped or ridge-shaped cross-section.

[0021] Furthermore, when multi-point coupling is used, the lengths of the waveguide sections at different coupling points may be the same or different.

[0022] A calculation method for a photon-based water storage tank based on the system includes the following steps:

[0023] (1) Encode the input signal u(t) and keep the signal value u(nΔt) within the time interval nΔt≤t<(n+1)Δt to obtain the discrete signal u(n);

[0024] (2) Input u(n) into the ring resonant cavity through the signal transmission waveguide. After resonance between the waveguide coupling region and the ring resonant cavity, the optical signal is obtained from the output end of each signal transmission waveguide.

[0025] (3) Detect the output ports at wavelength λ k Optical power values ​​at k = 1, 2, ..., N Where i represents the output port number;

[0026] (4) Virtual nodes are generated based on optical power values, and the reservoir output is calculated using weighted calculation:

[0027]

[0028] in, These are the output weights obtained through ridge regression training.

[0029] Furthermore, the time interval Δt in step (1) is dynamically adjusted according to the task requirements.

[0030] Furthermore, in step (3), virtual nodes are generated by combining wavelength division multiplexing with spatial nodes, and the total number of nodes is n×N.

[0031] The beneficial effects of this invention are as follows:

[0032] (1) By abandoning the delayed feedback structure and using the wavelength dimension to replace the time dimension to generate virtual nodes, the contradiction between the number of nodes and the processing time in traditional reservoir computing is completely resolved. The system's computing performance is no longer limited by time delay and can process high-density tasks in parallel.

[0033] (2) The single micro-ring architecture, combined with multi-port spatial nodes, enables an order-of-magnitude expansion of virtual nodes within a finite physical structure. A complete reservoir system can be constructed with only a single resonant cavity, significantly reducing device complexity and fabrication difficulty, and providing a feasible path for large-scale photonic integrated circuits.

[0034] (3) The virtual node topology is dynamically reconstructed by wavelength tuning to achieve adaptive configuration of reservoir parameters. The resonant response at different wavelengths provides a natural nonlinear transformation, which can complete complex feature mapping without the need to introduce additional nonlinear devices.

[0035] (4) In typical tasks such as time series prediction and speech classification, the system only requires basic optical components to achieve an accuracy level comparable to electronic computing, which verifies its application value in low power consumption and high real-time scenarios.

[0036] (5) Silicon-based processes are compatible with standard semiconductor manufacturing processes, and the simplified structure effectively avoids the challenges of multi-device alignment and calibration. The system is highly stable, insensitive to environmental disturbances, and significantly improves the feasibility of industrial deployment.

[0037] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0038] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein:

[0039] Figure 1 This is a schematic diagram of the structure of the present invention;

[0040] Figure 2 This is a schematic diagram illustrating the working principle of the present invention;

[0041] Figure 3 This is a schematic diagram of an embodiment of the reconfigurable photonic water storage tank based on a single microring resonator according to the present invention;

[0042] Figure 4 This is an experimental spectrum diagram of an embodiment of the reconfigurable photonic water storage tank based on a single microring resonator according to the present invention;

[0043] Figure 5 Comparison of the predicted waveform and the original waveform for Mackey-glass time series;

[0044] Figure 6 A plot of mean square error between the predicted waveform and the original waveform for a Mackey-glass time series;

[0045] Figure 7 The confusion matrix diagram for performing a classification task on a public dataset of Japanese vowels. Detailed Implementation

[0046] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0047] The accompanying drawings are for illustrative purposes only and are schematic diagrams, not actual pictures. They should not be construed as limiting the invention. To better illustrate the embodiments of the invention, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0048] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," "front," and "rear" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present invention. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0049] Figure 1 This is a schematic diagram of the present invention. Figure 2 This is a schematic diagram illustrating the working principle of the present invention. The working principle is as follows: First, u(t) is encoded so that its value remains u(nΔt) during the time interval nΔt ≤ t < (n+1)Δt, thus obtaining u(n). u(n) is then input to the photonic reservoir via signal transmission waveguide 1. Waveguides 1 through n serve as spatial nodes of the reservoir. Wavelength division multiplexing (WDM) is used to extract power information at different wavelengths from the outputs of the n transmission waveguides, which are then used as virtual nodes to calculate the final output. Nonlinear computation is provided during optical power detection.

[0050] Figure 3A schematic diagram of an embodiment of the reconfigurable photonic reservoir based on a single microring resonator of the present invention is provided. The signal transmission waveguide is a single-mode waveguide with a strip-shaped cross-section. Waveguide coupling region 1 is a single-point lateral coupling between a straight waveguide and a ring resonator. Waveguide coupling regions 2, 3, and 4 are two-point lateral couplings between a curved waveguide and a ring resonator. The optical signal is input from the signal transmission waveguide, and there are optical signal outputs at the straight-through end of the microring and the three download ends. The power transfer function of the straight-through end and the three download ends is calculated using the transfer matrix method, and the results are shown below.

[0051]

[0052] In the formula T t Let T be the power transfer function at the through end. d1 T d2 and T d3 These are the power transfer functions for the three download ends. For the cyclic phase shift, k1 is the coupling efficiency of coupling region 1, t1 is the transmission efficiency of coupling region 1, k2 and t2 are the coupling efficiency and transmission efficiency of coupling region 2, and so on. From the power transfer function above, it can be seen that when an optical signal is input, the output of the four ports is determined by the coupling efficiency and transmission efficiency of the four coupling regions, and the output values ​​are different for each. Therefore, different output ports can be used as nodes of the photon reservoir, and power values ​​of different wavelengths extracted from different output ports can be used as virtual nodes of the reservoir. When the input signal u(n) is input into the photon reservoir, assuming N nodes are extracted at each output port, then in this embodiment, the total number of nodes in the photon reservoir is 4N, and the total output is as described in the following formula.

[0053]

[0054] In the formula, y(n) represents the total output of the reservoir. For the i-th output port, λ k Sampling node at wavelength, These are the output weights trained using ridge regression.

[0055] Figure 4 The power transmission spectra of the four ports obtained in the experiment of this embodiment of the invention are shown. To verify the performance of the photonic reservoir implemented in this embodiment, a Mackey-glass time series (time delay of 17 bits) prediction task was performed one step ahead using these transmission spectra. The Mackey-glass time series u(n) (n=1000) was input into the photonic reservoir, and the output power of 10 wavelengths at equal intervals in the wavelength range of 1558nm to 1560nm at each port was sampled to obtain the power transmission spectrum. And use ridge regression to train the output weights Used to calculate the predicted signal waveform. Comparison of the predicted signal waveform with the original waveform. Figure 5 As shown. The mean square error between the predicted signal waveform and the original waveform is as follows. Figure 6 As shown. By Figure 6 It can be seen that the mean square error is basically around 10. -6 The order of magnitude, with a maximum value of approximately 2.5 × 10⁻⁶. -6 The mean square error of the full data is 0.0013, and the normalized mean square error is 0.0233, which shows that the photonic water storage scheme can predict Mackey-glass time series waveforms with high accuracy.

[0056] To verify the performance of the photonic reservoir computing system in performing classification tasks, we used the Japanese vowel dataset as a benchmark set. This dataset is a typical multivariate multi-classification task dataset, and its core features consist of the first 12 Mel-frequency cepstral coefficients (MFCCs) generated by nine different speakers when they pronounce the Japanese diphthong "ae". The classification goal is to accurately identify the corresponding speaker based on the MFCC features.

[0057] In the field of speech signal processing, MFCC features are widely used in speech recognition systems due to their excellent representational capabilities. The extraction process strictly follows the characteristics of human auditory perception, with the following steps: First, the power spectrum of the audio signal is calculated; then, the spectrum is graded using a Mel-scale filter bank; next, the logarithm of the energy in each Mel frequency band is taken; finally, the cepstral coefficients are obtained through discrete cosine transform. In this test, the MFCC feature sequences of all speech samples underwent frame averaging and normalization. The dataset contains 270 training samples (9 speakers, each providing 30 pronunciation samples) and 370 test samples (the same speakers as in the training samples, each providing 24–88 samples).

[0058] The preprocessed MFCC feature sequence is used as input to the photon reservoir, and the same method as in predicting the Mackey-glass time series is employed to obtain... Classification weights were obtained using ridge regression on the training set. Use this weight to classify each utterance in the test set. Figure 7 The classification results of the test set are presented in the form of a confusion matrix. We normalized the data to more clearly illustrate the classification accuracy. The diagonal elements show the percentage of correctly classified utterances for each speaker, while the off-diagonal elements show the percentage of incorrectly classified utterances. The calculated overall classification accuracy reached 87.3%, demonstrating that the reservoir system can complete the classification task with high precision.

[0059] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A photonic reservoir computing method for a single micro-ring resonator based reconfigurable photonic reservoir system, characterized by: Includes the following steps: S1: encode an input signal u ( t ) in a time interval n Δ t ≤ t <( n+ 1)Δ t , and keep the signal value u ( n Δ t ) in u ( n ) to obtain a discrete signal S2: Will u ( n The signal is input to the ring resonant cavity through the signal transmission waveguide. After resonance between the waveguide coupling region and the ring resonant cavity, the optical signal is obtained from the output end of each signal transmission waveguide. S3: detecting the optical power values at the wavelengths , k = 1,2,… N at each output port wherein i denotes the output port number; S4: Generate virtual nodes based on optical power values, and calculate the reservoir output through weighted calculation: wherein, is the output weight obtained by ridge regression training; The reconfigurable photonic water storage system based on a single microring resonator includes: n individual signal transmission waveguides, n A waveguide coupling region and a ring resonant cavity; wherein... n ≥1; The signal transmission waveguide is used for inputting and outputting optical signals; The waveguide coupling region connects the signal transmission waveguide and the ring resonant cavity, and is used to couple optical signals of a specific wavelength to the ring resonant cavity or the signal transmission waveguide, and change the signal optical power. The annular resonant cavity is used to resonate optical signals of a specific wavelength and transmit the optical signals to different waveguide coupling regions; the waveguide coupling regions adopt one of the following coupling methods: Lateral or vertical coupling between silicon-based straight waveguides and ring resonators; lateral or vertical coupling between silicon-based bent waveguides and ring resonators. Coupler types include: coupling between two straight waveguides, single-point or multi-point coupling between a straight waveguide and a curved waveguide, and single-point or multi-point coupling between two curved waveguides. Virtual nodes of the reservoir are generated by detecting the optical signal power information of different wavelengths in the output ports of each signal transmission waveguide; the structure of the reservoir does not use delay to generate the virtual nodes.

2. The calculation method for a photon reservoir according to claim 1, characterized in that: The signal transmission waveguide is a silicon-based single-mode waveguide or multi-mode waveguide, and its cross-section is ridge-shaped or strip-shaped.

3. The calculation method for a photon reservoir according to claim 1, characterized in that: The ring resonant cavity is circular, racetrack-shaped, or irregularly shaped, and its waveguide is a silicon-based single-mode waveguide or multi-mode waveguide with a strip-shaped or ridge-shaped cross-section.

4. The calculation method for a photon reservoir according to claim 1, characterized in that: When multi-point coupling is used, the lengths of the waveguide sections at different coupling points may be the same or different.

5. The calculation method for a photon reservoir according to claim 1, characterized in that: The time interval Δ in S1 t Adjust dynamically according to task requirements.

6. The calculation method for a photon reservoir according to claim 1, characterized in that: In S3, virtual nodes are generated by combining wavelength division multiplexing with spatial nodes, and the total number of nodes is [number missing]. n × N .

Images