Dispersion engineering method of ultra-flat soliton microcavity optical comb

Abstract

The invention discloses a dispersion engineering method of an ultra-flat soliton micro-cavity optical comb, which belongs to the technical field of optical frequency comb generation and comprises the following steps: S1, determining geometric structure parameters related to optimization of micro-ring cavity integrated dispersion; s2, constructing a data set required by neural network training; s3, establishing a full-connection neural network model; s4, predicting the geometric structure parameters of the micro-ring cavity for the target integrated dispersion curve through the model in the S3, the method is suitable for different photon integration platforms and micro-cavity structures, the actual micro-nano preparation process can be guided, the neural network model is used, the micro-ring cavity structure parameters are directly obtained through the target dispersion curve, and the accuracy of the micro-ring cavity structure parameters is improved. The complexity of parameter optimization is reduced, and the training time of the model is greatly shortened by preprocessing the data set.

Description

Technical Field

[0001] The present invention belongs to the technical field of optical frequency comb generation, and more specifically, particularly relates to a dispersion engineering method for an ultra-flat soliton microcavity optical comb. Background Art

[0002] Soliton micro-ring cavity optical comb (abbreviated as micro-comb) is based on the principle of Kerr nonlinear effect in micro-ring cavity. Compared with traditional mode-locked laser optical comb and fiber optical comb, it has the advantages of large comb tooth spacing, wide spectral range, miniaturization and portability. In the past 10 years, it has attracted extensive attention from researchers at home and abroad. This technology requires the establishment of a balance between gain and loss and a balance between dispersion and nonlinearity inside the optical micro-ring cavity. With the deepening of research, different forms of soliton micro-combs have gradually been experimentally confirmed, such as dark solitons and Raman solitons. At the same time, soliton micro-combs have been able to be realized on multiple photonic platforms and have shown attractive prospects in many fields.

[0003] Since there is a complex nonlinear relationship between the spectral envelope of the soliton microcomb and the dispersion characteristics of the microring cavity, the realization of a flat soliton microcomb has strict requirements on the dispersion of the microring cavity. Studies have shown that microring cavities with near-zero anomalous dispersion are more likely to be excited to produce flat soliton microcombs. The dispersion characteristics of a microring cavity are mainly determined by its material properties and geometric structure. The material of a microring cavity is relatively fixed, so its dispersion is usually controlled by adjusting the geometry of the microring cavity. This process is called dispersion engineering. Dispersion engineering usually uses a forward design method.

[0004] The genetic algorithm reported so far obtains the dispersion characteristics of the photonic crystal micro-ring cavity through inverse design, but it still needs to design the photonic crystal micro-ring cavity structure through forward design to match the optimized dispersion curve. Although this method can be extended to other photonic platforms, it is limited to the photonic crystal micro-ring cavity structure and is difficult to use for other micro-cavity structures, such as the commonly used micro-ring cavity. Summary of the invention

[0005] The main purpose of the present invention is to provide a dispersion engineering method for an ultra-flat soliton microcavity optical comb, aiming to solve the existing technical problems.

[0006] To achieve the above purpose, the technical solution adopted by the present invention is as follows:

[0007] A dispersion engineering method for an ultra-flat soliton microcavity optical comb comprises the following steps:

[0008] S1. Determine the geometric structure parameters involved in optimizing the integrated dispersion of the micro-ring cavity;

[0009] S2, build the data set required for neural network training;

[0010] S3, establish a fully connected neural network model;

[0011] S4. Predict the geometric parameters of the micro-ring cavity by integrating the dispersion curve for the target through the model of S3.

[0012] Preferably, the steps for acquiring the data set in S2 are as follows: combining the scanning ranges and corresponding step sizes of the optimized geometric structure parameters to obtain a preliminary data set; and screening the preliminary data set to obtain the data set required for training.

[0013] Preferably, the data set in S2 is divided into a test set and a training set in a ratio of 85:15.

[0014] Preferably, the model parameters need to be repeatedly adjusted during the establishment of the neural network model in S3, and the model parameters include model structure, learning rate, activation function and optimizer.

[0015] As a preferred method, the specific steps for screening the data in S2 are as follows:

[0016] S20, according to the objective function, obtain the transverse electric mode of the z-cut micro-ring cavity, which corresponds to the material refractive index of the e-light

[0017]

[0018] Where λ is the wavelength of light in micrometers;

[0019] S21. The physical field frequency scanning solution obtains the mode number μ of the micro-ring cavity at this frequency point. Through post-processing, the integrated dispersion (D int ) curve, the formula is as follows:

[0020]

[0021] Among them, ω μ is the angular frequency under different mode numbers, ω0 is the angular frequency of the pump light, and D1 / (2π) is the free spectrum range of the micro-ring cavity;

[0022] S22. Distorted data are found and deleted from the preliminary data set in S2.

[0023] As a preferred method, the data in the training data set needs to be preprocessed. The specific processing steps are as follows: In the neural network model, the input data is D int The integrated dispersion set of the thin-film lithium niobate micro-ring cavity after scaling the range of / (2π) to [-1,+1] GHz. The output data is the geometric feature set of the thin-film lithium niobate micro-ring cavity after logarithmization to the base 10.

[0024] As an advantage, it also includes the steps of:

[0025] S5. Use the open source solver pyLLE to simulate the generation of soliton micro-combs in the micro-ring cavity obtained in S4.

[0026] Preferably, pyLLE in S5 uses a split-step Fourier algorithm, as shown in the following formula:

[0027]

[0028] Among them, E(t,τ) and E in They represent the light field circulating in the microring cavity and the amplitude of the incident light, respectively. t is the slow-changing time, τ is the fast-changing time, and t R is the time for light to circulate in the cavity, γ is the Kerr nonlinear coefficient, α is the loss of light in the microring cavity, δ0 is the detuning between the mode angular frequency and the pump light angular frequency, and θ is the coupling coefficient between the straight waveguide and the microring cavity.

[0029] The present invention has the following beneficial effects:

[0030] 1. The present invention starts with parameter optimization design, and reversely designs the micro-ring cavity by using neural networks and near-zero target dispersion curves, obtains multiple micro-ring cavity geometric parameters that meet the requirements, and predicts ultra-flat soliton micro-combs. Therefore, the method proposed in the present invention is applicable to different photonic integration platforms and micro-cavity structures, and can guide the actual micro-nano preparation process;

[0031] 2. Compared with the current mainstream design idea - the inverse design process through genetic algorithm, it should be understood that it obtains a better integrated dispersion curve through reverse iteration based on the target optical comb spectrum.

[0032] The calculation time of this algorithm is long, which will lead to a slower convergence speed under the premise of large data volume. In addition, in the subsequent process, it is necessary to adopt forward design to match the optimized integrated dispersion by adjusting the photonic crystal micro-ring cavity structure, so the scope of application is narrow. The present invention uses a neural network model to directly obtain the micro-ring cavity structure parameters through the target dispersion curve, which reduces the complexity of parameter optimization, and greatly shortens the training time of the model by preprocessing the data set.

[0033] In addition to the advantages described above, the present invention has other purposes, features and advantages. The present invention will be further described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] The drawings constituting a part of this application are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0035] Figure 1Schematic diagram of the structure of the thin film lithium niobate micro-ring cavity used in the embodiment of the present invention;

[0036] Figure 2 Schematic diagram of the neural network model used in the present invention and its input and output data; from left to right are respectively an example of a dispersion curve input into the neural network model, a structural schematic diagram of the neural network model, micro-ring cavity structural parameters output by the neural network model, and a flat dispersion curve calculated by the structure;

[0037] Figure 3 It is a workflow diagram in the present invention;

[0038] Figure 4 are the relevant results obtained after the neural network used in the examples of the present invention is trained; (a) uses the mean square error value of the test set to measure the accuracy of the neural network model prediction; (b) uses the R 2 The value measures the accuracy of the neural network model prediction; (c) Figure shows a specific thin film lithium niobate micro-ring cavity structure (R = 110μm, W = 1.3μm, H = 0.7μm, θ = 60°, H slab =0.2μm), as the model training progresses, the predicted dispersion curve gradually approaches the inherent dispersion curve of the structure;

[0039] Figure 5 These are two near-zero dispersion curves obtained using a random number generation method in step S4 of the present invention.

[0040] Figure 6 It is the integrated dispersion curve of 8 thin-film lithium niobate micro-ring cavity structures in the frequency range of 185 to 205 THz predicted by using the input near-zero dispersion curve in the example of the present invention;

[0041] Figure 7 is the spectral curve of the flat soliton micro-comb obtained by adjusting the simulation parameters in step S5 of the present invention;

[0042] Figure 8 In the 8 thin-film lithium niobate micro-ring cavity structures obtained in step S4 of the present invention, the film thickness H and the unetched depth H of structure Str.3 are slab Fine-tune the relevant results; Figure (a) shows the H and H of structure Str.3. slab D obtained after fine-tuning the parameters int Curve changes; Figure (b) shows the soliton micro-comb results obtained under specific pump parameters when H increases by 20nm. Specific embodiments

[0043] The specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0044] The technical solutions in the embodiments of the present invention will be described below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0045] The present invention discloses a dispersion engineering method for an ultra-flat soliton microcavity optical comb, comprising the following steps:

[0046] S1. Determine the geometric structure parameters involved in optimizing the integrated dispersion of the micro-ring cavity;

[0047] S2, build the data set required for neural network training;

[0048] S3, establish a fully connected neural network model;

[0049] S4, predicting the geometric structure parameters of the micro-ring cavity by integrating the dispersion curve of the target through the model of S3;

[0050] S5. Use the open source solver pyLLE to simulate the generation of soliton micro-combs in the micro-ring cavity obtained in S4.

[0051] In this example, steps S1-S5 are executed to obtain an extremely flat soliton micro-comb with a center frequency of 193.4 THz, a comb power difference of less than 0.01 dB, and a spectral range covering 20 THz.

[0052] The design method described in the present invention is further described below in combination with a z-cut thin-film lithium niobate photonic platform and a thin-film lithium niobate micro-ring cavity;

[0053] Step S1: The determined geometric feature range and interval step information of the thin film lithium niobate micro-ring cavity are described as follows: and the structure of the micro-ring cavity is as shown in the attached Figure 1 As shown, the radius R ranges from 30 to 130 μm, with a step interval of 20 μm; the top width W ranges from 1 to 2 μm, with a step interval of 0.1 μm; the film thickness H ranges from 0.5 to 0.8 μm, with a step interval of 0.05 μm; the unetched depth H slab The range is 0.2-0.3 μm, and the step interval is 0.05 μm; the range of the side wall inclination angle θ is 50°-70°, and the step interval is 5°. By scanning the above parameters, the total number of micro-ring cavities to be calculated is 6930.

[0054] Step S2: Based on the z-cut thin-film lithium niobate photonic platform, simulate and calculate the integrated dispersion characteristics of the micro-ring cavity. Figure 1In the figure, the thickness of the lower cladding SiO2 is 2μm, the thin film lithium niobate micro-ring cavity structure mentioned in step S1 is located above the SiO2 layer, and the upper cladding is air. The wavelength of light input to the micro-ring cavity is 1550nm (frequency is 193.41THz), and the FEM simulation is used for frequency scanning calculation. The frequency range is 176.34~214.14THz, and the scanning step is 2THz. Thin film lithium niobate is an anisotropic material. The transverse electric mode in the z-cut thin film lithium niobate micro-ring cavity corresponds to the material refractive index of e-light, which is described by the Sellmeier equation as follows:

[0055]

[0056] Where λ is the wavelength of light in micrometers.

[0057] The frequency scanning solution of the physical field obtains the mode number μ of the micro-ring cavity at this frequency point. Based on the calculation results, the integrated dispersion (D int ) curve, the formula is as follows:

[0058]

[0059] Where μ is the mode number, ω μ is the angular frequency under different mode numbers, ω0 is the angular frequency of the pump light, and D1 / (2π) is the free spectrum range of the micro-ring cavity. int From the curve, we can get the high-order dispersion of the micro-ring cavity (D2, D3, D4, ...).

[0060] In the polynomial fitting process of this example, the selected mode number μ range is [-46, 64], the total number of modes is 111, and the frequency rounding range covered by this mode number range is 185 to 205 THz.

[0061] In this step, it was found that a small number of thin-film lithium niobate micro-ring cavities could not confine the internal light field, resulting in serious distortion of the simulation results. These cases were concentrated in the case of R 30μm, H and H slab There are 50 intervals with a difference of less than 0.25 μm, so after excluding the distorted results, the total number in the data set is 6880.

[0062] Step S3: Establish a fully connected neural network model, as shown in the attached Figure 2 As shown, the data in the dataset needs to be preprocessed first to narrow the difference in input / output data coverage space and improve the prediction accuracy of the model.

[0063] Specifically, in this neural network model, the input data is D int / (2π) Dispersion D of thin-film lithium niobate micro-ring cavity integrated after data range is scaled to [-1, +1] GHzint (-46)~D int (64), the output data is the geometric characteristics of the thin film lithium niobate micro-ring cavity after logarithmization to base 10 [R, W, H, H slab ,θ], the dataset is divided into training set and test set in the ratio of 85:15.

[0064] The neural network model needs to adjust parameters repeatedly to ensure the best prediction effect of the model. The whole process is as shown in the attached figure. Figure 3 As shown in the figure, the model parameters include model structure, learning rate, activation function, optimizer, etc. After a lot of attempts, the model parameters are as follows: the neural network contains 1 input layer, 1 output layer, and 4 hidden layers, each of which contains 512, 512, 1024, and 1024 neurons. The package size of the training set is 64, the learning rate is set to 0.0001, and after every 10 iterations, the learning rate will drop to 0.1 times the original value.

[0065] Sigmoid is set as the activation function, AdamW as the optimizer, and mean square error as the loss function. It should be noted that when calculating the mean square error, each geometric structure parameter is calculated separately and the comprehensive mean square error is calculated at the same time. Under such settings, it can ensure that the theoretically best prediction accuracy is obtained in the case of this data set, that is, the loss function is minimized. After the model prediction is completed, the internal parameters of the model are locked, and the accuracy of the model prediction is evaluated through the test set, as shown in the attached figure. Figure 4 shown.

[0066] Step S4: Using random number generation method to obtain two near-zero D int Curves, such as Figure 5 As shown in the figure, the dispersion means are 0 GHz and 0.05 GHz respectively, the standard deviation is 0.01 GHz, the horizontal axis is the mode number μ, the range is [-46, 64], and the vertical axis is the generated integrated dispersion value. int The curve is used as the input of the neural network model, and four predictions are performed to obtain the corresponding thin film lithium niobate micro-ring cavity structure parameters, a total of 8 (Str.1~Str.8), as shown in Table 1.

[0067] Based on the 8 structural parameters obtained, FEM simulation was used to obtain the D int Curve, as shown Figure 6 As shown. It can be seen that D int When the mean value is 0 GHz, the micro-ring cavity structures Str.3 and Str.4 can obtain a flat and near-zero D in the frequency range of 185 to 205 THz.int , and exhibits an anomalous dispersion state near the pump wavelength, so it is easy to excite the generation of soliton microcombs. int When the mean value is 0.05 GHz, it is difficult to obtain a flat dispersion structure.

[0068] Table 1 Predicted structural parameters of eight thin-film lithium niobate micro-ring cavities

[0069]

[0070] Step S5: Using the open source soliton microcomb solver pyLLE, the two microring cavity structures (Str.3 and Str.4) optimized in step S4 are used to calculate the D int The curve simulates the generation process and spectral shape of the soliton microcomb. pyLLE uses the split-step Fourier algorithm to numerically solve the LLE equation, as shown in the following equation:

[0071]

[0072] Among them, E(t,τ) and E in They represent the light field circulating in the micro-ring cavity and the amplitude of the incident light, respectively. t is the slow-changing time and τ is the fast-changing time (-t R / 2≤τ≤t R / 2), t R is the time it takes for light to circulate in the cavity, γ is the Kerr nonlinear coefficient, α is the loss of light in the microring cavity, δ0 is the detuning between the mode angular frequency and the pump light angular frequency, and θ is the coupling coefficient between the straight waveguide and the microring cavity. -1 Related to the fast changing time τ.

[0073] In pyLLE, the two thin-film lithium niobate micro-ring cavities (Str.3 and Str.4) obtained in step S4 were subjected to a D int The curve is input into it, and the pump light power (P in ) and intrinsic Q value (Q i ), the solver will calculate the output spectrum of the micro-ring cavity within this frequency range and the optical power variation inside the micro-ring cavity.

[0074] Attached Figure 7 It shows that by adjusting P in and Q i , the flat light comb is obtained. As can be seen from the figure, at Q i =7×106,P in= 800mW, the micro-ring cavity Str.3 produced a soliton optical comb with a power difference of less than 0.01dB in the frequency range of 180-205THz, and the flatness was excellent. i =1.3×106), a flat soliton microcomb with a comb power difference of less than 0.2 dB can still be achieved. This parameter reduces the requirement for the microring cavity loss and is easier to prepare in practice. The above results show that the method proposed in the present invention can predict a variety of microring cavity structure parameters for flat soliton microcomb generation, thereby providing guidance for the processing and preparation of microring cavities.

[0075] In the actual micro-nano processing, there are often deviations between the design parameters and the processing results, among which the height parameters of the micro-ring cavity are more difficult to control. Figure 8 The total film thickness H and the unetched depth H were studied. slab The impact on the integrated dispersion curve and the flatness of the optical comb when deviating from the optimal result (Str.3). It can be seen that when H and H slab When the change is 10nm, the D of the thin film lithium niobate micro-ring cavity int The curve will change accordingly, which will reduce the spectral flatness of the soliton microcomb. Therefore, the generation of flat soliton microcombs also places high demands on micro-nano fabrication technology.

[0076] In the description of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "center", "lateral", "up", "down", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc. is based on the orientation or positional relationship shown in the accompanying drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. In addition, the terms "first" and "second" are used only for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, unless otherwise specified, "several" means two or more. In addition, the term "including" and any variation thereof are intended to cover non-exclusive inclusions.

[0077] The present invention is described according to the embodiments. Without departing from the principle, the present device can also be modified and improved. It should be pointed out that any technical solution obtained by equivalent replacement or equivalent transformation falls within the protection scope of the present invention.

Claims

1. A dispersion engineering method for an ultra-flat soliton microcavity optical comb, characterized in that: The steps include: S1. Determine the geometric structure parameters involved in optimizing the integrated dispersion of the micro-ring cavity; S2, build the data set required for neural network training; S3, establish a fully connected neural network model; S4. Predict the geometric parameters of the micro-ring cavity by integrating the dispersion curve for the target through the model of S3.

2. The dispersion engineering method of an ultra-flat soliton microcavity optical comb according to claim 1, characterized in that: The steps for acquiring the data set in S2 are as follows: combining the scanning ranges and corresponding step sizes of the optimized geometric structure parameters to obtain a preliminary data set; and screening the preliminary data set to obtain the data set required for training.

3. The dispersion engineering method of an ultra-flat soliton microcavity optical comb according to claim 1, characterized in that: The dataset in S2 is divided into a test set and a training set in a ratio of 85:

15.

4. The dispersion engineering method of an ultra-flat soliton microcavity optical comb according to claim 1, characterized in that: In the process of establishing the neural network model in S3, it is necessary to repeatedly adjust the model parameters, which include model structure, learning rate, activation function and optimizer.

5. The dispersion engineering method of an ultra-flat soliton microcavity optical comb according to claim 2, characterized in that: The specific steps for filtering data in S2 are as follows: S20, according to the objective function, obtain the transverse electric mode of the z-cut micro-ring cavity, which corresponds to the material refractive index of the e-light Where λ is the wavelength of light in micrometers; S21. The physical field frequency scanning solution obtains the mode number μ of the micro-ring cavity at this frequency point. Through post-processing, the integrated dispersion (D int ) curve, the formula is as follows: Among them, ω μ is the angular frequency under different mode numbers, ω0 is the angular frequency of the pump light, and D1 / (2π) is the free spectrum range of the micro-ring cavity; S22. Distorted data are found and deleted from the preliminary data set in S2.

6. The dispersion engineering method of an ultra-flat soliton microcavity optical comb according to claim 5, characterized in that: It is necessary to preprocess the data in the data set required for training. The specific processing steps are as follows: In this neural network model, the input data is D int The integrated dispersion set of the thin-film lithium niobate micro-ring cavity after scaling the range of / (2π) to [-1,+1] GHz. The output data is the geometric feature set of the thin-film lithium niobate micro-ring cavity after logarithmization to the base 10.

7. A dispersion engineering method for an ultra-flat soliton microcavity optical comb according to any one of claims 1 to 6, characterized in that: Also includes the steps: S5. Use the open source solver pyLLE to simulate the generation of soliton micro-combs in the micro-ring cavity obtained in S4.

8. The dispersion engineering method of an ultra-flat soliton microcavity optical comb according to claim 7, characterized in that: In S5, pyLLE uses the split-step Fourier algorithm to numerically solve the LLE equation, as shown in the following formula: Among them, E(t,τ) and E in They represent the light field circulating in the microring cavity and the amplitude of the incident light, respectively. t is the slow-changing time, τ is the fast-changing time, and t R is the time for light to circulate in the cavity, γ is the Kerr nonlinear coefficient, α is the loss of light in the microring cavity, δ0 is the detuning between the mode angular frequency and the pump light angular frequency, and θ is the coupling coefficient between the straight waveguide and the microring cavity.

Images