A high-Q silicon-based photonic crystal nanobeam microcavity based on intelligent algorithm
Optimizing the gradient hole and cavity length of photonic crystal nanobeam microcavity through neural networks and genetic algorithms, solving the problem of difficulty in achieving high Q value under small mode volumes, achieving efficient parameter optimization and performance improvement, and is suitable for high-precision optics Sensing and filtering systems.
Patent Information
- Application Number
- CN202210017518.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-01-07
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2042-01-07
AI Technical Summary
It is difficult to achieve high Q-value silicon-based photonic crystal nanobeam microcavity under smaller mode volumes. The existing designs have problems with low parameter optimization efficiency and large cavity volume.
An intelligent algorithm optimization method combined with neural network and genetic algorithm is adopted to adjust the gradient hole distance and cavity length in the photonic crystal nanobeam microcavity to generate the optimal resonant cavity structure and improve the Q value. The specific steps include scanning the cavity length, generating random structures, segmenting the data set for training and testing, using neural network output as a fitness function of the genetic algorithm, iteratively optimizing structural parameters, and finally achieving high Q values.
The highest Q value is achieved under small mode volume conditions, improving the performance of photonic crystal nanobeam microcavity, breaking through the limitations of low parameter optimization efficiency and large cavity volume, and is suitable for high-precision optical sensing and filtering systems.
Smart Images

Figure CN114429081B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the fields of optical sensing and optical communication, and in particular to a high-Q silicon-based photonic crystal nanobeam microcavity based on an intelligent algorithm. Background Art
[0002] With the advent of the ultra-large-scale era of integrated circuits, the characteristic size of integrated circuits will develop in a more miniaturized direction. Processes and materials are the primary bottlenecks for the continued miniaturization of integrated circuits. Silicon-based photonic chips or silicon-based optoelectronic hybrid integrated chips are to integrate various photonic functional devices and even microelectronic devices on one chip. Therefore, the size of photonic devices should reach the size of microelectronic devices, that is, the control of photons should be achieved at the same scale of micro-nano and light wavelengths.
[0003] In order to solve this problem, photonic crystals have attracted much attention. The band structure characteristics of photonic crystals realize the effective regulation of photons. Moreover, due to the advantages of low transmission loss, high design freedom, low bending loss, rich functions, and easy integration, photonic crystals have become a good platform for the production of silicon-based micro-nano optoelectronic devices, and play an increasingly important role in silicon-based optoelectronic integrated circuits. Photonic crystal microcavities are formed by introducing point defects in the perfect photonic crystal structure. In the cavity, only the light field of light waves of a specific frequency that meet the resonance conditions will be resonantly enhanced. The volume of the photonic crystal microcavity is very small, and the energy in the cavity is very strong after resonance. Therefore, the light energy density in the cavity is very large, the interaction between light and matter in the cavity is greatly enhanced, and the optical nonlinear effect is significantly enhanced. Photonic crystal microcavities have many advantages, such as high quality factor, small mode volume, small size, easy integration, high design freedom, and strong interaction between light and matter in the cavity. At present, photonic crystal microcavities are widely used in optical filters, all-optical switches, low-threshold and thresholdless lasers, high-efficiency light sources, and high-sensitivity sensors due to their unique characteristics.
[0004] In the past decade, various schemes have been proposed to pursue high-performance photonic crystal nanobeam microcavities. The classic design method uses two tapered waveguides with tapered apertures and lattice periods in the middle of the mirrors on both sides of the microcavity, and obtains a one-dimensional photonic crystal nanobeam microcavity. By further optimizing the cavity length, a theoretical Q factor of 1.4×10 7 , the modulus volume is 0.39(λ / n) 3 The experimentally measured Q factor is 7.5×10 5. There are also photonic crystal nanobeam microcavities designed using a deterministic design method that can achieve high Q values. However, the resonant cavity mode volume is relatively large and requires 30 pairs of Gaussian mirrors, which increases the difficulty of device design and manufacturing. This design not only uses the excellent data analysis capabilities of machine learning to analyze the relationship between Q value and structural parameters, but also combines genetic algorithms to use their characteristics of better and faster exploration of optimal solutions to efficiently solve the parameter space. This design uses the machine learning-genetic algorithm method to optimize the photonic crystal nanobeam microcavity, and the theoretical Q factor is as high as 1.2×10 8 , the mode volume is only 0.32(λ / n) 3 .
[0005] In summary, the high-Q photonic crystal nanobeam microcavity based on machine learning-genetic algorithm optimization is the motivation for the invention, and has a huge market demand in optical communication systems. Summary of the invention
[0006] The technical problem to be solved by the present invention is to obtain a silicon-based photonic crystal nanobeam microcavity with a high Q value under a relatively small mode volume.
[0007] A high-Q silicon-based photonic crystal nanobeam microcavity based on an intelligent algorithm, wherein the Q-value silicon-based photonic crystal nanobeam microcavity is a resonant cavity, comprising two reflection regions and a gradient region, wherein the gradient region is located between the two reflection regions, and the gradient region comprises a plurality of gradient holes;
[0008] First, only the cavity length is scanned, and the size of the gradient hole increases progressively from the middle position of the gradient zone to both sides, and the silicon-based photonic crystal nanobeam microcavity at this time is determined to be the initial resonant cavity structure. On the basis of the initial resonant cavity structure, the gradient hole is fine-tuned to generate a random structure, and a data set of Q values related to random structure parameters is obtained, and the data set is divided into a training set and a test set;
[0009] Then a neural network is used, whose input is the cavity length and the distance between the gradient holes, and whose output is log 10 Q, use the training set and test set to train and test the neural network respectively to obtain the trained neural network;
[0010] Finally, the genetic algorithm is used to optimize the initial resonant cavity structure, and the output of the trained neural network is used as the fitness function of the genetic algorithm. It is iterated in the direction of high fitness to obtain the final converged log 10 Q, that is, to obtain the optimal resonant cavity structure parameters and the corresponding Q value, that is, to obtain a high-Q silicon-based photonic crystal nanobeam microcavity.
[0011] Furthermore, the initial resonant cavity structure is a suspended conical structure, and the radius of the gradual hole is 0.28 times the distance between two adjacent holes.
[0012] Furthermore, the calculation formula of the Q value is:
[0013]
[0014] Among them, ω 0 is the angular frequency and U(t) is the total energy stored in the resonant cavity.
[0015] Furthermore, the material of the resonant cavity is silicon, and a plurality of air holes are provided therein, the plurality of air holes are distributed in a bilaterally symmetrical manner, and the gradient hole is a part of the air holes.
[0016] Furthermore, the resonant cavity adopts an SOI platform, the waveguide width of the resonant cavity is 500nm, the height is 220nm, the lattice period is 460nm, and the radius of the air hole in the reflection area is 128.8nm.
[0017] Furthermore, the distance between the centers of two adjacent air holes in the reflection area of the resonant cavity is 329.8102 nm, the spacing of the gradient holes from inside to outside is 340.1814 nm, 359.7913 nm, 380.4111 nm, 400.3983 nm, 420.2000 nm, and 439.8526 nm, and the radius of the gradient holes is 0.28 times the corresponding hole spacing.
[0018] Furthermore, the neural network adopts an artificial neural network, which includes an input layer, an output layer and a hidden layer, and the number of neurons in the hidden layer is 6.
[0019] Furthermore, the random structural parameters include the distance between the gradient holes and the cavity length.
[0020] Furthermore, in the genetic algorithm, the distance between the gradient holes and the range of cavity length are set to [-1 nm, 1 nm].
[0021] The beneficial effect of the present invention is that the present invention not only uses the excellent data analysis ability of the neural network to obtain the relationship between the Q value and the structural parameters, but also cleverly combines the advantage of the genetic algorithm in efficiently seeking the optimal solution to optimize the photonic crystal nanobeam microcavity, thereby greatly improving the performance of the photonic crystal nanobeam microcavity. The neural network-genetic algorithm technology can effectively explore the relationship between multiple structural parameters and Q values of the microcavity, breaking through the limitations of low optimization efficiency and long optimization time of the photonic crystal nanobeam microcavity parameters. The optimized photonic crystal nanobeam microcavity has a small mode volume (~0.32 (λ / n) 3 ) conditions, the current highest Q value was achieved, which is conducive to the construction of on-chip high-precision optical sensing and filtering systems. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] The specific implementation modes of the present invention are further described in detail below in conjunction with the accompanying drawings.
[0023] Figure 1 This is a planar structural diagram of a high-Q silicon-based photonic crystal nanobeam microcavity.
[0024] Figure 2 It is the neural network model and the fitting diagrams of the training set and test set respectively.
[0025] Figure 3 This is the electric field diagram of the high-Q silicon-based photonic crystal nanobeam microcavity.
[0026] Figure 4 This is a comparison diagram of the electric field Gaussian fitting and the energy in the leakage region before and after optimization of the high-Q silicon-based photonic crystal nanobeam microcavity.
[0027] Figure 5 It is an iterative graph of genetic algorithm optimization of photonic crystal nanobeam microcavity. DETAILED DESCRIPTION
[0028] The present invention will be further described below in conjunction with the accompanying drawings and specific implementation methods.
[0029] This embodiment discloses a high-Q silicon-based photonic crystal nanobeam microcavity based on an intelligent algorithm, which is optimized based on a neural network and a genetic algorithm. By changing the distance of the gradient holes and the cavity length in the resonant cavity, the light field distribution in the resonant cavity is more in line with the Gaussian function model, which greatly improves the photonic crystal's ability to restrict light, and thus greatly improves the Q value of the photonic crystal microcavity. According to the above principle, the cavity length (the distance between the centers of the two central holes) is first scanned, and then the distance between the gradient holes decreases from the outside to the inside in an arithmetic progression to obtain an initial resonant cavity structure. By fine-tuning the distance between the gradient holes, the Q value changes significantly, so the gradient holes are further optimized to obtain a higher Q value. Considering that the neural network has excellent data processing capabilities, the neural network is combined to explore the relationship between the Q value and the structural parameters. The distance between the gradient holes and the cavity length in the initial resonant cavity is randomly changed to generate a random structure, and a data set of Q values about the structural parameters is obtained, and the data set is divided into a training set and a test set. Then the training set is used to train the neural network, and finally the genetic algorithm is used to explore the parameters, and finally the following is obtained. Figure 1 The high-Q silicon-based photonic crystal nanobeam cavity shown in the figure has a main structure of a resonant cavity with a progressive spacing of gradient holes, and the radius changes with the change of the hole spacing. The photonic crystal nanobeam cavity is in a suspended state and has a conical structure. The optimized structural parameters change very little, and the mode volume remains almost unchanged, maintaining a small value.
[0030] The quality factor (Q) of a photonic crystal nanobeam microcavity describes the ability of a photonic crystal nanobeam microcavity to restrict photons in time, that is, the length of the photon lifetime. It can also reflect the optical loss of the microcavity. The larger the optical Q value, the stronger the resonance enhancement of the resonant cavity to the light field, and the greater the light field intensity in the cavity under the same conditions. In a photonic crystal microcavity, the size of the Q value comes from the scattering loss. The larger the scattering loss, the smaller the Q value. Therefore, how to reduce the scattering loss is the key to improving the Q value. The Q value can be described as:
[0031]
[0032] where ω 0 is the angular frequency and U(t) is the total energy stored in the cavity.
[0033] By changing the distance between the gradient holes in the resonant cavity, the light field distribution in the resonant cavity is made to better meet the Gaussian function model, which greatly improves the ability of the photonic crystal to confine light, thereby greatly improving the Q value of the photonic crystal microcavity. First, only the cavity length is scanned, and then the distance between the 6 holes on the left and right is gradually changed in equal steps, that is, the first step is to determine the initial resonant cavity structure. Due to the fine-tuning of the gradient holes, the Q value has a significant change. In order to obtain better structural parameters and a higher Q value, machine learning methods are combined for further exploration. Based on the initial structure, a random structure is generated to obtain a data set of Q values and structural parameters. There are 30 air holes in total. The output of the neural network (i.e. log 10 Q) as the fitness function. The optimal distance between the two center holes of the resonant cavity is 329.8102nm, and the spacing of the gradient holes (the distance between the two centers) from the inside to the outside is 340.1814nm, 359.7913nm, 380.4111nm, 400.3983nm, 420.2000nm, 439.8526nm, and the radius of the gradient holes is 0.28 times the corresponding hole spacing.
[0034] Figure 2 is the neural network model and its fitting on the training set and test set. The neural network contains a hidden layer, an input layer and an output layer. The optimized structural parameters are used as the input of the neural network, and log 10 Q is used as the output, and the hidden layer contains 6 neurons. Among them, the distance between the gradient holes in the random structure and the cavity length (the distance between the centers of the two holes in the center) are used as its input (i.e. 7 inputs), log 10Q is used as the output (i.e. 1 output). The data set is divided into two parts: training set and test set according to the ratio of 7:3. Then the neural network is trained. In the training set and the test set, the data set is distributed near the fitting function. The prediction error of the neural network is 0.097, which is sufficient to predict the Q value of the microcavity. Finally, the genetic algorithm is used to optimize the structural parameters, and the predicted value of the neural network is used as the fitness of the genetic algorithm. It is iterated in the direction of high fitness. After a finite number of iterations, log 10 Q finally converges. And the Q value is verified to be 1.2×10 by the finite difference time domain method. 8 , reaching the highest value currently designed.
[0035] Figure 3 This is the electric field diagram of the optimized silicon-based photonic crystal nanobeam microcavity with the best Q value, that is, the electric field diagram of the high-Q silicon-based photonic crystal nanobeam microcavity. As can be seen from the figure, the light energy is concentrated in the geometric center of the photonic crystal nanobeam microcavity, and the light field is confined in the photonic crystal nanobeam microcavity.
[0036] Figure 4 This is a comparison of the electric field Gaussian fitting and the energy in the leakage zone of the high-Q silicon-based photonic crystal nanobeam microcavity before and after optimization. Compared with before optimization, the electric field Gaussian fitting of the optimized structure is better, more consistent with the Gaussian function model, and has a stronger ability to limit light energy. It can also be seen from the energy in the leakage zone that the energy in the leakage zone is significantly reduced after optimization, which means that the scattering loss of the microcavity has become smaller, indicating that the Q value of the microcavity has been significantly improved.
[0037] Figure 5 The machine learning-genetic algorithm technology was used to optimize the photonic crystal nanobeam microcavity iteratively. After 100 iterations, the fitness gradually converged to about 7.6, and the corresponding structural parameters were obtained. Finally, the time-domain finite difference method was used to verify the resonant cavity structural parameters, that is, the Q value of the optimized photonic crystal nanobeam microcavity was 1.2×10 8 , which is the highest Q value of photonic crystal nanobeam microcavity achieved under the same mode volume. The mode volume of the photonic crystal nanobeam microcavity obtained after optimization is 0.32 (λ / n) 3 , Q value and mode volume are both detection parameters of photonic crystal nanobeam microcavity. The mode volume describes the spatial restriction capability of photonic crystal nanobeam microcavity on photons. The smaller the mode volume, the smaller the area in which the resonant cavity is restricted. Under the condition of the same total energy in the resonant cavity, the greater the optical power density in the resonant cavity, the stronger the interaction between light and matter. The mode volume can be described as:
[0038]
[0039] Where ε(r) is the dielectric constant and E(r) is the electric field.
[0040] Through the technical solution of the present invention, the Q value of the photonic crystal nanobeam microcavity finally obtained is optimal, and the calculated mode volume is also small, which meets the design requirements in this field.
[0041] The beneficial effect of the present invention is that the present invention not only uses the excellent data analysis ability of the neural network to obtain the relationship between the Q value and the structural parameters, but also cleverly combines the advantage of the genetic algorithm in efficiently seeking the optimal solution to optimize the photonic crystal nanobeam microcavity, thereby greatly improving the performance of the photonic crystal nanobeam microcavity. The neural network-genetic algorithm technology can effectively explore the relationship between multiple structural parameters and Q values of the microcavity, breaking through the limitations of low optimization efficiency and long optimization time of the photonic crystal nanobeam microcavity parameters. The optimized photonic crystal nanobeam microcavity has a small mode volume (~0.32 (λ / n) 3 ) conditions, the current highest Q value was achieved, which is conducive to the construction of on-chip high-precision optical sensing and filtering systems.
[0042] The above are the technical principles and simulation results of the present invention, but those skilled in the art should understand that various changes in form and details of the present invention are within the scope of protection of the present invention without departing from the spirit and scope of the present invention as defined in the attached claims.
Claims
1. A high-Q silicon-based photonic crystal nanobeam microcavity based on an intelligent algorithm, characterized by: The silicon-based photonic crystal nanobeam microcavity is a resonant cavity, comprising two reflection zones and a gradient zone, wherein the gradient zone is located between the two reflection zones, and the gradient zone comprises a plurality of gradient holes; First, only the cavity length is scanned, and the size of the gradient hole increases progressively from the middle position of the gradient zone to both sides, and the silicon-based photonic crystal nanobeam microcavity at this time is determined to be the initial resonant cavity structure. On the basis of the initial resonant cavity structure, the gradient hole is fine-tuned to generate a random structure, and a data set of Q values related to random structure parameters is obtained, and the data set is divided into a training set and a test set; Then a neural network is used, whose input is the cavity length and the distance between the gradient holes, and whose output is log 10 Q, use the training set and test set to train and test the neural network respectively to obtain the trained neural network; Finally, the genetic algorithm is used to optimize the initial resonant cavity structure, and the output of the trained neural network is used as the fitness function of the genetic algorithm. It is iterated in the direction of high fitness to obtain the final converged log 10 Q, that is, to obtain the optimal resonant cavity structure parameters and the corresponding Q value.
2. The high-Q silicon-based photonic crystal nanobeam microcavity according to claim 1, characterized in that: The initial resonant cavity structure is a suspended conical structure.
3. The high-Q silicon-based photonic crystal nanobeam microcavity according to claim 1, characterized in that: The calculation formula of the Q value is: Where ω0 is the angular frequency and U(t) is the total energy stored in the resonant cavity.
4. The high-Q silicon-based photonic crystal nanobeam microcavity according to claim 1, characterized in that: The resonant cavity is made of silicon and has a plurality of air holes therein. The plurality of air holes are distributed in a bilaterally symmetrical manner, and the gradient hole is a part of the air holes.
5. The high-Q silicon-based photonic crystal nanobeam microcavity according to claim 1, characterized in that: The resonant cavity adopts an SOI platform, the waveguide width of the resonant cavity is 500nm, the height is 220nm, the lattice period is 460nm, and the radius of the air hole in the reflection area is 128.8nm.
6. The high-Q silicon-based photonic crystal nanobeam microcavity according to claim 1, characterized in that: The center distance between two adjacent air holes in the reflection area of the resonant cavity is 329.8102nm, and the spacing of the gradient holes from inside to outside is 340.1814nm, 359.7913nm, 380.4111nm, 400.3983nm, 420.2000nm, and 439.8526nm, respectively, and the radius of the gradient holes is 0.28 times the corresponding hole spacing.
7. The high-Q silicon-based photonic crystal nanobeam microcavity according to claim 1, characterized in that: The neural network adopts an artificial neural network, which includes an input layer, an output layer and a hidden layer, and the number of neurons in the hidden layer is 6.
8. The high-Q silicon-based photonic crystal nanobeam microcavity according to claim 1, characterized in that: The random structural parameters include the distance between the gradient holes and the cavity length.
9. The high-Q silicon-based photonic crystal nanobeam microcavity according to claim 1, characterized in that: In the genetic algorithm, the distance between the gradient holes and the range of cavity length are set to [-1nm, 1nm].
