Artificial neural network-based silicon photonic interlayer coupling structure design method

Abstract

The invention relates to a silicon photonic interlayer coupling structure design method based on an artificial neural network, and belongs to the technical field of artificial intelligence. The method comprises the following steps: acquiring a plurality of interlayer coupling structure data sets, performing normalization processing on the interlayer coupling structure data sets, and dividing the data sets; constructing a back propagation neural network; in the design process, a self-adaptive genetic algorithm is used for optimizing the weight and the threshold value of the back-propagation neural network; using a training set to train the back propagation neural network model; coupling structure parameters are input into the trained model, a coupling efficiency value is output, and whether the coupling structure meets design requirements or not is judged according to the coupling efficiency value; and continuously adjusting the structural parameters until the design requirements are met. According to the method, the design time of the interlayer coupling structure is shortened, the design difficulty of the interlayer coupling structure is reduced, and automation and intelligence of the design of the interlayer coupling structure are realized.

Description

Technical Field

[0001] The invention relates to a silicon photon interlayer coupling structure design method based on an artificial neural network, and belongs to the technical field of artificial intelligence. Background Art

[0002] Silicon photonics technology is an innovative technology in the information age. It is based on a silicon material platform with excellent comprehensive performance, uses photons as carriers for data transmission, and the related manufacturing processes are highly compatible with high-level complementary metal oxide semiconductor technology. Traditional design methods for photonic devices often use electromagnetic field numerical simulation methods such as finite element method and finite time-domain difference method. In traditional programs, we usually start with full-wave simulation of the initial design based on empirical knowledge, and then iteratively adjust the geometric / material parameters to approach the customer's specific requirements. This trial-and-error process has higher requirements on the computer's CPU and memory and is very time-consuming.

[0003] Artificial neural networks can establish implicit relationships between inputs (i.e., geometric / material parameters) and outputs (i.e., optical responses) to simulate the process of nonlinear neural conduction in the human body. With the help of well-trained artificial neural networks, the complex and time-consuming design process that relies heavily on numerical simulation and optimization can be bypassed. Summary of the invention

[0004] In view of the problem that the current design of interlayer coupling structures requires a long design time, the present invention proposes a silicon photonic interlayer coupling structure design method based on artificial neural networks. The present invention introduces genetic algorithms into neural networks, which shortens the design time of couplers and designs a more diversified and complex model structure that meets design requirements.

[0005] The technical solution of the present invention is: a method for designing a silicon photonic interlayer coupling structure based on an artificial neural network, the method comprising:

[0006] Step 1, obtaining a plurality of interlayer coupling structure data sets, normalizing the interlayer coupling structure data sets, and dividing the data sets; wherein the interlayer coupling structure data sets are generated according to the structural parameters of the interlayer coupling structure and the corresponding coupling efficiency;

[0007] Step 2, build a back propagation neural network;

[0008] Step 3: During the design process, the weights and thresholds of the back propagation neural network are optimized using the adaptive genetic algorithm (AGA).

[0009] Step 4, using the training set to train the back propagation (BP) neural network model;

[0010] Step 5. Input the coupling structure parameters into the trained model, and output the coupling efficiency value. According to the coupling efficiency value, judge whether the coupling structure meets the design requirements; continuously adjust the structural parameters until the design requirements are met, and end the design process.

[0011] Furthermore, in Step 1, the multiple interlayer coupling structures are single-layer interlayer coupling structures and double-layer interlayer coupling structures; the structural parameters of the single-layer interlayer coupling structure are length, width, thickness and spacing, and the structural parameters of the double-layer interlayer coupling structure are length and width.

[0012] Furthermore, the double-layer interlayer coupling structure is divided into a first layer Si 3 N 4 To the second layer Si 3 N 4 Interlayer coupling structure and the second layer Si 3 N 4 to the Si interlayer coupling structure.

[0013] Furthermore, in Step 2, the back propagation neural network includes a plurality of base layers connected in sequence, and an input layer and an output layer arranged at the input end and the output end of the base layer, and the base layer includes a fully connected layer and an activation layer connected in sequence.

[0014] Further, the base layer has three layers;

[0015] For a single-layer interlayer coupling structure;

[0016] The number of neurons in its input layer is 13.

[0017] The number of neurons in the first base layer is 12, the number of neurons in the second base layer is 30, the number of neurons in the third base layer is 12, and the number of neurons in the output layer is 2;

[0018] For the first layer Si 3 N 4 To the second layer Si 3 N 4 Interlayer coupling structure;

[0019] The number of neurons in its input layer is 8;

[0020] The number of neurons in the first base layer is 9, the number of neurons in the second base layer is 30, the number of neurons in the third base layer is 9, and the number of neurons in the output layer is 2;

[0021] For the second layer Si 3 N 4 to the Si interlayer coupling structure;

[0022] The number of neurons in its input layer is 6, the number of neurons in the first base layer is 10, the number of neurons in the second base layer is 30, the number of neurons in the third base layer is 10, and the number of neurons in the output layer is 2.

[0023] Furthermore, in Step 2, the learning calculation formula of the back propagation neural network is:

[0024]

[0025] Among them, w ji represents the connection weight between neuron i and neuron j; θ j represents the threshold of neuron i; y i represents the output of neuron i; x j represents the output of neuron j, that is, the input of neuron i, n represents the number of base layers, and f represents the tanh function.

[0026] Furthermore, in Step 3, the adaptive genetic algorithm calculation process is as follows:

[0027]

[0028] Where P c represents the crossover probability, P m represents the mutation probability, f max represents the maximum fitness value of the population, f avg represents the average fitness value of the population, f 0 represents the larger fitness value of the two individuals in the crossover, f' represents the fitness value of the mutant individual, k 1 , k 2 , k 3 , k 4 Represents any coefficient between 0 and 1;

[0029] The fitness function is:

[0030]

[0031] Represents the predicted output value of the neural network, y i Represents the true output value, k represents any coefficient between 0 and 1, and abs() represents the absolute value.

[0032] The present invention also provides an interlayer coupling structure, which is formed by using the above-mentioned method for designing multiple interlayer coupling structures of silicon photonics based on artificial neural networks.

[0033] The beneficial effects of the present invention are:

[0034] The present invention shortens the design time of the interlayer coupling structure, reduces the design difficulty of the interlayer coupling structure, and realizes the automation and intelligence of the interlayer coupling structure design;

[0035] The BP neural network of the present invention has strong adaptability and flexibility, can quickly adapt to new structures, and realize high-precision device design. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 This is a schematic diagram of the process of the silicon photonic interlayer coupling structure design method based on artificial neural network in this application;

[0037] Figure 2 This is a schematic diagram of the single-layer interlayer coupling structure of this application;

[0038] Figure 3 This is a schematic diagram of the double-layer interlayer coupling structure of this application;

[0039] Figure 4 This is a schematic diagram of the TE mode spectrum response of the single-layer interlayer coupling structure of this application;

[0040] Figure 5 This is a schematic diagram of the TM mode spectrum response of the single-layer interlayer coupling structure of this application;

[0041] Figure 6 For this application, the first layer of Si 3 N 4 To the second layer Si 3 N 4 Schematic diagram of TE mode spectral response of interlayer coupling structure;

[0042] Figure 7 For this application, the first layer of Si 3 N 4 To the second layer Si 3 N 4 Schematic diagram of TM mode spectrum response of interlayer coupling structure;

[0043] Figure 8 For this application, the second layer of Si 3 N 4 Schematic diagram of TE mode spectrum response of Si interlayer coupling structure;

[0044] Fig. 9 For this application, the second layer of Si 3 N 4 Schematic diagram of the TM mode spectral response of the Si interlayer coupling structure. DETAILED DESCRIPTION

[0045] like Figure 1-Figure 9As shown, in order to solve the problem that the current design of interlayer coupling structure requires a long design time, the present invention proposes a silicon photonic interlayer coupling structure design method based on artificial neural network. The present invention introduces genetic algorithm into neural network, and designs a more diversified and complex model structure that meets the design requirements while shortening the design time of the coupler.

[0046] The present invention proposes a method for designing a silicon photonic interlayer coupling structure based on an artificial neural network, the method comprising:

[0047] Step 1, obtaining a variety of interlayer coupling structure data sets, normalizing the interlayer coupling structure data sets, and dividing them into a training set, a test set, and a validation set according to a ratio of 14:3:3; wherein the interlayer coupling structure data set is generated according to the structural parameters of the interlayer coupling structure and the corresponding coupling efficiency;

[0048] Furthermore, in Step 1, the multiple interlayer coupling structures are single-layer interlayer coupling structures and double-layer interlayer coupling structures; the structural parameters of the single-layer interlayer coupling structure are length, width, thickness and spacing, and the structural parameters of the double-layer interlayer coupling structure are length and width.

[0049] Furthermore, the double-layer interlayer coupling structure is divided into a first layer Si 3 N 4 To the second layer Si 3 N 4 Interlayer coupling structure and the second layer Si 3 N 4 to the Si interlayer coupling structure.

[0050] The single-layer interlayer coupling structure used in the present invention is shown in Figure 2 , the input variables are wavelength 1.5μm~1.6μm, waveguide spacing GAP, ranging from 0.2μm to 0.6μm, silicon waveguide thickness H_SI, ranging from 0.1μm to 0.3μm, silicon nitride waveguide thickness H_SIN, ranging from 0.3μm to 0.7μm, waveguide length L1, ranging from 8μm to 12μm, L2 ranging from 0μm to 2 0μm, L3 ranges from 46μm to 66μm, L4 ranges from 50μm to 70μm, L5 ranges from 8μm to 12μm, waveguide width W1 ranges from 0.9μm to 1.1μm, W2 ranges from 0μm to 0.2μm, W3 ranges from 0μm to 0.2μm, W4 ranges from 0.2μm to 0.4μm, and the cladding material is silicon dioxide;

[0051] The double-layer interlayer coupling structure used in the present invention is shown in Figure 3 , Si 3 N 4 To the second layer Si 3N 4 Interlayer coupling structure, input variable waveguide spacing GAP1 is fixed to 0.2μm, silicon nitride waveguide thickness H_SIN_1 is fixed to 0.4μm, silicon nitride waveguide thickness H_SIN_2 is fixed to 0.4μm, waveguide length L1 is fixed to 0, waveguide width W2 is fixed to 1μm, W21 is fixed to 1μm, input variable wavelength, value range is 1.5μm~1.6μm, waveguide length L2, value range is 20μm~120μm, L3 value range is 20μm~160m, L4 value range is 0μm~70μm, L5 value range is 20μm~80μm, L6 value range is 20μm~120μm, waveguide width W1, value range is 0μm~0.6μm, W11.1 value range is 0μm~0.6μm. Cladding material is silicon dioxide;

[0052] The second layer Si 3 N 4 The waveguide spacing GAP2 to the Si interlayer coupling structure is fixed to 0.25μm, the waveguide length L5 is fixed to 56μm in the simulation of wavelength and L7, and is fixed to 106μm in the simulation of L8, L9, W3 and W4, the waveguide length L6 is fixed to 30μm in the simulation of wavelength and L7, and is fixed to 100μm in the simulation of L8, L9, W3 and W4, the silicon waveguide thickness H_SI is fixed to 0.22μm, and the waveguide width W5 is fixed to 0.5μm. Input variable wavelength, the value range is 1.5μm~1.6μm, waveguide length L7, the value range is 20μm~120μm, L8 value range is 0μm~120μm, L9 value range is 30μm~180μm, waveguide width W3, the value range is 0μm~0.6μm, W4 value range is 0μm~0.3μm. The cladding material is silica.

[0053] Step 2, build a back propagation neural network;

[0054] Furthermore, in Step 2, the back propagation neural network includes a plurality of base layers connected in sequence, and an input layer and an output layer arranged at the input end and the output end of the base layer, and the base layer includes a fully connected layer and an activation layer connected in sequence.

[0055] Further, the base layer has three layers;

[0056] For a single-layer interlayer coupling structure;

[0057] The number of neurons in its input layer is 13.

[0058] The number of neurons in the first base layer is 12, the number of neurons in the second base layer is 30, the number of neurons in the third base layer is 12, and the number of neurons in the output layer is 2;

[0059] For the first layer Si 3 N 4 To the second layer Si 3 N 4 Interlayer coupling structure;

[0060] The number of neurons in its input layer is 8;

[0061] The number of neurons in the first base layer is 9, the number of neurons in the second base layer is 30, the number of neurons in the third base layer is 9, and the number of neurons in the output layer is 2;

[0062] For the second layer Si 3 N 4 to the Si interlayer coupling structure;

[0063] The number of neurons in its input layer is 6, the number of neurons in the first base layer is 10, the number of neurons in the second base layer is 30, the number of neurons in the third base layer is 10, and the number of neurons in the output layer is 2.

[0064] Furthermore, in Step 2, the learning calculation formula of the back propagation neural network is:

[0065]

[0066] Among them, w ji represents the connection weight between neuron i and neuron j; θ j represents the threshold of neuron i; y i represents the output of neuron i; x j represents the output of neuron j, that is, the input of neuron i, n represents the number of base layers, and f represents the tanh function.

[0067] Step 3: During the design process, the weights and thresholds of the back propagation neural network are optimized using the adaptive genetic algorithm (AGA).

[0068] Furthermore, in Step 3, the adaptive genetic algorithm calculation process is as follows:

[0069]

[0070] Where P c represents the crossover probability, P m represents the mutation probability, f max represents the maximum fitness value of the population, f avg represents the average fitness value of the population, f 0represents the larger fitness value of the two individuals in the crossover, f' represents the fitness value of the mutant individual, k 1 , k 2 , k 3 , k 4 Represents any coefficient between 0 and 1;

[0071] The fitness function is:

[0072]

[0073] in, Represents the predicted output value of the neural network, y i Represents the true output value, k represents any coefficient between 0 and 1, and abs() represents the absolute value.

[0074] The present invention also provides an interlayer coupling structure, which is formed by using the above-mentioned method for designing multiple interlayer coupling structures of silicon photonics based on artificial neural networks.

[0075] Step 4, using the training set to train the back propagation (BP) neural network model;

[0076] Step 5. Input the coupling structure parameters into the trained model, and output the coupling efficiency value. According to the coupling efficiency value, judge whether the coupling structure meets the design requirements; continuously adjust the structural parameters until the design requirements are met, and end the design process.

[0077] In order to verify the prediction ability of the neural network, the present invention uses the neural network to predict the spectral response of the interlayer coupling structure. Figures 4 to 9 It is the TE and TM spectral response in the wavelength range of 1.5μm to 1.6μm. It can be seen that the spectral response predicted by the AGA-BP neural network (the trained model of the present invention) is roughly consistent with the true value trend. Through comparison and analysis, it can be seen that the AGABP neural network can more effectively predict the spectral response of the interlayer coupling structure.

[0078] The processed training set is input into the neural network of the present invention for training. The learning rate of the neural network is 0.0001, the minimum convergence error of the training target is 0.0001, and the maximum number of training times is 1000. The number of iterations of the genetic algorithm is set to 100, the population size is set to 50, and the parameter settings of the neural network are consistent with those of the conventional BP neural network. The initial crossover probability Pc is set to 0.6, and the initial mutation probability is set to 0.1. The results show that the mean square error can reach 5.77E-05. The first layer Si 3 N 4 To the second layer Si 3 N 4The mean square error of the interlayer coupling structure can reach 5.11E-04. 3 N 4 The mean square error of the Si interlayer coupling structure can reach 3.13E-05. The results show that this method is practical and neural networks can be used instead of lengthy full-wave simulations to perform positive design of interlayer coupler structures.

[0079] As shown in Table 1, the MSE of the single-layer interlayer coupling structure AGA-BP network is 5.77E-05. As shown in Table 2, the first layer Si 3 N 4 To the second layer Si 3 N 4 The MSE of the interlayer coupling structure AGA-BP network is 5.11E-04. As shown in Table 3, the second layer Si 3 N 4 The MSE of the AGA-BP network to the Si interlayer coupling structure error is 3.13E-05. The MSE of the AGA-BP neural network is smaller than that of the BP network and the GA-BP network. The performance of the AGA-BP network model is the best, better than the BP network and the GA-BP network.

[0080] Table 1 shows the single-layer interlayer coupling structure error

[0081]

[0082] Table 2 shows the first layer Si 3 N 4 To the second layer Si 3 N 4 Interlayer coupling structure error

[0083]

[0084] Table 3 shows the second layer Si 3 N 4 To Si interlayer coupling structure error

[0085]

[0086] The specific implementation modes of the present invention are described in detail above in conjunction with the accompanying drawings, but the present invention is not limited to the above implementation modes, and various changes can be made within the knowledge scope of ordinary technicians in this field without departing from the purpose of the present invention.

Claims

1. A method for designing silicon photonic interlayer coupling structures based on artificial neural networks, characterized in that: The method comprises: Step 1, obtaining a plurality of interlayer coupling structure data sets, normalizing the interlayer coupling structure data sets, and dividing the data sets; wherein the interlayer coupling structure data sets are generated according to the structural parameters of the interlayer coupling structure and the corresponding coupling efficiency; Step 2, build a back propagation neural network; Step 3: During the design process, an adaptive genetic algorithm is used to optimize the weights and thresholds of the back propagation neural network. Step 4, using the training set to train the back propagation neural network model; Step 5. Input the coupling structure parameters into the trained model, and output the coupling efficiency value. According to the coupling efficiency value, judge whether the coupling structure meets the design requirements; continuously adjust the structural parameters until the design requirements are met, and end the design process.

2. The method for designing silicon photonic interlayer coupling structures based on artificial neural networks according to claim 1, characterized in that: In the Step 1, the plurality of interlayer coupling structures are a single-layer interlayer coupling structure and a double-layer interlayer coupling structure; The structural parameters of the single-layer interlayer coupling structure are length, width, thickness and spacing, and the structural parameters of the double-layer interlayer coupling structure are length and width.

3. The method for designing silicon photonic interlayer coupling structures based on artificial neural networks according to claim 2, characterized in that: The double-layer interlayer coupling structure is divided into a first-layer Si3N4 to second-layer Si3N4 interlayer coupling structure and a second-layer Si3N4 to Si interlayer coupling structure.

4. The method for designing silicon photonic interlayer coupling structures based on artificial neural networks according to claim 2, characterized in that: In the Step 2, the back propagation neural network includes a plurality of base layers connected in sequence, and an input layer and an output layer arranged at the input end and the output end of the base layer, and the base layer includes a fully connected layer and an activation layer connected in sequence.

5. The method for designing silicon photonic interlayer coupling structures based on artificial neural networks according to claim 4, characterized in that: The base layer has three layers; For a single-layer interlayer coupling structure; The number of neurons in its input layer is 13. The number of neurons in the first base layer is 12, the number of neurons in the second base layer is 30, the number of neurons in the third base layer is 12, and the number of neurons in the output layer is 2; For the interlayer coupling structure from the first layer Si3N4 to the second layer Si3N4; The number of neurons in its input layer is 8; The number of neurons in the first base layer is 9, the number of neurons in the second base layer is 30, the number of neurons in the third base layer is 9, and the number of neurons in the output layer is 2; For the second layer Si3N4 to Si interlayer coupling structure; The number of neurons in its input layer is 6, the number of neurons in the first base layer is 10, the number of neurons in the second base layer is 30, the number of neurons in the third base layer is 10, and the number of neurons in the output layer is 2.

6. The method for designing silicon photonic interlayer coupling structures based on artificial neural networks according to claim 1, characterized in that: In Step 2, the learning calculation formula of the back propagation neural network is: Among them, w ji represents the connection weight between neuron i and neuron j; θ j represents the threshold of neuron i; y i represents the output of neuron i; x j represents the output of neuron j, that is, the input of neuron i, n represents the number of base layers, and f represents the tanh function.

7. The method for designing silicon photonic interlayer coupling structures based on artificial neural networks according to claim 1, characterized in that: In Step 3, the adaptive genetic algorithm calculation process is as follows: Where P c represents the crossover probability, P m represents the mutation probability, f max represents the maximum fitness value of the population, f avg represents the average fitness value of the population, f0 represents the larger fitness value of the two crossed individuals, f' represents the fitness value of the mutant individual, k1, k2, k3, k4 represent any coefficients between 0 and 1; The fitness function is: Represents the predicted output value of the neural network, y i Represents the true output value, k represents any coefficient between 0 and 1, and abs() represents the absolute value.

8. An interlayer coupling structure, characterized in that: The interlayer coupling structure is formed by using the method for designing multiple interlayer coupling structures of silicon photonics based on artificial neural networks as described in any one of claims 1 to 7.

Images