MOM capacitor modeling method based on convolutional neural network

Abstract

The invention belongs to the technical field of MOM capacitor modeling simulation, and particularly discloses an MOM capacitor modeling method based on a convolutional neural network, and the method comprises the steps: S1, obtaining sample data through electromagnetic simulation software; s2, performing normalization processing on the sample data to obtain normalized sample data; s3, constructing a convolutional neural network model, wherein the convolutional neural network model comprises an input layer, a convolutional layer, a batch normalization layer, an activation function layer and a full connection layer which are connected in sequence; s4, training the convolutional neural network model; and S5, normalizing the structure parameter value and the simulation frequency value of the to-be-tested MOM capacitor, inputting the normalized structure parameter value and simulation frequency value into the trained convolutional neural network model, and carrying out reverse normalization on the output value to obtain a predicted value of the performance parameter of the to-be-tested MOM capacitor. According to the method, the modeling period is shortened, and the prediction precision of the MOM capacitor model on the capacitor performance parameters is improved. The method is suitable for predicting the performance parameters of the MOM capacitor.

Description

Technical Field

[0001] The invention belongs to the technical field of MOM capacitor modeling and simulation, and specifically relates to a MOM capacitor modeling method based on a convolutional neural network. Background Art

[0002] With the continuous development of RF and microwave integrated circuit technology, electronic devices with new processes and new materials continue to emerge. MOM (Metal-Oxide-Metal) capacitors are a common passive component in integrated circuit design. They consist of two layers of metal plates and an insulating medium sandwiched in between. MOM capacitors are widely used in various high-precision, high-performance RF and microwave integrated circuits because of their low cost, high Q value, and high integration. Therefore, establishing a model that can characterize the characteristics of MOM capacitors is of great significance for the simulation and practical application of MOM capacitors.

[0003] The modeling cycle and model accuracy of conventional MOM capacitor models are often contradictory. The modeling accuracy is low and cannot meet the modeling requirements of the device. How to shorten the modeling cycle and improve the accuracy of the MOM capacitor model has always been one of the key points and hotspots of current research in the RF and microwave fields. Summary of the invention

[0004] The object of the present invention is to provide a kind of MOM capacitor modeling method based on convolutional neural network, to shorten the modeling cycle, improve the prediction accuracy of MOM capacitor model to capacitor performance parameters.

[0005] In order to achieve the above-mentioned purpose, the technical methods adopted by the present invention are as follows: A MOM capacitor modeling method based on a convolutional neural network comprises the following steps performed in sequence: S1. Obtaining structural parameter values ​​and simulation frequency values ​​of multiple groups of MOM capacitors and corresponding performance parameter values ​​as sample data through electromagnetic simulation software; S2. normalizing the sample data to obtain normalized sample data; S3. Construct a convolutional neural network model, which includes an input layer, a convolution layer, a batch normalization layer, an activation function layer, and a fully connected layer connected in sequence; S4, using the structural parameter values ​​and simulation frequency values ​​in the normalized sample data as inputs of the convolutional neural network model, using the performance parameter values ​​in the normalized sample data as outputs of the convolutional neural network model, training the convolutional neural network model, and obtaining a trained convolutional neural network model; S5, after the structural parameter value and the simulation frequency value of the MOM capacitor to be measured are normalized, input to the trained convolutional neural network model, output the corresponding normalized performance parameter value, then the normalized performance parameter value is denormalized, and the predicted value of the MOM capacitor performance parameter to be measured is obtained.

[0006] As a limitation: the structural parameters of MOM capacitors include finger length, finger width and finger spacing, and the performance parameters include capacitance and quality factor.

[0007] As further restrictions: the width and height of the input layer of the convolutional neural network model are set to 1, and the number of channels is set to 4; the number of convolution kernels of the convolution layer is set to 9, the size of the convolution kernel is 1×1×4, padding=0, step size=1, and the initial weights and biases of the convolution kernel are generated randomly; the batch size of the batch normalization layer is set to 128; the activation function used in the activation function layer is the Leaky ReLU function; the number of neurons in the fully connected layer is 2, and the fully connected layer serves as the output layer of the convolutional neural network model.

[0008] As a further limitation: the training method of the convolutional neural network model in step S4 uses the Adam optimization algorithm to iteratively optimize the convolutional neural network model to obtain a trained convolutional neural network model.

[0009] Due to the adoption of the above scheme, the present invention has the following beneficial effects compared with the prior art: The invention provides a MOM capacitor modeling method based on a convolutional neural network. By constructing a convolutional neural network model of a MOM capacitor and utilizing the learning ability of a neural network to learn the relationship between a structural parameter and a performance parameter of a MOM capacitor, the performance parameters of a MOM capacitor to be tested can be quickly predicted, the modeling cycle is shortened, and the prediction accuracy of the MOM capacitor model for the MOM capacitor performance parameters is improved through training and learning of a large amount of sample data; the sample data is normalized, so that the convergence speed of the convolutional neural network model of the MOM capacitor can be accelerated, the robustness of the convolutional neural network model is improved, and the prediction accuracy is further improved; and by adding a batch normalization layer to the convolutional neural network model, the training and convergence speed of the convolutional neural network model are accelerated, the gradient propagation is improved, the generalization ability of the model is improved, and the prediction accuracy is further improved.

[0010] The present invention is applicable to the prediction of MOM capacitor performance parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0012] Figure 1The present invention is a flowchart of a MOM capacitor modeling method based on a convolutional neural network according to an embodiment of the present invention; Figure 2 It is the structural parameter schematic diagram of MOM capacitor; Figure 3 The capacitance curve of the MOM capacitor to be measured obtained by electromagnetic simulation is compared with the capacitance curve obtained by adopting the multilayer perceptron neural network and the present embodiment method; Figure 4 The figure is a comparison diagram of the quality factor curve of the MOM capacitor to be measured obtained by electromagnetic simulation and the quality factor curve obtained by using the multilayer perceptron neural network and the method of this embodiment. DETAILED DESCRIPTION

[0013] The present invention is further described below in conjunction with embodiments, but those skilled in the art should understand that the present invention is not limited to the following embodiments, and any improvements and equivalent changes made based on the specific embodiments of the present invention are within the scope of protection of the claims of the present invention.

[0014] Embodiment A MOM capacitor modeling method based on convolutional neural network A MOM capacitor modeling method based on convolutional neural network, such as Figure 1 As shown, the following steps are performed in sequence: S1. Obtain the structural parameter values ​​and simulation frequency values ​​of multiple groups of MOM capacitors and the corresponding performance parameter values ​​through electromagnetic simulation software ADS software or HFSS software as sample data; the structural parameters of MOM capacitors are as follows: Figure 2 As shown, including the finger length L C , Finger Width W C and interdigitation spacing S C , performance parameters include capacitance C and quality factor Q; S2. normalizing the sample data to obtain normalized sample data; S3. Construct a convolutional neural network model, which includes an input layer, a convolution layer, a batch normalization layer, an activation function layer and a fully connected layer connected in sequence; the width and height of the input layer of the convolutional neural network model are set to 1, and the number of channels is set to 4, which are used to input the normalized finger length, finger width and finger spacing of the MOM capacitor, as well as the normalized simulation frequency value; the number of convolution kernels of the convolution layer is set to 9, the size of the convolution kernel is 1×1×4, padding=0, step size=1, and the initial weight and bias of the convolution kernel are generated randomly; the batch size of the batch normalization layer is set to 128; the activation function used in the activation function layer is the Leaky ReLU function, and the negative slope of the Leaky ReLU function is set to 0.01; the number of neurons in the fully connected layer is 2, and the fully connected layer serves as the output layer of the convolutional neural network model to output the normalized structural parameter values ​​of the MOM capacitor; S4. The structural parameter values ​​and simulation frequency values ​​in the normalized sample data are used as the input of the convolutional neural network model, and the performance parameter values ​​in the normalized sample data are used as the output of the convolutional neural network model to train the convolutional neural network model. In this embodiment, the Adam optimization algorithm is used to iteratively optimize the convolutional neural network model. The number of training times is set to 1000 times, the initial learning rate is set to 0.01, and the learning rate is attenuated by half after every 250 training times. The other settings are all the default settings of the algorithm to obtain a trained convolutional neural network model. S5, after the structural parameter value and the simulation frequency value of the MOM capacitor to be measured are normalized, input to the trained convolutional neural network model, output the corresponding normalized performance parameter value, then the normalized performance parameter value is denormalized, and the predicted value of the MOM capacitor performance parameter to be measured is obtained.

[0015] The structural parameters of the MOM capacitor to be tested are input into a trained multilayer perceptron neural network to predict the performance parameters. The multilayer perceptron neural network includes an input layer, a hidden layer and an output layer. The number of neurons in the input layer is 4, the number of neurons in the hidden layer is 9, and the number of neurons in the output layer is 2. The sigmoid function is used as the activation function of the hidden layer neurons, and the gradient descent algorithm is used as the parameter training method of the multilayer perceptron neural network to obtain the predicted values ​​of the performance parameters of the MOM capacitor to be tested by the multilayer perceptron neural network.

[0016] The structural parameters of the MOM capacitor to be tested are input into the electromagnetic simulation software for simulation. The simulation frequency value interval is (0.5, 10), and the unit is GHz. The capacitance curve and quality factor curve of the MOM capacitor to be tested are obtained. The capacitance curve obtained by electromagnetic simulation is compared with the capacitance curve obtained by using the multilayer perceptron neural network and the method of this embodiment. Figure 3As shown, the quality factor curve obtained by electromagnetic simulation is compared with the quality factor curve obtained by using the multilayer perceptron neural network and the method of this embodiment. Figure 4 As shown. Figure 3 It can be seen from the figure that the capacitance curve of the MOM capacitor to be measured obtained by the present implementation method is closer to the result of electromagnetic simulation than the capacitance curve of the MOM capacitor to be measured obtained by the multilayer perceptron neural network. The error between the capacitance curve of the MOM capacitor to be measured obtained by the present implementation method and the electromagnetic simulation result is 1.37%; Figure 4 It can be seen that the quality factor curve of the MOM capacitor to be measured obtained by adopting the present implementation method is closer to the result of electromagnetic simulation than the quality factor curve of the MOM capacitor to be measured obtained by adopting the multilayer perceptron neural network, and the error between the quality factor curve of the MOM capacitor to be measured obtained by adopting the present implementation method and the electromagnetic simulation result is 1.53%.

Claims

1. A MOM capacitor modeling method based on convolutional neural network, characterized in that, The process includes the following steps: S1. Obtaining structural parameter values ​​and simulation frequency values ​​of multiple groups of MOM capacitors and corresponding performance parameter values ​​as sample data through electromagnetic simulation software; S2. normalizing the sample data to obtain normalized sample data; S3. Construct a convolutional neural network model, which includes an input layer, a convolution layer, a batch normalization layer, an activation function layer, and a fully connected layer connected in sequence; S4, using the structural parameter values ​​and simulation frequency values ​​in the normalized sample data as inputs of the convolutional neural network model, using the performance parameter values ​​in the normalized sample data as outputs of the convolutional neural network model, training the convolutional neural network model, and obtaining a trained convolutional neural network model; S5, after the structural parameter value and the simulation frequency value of the MOM capacitor to be measured are normalized, input to the trained convolutional neural network model, output the corresponding normalized performance parameter value, then the normalized performance parameter value is denormalized, and the predicted value of the MOM capacitor performance parameter to be measured is obtained.

2. a kind of MOM capacitor modeling method based on convolutional neural network according to claim 1, is characterized in that, The structural parameters of MOM capacitors include finger length, finger width and finger spacing, and the performance parameters include capacitance and quality factor.

3. a kind of MOM capacitor modeling method based on convolutional neural network according to claim 2, it is characterised in that, The width and height of the input layer of the convolutional neural network model are set to 1, and the number of channels is set to 4; the number of convolution kernels of the convolution layer is set to 9, the size of the convolution kernel is 1×1×4, padding=0, step size=1, and the initial weight and bias of the convolution kernel are generated randomly; the batch size of the batch normalization layer is set to 128; the activation function used in the activation function layer is the Leaky ReLU function; the number of neurons in the fully connected layer is 2, and the fully connected layer serves as the output layer of the convolutional neural network model.

4. according to a kind of MOM capacitance modeling method based on convolutional neural network described in any one of claim 1-3, it is characterised in that, The training method of the convolutional neural network model in step S4 uses the Adam optimization algorithm to iteratively optimize the convolutional neural network model to obtain a trained convolutional neural network model.

Images