Method, system, device and medium for establishing hall device verilog-a simulation model based on artificial neural network

Abstract

The application discloses a Hall device Verilog-A simulation model establishment method, system, device and medium based on an artificial neural network, and the method comprises the following steps: establishing a Hall device equivalent model, decomposing the Hall device into a plurality of equivalent devices according to the established equivalent model, obtaining electrical parameters of the equivalent devices, obtaining trained network model parameters, extracting neural network model parameters, describing the neural network model in Verilog-A according to the extracted neural network model parameters and standardization and normalization of the neural network model, obtaining a neural network model described based on the Verilog-A language, performing behavior level modeling on connection relationships between output data of the neural network model described based on the Verilog-A language and the equivalent model established in step 1, and obtaining a Hall device Verilog-A simulation model based on the artificial neural network; the system, the device and the medium are used for realizing the Hall device Verilog-A simulation model establishment method based on the artificial neural network; and the application has the characteristics of simple model structure, low modeling difficulty, short modeling cycle and simple modeling process.

Description

Technical Field

[0001] This invention relates to the field of microelectronic device simulation technology, specifically to a method, system, device, and medium for establishing a Verilog-A simulation model of a Hall device based on an artificial neural network. Background Technology

[0002] Currently, semiconductor device modeling mainly involves three steps: 1. Analyzing the process and structure used in the device to find suitable equations to describe its characteristics; 2. Substituting process parameters into the equations and extracting model parameters from the equations by fitting measurement data; 3. Forming a Spice model or Verilog-A model for circuit simulation. For semiconductor devices such as MOSFETs, BJTs, and Diodes, there are mostly standard and usable equations, such as BSIM3 and BSIM, but such a standard is lacking for Hall devices. Current models used to describe the characteristics of Hall devices are either too complex or too idealized, making it difficult to accurately describe Hall devices. Furthermore, the physical model of the device is related to the process and structure used, requiring knowledge of the corresponding process characteristics and structural features, as well as TCAD modeling with the appropriate process parameters. All of this necessitates communication with the process manufacturer. Therefore, the existing technology for building Hall device simulation models is costly to learn, inefficient, and time-consuming.

[0003] Patent application with publication number [CN102236736A] discloses "a circuit simulation model of a cross-shaped CMOS integrated Hall magnetic sensor," providing a circuit simulation model of a cross-shaped CMOS integrated Hall magnetic sensor that can be simulated on a general electronic circuit simulator. The model consists of a centrally symmetric network composed of 12 nonlinear N-well resistors, 8 PN junction capacitors, and 4 current-controlled voltage sources. The cross-shaped device is divided into a central region and four interdigitated regions. The active region of the central region is represented by an RH-RD-CB network, and the active regions of the interdigitated regions are represented by an RF-CF network. Due to the numerous physical and geometric effects of the Hall sensor that need to be considered, the model has the disadvantages of complex modeling and a narrow application range.

[0004] Patent application [CN110991062A] discloses "a method for implementing a Verilog-A simulation model of an OLED," which includes the following steps: 1) acquiring the current-voltage and capacitance-voltage data of the device; 2) establishing a curve fitting polynomial based on the data; 3) establishing a Verilog-A expression to obtain the simulated OLED model; 4) performing curve fitting for current-voltage and capacitance-voltage respectively; and 5) validating the model. Since polynomial fitting is used to fit the characteristic curves of the device, multivariate polynomial regression requires establishing the relationship between multiple independent input and output variables. Furthermore, nonlinear fitting requires piecewise fitting of the characteristic curves, necessitating a certain understanding of the input-output relationship. This method is complex and has a long modeling cycle. Summary of the Invention

[0005] To overcome the shortcomings of the prior art, the present invention aims to provide a method, system, device, and medium for establishing a Verilog-A simulation model of a Hall device based on an artificial neural network. By training the artificial neural network, the relationship between the input of the Hall device and the equivalent model parameters is fitted. The Verilog-A language is used to describe the neural network model and the equivalent model of the Hall device, resulting in a simulation model that can be used for circuit design. This model has the advantages of simple model structure, low modeling difficulty, short modeling cycle, and simple modeling process.

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

[0007] A method for establishing a Verilog-A simulation model of a Hall device based on an artificial neural network includes the following steps:

[0008] Step 1: Establish an equivalent model of the Hall device, and decompose the Hall device into several equivalent devices based on the established equivalent model;

[0009] Step 2: Use the van der Bauer method to measure the characteristic curves of the Hall device under different measurement conditions, and calculate the electrical parameters of the equivalent device obtained in Step 1 based on the characteristic curves.

[0010] Step 3: Record the measurement conditions applied to the Hall device in Step 2 and the electrical parameters of the equivalent device under these conditions as recorded data. The applied measurement conditions serve as the features of the recorded data, and the electrical parameters of the equivalent device serve as the labels of the recorded data. The recorded data is then standardized and normalized to form a dataset.

[0011] Step 4: Divide the dataset obtained in Step 3 into a training set, a validation set, and a test set. Input the training set into the neural network model to iteratively optimize the neural network model parameters. After training, input the validation set into the trained model for validation and optimize the hyperparameters of the neural network model. Determine whether to stop training based on the error until the trained network model parameters are obtained. The test set is used to evaluate the final performance of the model.

[0012] Step 5: Extract the neural network model parameters obtained in Step 4. Standardize and normalize the extracted neural network model parameters and the neural network model and describe them in Verilog-A to obtain a neural network model described in Verilog-A. Perform behavioral-level modeling on the connection relationship between the output data of the neural network model described in Verilog-A and the equivalent model established in Step 1 to obtain the Verilog-A simulation model of the Hall device based on the artificial neural network.

[0013] The equivalent device in step 1 is one or more of the following: resistor, capacitor, inductor, voltage source, and current source.

[0014] The measurement conditions in step 2 include temperature, magnetic field, and bias current. The electrical parameters of the equivalent device can be resistance, capacitance, inductance, voltage, or current, depending on the equivalent device.

[0015] The neural network model in step 4 includes an input layer, an output layer, and one or more hidden layers. Each layer contains one or more neurons. The number of neurons in the input layer corresponds to the features of the dataset obtained in step 3, and the number of neurons in the output layer corresponds to the labels of the dataset obtained in step 3. The number of hidden layers and the number of neurons in each layer are empirical values. The value of a single neuron in the hidden and output layers is obtained by weighted summation of the values ​​of the n neurons in the previous layer and transformation using an activation function f(x). The calculation formula is as follows:

[0016]

[0017] Among them, w i Let x be the weight from the i-th neuron in the previous layer to this neuron. i The input to the i-th neuron is b, which is the bias from the previous layer to this layer. The activation function f(x) can be one of ReLU, Sigmoid, or Tanh.

[0018] The specific process of step 4 is as follows:

[0019] Step 4.1: Divide the dataset obtained in Step 3 into a training set, a validation set, and a test set;

[0020] Step 4.2: Input the training set from Step 4.1 into the neural network model for iterative training. Calculate the error for each training iteration based on the loss function. The formula for calculating the error is:

[0021]

[0022] Where n is the number of samples in the training sample set, y i For predicted values, For tags;

[0023] Step 4.3: During each iteration of training, the weights and biases of the neural network model are updated based on the backpropagation error obtained in Step 4.2. An optimization algorithm is used to optimize the learning rate of the neural network model. After each iteration of training is completed, the validation set is input into the trained model for validation. Based on the validation results of each iteration, it is determined whether the average relative error of the latest trained model validation result is lower than the set value. If so, the latest trained model is taken as the trained neural network model. If not, Step 4.2-4.3 is repeated until the requirements are met, and the trained neural network model is obtained.

[0024] Step 4.4: Input the test set into the trained neural network model obtained in step 4.3 to obtain the final test results of the model, which are used to evaluate the final performance of the model.

[0025] The specific process of step 5 is as follows:

[0026] Step 5.1: Extract the model parameters of the trained neural network model obtained in Step 4. The model parameters include the weights of the connections between the layers in the model and the biases of each layer.

[0027] Step 5.2: When using Verilog-A to describe the Hall device simulation model, firstly, the input variables of the simulation model are preprocessed in the same way as in step 3; then, according to the structure of the neural network in step 4, the model parameters extracted in step 5.1 and equation (5) are described using Verilog-A, and the output data of the neural network model are normalized and standardized in the same way as in step 3 in Verilog-A to obtain the neural network model described by Verilog-A.

[0028] Step 5.3: The connection relationship between the output data of the neural network model based on Verilog-A obtained in Step 5.2 and the equivalent model established in Step 1 is described in Verilog-A at the behavioral level, and finally a Hall device model that can be used for circuit design simulation is obtained.

[0029] A system for establishing a Verilog-A simulation model of a Hall device based on an artificial neural network includes:

[0030] The Hall device equivalent model establishment module is used to establish the equivalent model of the Hall device and decompose the Hall device into several equivalent devices based on the established equivalent model.

[0031] Data measurement module: Uses the van der Bauer method to measure the characteristic curves of the Hall device under different measurement conditions, and calculates the electrical parameters of the equivalent device based on the characteristic curves;

[0032] The dataset creation module is used to record the measurement conditions applied to the Hall device and the electrical parameters of the equivalent device under those conditions as recorded data. The applied measurement conditions serve as the features of the recorded data, and the electrical parameters of the equivalent device serve as the labels of the recorded data. The recorded data is preprocessed to form a dataset.

[0033] Neural network model training module: The dataset obtained in step 3 is divided into training set, validation set and test set. The training set is input into the neural network model to iteratively optimize the neural network model parameters. After one round of training, the validation set is input into the trained model for validation, optimize the hyperparameters of the neural network model, and determine whether to stop training based on the error until the trained network model parameters are obtained. The test set is used to evaluate the final effect of the model.

[0034] The Verilog-A simulation model building module is used to extract the neural network model parameters obtained in step 4, standardize and normalize the extracted neural network model parameters and the neural network model, and describe them in Verilog-A to obtain a neural network model described in Verilog-A. The module then performs behavioral-level modeling on the connection relationship between the output data of the neural network model described in Verilog-A and the equivalent model established in step 1 to obtain the Verilog-A simulation model of the Hall device based on the artificial neural network.

[0035] A device for establishing a Verilog-A simulation model of a Hall effect device based on an artificial neural network includes:

[0036] Memory: Used to store the computer program that implements the method for establishing a Verilog-A simulation model of a Hall device based on an artificial neural network;

[0037] Processor: Used to implement a method for establishing a Verilog-A simulation model of a Hall device based on an artificial neural network when executing the computer program.

[0038] A computer-readable storage medium storing a computer program that, when executed by a processor, enables a method for establishing a Verilog-A simulation model of a Hall device based on an artificial neural network.

[0039] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0040] 1. This invention achieves the fitting of the relationship between the input of the Hall device and the equivalent model parameters by training a neural network. After training with a large amount of data, the Hall device Verilog-A simulation model based on the artificial neural network is generated. Therefore, it has the characteristics of data-driven adaptability, does not rely on theoretical formulas, and the implementation process can be automated, with a short modeling cycle and simple modeling process.

[0041] 2. Because of its data-driven adaptability, this invention treats the device under test as a black box. The measured data can be used to build a model without relying on the derivation of theoretical formulas. Therefore, there is no need to consider the internal process and structure of the device. After describing it using Verilog-A language, a simulation model that can be used for circuit design can be obtained. It has the characteristics of simple model structure and low modeling difficulty.

[0042] 4. Because this invention uses Verilog-A language to describe the neural network model and the equivalent model of the Hall device, it can be automated, thus shortening the modeling cycle and simplifying the modeling process.

[0043] In summary, this invention achieves the fitting of the relationship between the input of the Hall device and the equivalent model parameters by training an artificial neural network, and uses Verilog-A language to describe the neural network model and the equivalent model of the Hall device, resulting in a simulation model that can be used for circuit design. It has the characteristics of simple model structure, low modeling difficulty, short modeling cycle and simple modeling process. Attached Figure Description

[0044] Figure 1 This is a flowchart of the method of the present invention.

[0045] Figure 2 This is a schematic diagram of the equivalent model of the horizontal Hall device in an embodiment of the present invention.

[0046] Figure 3 This is a schematic diagram of the artificial neural network model structure according to an embodiment of the present invention.

[0047] Figure 4 This is a comparison chart of the output results of the neural network model in this embodiment of the invention with the test set, wherein... Figure 4 (a) is V H The comparison chart, Figure 4 (b) is R V Comparison chart, Figure 4 (c) is R D The comparison chart.

[0048] Figure 5This is a comparison chart of simulated values ​​and measured values ​​output by the simulation model in an embodiment of the present invention, wherein... Figure 5 (a) is a graph showing the change of Hall voltage with temperature. Figure 5 (b) is a graph showing the change of input impedance with temperature. Detailed Implementation

[0049] The present invention will now be described in detail with reference to the accompanying drawings, using a horizontal Hall effect device as an example.

[0050] See Figure 1 A method for establishing a Verilog-A simulation model of a Hall device based on an artificial neural network includes the following steps:

[0051] See Figure 2 Step 1: Establish an equivalent model of the horizontal Hall device. Based on the established equivalent model, decompose the horizontal Hall device into several equivalent devices. The equivalent model of the horizontal Hall device consists of a centrally symmetrical structural network composed of 8 resistors and four controlled voltage sources. A, B, C, and D represent the four contact holes of the horizontal Hall device. R1 represents the resistance between two opposite contact holes A and C, B and D. R2 represents the resistance between two adjacent contact holes A and B, B and C, C and D, and D and A. The four controlled voltage sources CCVS1, CCVS2, CCVS3, and CCVS4 represent the Hall voltage generated by the Hall device under the action of a magnetic field.

[0052] Step 2: Measure the characteristic parameters of the Hall device using the van der Bauer method. Measure the Hall effect and resistance of the Hall device under different temperatures, magnetic fields, and bias currents. The test conditions are: temperature T -40℃, -20℃, 0℃, 25℃, 50℃, 85℃, 100℃, and 125℃; magnetic field B range -100mT to 100mT with a step size of 20mT; bias current Ibias range 0 to 100µA with a step size of 10µA. Calculate the IV curve and Hall voltage V based on the measurements. D The resistance R between the two ports of the Hall device can be calculated from the IV curve. V and R D Based on the equivalent model established in this example, the electrical parameters of the equivalent device, namely the resistance values ​​of R1 and R2, are calculated. The calculation formulas for R1 and R2 are as follows:

[0053]

[0054]

[0055] Among them, R V R is the input impedance of the horizontal Hall effect relative to the two contact holes; DThe input impedance is the input impedance of two adjacent contact holes in the horizontal Hall effect.

[0056] Step 3: Apply the measurement conditions to the Hall device in Step 2: temperature T, magnetic field B, and bias current Ibias, as well as the electrical parameters R1, R2, and V of the equivalent device under these conditions. H The data is recorded, with the applied measurement conditions serving as characteristics of the recorded data, and the electrical parameters of the equivalent device serving as labels for the recorded data. Temperature T, magnetic field B, and bias current Ibias are normalized, and R1, R2, and Hall voltage V are also included. H Normalization is performed to form a dataset;

[0057] The methods for standardization and normalization are shown in equations (3) and (4):

[0058]

[0059]

[0060] Where μ is the mean of the sample data, σ is the standard deviation of the sample data, and y max y represents the maximum value of the sample data. min It is the minimum value of the sample data.

[0061] Step 4: Divide the dataset obtained in Step 3 into a training set, a validation set, and a test set. Input the training set into the neural network model to iteratively optimize the neural network model parameters. After training, input the validation set into the trained model to obtain the hyperparameters of the neural network model.

[0062] See Figure 3 , is the neural network model structure used in this example. The selected neural network has 3 hidden layers, with 32, 16 and 8 neurons in the 3 hidden layers, respectively. The ReLU function is used as the activation function between the current layer and the next layer in each hidden layer.

[0063] The neural network model in step 4 consists of an input layer, an output layer, and one or more hidden layers. The input layer corresponds to the features of the dataset in step 3, and the output layer corresponds to the labels of the dataset in step 3. The number of hidden layers and the number of neurons in each layer are empirical values. The value of a single neuron in each layer is calculated by weighting and summing the values ​​of the n neurons in the previous layer and then using the ReLU function as the activation function for non-linear transformation. The calculation formula is as follows:

[0064]

[0065] Among them, w i Let x be the weight from the i-th neuron in the previous layer to this neuron. iThe input of the i-th neuron, b is the bias from the previous layer to this layer;

[0066] The specific process of step 4 is as follows:

[0067] Step 4.1: Divide the dataset obtained in Step 3 into a training set, a validation set, and a test set in a ratio of 8:1:1;

[0068] Step 4.2: Input the training set from Step 4.1 into the neural network model for training. Set 32 ​​data samples to be captured per training session. Calculate the error for each training session based on the loss function. The error calculation formula is as follows:

[0069]

[0070] Where n is the number of samples in the training sample set, y i For predicted values, For tags;

[0071] Step 4.3: During each iteration of training, the weights and biases of the neural network model are updated based on the backpropagation error obtained in Step 4.2. An optimization algorithm (such as Adam or SGD) is used to optimize the learning rate of the neural network. After each iteration of training, the validation set is input into the model for validation. Based on the validation results of each iteration, it is determined whether the average relative error of the model is lower than 0.005. If so, the latest trained model is taken as the trained neural network model. If not, Step 4.2-4.3 is repeated continuously until the expected result is met, and the trained neural network model is obtained.

[0072] Step 4.4: Input the test set into the trained neural network model obtained in Step 4.3 to obtain the final test results of the model, which are used to evaluate the final performance of the model.

[0073] Step 5: Extract the neural network model parameters obtained in Step 4. Based on the extracted neural network model parameters and the neural network model, standardize and normalize them. Describe Equations (3), (4) and (5) in Verilog-A to obtain a neural network model described in Verilog-A. Input the equivalent parameters obtained from the neural network model described in Verilog-A with the connection relationship of the equivalent model established in Step 1 to perform behavioral-level modeling and obtain the Verilog-A simulation model of the Hall device based on the artificial neural network.

[0074] The specific process of step 5 is as follows:

[0075] Step 5.1: Extract the model parameters of the neural network model obtained in Step 4. The model parameters include the weights w between the layers in the model and the biases b of each layer.

[0076] Step 5.2: When using Verilog-A to describe the Hall device simulation model, first, the input data of the simulation model is standardized in the same way as in step 3; then, according to the structure of the neural network in step 4, the parameters w and b extracted in step 5.1 and equation (5) are described using Verilog-A; the output data of the neural network model is denormalized to obtain the neural network model described by Verilog-A; the denormalization formula is as follows:

[0077] y new =(y max -y min )y+y min #(7)

[0078] Where y is the data to be inversely normalized, y max y represents the maximum value of the sample data. min It is the minimum value of the sample data, y new The data is after inverse normalization;

[0079] Step 5.3: Based on the connection relationship between the output of the Verilog-A-based neural network model obtained in Step 5.2 and the equivalent model established in Step 1, perform a behavioral-level description in Verilog-A to finally obtain a Hall device model that can be used for circuit design simulation.

[0080] See Figure 4 , Figure 4 This displays the degree of deviation between the neural network model's predicted values ​​and the actual measured values ​​on the test set. The horizontal axis represents the actual measured values, and the vertical axis represents the predicted values. The distance of the black dots from the diagonal line indicates the degree of deviation. Figure 4 (a) represents the Hall voltage V H The black dots are closely distributed near the diagonal, and the calculated average relative error is 0.0031, indicating that the neural network model accurately reflects the Hall voltage V. H The prediction effect is relatively good; Figure 4 (b) represents resistance R V The black dots are closely distributed near the diagonal, and the calculated average relative error is 0.0021, indicating that the neural network model has good performance on R. V The prediction effect is relatively good; Figure 4 (c) represents R D The black dots are closely distributed near the diagonal, and the calculated average relative error is 0.0018, indicating that the neural network model has good performance on R. D The prediction effect is good. In summary, the Verilog-A simulation model of Hall device based on artificial neural network established by this invention has a good prediction effect.

[0081] See Figure 5 , Figure 5 (a) represents the Hall voltage V at different temperatures in the established simulation model. H The comparison between the temperature change curve and the measured values ​​shows that the error between the characteristic curve model output and the actual data is 0.93%. Figure 5 (b) The resistance R between the two ports of the Hall device in the established simulation model at different temperatures. V and R D The comparison between the temperature change curve and the measured value shows that the error between the characteristic curve model output and the actual data is 0.15% and 0.13%, respectively. This indicates that the prediction results of the Hall device Verilog-A simulation model based on artificial neural network established by this invention are close to the actual data and the prediction results are accurate.

[0082] A system for establishing a Verilog-A simulation model of a Hall device based on an artificial neural network includes:

[0083] The Hall device equivalent model establishment module is used to establish the equivalent model of the Hall device and decompose the Hall device into several equivalent devices based on the established equivalent model.

[0084] Data measurement module: Uses the van der Bauer method to measure the characteristic curves of the Hall device under different measurement conditions, and calculates the electrical parameters of the equivalent device based on the characteristic curves;

[0085] The dataset creation module is used to record the measurement conditions applied to the Hall device and the electrical parameters of the equivalent device under those conditions as recorded data. The applied measurement conditions serve as the features of the recorded data, and the electrical parameters of the equivalent device serve as the labels of the recorded data. The recorded data is preprocessed to form a dataset.

[0086] Neural network model training module: The dataset obtained in step 3 is divided into training set, validation set and test set. The training set is input into the neural network model to iteratively optimize the neural network model parameters. After one round of training, the validation set is input into the trained model for validation, optimize the hyperparameters of the neural network model, and determine whether to stop training based on the error until the trained network model parameters are obtained. The test set is used to evaluate the final effect of the model.

[0087] The Verilog-A simulation model building module is used to extract the neural network model parameters obtained in step 4, standardize and normalize the extracted neural network model parameters and the neural network model, and describe them in Verilog-A to obtain a neural network model described in Verilog-A. The module then performs behavioral-level modeling on the connection relationship between the output data of the neural network model described in Verilog-A and the equivalent model established in step 1 to obtain the Verilog-A simulation model of the Hall device based on the artificial neural network.

[0088] A device for establishing a Verilog-A simulation model of a Hall effect device based on an artificial neural network includes:

[0089] Memory: Used to store the computer program that implements the method for establishing a Verilog-A simulation model of a Hall device based on an artificial neural network;

[0090] Processor: Used to implement a method for establishing a Verilog-A simulation model of a Hall device based on an artificial neural network when executing the computer program.

[0091] A computer-readable storage medium storing a computer program that, when executed by a processor, enables a method for establishing a Verilog-A simulation model of a Hall device based on an artificial neural network.

Claims

1. A method for establishing a Verilog-A simulation model of a Hall device based on an artificial neural network, characterized in that, Includes the following steps: Step 1: Establish an equivalent model of the Hall device, and decompose the Hall device into several equivalent devices based on the established equivalent model; Step 2: Use the van der Bauer method to measure the characteristic curves of the Hall device under different measurement conditions, and calculate the electrical parameters of the equivalent device obtained in Step 1 based on the characteristic curves. Step 3: Record the measurement conditions applied to the Hall device in Step 2 and the electrical parameters of the equivalent device under these conditions as recorded data. The applied measurement conditions serve as the features of the recorded data, and the electrical parameters of the equivalent device serve as the labels of the recorded data. The recorded data is then standardized and normalized to form a dataset. Step 4: Divide the dataset obtained in Step 3 into a training set, a validation set, and a test set. Input the training set into the neural network model to iteratively optimize the neural network model parameters. After training, input the validation set into the trained model for validation and optimize the hyperparameters of the neural network model. Determine whether to stop training based on the error until the trained network model parameters are obtained. The test set is used to evaluate the final performance of the model. The neural network model comprises an input layer, an output layer, and one or more hidden layers. Each layer contains one or more neurons. The number of neurons in the input layer corresponds to the features of the dataset obtained in step 3, and the number of neurons in the output layer corresponds to the labels of the dataset obtained in step 3. The number of hidden layers and the number of neurons in each layer are empirical values. The value of a single neuron in the hidden and output layers is obtained by weighted summation of the values ​​of the n neurons in the previous layer, and then transformed by an activation function f(x). The calculation formula is as follows: in, Let be the weight from the i-th neuron in the previous layer to this neuron. The input to the i-th neuron, The bias from the previous layer to this layer is defined by the activation function f(x), which is one of ReLU, Sigmoid, or Tanh. The specific process is as follows: Step 4.1: Divide the dataset obtained in Step 3 into a training set, a validation set, and a test set; Step 4.2: Input the training set from Step 4.1 into the neural network model for iterative training. Calculate the error for each training iteration based on the loss function. The formula for calculating the error is: Where n is the number of samples in the training sample set. For predicted values, For tags; Step 4.3: During each iteration of training, the weights and biases of the neural network model are updated based on the backpropagation error obtained in Step 4.

2. An optimization algorithm is used to optimize the learning rate of the neural network model. After each iteration of training is completed, the validation set is input into the trained model for validation. Based on the validation results of each iteration, it is determined whether the average relative error of the latest trained model validation result is lower than the set value. If so, the latest trained model is taken as the trained neural network model. If not, Step 4.2-4.3 is repeated until the requirements are met, and the trained neural network model is obtained. Step 4.4: Input the test set into the trained neural network model obtained in Step 4.3 to obtain the final test results of the model, which are used to evaluate the final performance of the model. Step 5: Extract the neural network model parameters obtained in Step 4. Standardize and normalize the extracted neural network model parameters and the neural network model and describe them in Verilog-A to obtain a neural network model described in Verilog-A. Perform behavioral-level modeling on the connection relationship between the output data of the neural network model described in Verilog-A and the equivalent model established in Step 1 to obtain the Verilog-A simulation model of the Hall device based on the artificial neural network.

2. The method for establishing a Verilog-A simulation model of a Hall device based on an artificial neural network according to claim 1, characterized in that, The equivalent device in step 1 is one or more of the following: resistor, capacitor, inductor, voltage source, and current source.

3. The method for establishing a Verilog-A simulation model of a Hall device based on an artificial neural network according to claim 1, characterized in that, The measurement conditions in step 2 include temperature, magnetic field, and bias current. The electrical parameters of the equivalent device are resistance, capacitance, inductance, voltage, or current, depending on the equivalent device.

4. The method for establishing a Verilog-A simulation model of a Hall device based on an artificial neural network according to claim 1, characterized in that, The specific process of step 5 is as follows: Step 5.1: Extract the model parameters of the trained neural network model obtained in Step 4. The model parameters include the weights of the connections between the layers in the model and the biases of each layer. Step 5.2: When using Verilog-A to describe the Hall device simulation model, firstly, the input variables of the simulation model are preprocessed in the same way as in step 3; then, according to the structure of the neural network in step 4, the model parameters extracted in step 5.1 and equation (5) are described using Verilog-A, and the output data of the neural network model are normalized and normalized in the same way as in step 3 in Verilog-A to obtain the neural network model described by Verilog-A. Step 5.3: The connection relationship between the output data of the neural network model based on Verilog-A obtained in Step 5.2 and the equivalent model established in Step 1 is described in Verilog-A at the behavioral level, and finally a Hall device model that can be used for circuit design simulation is obtained.

5. A system for establishing a Verilog-A simulation model of a Hall device based on an artificial neural network, used to implement the method described in claim 1, characterized in that, include: The Hall device equivalent model establishment module is used to establish the equivalent model of the Hall device and decompose the Hall device into several equivalent devices based on the established equivalent model. Data measurement module: Uses the van der Bauer method to measure the characteristic curves of Hall devices under different measurement conditions, and calculates the electrical parameters of the equivalent device based on the characteristic curves; The dataset creation module is used to record the measurement conditions applied to the Hall device and the electrical parameters of the equivalent device under those conditions as recorded data. The applied measurement conditions serve as the features of the recorded data, and the electrical parameters of the equivalent device serve as the labels of the recorded data. The recorded data is preprocessed to form a dataset. Neural network model training module: The dataset obtained in step 3 is divided into training set, validation set and test set. The training set is input into the neural network model to iteratively optimize the neural network model parameters. After one round of training, the validation set is input into the trained model for validation, optimize the hyperparameters of the neural network model, and determine whether to stop training based on the error until the trained network model parameters are obtained. The test set is used to evaluate the final effect of the model. The Verilog-A simulation model building module is used to extract the neural network model parameters obtained in step 4, standardize and normalize the extracted neural network model parameters and the neural network model, and describe them in Verilog-A to obtain a neural network model described in Verilog-A. The module then performs behavioral-level modeling on the connection relationship between the output data of the neural network model described in Verilog-A and the equivalent model established in step 1 to obtain the Verilog-A simulation model of the Hall device based on the artificial neural network.

6. A device for establishing a Verilog-A simulation model of a Hall effect device based on an artificial neural network, characterized in that, include: Memory: for storing a computer program that implements the method for establishing a Verilog-A simulation model of a Hall device based on an artificial neural network as described in any one of claims 1-4; Processor: Used to implement the Verilog-A simulation model establishment method for Hall devices based on artificial neural networks as described in any one of claims 1-4 when executing the computer program.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, enables the implementation of the Verilog-A simulation model establishment method for a Hall device based on an artificial neural network as described in any one of claims 1-4.

Images