SiC MOSFET I-V characteristic modeling method and system based on physical information neural network

Abstract

The invention relates to the technical field of SiC MOSFETs, particularly discloses a SiC MOSFET I-V characteristic modeling method and system based on a physical information neural network, and aims at solving the problem that an existing SiC MOSFET simulation model cannot give consideration to high precision, high speed and practicability. Loss between a calculated value of a physical model and a predicted value of the neural network is added into a loss function of the neural network, and the model can be efficiently and accurately trained only through a small amount of data. According to the method and the system, the internal advantages of the physical neural network are fully utilized, and the actual physical behavior information of the device is embedded into the model, so that a modeling result is highly matched with experimental data in numerical value, strict physical interpretation is shown in theory, and key technical support is provided for industrial application of the silicon carbide power device.

Description

Technical Field

[0001] The present invention relates to the technical field of SiC MOSFET, and in particular to a SiC MOSFET I-V characteristic modeling method and system based on a physical information neural network. Background Art

[0002] Silicon carbide metal oxide semiconductor field-effect transistors (SiC MOSFETs) are a key material for the development of the third-generation semiconductor industry in power electronics. Their exceptional physical and electronic properties (such as high voltage and high temperature resistance and low conduction losses) enable them to meet the stringent requirements of high efficiency, miniaturization, and lightweight designs. Currently, SiC MOSFETs have been widely used in switching power supplies, motor drives, and photovoltaic power generation, and have become a focus of research. Although SiC MOSFETs have significantly improved the efficiency and power density of power electronics systems, their practical application still faces multiple challenges. During high-speed switching transients, high dv / dt (voltage change rate) and di / dt (current change rate) can easily induce electromagnetic interference (EMI), threatening the reliability and safety of converters and related equipment. Furthermore, high dv / dt stress can cause gate oxide breakdown, reducing device life and stability. The thermal stress generated by high operating temperatures and rapid switching increases the requirements for packaging design and heat dissipation solutions to prevent localized overheating and thermal failure. Drive circuit design demands extremely high response speed and accuracy, necessitating precise control of gate drive timing to suppress overshoot and ringing. Building an accurate SiC MOSFET model not only effectively simulates high-frequency switching transients, parasitic parameters, and thermal coupling effects, fundamentally suppressing sudden voltage and current changes and resonant noise, thereby reducing EMI risks, but also provides solid theoretical support for addressing thermal stress issues and optimizing drive circuit design.

[0003] Existing SiC MOSFET simulation models are primarily categorized into three types: numerical models, physical models, and behavioral models. Numerical models solve device equations numerically, simulating various physical effects such as recombination and bandgap shrinkage, making them suitable for solving complex problems. However, their applicability is limited and they rely on simulation tools such as SILVACO, Sentaurus TCAD, and MEDICI. Physical models, by describing the physical mechanisms of carrier evolution within the device, can accurately characterize the device's electrical characteristics and reveal its underlying mechanisms, making them suitable for in-depth research and practical applications. However, while physical models offer high simulation accuracy and consistent behavior, their slow simulation speed limits their widespread application. Behavioral models, based on equivalent circuits, offer advantages such as a simple structure and fast computational speed, making them particularly suitable for transient circuit simulation. However, behavioral models lack physical information, are less versatile, and their parameter extraction relies on expert experience. Summary of the Invention

[0004] The present invention provides a SiC MOSFET IV characteristic modeling method and system based on a physical information neural network, which solves the technical problem that existing SiC MOSFET simulation models have limited applicability, slow simulation speed, or low versatility, and cannot achieve high precision, high speed, and practicality at the same time.

[0005] To solve the above technical problems, the present invention provides a SiC MOSFET IV characteristic modeling method based on a physical information neural network, comprising the following steps:

[0006] Get SiC MOSFET at different gate-source voltages V GS , different drain-source voltage V DS The drain-source current I DS With V DS The relevant characteristic waveforms are used as the original data set;

[0007] Fitting the physical model of SiC MOSFET using the original data set to obtain optimal model parameters;

[0008] Constructing a physical information neural network model based on the neural network and the physical model;

[0009] Using the original data set to train and verify the constructed physical information neural network model;

[0010] The verified physical information neural network model is used to GS 、V DS Input value is predicted and output is I DS Predicted value;

[0011] Based on multiple groups of V GS 、V DS Input value and I DS The corresponding relationship of the predicted value is generated in V GS Under input value, V DS Input value and I DS IV characteristic waveforms between predicted values.

[0012] Furthermore, the constructed physical information neural network model includes a neural network and the physical model; the loss of the physical information neural network model includes data loss and physical losses The data loss is based on the I output of the neural network DS Predicted value and I DS Real value calculation; the physical loss is based on the I output of the neural network DS The predicted values and the physical model are based on V GS 、VDS I calculated from the input value DS Physical value calculation.

[0013] Furthermore, the loss function of the physical information neural network model is designed as data loss and physical losses Multiply by their respective weights and add the sum of the adjusted balance terms.

[0014] Furthermore, data loss The weight design is physical losses The weight design is The adjustment balancing term is designed as logσ u +logσ f , σ u , σ f denote the first trainable parameter and the second trainable parameter respectively.

[0015] Furthermore, the neural network includes an input layer, a hidden layer and an output layer; the input layer has two neurons, corresponding to V GS 、V DS Input value; the hidden layer has 10 neurons; the output layer has 1 neuron; the hidden layer uses the ReLU activation function; the output layer uses the linear activation function.

[0016] Furthermore, the SiC MOSFET is obtained at different gate-source voltages V GS , different drain-source voltage V DS The drain-source current I DS With V DS Related characteristic waveforms, specifically including the steps:

[0017] Obtain a simulation model of SiC MOSFET and build an IV characteristic simulation test circuit based on the simulation model; in this circuit, the junction temperature T of SiC MOSFET is pre-set. j , and the DC voltage V DS and V GS The initial value, step size and final value of to simulate different working conditions;

[0018] Export the simulation data of the IV characteristic simulation test circuit under different working conditions, and convert the simulation data into the IV characteristic simulation data of SiC MOSFET under different working conditions;

[0019] The IV characteristic simulation data is processed into three columns [V GS ,I DS ,V DS ] of CSV data.

[0020] Furthermore, the physical model adopts the EKV model, and the mathematical model of the EKV model is expressed as follows: DS =f(V GS ,V DS ,g m ,φ t ,κ S ,V th ,k,λ,n,α),f(V GS ,V DS ,g m ,φ t ,κ S ,V th ,k,λ,n,α) represents the same as V GS ,V DS , transconductance g m 、Thermovoltageφ t , subthreshold slope k S , threshold voltage V th , a function related to the power law index k, the channel length modulation coefficient λ, the first linear region parameter n and the second linear region parameter α.

[0021] Furthermore, the parameter g is determined by particle swarm optimization algorithm. m ,φ t ,κ S ,V th ,the optimal value of k,λ,n,α.

[0022] Furthermore, the junction temperature T j Fixed at 25°C, V GS The value range is [8V, 20V], V DS The value range is [0V, 40V], V GS and V DS The step size is 1V.

[0023] The present invention also provides a SiC MOSFET IV characteristic modeling system based on physical information neural network, which includes: a data set construction module, a physical model fitting module, a network model construction module, a model training and verification module and a model application module;

[0024] The dataset building module is used to obtain the SiC MOSFET at different gate-source voltages V GS , different drain-source voltage V DS The drain-source current I DS With V DS The relevant characteristic waveforms are used as the original data set;

[0025] The physical model fitting module is used to fit the physical model of SiC MOSFET using the original data set to obtain optimal model parameters;

[0026] The network model building module is used to build a physical information neural network model based on the neural network and the physical model;

[0027] The model training and verification module is used to train and verify the constructed physical information neural network model using the original data set;

[0028] The model application module is used to use the verified physical information neural network model to GS 、V DS Input value is predicted and output is I DS Predicted values, and based on multiple groups of V GS 、V DS Input value and I DS The corresponding relationship of the predicted value is generated in V GS Under input value, V DS Input value and I DS IV characteristic waveforms between predicted values.

[0029] The present invention provides a SiC MOSFET IV characteristic modeling method and system based on a physical information neural network. This method addresses the inability of existing SiC MOSFET simulation models to achieve high precision, high speed, and practicality. By integrating the physical model with the neural network, the loss function of the neural network incorporates the loss between the calculated values of the physical model and the predicted values of the neural network. This allows for efficient and accurate model training using only a small amount of data. This method and system leverages the inherent advantages of physical neural networks, embedding information about the device's actual physical behavior into the model. This ensures that the modeling results not only highly align numerically with experimental data but also exhibit a rigorous physical interpretation in theory, providing key technical support for the industrial application of silicon carbide power devices. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 Flowchart of a SiC MOSFET IV characteristic modeling method based on a physical information neural network provided by an embodiment of the present invention;

[0031] Figure 2 This is a schematic diagram of an IV characteristic simulation test circuit provided by an embodiment of the present invention;

[0032] Figure 3 1 is a waveform diagram of IV characteristic simulation in LTspice provided by an embodiment of the present invention;

[0033] Figure 4 is a flow chart of the particle swarm optimization algorithm provided by an embodiment of the present invention;

[0034] Figure 5This is a framework diagram of a physical information neural network model provided by an embodiment of the present invention;

[0035] Figure 6 is a flow chart for determining parameters of loss weights provided by an embodiment of the present invention;

[0036] Figures 7 to 19 V provided by the embodiment of the present invention GS IV characteristic waveform when the voltage is equal to 8 to 20. DETAILED DESCRIPTION

[0037] The following describes the embodiments of the present invention in detail with reference to the accompanying drawings. The embodiments are provided for illustrative purposes only and are not to be construed as limiting the present invention. The accompanying drawings are provided for reference and illustration only and do not constitute a limitation on the scope of protection of the present invention. Many changes may be made to the present invention without departing from the spirit and scope of the present invention.

[0038] The SiC MOSFET IV characteristic modeling method based on physical information neural network provided by the embodiment of the present invention is as follows: Figure 1 As shown in the flowchart, the steps include:

[0039] S1, obtain SiC MOSFET at different gate-source voltages V GS , different drain-source voltage V DS The drain-source current I DS With V DS The relevant characteristic waveforms are used as the original data set;

[0040] S2. Fit the physical model of SiC MOSFET using the original data set to obtain the optimal model parameters;

[0041] S3. Construct a physical information neural network model based on the neural network and physical model;

[0042] S4. Use the original data set to train and verify the constructed physical information neural network model;

[0043] S5. Use the verified physical information neural network model to GS 、V DS Input value is predicted and output is I DS Predicted value;

[0044] S6, based on multiple groups of V GS 、V DS Input value and I DS The corresponding relationship of the predicted value is generated in V GS Under input value, V DS Input value and I DS IV characteristic waveforms between predicted values.

[0045] In this embodiment, step S1 specifically includes the following steps:

[0046] S11. Obtain the simulation model (SPICE model) of SiC MOSFET from the SiC MOSFET manufacturer and import the simulation model into LTspice to build the IV characteristic simulation test circuit (such as Figure 2 As shown); In this circuit, the junction temperature T of the SiC MOSFET is preset. j , and the DC voltage V DS and V GS The initial value, step size and final value of to simulate different working conditions;

[0047] S12, export the simulation data of IV characteristic simulation test circuit under different working conditions, the simulation results are as follows Figure 3 As shown; then the Python script is used to convert the simulation data into IV characteristic simulation data (TXT data) of SiC MOSFET under different working conditions;

[0048] S13, use Python script to process IV characteristic simulation data into three columns [V GS ,I DS ,V DS ] of CSV data.

[0049] In step S11, this embodiment sets the junction temperature T j Fixed at 25°C, V GS The value range is [8V, 20V], V DS The value range is [0V, 40V], V GS and V DS The step size is 1V, so there are 13×41=533 simulated working conditions, as shown in Table 1 below.

[0050] Table 1

[0051]

[0052] In step S2 of this embodiment, the physical model is the EKV model, and the mathematical model of the EKV model is expressed as follows:

[0053] I DS =f(V GS ,V DS ,g m ,φt,κ S ,V th ,k,λ,n,α)

[0054] Among them, f(V GS ,V DS ,g m,φ t ,κ S ,V th ,k,λ,n,α) represents the same as V GS ,V DS , transconductance gm, thermal voltage φt, subthreshold slope κ S , threshold voltage V th , a function related to the power law index k, the channel length modulation coefficient λ, the first linear region parameter n and the second linear region parameter α.

[0055] Specifically, the mathematical model of the EKV model is expressed as:

[0056]

[0057] In this embodiment, the EKV model is fitted by the particle swarm optimization algorithm to obtain the optimal parameters. The implementation process of the particle swarm optimization algorithm is as follows: Figure 4 As shown, the specific steps include:

[0058] Initialize the position and velocity of the particle;

[0059] Evaluate the fitness of each particle;

[0060] Update individual best position;

[0061] Update the global best position;

[0062] Update particle velocity and position;

[0063] Determine whether the maximum number of iterations has been reached. If so, return the optimal parameters. Otherwise, proceed to the next iteration.

[0064] Finally, after algorithm fitting, the optimal parameters of the EKV model are shown in Table 2, and the root mean square error (RMSE) of the fitting effect is 3.581112.

[0065] Table 2 Parameter values of the EKV model

[0066] parameter describe Numerical <![CDATA[g m ]]> transconductance 10.450058980500728 <![CDATA[Φ t ]]> Thermal voltage 0.9259 <![CDATA[κ S ]]> Subthreshold slope 10.8392938328224541 <![CDATA[V th ]]> Threshold voltage 8.460410149433798 k Power law exponent 1.2103015114660944 λ Channel length modulation coefficient 1e-09 n Linear region parameters 1.2 α Linear region parameters 0.5

[0067] In step S3 of this embodiment, the physical information neural network model includes a neural network and a physical model. Figure 5 As shown in the structure diagram, the neural network consists of an input layer, a hidden layer, and an output layer. The input layer has two neurons x0 and x1, corresponding to V GS 、V DS Input value. The hidden layer has 10 neurons (h0 to h9). The output layer has 1 neuron y0, which outputs I DS The weights between x0 and x1 and h0 to h9 are w 00 to w19 , the weight between h0 to h9 and y0 is w o1 to w o9 The weight matrix and bias of the neural network are represented as w and b respectively. The hidden layer uses the ReLU activation function. The output layer uses the linear activation function.

[0068] In this embodiment, the loss of the physical information neural network model includes data loss and physical losses Data loss is based on the I output of the neural network DS Predicted value and I DS Real value calculation; physical loss is based on the I output of the neural network DS Predicted values and physical models are based on V GS 、V DS I calculated from the input value DS Physical value calculation. Data loss and physical losses The mean square error (MSE) is used as the loss function.

[0069] Specifically, the loss function of the physical information neural network model is designed as data loss and physical losses Multiplying by their respective weights and adding the sum of the adjustment balance items, the formula is expressed as:

[0070]

[0071] Among them, σ u , σ f Denote the first trainable parameter and the second trainable parameter, respectively, which are used to control the weights of data loss and physical loss.

[0072] In step S4 of this embodiment, the original data set is divided into a training set and a validation set in a ratio of 8:2.

[0073] In step S5 of this embodiment, the physical information neural network model is trained using the training set, and the trained physical information neural network model is verified using the verification set, and the model parameters with the best verification loss are selected. During the training process, the model automatically adjusts the parameter σ by minimizing the total loss function. u , σ f The value of , thereby dynamically balancing the weights of data loss and physical loss, and achieving optimal coordination between the two. The training process of the model is as follows Figure 6 During the validation process, the model evaluation results are shown in Table 3.

[0074] Table 3 Model evaluation results

[0075] Evaluation Metrics Assessment Value <![CDATA[Coefficient of determination (R 2 )]]> 0.9987 Root mean square error (RMSE) 0.0230 Mean Absolute Error (MAE) 0.0181

[0076] It should be noted that before model training, the training set and validation set need to be normalized (using maximum and minimum value normalization) and normalized to between [-1, 1].

[0077] In steps S5 and S6 of this embodiment, the trained neural network model is used to normalize the input data [V GS ,V DS ] to predict, get the normalized output result, reverse the normalized output result back to the original data range, and then get the I DS The predicted data are finally combined to output the IV characteristic waveform of SiC MOSFET. The effect of model fitting is as follows Figures 7 to 19 shown. Figures 7 to 19 V GS The IV characteristic waveform when the value is equal to 8 to 20 (step size is 1). Figures 7 to 19 It can be seen that the model prediction is relatively close to the real data, and the physical information neural network model provided by the present invention has high modeling accuracy.

[0078] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially, or in a different order, as long as the desired results of the technical solution of the present invention can be achieved. This embodiment is not limited here.

[0079] In order to implement the above method, an embodiment of the present invention further provides a SiCMOSFET IV characteristic modeling system based on a physical information neural network, which includes a data set construction module, a physical model fitting module, a network model construction module, a model training and verification module, and a model application module;

[0080] The dataset building module is used to obtain the SiC MOSFET at different gate-source voltages V GS , different drain-source voltage V DS The drain-source current I DS With V DS The relevant characteristic waveforms are used as the original data set;

[0081] The physical model fitting module is used to fit the physical model of SiC MOSFET using the original data set to obtain the optimal model parameters;

[0082] The network model building module is used to build a physical information neural network model based on the neural network and the physical model;

[0083] The model training and verification module is used to train and verify the constructed physical information neural network model using the original data set;

[0084] The model application module is used to use the verified physical information neural network model to GS 、V DS Input value is predicted and output is I DS Predicted values, and based on multiple groups of V GS 、V DS Input value and I DS The corresponding relationship of the predicted value is generated in V GS Under input value, V DS Input value and I DS IV characteristic waveforms between predicted values.

[0085] The specific work of each module has been described in detail in the above method and will not be repeated here.

[0086] It is emphasized that the embodiments described herein can be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with embodiments of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0087] Computer programs for implementing the methods and systems of the present invention can be written in any combination of one or more programming languages and stored in a computer-readable storage medium. These computer programs can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The computer program can be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0088] Computer-readable storage media can be tangible media that can contain or store a computer program for use with an instruction execution system, device or equipment or used in combination with an instruction execution system, device or equipment. Computer-readable storage media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, computer-readable storage media can be a machine-readable signal medium. More specific examples of machine-readable storage media can include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, compact disc read-only memories (CD ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0089] To demonstrate the superiority of the physical information neural network model over other modeling methods, the speed and accuracy of the model of the present invention were compared with the EKV model. The results are shown in Table 4 below. In Table 4, N represents the calculation speed of the EKV model. As can be seen from the data in Table 4, the method of the present invention outperforms the EKV model in both speed and accuracy.

[0090] Table 4 Comparison results between the model of the present invention and the EKV model

[0091] Model of the present invention EKV model Root mean square error 0.0230 3.581112 Computation speed in C language environment 5.6N N

[0092] The method of the present invention is compared with the ANN method in terms of accuracy and speed under the condition of controlling the training data set. The specific implementation is as follows: first, use the data set of 1-2), and continue to maintain the training and verification ratio of this data set at 8:2 and randomly divide it. Then the part of the verification set remains unchanged, and the part of the training set is reduced to train the model of the present invention and the ANN (artificial neural network) model respectively. The neural network comparison effect of the two models is shown in Table 5, where V in the table represents the calculation speed of the ANN model. The training rounds of the two models are set to 100 times, and the accuracy comparison effect of reducing the training set size is shown in Table 6. It can be seen from Tables 5 and 6 that as the training set is reduced, the accuracy of the ANN model decreases significantly, and the accuracy of the model of the present invention decreases slightly.

[0093] Table 5 Comparison results between the model of the present invention and the ANN model

[0094] Model of the present invention ANN model Number of neurons in the output layer 2 2 Number of hidden layers 1 1 Number of hidden layer neurons 10 10 Number of neurons in the output layer 1 1 Hidden layer activation function ReLU ReLU Output layer activation function Linear Linear Model calculation speed V V

[0095] Table 6 Accuracy comparison between the proposed model and the ANN model

[0096] training set Root mean square error of the model of the present invention ANN model root mean square error 100% training set 0.0554 0.0808 90% training set 0.0750 0.1037 80% training set 0.0847 0.1006 70% training set 0.0834 0.1129 60% training set 0.1014 0.1396 50% training set 0.0953 0.1507 40% training set 0.1050 0.1549 30% training set 0.1093 0.2754 20% training set 0.1509 0.2970 10% training set 0.2776 0.7593

[0097] In summary, the embodiments of the present invention provide a method and system for modeling SiC MOSFET IV characteristics based on a physical information neural network. This method addresses the problem that existing SiC MOSFET simulation models cannot achieve high precision, high speed, and practicality at the same time. The method fuses the physical model with the neural network, and incorporates the loss between the calculated value of the physical model and the predicted value of the neural network into the loss function of the neural network. This allows for efficient and accurate model training with only a small amount of data. This method and system fully leverage the inherent advantages of physical neural networks, embedding the device's actual physical behavior information into the model. This ensures that the modeling results are not only highly consistent with experimental data numerically, but also exhibit a rigorous physical interpretation in theory, providing key technical support for the industrial application of silicon carbide power devices.

[0098] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be considered as equivalent replacement methods and are included in the scope of protection of the present invention.

Claims

1. A SiC MOSFET IV characteristic modeling method based on physical information neural network, characterized in that: Including steps: Get SiC MOSFET at different gate-source voltages V GS , different drain-source voltage V DS The drain-source current I DS With V DS The relevant characteristic waveforms are used as the original data set; Fitting the physical model of SiC MOSFET using the original data set to obtain optimal model parameters; Constructing a physical information neural network model based on the neural network and the physical model; Using the original data set to train and verify the constructed physical information neural network model; The verified physical information neural network model is used to GS 、V DS Input value is predicted and output is I DS Predicted value; Based on multiple groups of V GS 、V DS Input value and I DS The corresponding relationship of the predicted value is generated in V GS Under input value, V DS Input value and I DS IV characteristic waveforms between predicted values.

2. The SiC MOSFET IV characteristic modeling method based on physical information neural network according to claim 1, characterized in that: The constructed physical information neural network model includes a neural network and the physical model; the loss of the physical information neural network model includes data loss and physical losses The data loss is based on the I output of the neural network DS Predicted value and I DS Real value calculation; the physical loss is based on the I output of the neural network DS The predicted values and the physical model are based on V GS 、V DS I calculated from the input value DS Physical value calculation.

3. The SiC MOSFET IV characteristic modeling method based on physical information neural network according to claim 2, characterized in that: The loss function of the physical information neural network model is designed to be data loss and physical losses Multiply by their respective weights and add the sum of the adjusted balance terms.

4. The SiC MOSFET IV characteristic modeling method based on physical information neural network according to claim 3, characterized in that: Data loss The weight design is physical damage The weight design is The adjustment balancing term is designed as logσ u +logσ f , σ u , σ f denote the first trainable parameter and the second trainable parameter respectively.

5. The SiC MOSFET IV characteristic modeling method based on physical information neural network according to claim 4, characterized in that: The neural network includes an input layer, a hidden layer and an output layer; the input layer has two neurons corresponding to V GS 、V DS Input value; the hidden layer has 10 neurons; the output layer has 1 neuron; the hidden layer uses the ReLU activation function; the output layer uses the linear activation function.

6. The SiC MOSFET IV characteristic modeling method based on physical information neural network according to any one of claims 1 to 5, characterized in that: Get SiC MOSFET at different gate-source voltages V GS , different drain-source voltage V DS The drain-source current I DS With V DS Related characteristic waveforms, specifically including the steps: Obtain a simulation model of SiC MOSFET and build an IV characteristic simulation test circuit based on the simulation model; in this circuit, the junction temperature T of SiC MOSFET is pre-set. j , and the DC voltage V DS and V GS The initial value, step size and final value of to simulate different working conditions; Export the simulation data of the IV characteristic simulation test circuit under different working conditions, and convert the simulation data into the IV characteristic simulation data of SiC MOSFET under different working conditions; The IV characteristic simulation data is processed into three columns [V GS ,I DS ,V DS ] of CSV data.

7. The SiC MOSFET IV characteristic modeling method based on physical information neural network according to claim 6, characterized in that: The physical model adopts the EKV model, and the mathematical model of the EKV model is expressed as follows: DS =f(V GS ,V DS ,g m ,φ t ,k S ,V th ,k,λ,n,α),f(V GS ,V DS ,g m ,φ t ,κ S ,V th ,k,λ,n,α) represents the same as V GS ,V DS , transconductance g m 、Thermovoltageφ t , subthreshold slope κ S , threshold voltage V th , a function related to the power law index k, the channel length modulation coefficient λ, the first linear region parameter n and the second linear region parameter α.

8. The SiC MOSFET IV characteristic modeling method based on physical information neural network according to claim 7, characterized in that: Determine the parameter g by particle swarm optimization algorithm m ,φ t ,κ S ,V th ,the optimal value of k,λ,n,α.

9. The SiC MOSFET IV characteristic modeling method based on physical information neural network according to claim 6, characterized in that: Junction temperature T j Fixed at 25°C, V GS The value range is [8V, 20V], V DS The value range is [0V, 40V], V GS and V DS The step size is 1V.

10. SiC MOSFET IV characteristic modeling system based on physical information neural network, characterized by: It includes data set construction module, physical model fitting module, network model construction module, model training and verification module and model application module; The dataset building module is used to obtain the SiC MOSFET at different gate-source voltages V GS , different drain-source voltage V DS The drain-source current I DS With V DS The relevant characteristic waveforms are used as the original data set; The physical model fitting module is used to fit the physical model of SiC MOSFET using the original data set to obtain optimal model parameters; The network model building module is used to build a physical information neural network model based on the neural network and the physical model; The model training and verification module is used to train and verify the constructed physical information neural network model using the original data set; The model application module is used to use the verified physical information neural network model to GS 、V DS Input value is predicted and output is I DS Predicted values, and based on multiple groups of V GS 、V DS Input value and I DS The corresponding relationship of the predicted value is generated in V GS Under input value, V DS Input value and I DS IV characteristic waveforms between predicted values.

Images