A power device electro-thermal analysis method based on a physical information graph neural network
By using a thermal analysis method based on physical information graph neural networks, the problems of finite element simulation convergence difficulty and high computational cost in the electrothermal analysis of power devices are solved, achieving efficient and accurate electrothermal analysis and supporting multi-scale device design.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHANGHAI INST OF MICROSYSTEM & INFORMATION TECH CHINESE ACAD OF SCI
- Filing Date
- 2026-03-24
- Publication Date
- 2026-07-31
AI Technical Summary
Existing electrothermal analysis methods for power devices suffer from difficulties in finite element simulation convergence and high computational costs during the design phase. They also rely on a large number of simulations and iterative empirical parameters, consuming significant human and computational resources.
A thermal analysis method based on physical information graph neural network is adopted. The grid data of power device is converted into nodes and edges of graph structure by encoder, and features are embedded. The node features are iteratively updated by multilayer perceptron with multilayer message passing network and residual connection. The electrothermal field distribution results are obtained by decoder mapping. The electrothermal equation constraints and physical information are fused to construct thermal analysis model.
It significantly improves modeling efficiency, shortens the design cycle, enhances prediction accuracy and generalization ability, reduces computational resource consumption, and supports multi-scale analysis and adaptation to different materials, structures, and bias scenarios.
Smart Images
Figure CN122490977A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of microelectronics and solid-state electronics, and in particular to a method for electrothermal analysis of power devices based on physical information graph neural networks. Background Technology
[0002] Power converters face severe reliability challenges under abnormal conditions, especially in non-clamped inductor switching and short-circuit events, where Joule heating-induced temperature rise is the primary failure mechanism. Currently, electrothermal analysis methods for power devices mainly focus on single-cell behavior or treating them as uniform heat sources for package-level finite element thermal simulation. However, multi-cell parallel simulation faces difficulties in finite element convergence and high computational costs, limiting electrothermal reliability analysis during the design phase. Furthermore, existing device structure design methods heavily rely on extensive simulations and iterative empirical parameters, consuming significant human and computational resources. Summary of the Invention
[0003] The technical problem to be solved by the present invention is to provide a power device electrothermal analysis method based on physical information graph neural network, which can realize efficient analysis of the static characteristics and transient electrothermal properties of power devices.
[0004] The technical solution adopted by this invention to solve its technical problem is: to provide a power device electrothermal analysis method based on physical information graph neural network, including the following steps:
[0005] Obtain the mesh data of the power device generated by the simulation software;
[0006] The mesh data of the power device structure is input into the thermal analysis model to obtain the electrothermal field distribution results of the power device; wherein, the thermal analysis model is constructed based on a physical information graph neural network and incorporates electrothermal equation constraints, including:
[0007] The encoder section is used to convert the grid data of the power device into nodes and edges of a graph structure, and embed features to obtain node features;
[0008] The processor section is used to iteratively update the node features through a multilayer perceptron with a multilayer messaging network and residual connections;
[0009] The decoder section is used to map the updated node features through a multilayer perceptron to obtain the electrothermal field distribution results of the power device.
[0010] The encoder part classifies the grid data of the power device into electrical attributes, thermal attributes and coordinate attributes, and embeds them uniformly as node features. At the same time, it broadcasts the external bias and position mark as global node information to the entire physical information graph neural network.
[0011] The thermal analysis model is trained using a dataset that covers various materials, structures, biases, and scales.
[0012] The loss function of the thermal analysis model is expressed as: ,in, The loss function of the thermal analysis model. The data loss term is represented as: , Represents the training dataset The amount of data in it This indicates the predicted electrothermal field distribution. This represents the true value of the electrothermal field. Indicates time, This represents a vector of operating parameters for power devices. This represents the training dataset; The loss term in the physical equation is expressed as: , Indicates the number of node features. This represents the node weight priority coefficient. This indicates the smoothness of the predicted electrothermal field. Represents the regularization coefficient. The PDE loss term is represented as: , Nonlinear operators representing the evolution of physical quantities Indicates carrier mobility; , , All of these are learnable standard deviation parameters.
[0013] The technical solution adopted by this invention to solve its technical problem is: to provide a power device electrothermal analysis device based on a physical information graph neural network, comprising:
[0014] The acquisition module is used to acquire the grid data of the power devices generated by the simulation software;
[0015] The analysis module is used to input the mesh data of the power device structure into the thermal analysis model to obtain the electrothermal field distribution results of the power device; wherein, the thermal analysis model is constructed based on a physical information graph neural network and incorporates electrothermal equation constraints, including:
[0016] The encoder section is used to convert the grid data of the power device into nodes and edges of a graph structure, and embed features to obtain node features;
[0017] The processor section is used to iteratively update the node features through a multilayer perceptron with a multilayer messaging network and residual connections;
[0018] The decoder section is used to map the updated node features through a multilayer perceptron to obtain the electrothermal field distribution results of the power device.
[0019] The encoder part classifies the grid data of the power device into electrical attributes, thermal attributes and coordinate attributes, and embeds them uniformly as node features. At the same time, it broadcasts the external bias and position mark as global node information to the entire physical information graph neural network.
[0020] The thermal analysis model is trained using a dataset that covers various materials, structures, biases, and scales.
[0021] The loss function of the thermal analysis model is expressed as: ,in, The loss function of the thermal analysis model. The data loss term is represented as: , Represents the training dataset The amount of data in it This indicates the predicted electrothermal field distribution. This represents the true value of the electrothermal field. Indicates time, This represents a vector of operating parameters for power devices. This represents the training dataset; The loss term in the physical equation is expressed as: , Indicates the number of node features. This represents the node weight priority coefficient. This indicates the smoothness of the predicted electrothermal field. Represents the regularization coefficient. The PDE loss term is represented as: , Nonlinear operators representing the evolution of physical quantities Indicates carrier mobility; , , All of these are learnable standard deviation parameters.
[0022] The technical solution adopted by the present invention to solve its technical problem is: to provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-mentioned power device electrothermal analysis method based on physical information graph neural network.
[0023] The technical solution adopted by the present invention to solve its technical problem is: to provide a computer-readable storage medium on which a computer program is stored, wherein when the computer program is executed by a processor, the steps of the above-mentioned power device electrothermal analysis method based on physical information graph neural network are implemented.
[0024] Beneficial effects
[0025] Due to the adoption of the above-mentioned technical solutions, this invention has the following advantages and positive effects compared with the prior art: The thermal analysis model proposed in this invention integrates the physical constraints of the electrothermal equation, significantly improving modeling efficiency and greatly increasing the inference speed compared with the conventional finite element method, far exceeding traditional simulation tools, and significantly shortening the design cycle of power devices; Under various operating conditions, the prediction accuracy of the thermal analysis model proposed in this invention is better than other neural network frameworks, and the generalization of the model is enhanced by physical constraints. At the same time, relying on the directness of graph networks, the thermal analysis model proposed in this invention does not require additional training data and retraining process, and can directly infer the electrothermal behavior of devices with new structural parameters. It supports multi-scale analysis from single cell to chip level, adapts to different materials, structures and bias scenarios, has strong generalization ability, and effectively reduces the consumption of human and computing resources. Attached Figure Description
[0026] Figure 1 This is a flowchart of the power device electrothermal analysis method based on a physical information graph neural network according to the first embodiment of the present invention.
[0027] Figure 2 This is a schematic diagram of the architecture of the thermal analysis model in the first embodiment of the present invention;
[0028] Figure 3 This is a schematic diagram of the network structure of the thermal analysis model in the first embodiment of the present invention;
[0029] Figure 4 This is a schematic diagram of grid data embedding in the first embodiment of the present invention;
[0030] Figure 5 This is a schematic diagram of multi-scale model verification in the first embodiment of the present invention;
[0031] Figure 6 This is a schematic diagram of dynamic performance comparison verification in the first embodiment of the present invention;
[0032] Figure 7 This is a schematic diagram of static performance comparison verification in the first embodiment of the present invention;
[0033] Figure 8 This is a schematic diagram illustrating the speed improvement effect in the first embodiment of the present invention. Detailed Implementation
[0034] The present invention will be further illustrated below with reference to specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be understood that after reading the teachings of this invention, those skilled in the art can make various alterations or modifications to the invention, and these equivalent forms also fall within the scope defined by the appended claims.
[0035] The first embodiment of the present invention relates to a power device electrothermal analysis method based on a physical information graph neural network, such as... Figure 1 As shown, it includes the following steps:
[0036] Step 1: Obtain the grid data of the power device generated by the simulation software.
[0037] Step 2: Input the mesh data of the power device structure into the thermal analysis model to obtain the electrothermal field distribution results of the power device.
[0038] like Figure 2 As shown, the thermal analysis model is constructed based on a Physical Information Graph Neural Network (PIGNN) and incorporates electrothermal equation constraints, including:
[0039] The encoder section is used to convert the grid data of the power device into nodes and edges of a graph structure, and embed features to obtain node features;
[0040] The processor section is used to iteratively update the node features through a multilayer perceptron with a multilayer messaging network and residual connections;
[0041] The decoder section is used to map the updated node features through a multilayer perceptron to obtain the electrothermal field distribution results of the power device.
[0042] Figure 3 The diagram illustrates the dimensional configuration of each layer in the thermal analysis model, ensuring the efficiency of the thermal analysis model in feature transfer and spatial relationship modeling.
[0043] like Figure 4 As shown, in this embodiment, the encoder classifies the grid data of the power device into electrical, thermal, and coordinate attributes, and embeds them uniformly as node features. Simultaneously, it broadcasts external biases and positional markers as global node information to the entire graph network. The dataset built during training of the thermal analysis model in this embodiment covers various materials, structures, biases, and scales to ensure the generalization ability of the thermal analysis model.
[0044] The loss function of the thermal analysis model in this embodiment is expressed as:
[0045] ;
[0046] in, The loss function of the thermal analysis model. The data loss term is represented as: , Represents the training dataset The amount of data in it This indicates the predicted electrothermal field distribution. This represents the true value of the electrothermal field. Indicates time, This represents a vector of operating parameters for power devices. This represents the training dataset; The loss term in the physical equation is expressed as: , Indicates the number of node features. This represents the node weight priority coefficient. This indicates the smoothness of the predicted electrothermal field. Represents the regularization coefficient. The PDE loss term is represented as: , Nonlinear operators representing the evolution of physical quantities Indicates carrier mobility; , , All of these are learnable standard deviation parameters.
[0047] Therefore, this implementation method employs physical equation loss terms with adaptive L2 regularization and data loss terms with adaptive heteroscedasticity weighting, thereby ensuring the training convergence and stability of the thermal analysis model.
[0048] Figure 5 The results demonstrate that during short-circuit transients, the thermal analysis model accurately predicts the temperature time series and electric field component distribution under different voltage biases.
[0049] like Figure 6 As shown, compared with the baseline models of physical information neural networks and graph neural networks, the prediction error of the thermal analysis model in this embodiment is consistently below 1.5% in the transient prediction of the electrothermal characteristics of power devices during short circuits, and the cumulative error over time is extremely small, demonstrating its superiority in time-series temperature response prediction. Figure 7 As shown, compared with the baseline models of physical information neural networks and graph neural networks, the average prediction error of the thermal analysis model in this embodiment is consistently below 5% for various static electrical characteristics of power devices.
[0050] like Figure 8 As shown, compared with the conventional finite element method, the inference speed of the thermal analysis model in this embodiment is increased by more than 5,000 times in the electrothermal coupling simulation of devices with different node scales.
[0051] It is easy to see that the thermal analysis model proposed in this invention integrates the physical constraints of the electrothermal equation, significantly improving modeling efficiency and inference speed compared to conventional finite element methods, far exceeding traditional simulation tools and greatly shortening the design cycle of power devices. Under various operating conditions, the thermal analysis model proposed in this invention has better prediction accuracy than other neural network frameworks, and enhances the model's generalization ability through physical constraints. At the same time, relying on the directness of graph networks, the thermal analysis model proposed in this invention does not require additional training data and retraining process, and can directly infer the electrothermal behavior of devices with new structural parameters. It supports multi-scale analysis from single cell to chip level, adapts to different materials, structures and bias scenarios, has strong generalization ability, and effectively reduces the consumption of human and computing resources.
[0052] The second embodiment of the present invention relates to a power device electrothermal analysis device based on a physical information graph neural network, comprising:
[0053] The acquisition module is used to acquire the grid data of the power devices generated by the simulation software;
[0054] The analysis module is used to input the mesh data of the power device structure into the thermal analysis model to obtain the electrothermal field distribution results of the power device; wherein, the thermal analysis model is constructed based on a physical information graph neural network and incorporates electrothermal equation constraints, including:
[0055] The encoder section is used to convert the grid data of the power device into nodes and edges of a graph structure, and embed features to obtain node features;
[0056] The processor section is used to iteratively update the node features through a multilayer perceptron with a multilayer messaging network and residual connections;
[0057] The decoder section is used to map the updated node features through a multilayer perceptron to obtain the electrothermal field distribution results of the power device.
[0058] The encoder part classifies the grid data of the power device into electrical attributes, thermal attributes and coordinate attributes, and embeds them uniformly as node features. At the same time, it broadcasts the external bias and position mark as global node information to the entire physical information graph neural network.
[0059] The thermal analysis model is trained using a dataset that covers various materials, structures, biases, and scales.
[0060] The loss function of the thermal analysis model is expressed as: ,in, The loss function of the thermal analysis model. The data loss term is represented as: , Represents the training dataset The amount of data in it This indicates the predicted electrothermal field distribution. This represents the true value of the electrothermal field. Indicates time, This represents a vector of operating parameters for power devices. This represents the training dataset; The loss term in the physical equation is expressed as: , Indicates the number of node features. This represents the node weight priority coefficient. This indicates the smoothness of the predicted electrothermal field. Represents the regularization coefficient. The PDE loss term is represented as: , Nonlinear operators representing the evolution of physical quantities Indicates carrier mobility; , , All of these are learnable standard deviation parameters.
[0061] The third embodiment of the present invention relates to an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the power device electrothermal analysis method based on physical information graph neural network of the first embodiment.
[0062] The fourth embodiment of the present invention relates to a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the power device electrothermal analysis method based on a physical information graph neural network of the first embodiment.
[0063] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.
[0064] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0065] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction methods implemented in a process. Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0066] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0067] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.
Claims
1. A method for electrothermal analysis of power devices based on physical information graph neural networks, characterized in that, Includes the following steps: Obtain the mesh data of the power device generated by the simulation software; The mesh data of the power device structure is input into the thermal analysis model to obtain the electrothermal field distribution results of the power device; wherein, the thermal analysis model is constructed based on a physical information graph neural network and incorporates electrothermal equation constraints, including: The encoder section is used to convert the grid data of the power device into nodes and edges of a graph structure, and embed features to obtain node features; The processor section is used to iteratively update the node features through a multilayer perceptron with a multilayer messaging network and residual connections; The decoder section is used to map the updated node features through a multilayer perceptron to obtain the electrothermal field distribution results of the power device.
2. The power device electrothermal analysis method based on physical information graph neural network according to claim 1, characterized in that, The encoder part classifies the grid data of the power device into electrical attributes, thermal attributes and coordinate attributes, and embeds them uniformly as node features. At the same time, it broadcasts the external bias and position mark as global node information to the entire physical information graph neural network.
3. The power device electrothermal analysis method based on physical information graph neural network according to claim 1, characterized in that, The thermal analysis model is trained using a dataset that covers various materials, structures, biases, and scales.
4. The power device electrothermal analysis method based on physical information graph neural network according to claim 1, characterized in that, The loss function of the thermal analysis model is expressed as: ,in, The loss function of the thermal analysis model. The data loss term is represented as: , Represents the training dataset The amount of data in it This indicates the predicted electrothermal field distribution. This represents the true value of the electrothermal field. Indicates time, This represents a vector of operating parameters for power devices. This represents the training dataset; The loss term in the physical equation is expressed as: , Indicates the number of node features. This represents the node weight priority coefficient. This indicates the smoothness of the predicted electrothermal field. Represents the regularization coefficient. The PDE loss term is represented as: , Nonlinear operators representing the evolution of physical quantities Indicates carrier mobility; , , All of these are learnable standard deviation parameters.
5. A power device electrothermal analysis device based on a physical information graph neural network, characterized in that, include: The acquisition module is used to acquire the grid data of the power devices generated by the simulation software; The analysis module is used to input the mesh data of the power device structure into the thermal analysis model to obtain the electrothermal field distribution results of the power device; wherein, the thermal analysis model is constructed based on a physical information graph neural network and incorporates electrothermal equation constraints, including: The encoder section is used to convert the grid data of the power device into nodes and edges of a graph structure, and embed features to obtain node features; The processor section is used to iteratively update the node features through a multilayer perceptron with a multilayer messaging network and residual connections; The decoder section is used to map the updated node features through a multilayer perceptron to obtain the electrothermal field distribution results of the power device.
6. The power device electrothermal analysis device based on physical information graph neural network according to claim 5, characterized in that, The encoder part classifies the grid data of the power device into electrical attributes, thermal attributes and coordinate attributes, and embeds them uniformly as node features. At the same time, it broadcasts the external bias and position mark as global node information to the entire physical information graph neural network.
7. The power device electrothermal analysis device based on physical information graph neural network according to claim 5, characterized in that, The thermal analysis model is trained using a dataset that covers various materials, structures, biases, and scales.
8. The power device electrothermal analysis device based on physical information graph neural network according to claim 5, characterized in that, The loss function of the thermal analysis model is expressed as: ,in, The loss function of the thermal analysis model. The data loss term is represented as: , Represents the training dataset The amount of data in it This indicates the predicted electrothermal field distribution. This represents the true value of the electrothermal field. Indicates time, This represents a vector of operating parameters for power devices. This represents the training dataset; The loss term in the physical equation is expressed as: , Indicates the number of node features. This represents the node weight priority coefficient. This indicates the smoothness of the predicted electrothermal field. Represents the regularization coefficient. The PDE loss term is represented as: , Nonlinear operators representing the evolution of physical quantities Indicates carrier mobility; , , All of these are learnable standard deviation parameters.
9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the power device electrothermal analysis method based on physical information graph neural network as described in any one of claims 1-4.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the power device electrothermal analysis method based on physical information graph neural network as described in any one of claims 1-4.