Cross-scale electrothermal performance prediction method based on double-machine learning model data generation

Abstract

The invention discloses a cross-scale electrothermal performance prediction method based on dual-machine learning model data generation. The method comprises the following steps: determining actual input parameters such as a semiconductor device type, a packaging type, semiconductor device-packaging level structure parameters, electrical excitation and environmental factors and an iterative input parameter range; collecting electrical and thermal data sets; respectively selecting and training an electrical machine learning model and a thermal machine learning model; analyzing the electrical data set and the thermal data set, setting an iteration rule, and carrying out electrothermal coupling iteration based on an electrical and thermal dual-machine learning model to generate an electrothermal coupling data set; and an electrothermal coupling machine learning model is selected and trained, and rapid prediction of the cross-scale electrothermal performance from actual input parameters to the semiconductor device is realized. According to the method, the problem of complex cross-platform data conversion and transmission of traditional device-level and packaging-level simulation is effectively solved, and the data generation efficiency and the final model training precision are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of electrothermal performance prediction of semiconductor devices, and in particular to a cross-scale electrothermal performance prediction method based on dual machine learning model data generation. Background Art

[0002] Power semiconductor devices are widely used in automotive, avionics, renewable energy and other fields. To meet the needs of practical applications, such devices usually need to be packaged to achieve mechanical support, electrical interconnection and thermal management functions. However, cross-scale self-heating effects at the device-package level can have a significant impact on the device's electrothermal behavior, leading to device performance degradation, and thus affecting system stability and reliability. As the power density of semiconductor devices continues to increase, cross-scale self-heating effects are becoming more and more significant. Assessing the changes in electrothermal performance affected by device-package structural parameters and working environment factors has become an important basis for achieving device-package coordinated optimization and system performance improvement.

[0003] In the prior art, device-level electrothermal simulation usually ignores the impact of package-level structural parameters and environmental factors on the electrothermal behavior of the device, and package-level thermal simulation also does not involve the feedback effect of the internal structural parameters of the device and its temperature on the electrical performance, making it difficult to fully reflect the device-package-level electrothermal coupling characteristics under actual application conditions. Some studies have proposed to jointly simulate device-level electrical simulation software with package-level thermal simulation software, and evaluate cross-scale electrothermal effects through iterative coupling of power loss and chip temperature. However, due to the lack of a unified data interface between multiple simulation platforms, power loss and temperature data need to be frequently transferred and converted between different simulation software, resulting in cumbersome data interaction, complex simulation process, low computational efficiency, and restricting engineering applications. In order to improve the efficiency of cross-scale electrothermal performance prediction, the invention with publication number CN118364728B discloses a device electrothermal stress extraction method based on machine learning and cross-scale iterative coupling. By constructing a high-quality data set for training a machine learning model, the device-package cross-scale electrothermal performance prediction is achieved. This method effectively improves the prediction accuracy, but it still has the following shortcomings: due to the machine learning model's reliance on a large amount of high-quality training data, the data generation stage still requires repeated device-package-level joint calculation iterations, resulting in a long data collection cycle and low efficiency. In addition, the calculation cannot be completed under extreme working conditions or when some inputs are missing, limiting the wide applicability of this method. Summary of the Invention

[0004] The purpose of this invention is to provide a cross-scale electrothermal performance prediction method based on dual machine learning model data generation. Through the coupled iteration of electrical and thermal dual machine learning models, it replaces the traditional device-level and package-level simulation processes, avoids the frequent transmission and conversion of complex data between simulation platforms, and can quickly generate multiple sets of high-quality electrothermal coupling data in a short time, significantly improving the data collection efficiency and prediction accuracy of the electrothermal coupling machine learning model.

[0005] In order to achieve the above technical objectives, the technical solution adopted by the present invention is:

[0006] A cross-scale electrothermal performance prediction method based on dual machine learning model data generation, the method comprising the following steps:

[0007] S1, determining the input parameters related to the semiconductor device-level structure, package-level structure, electrical excitation and environmental factors that are relevant to the cross-scale electrothermal performance of the semiconductor device, and the data range of each input parameter;

[0008] S2, simulating the electrical performance parameter values ​​of the semiconductor device under different device-level structural parameters, electrical excitation and chip temperature combinations to generate an electrical data set; simulating the chip temperature values ​​of the semiconductor device under different package-level structural parameters, environmental factors and power loss combinations to generate a thermal data set;

[0009] S3, respectively constructing and training an electrical machine learning model and a thermal machine learning model; the input parameters of the electrical machine learning model include the device type, device-level structural parameters, electrical excitation, and chip temperature of the semiconductor device, and outputting the electrical performance parameters of the semiconductor device. Then, based on the electrical performance parameter values ​​of the semiconductor device and the set electrical excitation, the power loss value is obtained through the power loss calculation formula; the input parameters of the thermal machine learning model include the package type, package-level structural parameters, environmental factors, and power loss value of the semiconductor device, and the output parameter is the chip temperature of the semiconductor device; wherein, the electrical performance parameters of the semiconductor device include initial current or initial voltage and steady-state current or steady-state voltage;

[0010] S4, analyzing the electrical data set and the thermal data set obtained in step S2 and setting iteration rules, performing electrothermal coupling iteration based on the electrical and thermal dual machine learning model to generate an electrothermal coupling data set;

[0011] S5, selecting and training an electrothermal coupling machine learning model based on the electrothermal coupling dataset; the input parameters of the electrothermal coupling machine learning model are the device-level structure, package-level structure, electrical excitation, and environmental factor-related input parameters of the semiconductor device; the output parameters are the cross-scale electrothermal performance of the semiconductor device, including temperature state, steady-state chip temperature, initial state current or initial state voltage, and steady-state current or steady-state voltage;

[0012] S6, imports the input parameters of the actual operation process into the trained electrothermal coupling machine learning model to quickly predict the cross-scale electrothermal performance of semiconductor devices.

[0013] Furthermore, in step S1, the device-level structure-related parameters are all structural parameters covered by nanometer / micrometer-level devices, including device type, drift region doping concentration, P-Well doping concentration, drift region thickness, channel length and chip area; the package-level structure-related parameters are millimeter / centimeter-level packaging parameters, including package type, thermal conductivity of the adhesive layer, thermal conductivity of the plastic packaging material and chip area, as well as structural parameters of external devices that affect heat dissipation, and external devices include but are not limited to heat sinks and printed circuit boards; electrical excitation is the electrical condition applied to the semiconductor device, including the voltage bias or current bias borne by the semiconductor device; environmental factors include but are not limited to ambient temperature, ambient radiation size and ambient convection state.

[0014] Furthermore, in step S2, the electrical performance parameters include but are not limited to the current value of the semiconductor device, the voltage value of the semiconductor device and the on-resistance value of the semiconductor device.

[0015] Furthermore, in step S3, the electrical machine learning model and the thermal machine learning model adopt regression models, including but not limited to one or a corresponding combination of deep neural network, Gaussian process regression, support vector machine and random forest.

[0016] Step S4 further comprises:

[0017] S41, determining a temperature limit threshold of the semiconductor device of the corresponding type according to the type of the semiconductor device, and using it as the temperature threshold for determining an over-temperature state in the iterative process;

[0018] S42, setting an iteration threshold for determining the convergence condition of the iterative process. When the ratio of the difference between two adjacent predicted temperature values ​​to the previous round of predicted temperature value is less than the iteration threshold, the iterative process is determined to have reached convergence and the iterative calculation is stopped.

[0019] S43, generating multiple groups of actual input parameter combinations based on the data range of each input parameter determined in step S1;

[0020] S44, based on the generated actual input parameter combination, calling the trained electrical machine learning model to predict the corresponding electrical performance parameter value of the semiconductor device based on the device type, device-level structural parameters, electrical stimulus, and chip temperature, where the initial value of the chip temperature is the ambient temperature;

[0021] S45, obtaining a power loss value by using a power loss calculation formula based on the obtained electrical performance parameter value of the semiconductor device and the set electrical excitation;

[0022] S46, based on the package type, package-level structural parameters, environmental factor parameters, and the calculated power loss value, calling the trained thermal machine learning model to predict the chip temperature;

[0023] S47, based on the obtained chip temperature, implementing adaptive cross-scale electrothermal coupling regulation according to a set temperature criterion, specifically including:

[0024] S471, determine whether the obtained chip temperature is greater than a preset temperature threshold. If so, output an over-temperature status flag and proceed to step S48; if not, execute step S472;

[0025] S472: Determine whether the difference between the obtained chip temperature and the previous chip temperature is less than a set iteration threshold. If so, output a normal temperature status tag, record the steady-state chip temperature, initial current or initial voltage, and steady-state current or steady voltage, and proceed to step S48. If so, substitute the chip temperature value as input again into step S44 and continue iteration.

[0026] S48, repeating steps S44 to S47 until all actual input parameter combinations generated in step S43 have completed iteration, thus completing the generation of the electrothermal coupling data set.

[0027] Step S5 further comprises:

[0028] S51, selecting a regression model as an electrothermal coupling machine learning model, where the input parameters of the electrothermal coupling machine learning model are device type, package type, device-level structural parameters, package-level structural parameters, electrical excitation, and environmental factors of the semiconductor device, and the output parameters are temperature state, steady-state chip temperature, initial-state current or initial-state voltage, and steady-state current or steady-state voltage;

[0029] The electrothermal coupling machine learning model adopts a multi-branch structure. According to the input characteristics, the device type, device-level structural parameters and electrical excitation are used as the first input branch and connected to the first fully connected layer group. The package type, package-level structural parameters and environmental factors are used as the second input branch and connected to the second fully connected layer group. The outputs of the two branches are then connected to the unified fusion layer, and the temperature state, steady-state chip temperature, initial current or initial voltage and steady-state current or steady voltage are finally output. Residual connection modules and attention mechanism modules are embedded in the first fully connected layer group, the second fully connected layer group and the unified fusion layer to enhance the model's ability to extract dynamic weights of complex coupling relationships and key features. The number, embedding position and specific form of the residual connection modules and attention mechanism modules are configured according to data complexity, model depth and task type, and are applied to each branch or unified fusion layer individually or in combination.

[0030] S52, dividing the electrothermal coupling data set obtained in step S4 into a training set, a validation set, and a test set according to a set ratio;

[0031] S53, based on the type of the electrothermal coupling data set obtained in step S4, all data in the data set are standardized, normalized, or logarithmized; wherein the two temperature states of normal temperature and overtemperature are converted into two different values ​​respectively;

[0032] S54, using the processed training set data to train the electrothermal coupling machine learning model to obtain the machine learning model weight value; using the processed validation set data or test set data to verify or test the trained electrothermal coupling machine learning model.

[0033] Step S6 further comprises:

[0034] S61, standardizing, normalizing, or logarithmizing the input actual operating data related to the device type, package type, device-level structural parameters, package-level structural parameters, electrical stimulus, and environmental factors of the semiconductor device;

[0035] S62, inputting the actual operating data after data processing into the trained electrothermal coupling machine learning model to predict the chip temperature state, initial current or initial voltage, steady-state current or steady-state voltage and steady-state chip temperature.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] First, the cross-scale electrothermal performance prediction method based on dual machine learning model data generation of the present invention considers generating electrical data sets and thermal data sets based on the principles of electrical and thermal performance, and constructs electrical and thermal machine learning models respectively. Through model iteration, the data generation efficiency can be improved and the data collection time can be significantly shortened. At the same time, on the basis of proposing a dual machine learning model coupled iterative data generation method, the electrical and thermal machine learning model structure is further optimized, the model's ability to identify and screen coupled feature data is improved, the interference of abnormal data is reduced, and the generated data is more adapted to the coupled iterative process, which effectively solves the technical problem that the two types of data are independently constructed, easily contain some data that is not suitable for coupled iteration, and are easily prone to non-convergence of iteration and large prediction errors when directly used for coupled iteration.

[0038] Second, the cross-scale electrothermal performance prediction method based on dual machine learning model data generation of the present invention sets reasonable over-temperature thresholds and iteration stop criteria based on the characteristics of semiconductor devices, ensures the convergence and prediction stability of the iterative process, and effectively solves the problems of iterative instability and low efficiency.

[0039] Third, the present invention's cross-scale electrothermal performance prediction method, based on dual machine learning model data generation, can effectively simulate the cross-scale electrothermal coupling effects between nano / micron-scale device structures and millimeter / centimeter-scale packaging and heat dissipation structures through iterative coupling calculations of the electrical machine learning model and the thermal machine learning model. This avoids the problems associated with complex data transfer and format conversion between simulation platforms in the prior art, significantly improving the efficiency and quality of electrothermal coupling data generation. In particular, the innovative introduction of residual connections and attention mechanisms into the electrical and thermal machine learning models enhances the model's ability to extract complex coupling relationship features, effectively alleviating the overfitting or key information loss issues that can easily occur in single-layer networks, and further improving the stability and accuracy of the prediction results.

[0040] Fourth, the present invention's cross-scale electrothermal performance prediction method based on dual machine learning model data generation, and the constructed electrothermal coupling machine learning model, can achieve rapid and accurate prediction of the cross-scale electrothermal performance of semiconductor devices based on device type, package type, device-level structural parameters, package-level structural parameters, electrical excitation, and environmental factors. Furthermore, an innovative multi-branch structure is set up, and the micron-level device structure and electrical excitation parameters, package structure, and environmental factor parameters are respectively input into independent branch networks, and then feature fusion is completed in a unified fusion layer, thereby improving the model's adaptability to the differential expression of multi-scale features. And according to the actual data distribution characteristics, the residual connection and attention mechanism are flexibly configured in each branch and fusion layer to further enhance the model's ability to fit complex electrothermal coupling laws, effectively improving the model's prediction accuracy and generalization ability.

[0041] Fifth, the cross-scale electrothermal performance prediction method based on dual machine learning model data generation of the present invention fully considers the nano / micron-level device structural parameters and millimeter / centimeter-level packaging and other heat dissipation structural parameters, while also achieving a full-process rate improvement from the data collection process to the evaluation process, providing an efficient tool for device-package-level cross-scale electrothermal performance evaluation, which is beneficial to device-package collaborative design optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 Schematic diagram of the flow chart of the cross-scale electrothermal performance prediction method based on dual machine learning model data generation of the present invention.

[0043] Figure 2 Schematic diagram of the silicon-based VDMOSFET device-package structure according to an embodiment of the present invention; wherein, (a) is a device-level structure diagram of the embodiment, (b) is a TO-247 package-level structure diagram, (c) is a TO-263 package-level structure diagram, (d) is a TO-220 package-level structure diagram, and (e) is a cross-sectional view of the package structure of the embodiment.

[0044] Figure 3This is a structural diagram of the dual machine learning model used in data generation by the dual machine learning model in an embodiment of the present invention; wherein, (a) is a schematic diagram of the electrical machine learning model structure, and (b) is a schematic diagram of the thermal machine learning model structure.

[0045] Figure 4 This is a schematic diagram showing the comparison results of the average time taken to obtain different data sets through dual machine learning models and simulation in an embodiment of the present invention.

[0046] Figure 5 Schematic diagram of the structure of the electrothermal coupling machine learning model in an embodiment of the present invention.

[0047] Figure 6 This is a schematic diagram of the average time comparison results of electrothermal performance under different data sets evaluated through an electrothermal coupling machine learning model and simulation in an embodiment of the present invention. DETAILED DESCRIPTION

[0048] The embodiments of the present invention are described in further detail below with reference to the accompanying drawings.

[0049] like Figure 1 As shown, a cross-scale electrothermal performance prediction method based on dual machine learning model data generation includes the following steps.

[0050] Step 1: Determine the actual input parameters and iterative input parameter ranges, including semiconductor device type, package type, semiconductor device-package level structural parameters, electrical excitation, and environmental factors. Specifically, it includes:

[0051] Based on the semiconductor device type and package type, theoretical knowledge or numerical simulation is used to analyze the impact of actual input parameters and iterative input parameters, such as semiconductor device-package-level structural parameters, electrical excitations, and environmental factors, on the cross-scale electrothermal performance of semiconductor devices. The parameters with the most significant impact on electrothermal performance and their data ranges are identified. The actual input parameters include semiconductor device type, package type, semiconductor device-package-level structural parameters, electrical excitations, and environmental factors. Device-level structural parameters include all structural parameters covering nanometer / micrometer-scale devices. Package-level structural parameters include all structural parameters of the millimeter / centimeter-scale package and its external components that affect heat dissipation (including but not limited to heat sinks and printed circuit boards). Electrical excitations are electrical conditions applied to the semiconductor device, including the voltage bias or current it experiences. Environmental factors include but are not limited to ambient temperature (specifically, ambient temperature also refers to the initial chip temperature), ambient radiation intensity, and ambient convection conditions. Iterative input parameters are the iterative input parameters of the electrical and thermal machine learning models during the electrothermal coupling iteration process, specifically including chip temperature for the electrical machine learning model and power loss for the thermal machine learning model. For example, for silicon-based VDMOSFETs, which are micron-scale devices, structural parameters include but are not limited to gate oxide thickness, channel length, channel doping concentration, drift region thickness, and drift region doping concentration. Package-level structural parameters include all structural parameters of the millimeter / centimeter-level package and its external components that affect heat dissipation (including but not limited to heat sinks and printed circuit boards). For the TO-247 package type, which is a millimeter-level package, structural parameters include but are not limited to the thermal conductivity of the plastic compound, the thermal conductivity of the adhesive layer, the lead frame material, and the pin material; for pure aluminum heat sinks, structural parameters include but are not limited to the length, width, and height of the heat sink, the number of fins, and the fin spacing; for printed circuit boards, structural parameters include but are not limited to the length, width, and height of the printed board and the proportion of copper material.

[0052] This embodiment takes silicon-based VDMOSFET as an example, which is a micron-level device. In this embodiment, according to the type of semiconductor device, the package-level structure is selected as three types of millimeter-level TO-220, TO263 and TO-247, and does not include heat sinks and printed circuit boards. The semiconductor device-package-level structure diagram in this embodiment is as follows: Figure 2 As shown. The silicon-based VDMOSFET device structure is as follows Figure 2 As shown in (a), it includes the drain electrode Drain, substrate N+Substrate, drift region N-Drift, P-type doped drift region P-Well, heavily doped P-type region P+, heavily doped N-type drift region N+, gate electrode Gate and source electrode Source. The semiconductor device package types are TO-220, TO263 and TO-247. Figure 2 As shown in (b), (c) and (d), its cross-sectional view is as follows Figure 2The three packaging structures include pins (electrodes of three pins, from left to right are gate (G), drain (D) and source (S)), plastic package, chip ( Figure 2 The device structure shown in (a) is composed of multiple devices connected in parallel), a bonding layer, and a lead frame. It is worth noting that the package leads have little impact on temperature and are therefore ignored in this example. Through theoretical knowledge and numerical simulation analysis, the actual input parameters, including silicon-based VDMOSFET device-package-level structural parameters, electrical excitation, and environmental factors, as well as the parameters with the greatest impact on electrical and thermal performance among the iterative input parameters, and their data ranges, were determined, as shown in Table 1.

[0053] Table 1: Related parameters and data ranges of this embodiment

[0054] Step 2: Collect electrical and thermal data sets.

[0055] In this embodiment, the electrical dataset is a dataset that obtains the current value of the corresponding semiconductor device based on the device type, device-level structural parameters, electrical excitation, and chip temperature; the thermal dataset is a dataset that obtains the chip temperature value of the corresponding semiconductor device based on the package type, package-level structural parameters, ambient temperature, and power loss. The process of obtaining the electrical dataset is as follows: based on the device type, under different combinations of device-level structural parameters, electrical excitation, and chip temperature, the electrical performance parameter values ​​of the semiconductor device are obtained through device-level electrical operations such as circuit simulation software, TCAD tools, and mathematical or physical calculation formulas to form the electrical dataset. The process of obtaining the thermal dataset is as follows: based on the package type, under different combinations of package-level structural parameters, environmental factors, and power loss, the chip temperature values ​​of the semiconductor device are obtained through package-level thermal operations such as thermal simulation software and thermal network models to form the thermal dataset.

[0056] In this embodiment, within the parameter ranges given in step 1, the semiconductor device current values ​​were first obtained using the TCAD tool Silvaco software, based on various combinations of device-level structural parameters, electrical excitation, and chip temperature. This data set contained 1,401,148 data points. Furthermore, within the parameter ranges given in step 1, the semiconductor device chip temperature values ​​were obtained using the thermal simulation tool ANSYS Icepak software, based on various combinations of package-level structural parameters, ambient temperature, and power loss. This data set contained 11,760 data points.

[0057] Step 3: Select and train an electrical machine learning model and a thermal machine learning model based on the electrical data set and thermal data set obtained in step 2. In this embodiment, the steps include:

[0058] Step 3.1. Select a machine learning model: Based on the electrical and thermal data sets obtained in step 2, the selected electrical machine learning model and thermal machine learning model both use regression models, including but not limited to one or a corresponding combination of deep neural networks, Gaussian process regression, support vector machines, and random forests.

[0059] Step 3.2, data set classification: The electrical data set and thermal data set obtained in step 2 are divided into training set, validation set and test set according to the ratio of 7:1.5:1.5 respectively.

[0060] Step 3.3, data set processing: According to the type of electrical and thermal data sets obtained in step 2, all data in the electrical and thermal data sets are standardized, normalized, and logarithmicized.

[0061] Step 3.4, machine learning model construction: According to the machine learning model selected in step 3.1, set the electrical machine learning model and thermal machine learning model structure and model input and output parameters respectively. Among them, the input parameters of the electrical machine learning model include device type, device-level structural parameters, electrical excitation and chip temperature, and the output parameter is the current value of the semiconductor device; the input parameters of the thermal machine learning model include package type, package-level structural parameters, ambient temperature and power loss, and the output parameter is the chip temperature value of the semiconductor device. According to the complexity of the problem and data, two fully connected layers, one residual block layer and one attention mechanism layer are constructed in the electrical and thermal machine learning models respectively to dynamically adjust the feature importance weights and improve the model convergence speed and performance. The constructed model structure is as follows Figure 3As shown in the figure, the electrical machine learning model architecture includes an input layer, an output layer, a fully connected layer, a residual connection module, and an attention mechanism module. The input layer receives input data such as device type, device-level structural parameters, electrical stimulus, and chip temperature. The output layer predicts the electrical performance parameters of semiconductor devices. Based on actual needs, a multi-layer architecture consisting of fully connected layers, residual connection modules, and attention mechanism modules is configured. The fully connected layer extracts the underlying nonlinear mapping features from the input parameters. The residual connection module captures the complex coupling relationships between device type, device-level structural parameters, electrical stimulus, and chip temperature, preventing gradient vanishing and information loss in deep networks and improving the model's ability to fit high-dimensional complex features. The attention mechanism module dynamically assigns weights based on the importance of different input features to the electrical performance prediction results, focusing on input features that significantly impact the semiconductor device's electrical performance, thereby improving the model's feature representation and prediction accuracy. The thermal machine learning model architecture includes an input layer, an output layer, a fully connected layer, a residual connection module, and an attention mechanism module. The input layer receives input data such as package type, package-level structural parameters, environmental factors, and power loss. The output layer predicts the chip temperature of the semiconductor device. Based on actual needs, a multi-layer structure consisting of fully connected layers, residual connection modules, and attention mechanism modules is set up. The fully connected layers extract the basic nonlinear relationship characteristics in the input parameters; the residual connection modules are used to capture the complex coupling characteristics between package type, package-level structural parameters, environmental factors, and power loss, improving the model's ability to fit the multi-physics field thermal coupling effects; the attention mechanism module dynamically adjusts the input feature weights based on the sensitivity of different input parameters to chip temperature changes, highlighting parameters with greater temperature influence, and improving the model's prediction accuracy. In the electrical machine learning model and thermal machine learning model, the number, embedding position, and specific form of the fully connected layers, residual connection modules, and attention mechanism modules can be flexibly configured based on data characteristics, model depth, and prediction requirements, and can be applied individually or in combination to the electrical machine learning model or thermal machine learning model.

[0062] Step 3.5, machine learning model training: The machine learning model constructed in step 3.4 is trained separately using the training set data processed in step 3.3 to obtain the machine learning model weight value.

[0063] Step 3.6, machine learning model verification and testing: Verify or test the trained electrical machine learning model and thermal machine learning model using the validation set data or test set data processed in step 3.3.

[0064] Step 4: Analyze the electrical and thermal data sets obtained in step 2 and set selection rules. Perform electrothermal coupling iteration based on the electrical and thermal dual machine learning model to generate an electrothermal coupling data set.

[0065] In this embodiment, based on the application characteristics of the semiconductor device itself, a temperature threshold (expressed as T th ), so that the temperature threshold can reasonably reflect the thermal reliability requirements of this type of device under actual application conditions, ensuring that the prediction results have good engineering applicability and reliability; set an iteration threshold (expressed as ε) for judging the convergence condition of the iterative process. When the ratio of the difference between two adjacent predicted temperature values ​​to the previous round of predicted temperature value is less than the iteration threshold, it is determined that the iterative process has reached convergence and the iterative calculation is stopped, thereby ensuring that the prediction process has good convergence and rationality.

[0066] The research object of this embodiment is silicon-based VDMOSFET, whose maximum withstand temperature is 150°C, so the temperature threshold T in the iteration process is set to th is set to 150℃; the iteration threshold ε in the iteration process is set to 1×10 -3 , ensuring that the prediction process has good convergence.

[0067] In this embodiment, the electrothermal coupling dataset is a dataset of corresponding cross-scale electrothermal performance, namely, temperature state, initial current, steady-state current, and steady-state chip temperature, obtained from actual input parameters, namely, semiconductor device type, package type, semiconductor device-package level structural parameters, electrical excitation, and ambient temperature. Specifically, the steps include:

[0068] Step 4.1: Generate multiple sets of actual input parameter combinations based on the actual input parameter range determined in step 1. In this embodiment, within the parameter range given in Table 1, a set of actual input parameters is selected as an example and the following steps are performed, as shown in Table 2.

[0069] Table 2 The first set of actual input parameters

[0070] Step 4.2: Based on the set of input parameters given in step 4.1, according to the device-level structural parameters, electrical excitation and chip temperature (the ambient temperature in this case) ), using the electrical machine learning model trained in step 3, the current value of the semiconductor device (expressed as I DS ), at this time, I DS =0.95A.

[0071] Step 4.3: Based on the semiconductor device current value obtained in step 4.2, combined with the set electrical excitation conditions (applying a constant gate and drain voltage bias V GS =10V, V DS =0.8V), and the power loss value (expressed as P) is obtained by the power loss calculation formula. loss ), P loss =IDS ×V DS =0.95A×0.8V=0.76W.

[0072] Step 4.4: Based on the package-level structural parameters and ambient temperature parameters determined in step 4.1 and the power loss value obtained in step 4.3, the thermal machine learning model trained in step 3 is called to predict the chip temperature (expressed as ),at this time .

[0073] Step 4.5: Based on the chip temperature obtained in step 4.4, adaptive cross-scale electrothermal coupling adjustment is implemented according to the set temperature criterion, specifically including:

[0074] Step 451: Since only one device type is considered in this embodiment, the corresponding T th The value will only be 150℃;

[0075] The chip temperature value obtained in step 452 and step 4.4 is 89.11° C., which is lower than the temperature threshold, so the process proceeds to step 453 .

[0076] Step 453: In this embodiment, the deviation threshold ε is 1×10 -3 Calculate the difference between the chip temperature obtained in step 4.4 and the chip temperature in the previous round , which is greater than 1×10 -3 Therefore, the chip temperature value of 89.11℃ is input into step 4.2 as the input parameter of the electrical machine learning model. -3 In the output parameter temperature state, "normal temperature" is output, in the steady-state chip temperature, "77.07℃" is output, in the initial current, "0.95A" is output, and in the steady-state current, "0.76A" is output, thus completing the collection of a set of data.

[0077] Step 4.6: Repeat until all actual input parameter combinations in step 4.1 are fully iterated to complete the generation of the electrothermal coupling data set.

[0078] In this embodiment, the average time of obtaining different data sets by dual machine learning models and simulation is compared, such as Figure 4 shown.

[0079] Step 5: Select and train an electrothermal coupling machine learning model based on the data set obtained in step 4. In this embodiment, the steps include:

[0080] Step 5.1. Select a machine learning model: Based on the electrothermal coupling data set obtained in step 4, the electrothermal coupling machine learning model selected is a deep neural network model.

[0081] Step 5.2, data set classification: The electrothermal coupling data set obtained in step 4 is divided into a training set, a validation set, and a test set in a ratio of 7:1.5:1.5.

[0082] Step 5.3, Dataset Processing: Based on the electrothermal coupling dataset type obtained in Step 4, all data in the dataset is normalized and log-logged. Specifically, since the temperature state parameters recorded during the electrothermal coupling dataset collection in Step 4 are "normal temperature" and "overtemperature," the model training process requires converting these two state records from text to numerical values. In this example, the "normal temperature" label is assigned a value of "1," and the "overtemperature" label is assigned a value of "0."

[0083] Step 5.4, machine learning model construction: According to the machine learning model selected in step 5.1, set the electrothermal coupling machine learning model structure and model input and output parameters. Among them, the input parameters of the electrothermal coupling machine learning model are device type, package type, semiconductor device-package level structure parameters, electrical excitation and ambient temperature, and the output parameters are temperature state, steady-state chip temperature, initial current and steady-state current. According to the complexity of the problem and data, we constructed three fully connected layers in the electrothermal coupling machine learning model, with device type / micron-level device structure parameters / electrical excitation as the first input branch, connected to the first fully connected layer group, and package type / package structure parameters and / ambient temperature as the second input branch, connected to the second fully connected layer group, and then the outputs of the two branches are connected to a unified fusion layer, and finally the temperature state, steady-state chip temperature, initial current and steady-state current are output. This model structure can better distinguish input parameters of different scales, avoid information coupling interference caused by large data differences, and bring about model inaccuracy problems. The constructed model structure is as follows Figure 5 shown.

[0084] Step 5.5: Machine learning model training: Train the machine learning model constructed in step 5.4 using the training set data processed in step 5.3.

[0085] Step 5.6, machine learning model verification and testing: Verify or test the electrothermal coupling machine learning model trained in step 5.5 using the validation set data or test set data processed in step 5.3.

[0086] Step 6: Using the electrothermal coupling machine learning model described in step 5, a rapid prediction of the cross-scale electrothermal performance of semiconductor devices based on actual input parameters is achieved. In this embodiment, the following steps are specifically included:

[0087] Step 6.1. Input parameter data processing: In this embodiment, a set of actual input parameters within the range of Table 1 is randomly selected, as shown in Table 3. The input semiconductor device-package level structural parameters, electrical excitation, and ambient temperature are normalized and logarithmized.

[0088] Table 3 The second set of actual input parameters

[0089] Step 6.2, device electrothermal performance prediction: Input the parameters after data processing in step 6.1 into the electrothermal coupling machine learning model trained in step 5 to achieve rapid prediction of the corresponding chip temperature state, initial current or initial voltage, steady-state current or steady-state voltage and steady-state chip temperature, that is, the temperature state is "normal temperature", the steady-state chip temperature is "56.94℃", the initial current is "0.25A", and the steady-state current is "0.23A".

[0090] In this embodiment, the average time of obtaining different data sets by the electrothermal coupling machine learning model and simulation is compared, such as Figure 6 As shown. The cross-scale electrothermal performance prediction method based on dual machine learning model data generation proposed in this invention fully considers the cross-scale electrothermal coupling effect at the device-package level, and effectively solves the complex problem of cross-platform data conversion and transmission in traditional device-level and package-level simulations, significantly improving data generation efficiency and final model training accuracy. The final constructed electrothermal coupling machine learning model can realize the prediction of cross-scale electrothermal performance from semiconductor device-package level structural parameters, electrical excitation and environmental factors, with the advantages of high prediction accuracy, fast calculation speed, reduced experimental manpower and cost, etc., which is conducive to the joint optimization design of device-package level.

[0091] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present application.

[0092] Obviously, those skilled in the art may make various changes and modifications to this application without departing from the spirit and scope of this application. Thus, if these modifications and variations of this application fall within the scope of the claims of this application and their equivalents, this application is intended to include these modifications and variations.

Claims

1. A cross-scale electrothermal performance prediction method based on dual machine learning model data generation, characterized in that: The method comprises the following steps: S1, determining the input parameters related to the semiconductor device-level structure, package-level structure, electrical excitation and environmental factors that are relevant to the cross-scale electrothermal performance of the semiconductor device, and the data range of each input parameter; S2, simulating the electrical performance parameter values ​​of the semiconductor device under different device-level structural parameters, electrical excitation and chip temperature combinations to generate an electrical data set; simulating the chip temperature values ​​of the semiconductor device under different package-level structural parameters, environmental factors and power loss combinations to generate a thermal data set; S3, respectively constructing and training an electrical machine learning model and a thermal machine learning model; the input parameters of the electrical machine learning model include the device type, device-level structural parameters, electrical excitation, and chip temperature of the semiconductor device, and outputting the electrical performance parameters of the semiconductor device. Then, based on the electrical performance parameter values ​​of the semiconductor device and the set electrical excitation, the power loss value is obtained through the power loss calculation formula; the input parameters of the thermal machine learning model include the package type, package-level structural parameters, environmental factors, and power loss value of the semiconductor device, and the output parameter is the chip temperature of the semiconductor device; wherein, the electrical performance parameters of the semiconductor device include initial current or initial voltage and steady-state current or steady-state voltage; S4, analyzing the electrical data set and the thermal data set obtained in step S2 and setting iteration rules, performing electrothermal coupling iteration based on the electrical and thermal dual machine learning model to generate an electrothermal coupling data set; S5, selecting and training an electrothermal coupling machine learning model based on the electrothermal coupling dataset; the input parameters of the electrothermal coupling machine learning model are the device-level structure, package-level structure, electrical excitation, and environmental factor-related input parameters of the semiconductor device; the output parameters are the cross-scale electrothermal performance of the semiconductor device, including temperature state, steady-state chip temperature, initial state current or initial state voltage, and steady-state current or steady-state voltage; S6, imports the input parameters of the actual operation process into the trained electrothermal coupling machine learning model to quickly predict the cross-scale electrothermal performance of semiconductor devices.

2. The cross-scale electrothermal performance prediction method based on dual machine learning model data generation according to claim 1 is characterized in that: In step S1, the device-level structure-related parameters are all structural parameters covered by nanometer / micrometer-level devices, including device type, drift region doping concentration, P-Well doping concentration, drift region thickness, channel length and chip area; the package-level structure-related parameters are millimeter / centimeter-level package parameters, including package type, adhesive layer thermal conductivity, plastic packaging material thermal conductivity and chip area, as well as external device structure parameters that affect heat dissipation, and external devices include but are not limited to heat sinks and printed circuit boards; electrical excitation is the electrical condition applied to the semiconductor device, including the voltage bias or current bias borne by the semiconductor device; environmental factors include but are not limited to ambient temperature, ambient radiation size and ambient convection state.

3. The cross-scale electrothermal performance prediction method based on dual machine learning model data generation according to claim 1 is characterized in that: In step S2 , the electrical performance parameters include but are not limited to the current value of the semiconductor device, the voltage value of the semiconductor device, and the on-resistance value of the semiconductor device.

4. The cross-scale electrothermal performance prediction method based on dual machine learning model data generation according to claim 1 is characterized in that: In step S3, the electrical machine learning model and the thermal machine learning model adopt regression models, including but not limited to one or a corresponding combination of deep neural network, Gaussian process regression, support vector machine and random forest.

5. The cross-scale electrothermal performance prediction method based on dual machine learning model data generation according to claim 1 is characterized in that: Step S4 further comprises: S41, determining a temperature limit threshold of the semiconductor device of the corresponding type according to the type of the semiconductor device, and using it as the temperature threshold for determining an over-temperature state in the iterative process; S42, setting an iteration threshold for determining the convergence condition of the iterative process. When the ratio of the difference between two adjacent predicted temperature values ​​to the previous round of predicted temperature value is less than the iteration threshold, the iterative process is determined to have reached convergence and the iterative calculation is stopped. S43, generating multiple groups of actual input parameter combinations based on the data range of each input parameter determined in step S1; S44, based on the generated actual input parameter combination, calling the trained electrical machine learning model to predict the corresponding electrical performance parameter value of the semiconductor device based on the device type, device-level structural parameters, electrical stimulus, and chip temperature, where the initial value of the chip temperature is the ambient temperature; S45, obtaining a power loss value by using a power loss calculation formula based on the obtained electrical performance parameter value of the semiconductor device and the set electrical excitation; S46, based on the package type, package-level structural parameters, environmental factor parameters, and the calculated power loss value, calling the trained thermal machine learning model to predict the chip temperature; S47, based on the obtained chip temperature, implementing adaptive cross-scale electrothermal coupling regulation according to a set temperature criterion, specifically including: S471, determine whether the obtained chip temperature is greater than a preset temperature threshold. If so, output an over-temperature status flag and proceed to step S48; if not, execute step S472; S472: Determine whether the difference between the obtained chip temperature and the previous chip temperature is less than a set iteration threshold. If so, output a normal temperature status tag, record the steady-state chip temperature, initial current or initial voltage, and steady-state current or steady voltage, and proceed to step S48. If so, substitute the chip temperature value as input again into step S44 and continue iteration. S48, repeating steps S44 to S47 until all actual input parameter combinations generated in step S43 have completed iteration, thus completing the generation of the electrothermal coupling data set.

6. The cross-scale electrothermal performance prediction method based on dual machine learning model data generation according to claim 1, characterized in that: Step S5 further comprises: S51, selecting a regression model as an electrothermal coupling machine learning model, where the input parameters of the electrothermal coupling machine learning model are device type, package type, device-level structural parameters, package-level structural parameters, electrical excitation, and environmental factors of the semiconductor device, and the output parameters are temperature state, steady-state chip temperature, initial-state current or initial-state voltage, and steady-state current or steady-state voltage; The electrothermal coupling machine learning model adopts a multi-branch structure. According to the input characteristics, the device type, device-level structural parameters and electrical excitation are used as the first input branch and connected to the first fully connected layer group. The package type, package-level structural parameters and environmental factors are used as the second input branch and connected to the second fully connected layer group. The outputs of the two branches are then connected to the unified fusion layer, and the temperature state, steady-state chip temperature, initial current or initial voltage and steady-state current or steady voltage are finally output. Residual connection modules and attention mechanism modules are embedded in the first fully connected layer group, the second fully connected layer group and the unified fusion layer to enhance the model's ability to extract dynamic weights of complex coupling relationships and key features. The number, embedding position and specific form of the residual connection modules and attention mechanism modules are configured according to data complexity, model depth and task type, and are applied to each branch or unified fusion layer individually or in combination. S52, dividing the electrothermal coupling data set obtained in step S4 into a training set, a validation set, and a test set according to a set ratio; S53, based on the type of the electrothermal coupling data set obtained in step S4, all data in the data set are standardized, normalized, or logarithmized; wherein the two temperature states of normal temperature and overtemperature are converted into two different values ​​respectively; S54, using the processed training set data to train the electrothermal coupling machine learning model to obtain the machine learning model weight value; using the processed validation set data or test set data to verify or test the trained electrothermal coupling machine learning model.

7. The cross-scale electrothermal performance prediction method based on dual machine learning model data generation according to claim 1, characterized in that: Step S6 further comprises: S61, standardizing, normalizing, or logarithmizing the input actual operating data related to the device type, package type, device-level structural parameters, package-level structural parameters, electrical stimulus, and environmental factors of the semiconductor device; S62, inputting the actual operating data after data processing into the trained electrothermal coupling machine learning model to predict the chip temperature state, initial current or initial voltage, steady-state current or steady-state voltage and steady-state chip temperature.

Images