A method and system for predicting the aging lifetime of SiC MOSFETs

Abstract

The application discloses an aging life prediction method and system for SiC MOSFET, relates to the technical field of silicon carbide field effect tube aging life prediction, and comprises the following steps: acquiring an initial life time data set corresponding to each experimental point combination; constructing a response surface model according to the initial life time data set by using a response surface analysis method; performing residual analysis on the response surface model to obtain qualified life time data that meets model adaptability and hypothesis condition compliance; constructing an aging life prediction initial model based on the coupling effect between aging influence factors and combining the response surface model; training the aging life prediction initial model by using the qualified life time data to obtain an aging life prediction model; and outputting an aging life prediction result of the SiC MOSFET according to real-time aging influence factor data and the aging life prediction model. The application has the beneficial effects that the experimental data demand is reduced and the aging life prediction accuracy is improved simultaneously.

Description

Technical Field

[0001] This application relates to the field of silicon carbide field-effect transistor aging lifetime prediction technology, and in particular to an aging lifetime prediction method and system for SiC MOSFETs. Background Technology

[0002] SiC MOSFETs (Silicon Carbide Field-Effect Transistors), as a new generation of wide-bandgap power devices, exhibit superior electrical performance under harsh conditions such as high temperature, high frequency, and high voltage due to their high breakdown field strength, high thermal conductivity, and wide bandgap characteristics, gradually becoming core components of high-efficiency power electronic equipment. However, compared with the device's inherent performance, the reliability of the encapsulation and insulating materials has become a key factor restricting its long-term stable operation. Existing commercial SiC MOSFETs mostly use organic polymer materials such as epoxy resin as encapsulation and insulating media. When the device is in a high junction temperature environment of around 175°C, accompanied by high stress of tens of MPa, the electric field, thermal field, and mechanical stress caused by differences in thermal expansion will interact, leading to degradation or even failure of the dielectric properties of the encapsulation material. Once the insulating layer breaks down or deteriorates, it will not only reduce device efficiency but may also trigger thermal runaway, causing serious damage to the entire system.

[0003] Current research on the lifespan prediction of encapsulated insulating materials largely relies on single-factor accelerated aging experiments or empirical formulas. These methods typically establish a corresponding lifespan model by changing one of the factors, such as voltage, temperature, or stress. However, in practical applications, voltage, stress, and temperature are not independent but exhibit significant coupling effects. Single-factor models struggle to reveal complex interactions, leading to deviations between predicted results and actual service life. Furthermore, traditional full-factor experimental designs, to comprehensively consider all main effects, their quadratic terms, and interaction terms, require an exponentially increasing number of experimental combinations, consuming significant time and costs and placing a heavy burden on experimental platforms and sample preparation.

[0004] Furthermore, under normal operating conditions, insulation life often far exceeds the experimentally acceptable period, making it extremely difficult to obtain valid failure data. To overcome the problem of excessively long lifespans, researchers typically employ accelerated aging methods, but this still requires numerous repeated experiments to ensure statistically significant confidence levels. This further exacerbates the experimental workload and resource consumption. Meanwhile, the distribution characteristics of lifetime data are complex; directly fitting the raw results often fails to yield robust prediction formulas. Therefore, it is necessary to combine appropriate statistical distributions and mathematical models to ensure reliability.

[0005] In summary, existing lifetime prediction methods still have significant shortcomings when dealing with the high voltage, high stress, and high temperature multi-field coupling conditions of SiC MOSFETs. These shortcomings are mainly reflected in high experimental costs, poor statistical reliability, and lack of model universality. Summary of the Invention

[0006] This application addresses the problem in existing technologies that fail to balance the demand for experimental data with accuracy in predicting the lifetime of SiC MOSFETs. It provides a method and system for predicting the aging lifetime of SiC MOSFETs. By combining experimental design and response surface methodology, it achieves experiments on the single and interactive effects of aging influencing factors with a smaller number of experimental groups. At the same time, it uses residual analysis to select reliable training data for training the aging lifetime prediction model, avoiding the problem of training bias caused by errors due to a small amount of experimental data. This simultaneously reduces the demand for experimental data and improves the accuracy of aging lifetime prediction.

[0007] To achieve the above technical objectives, this application provides a technical solution for predicting the aging lifetime of SiC MOSFETs, comprising the following steps: determining the aging influencing factors affecting the lifetime of SiC MOSFETs; selecting experimental point combinations of aging influencing factors based on experimental design methods, conducting lifetime tests on SiC MOSFETs according to the experimental point combinations, and obtaining an initial lifetime time data set corresponding to each experimental point combination; constructing a response surface model related to lifetime and aging influencing factors based on the initial lifetime time data set using response surface analysis; performing residual analysis on the lifetime prediction value output by the response surface model and the initial lifetime time data set, and obtaining qualified lifetime time data whose residual analysis results satisfy model fit and assumption compliance; constructing an initial aging lifetime prediction model based on the coupling effect between aging influencing factors and the response surface model; training the initial aging lifetime prediction model with qualified lifetime time data to obtain an aging lifetime prediction model; and outputting the aging lifetime prediction result of SiC MOSFETs based on real-time aging influencing factor data and the aging lifetime prediction model.

[0008] Furthermore, the experimental point combination for selecting aging influencing factors based on experimental design method includes: selecting several experimental parameters of voltage, stress and temperature according to an arithmetic sequence based on experimental design method, and coding the experimental parameters horizontally to obtain the experimental point combination.

[0009] Furthermore, the step of performing lifetime testing on SiC MOSFETs based on experimental point combinations and obtaining the initial lifetime time data set corresponding to each experimental point combination also includes: performing lifetime testing on SiC MOSFETs based on experimental point combinations according to a preset number of experiments to obtain a first experimental set corresponding to each experimental point combination; wherein the preset number of experiments is greater than or equal to two; fitting the lifetime distribution of the first experimental set using a Weibull distribution, and obtaining the initial lifetime time data based on the lifetime distribution fitting result and a preset failure probability.

[0010] Furthermore, the method of constructing a response surface model related to lifespan and aging influencing factors based on the initial lifespan time data set using response surface analysis includes: constructing a quadratic polynomial model based on voltage, stress, and temperature; and substituting the initial lifespan time data set into the quadratic polynomial model using the least squares method to obtain the response surface model.

[0011] Furthermore, the residual analysis of the lifetime prediction value output by the response surface model and the initial lifetime time data set includes: using the voltage, stress, and temperature corresponding to all experimental point combinations as inputs to the response surface model, and outputting lifetime prediction values; calculating the residual between the actual lifetime and the lifetime prediction value based on the actual lifetime corresponding to the experimental point combinations in the initial lifetime time data set; constructing a normal probability plot, a residual-fit value plot, and a residual order plot based on the distribution of the residuals and the experimental order, and performing residual analysis based on the normal probability plot, the residual-fit value plot, and the residual order plot.

[0012] Furthermore, the construction of the normal probability map, residual-fit value map, and residual order map based on the distribution of residuals and the experimental order includes: calculating the theoretical quantile corresponding to each residual according to the residual size and constructing the normal probability map; constructing the residual-fit value map based on the lifetime prediction value and the residuals; and constructing the residual order map based on the residuals and the corresponding experimental order.

[0013] Furthermore, the residual analysis based on the normal probability plot, the residual-fit value plot, and the residual order plot includes: determining whether the normality assumption is satisfied based on the degree of deviation between the distribution of residuals on the normal probability plot and the fitted line; determining whether the homogeneity of variance assumption is satisfied based on the degree of dispersion of the residual-fit value data points on the residual-fit value plot; and determining whether the independence assumption is satisfied based on the degree of dispersion of the residual-order data points on the residual-fit value plot.

[0014] Furthermore, the initial model for predicting aging life based on the coupling effect between aging influencing factors and the response surface model includes: constructing the initial model for predicting aging life based on the dynamic coupling effect of voltage and stress, the dynamic coupling effect of stress and temperature, and the response surface model.

[0015] Furthermore, the process includes: selecting a set of retained sample data and randomly dividing it into a training set and a validation set; calculating the validation determination coefficient of the aging life prediction model using the validation set data and obtaining the fitted value of the validation determination coefficient; calculating the training determination coefficient of the aging life prediction model using the training set data and obtaining the fitted value of the training determination coefficient; outputting the parameter fitting reliability evaluation result of the aging life prediction model based on the comparison results of the fitted value of the validation determination coefficient, the fitted value of the training determination coefficient, and the preset fitted value; performing residual analysis based on the residuals of the validation set data and the training set data and outputting the accuracy evaluation result of the aging life prediction model; and determining whether to reconstruct the aging life prediction model based on the parameter fitting reliability evaluation result and the accuracy evaluation result.

[0016] Another technical solution provided in this application is an aging lifetime prediction system for SiC MOSFETs, used to implement the method described above, comprising: an experimental unit for determining aging influencing factors affecting the lifetime of SiC MOSFETs, selecting experimental point combinations of aging influencing factors based on experimental design methods, performing lifetime tests on SiC MOSFETs according to the experimental point combinations, and obtaining an initial lifetime time data set corresponding to each experimental point combination; a model building unit for constructing a response surface model related to lifetime and aging influencing factors using response surface analysis methods based on the initial lifetime time data set, performing residual analysis on the lifetime prediction value output by the response surface model and the initial lifetime time data set, obtaining qualified lifetime time data whose residual analysis results satisfy model fit and assumption condition compliance, constructing an initial aging lifetime prediction model based on the coupling effect between aging influencing factors and the response surface model, training the initial aging lifetime prediction model with qualified lifetime time data, and obtaining an aging lifetime prediction model; and a lifetime prediction unit for outputting the aging lifetime prediction result of SiC MOSFETs based on real-time aging influencing factor data and the aging lifetime prediction model.

[0017] The beneficial effects of this application are as follows: 1. By combining experimental design and response surface methodology, a quadratic polynomial model is used to cover the main effects and interaction terms of aging influencing factors. This allows for experiments on single and interactive effects of aging influencing factors with fewer experimental groups, improving experimental efficiency and reducing R&D costs. Simultaneously, residual analysis of the lifetime prediction values ​​output by the response surface model and the initial lifetime time data is used to screen out qualified lifetime time data that meet the fit and assumption conditions, eliminating biased data and avoiding training distortion of the aging lifetime prediction model. Furthermore, an initial aging lifetime prediction model is jointly constructed based on the coupling effect and the response surface model, ensuring the model is adapted to the actual application of SiC MOSFETs (silicon carbide field-effect transistors) with multiple factors acting together, improving the accuracy of SiC MOSFET aging lifetime prediction and reducing the difficulty of constructing the aging lifetime prediction model.

[0018] 2. The cumulative distribution function of the Weibull distribution is used to describe the three stages of semiconductor device failure: early failure caused by manufacturing defects, random failure caused by accidental factors, and wear-out failure caused by aging. This method can adapt to the failure pattern of semiconductor devices throughout their entire life cycle, enabling the acquisition of continuous lifetime distribution through statistical fitting with less experimental data, thereby improving the accuracy and reliability of analysis with limited experimental data. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating the aging lifetime prediction method for SiC MOSFETs used in this application.

[0020] Figure 2 This is a normal probability diagram for one embodiment of the aging lifetime prediction method for SiC MOSFETs in this application;

[0021] Figure 3 This is a residual-fit value plot for one embodiment of the aging lifetime prediction method for SiC MOSFETs in this application;

[0022] Figure 4 This is a residual sequence diagram for one embodiment of the aging lifetime prediction method for SiC MOSFETs in this application. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description of this application is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely one preferred embodiment of this application and are only used to explain this application. They do not limit the scope of protection of this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0024] like Figure 1 As shown, the aging lifetime prediction method for SiC MOSFETs includes the following steps:

[0025] Identify the aging factors that affect the lifespan of SiC MOSFETs;

[0026] Based on the experimental design method, experimental point combinations of aging influencing factors are selected, and life tests are performed on SiC MOSFETs according to the experimental point combinations to obtain the initial life time data set corresponding to each experimental point combination.

[0027] A response surface model relating lifetime to aging influencing factors was constructed using the initial lifetime time data set based on response surface analysis.

[0028] Residual analysis was performed on the lifetime prediction values ​​output by the response surface model and the initial lifetime time data set to obtain qualified lifetime time data that meet the model fit and assumption conditions.

[0029] Based on the coupling effect among aging influencing factors, an initial model for aging lifetime prediction is constructed by combining a response surface model.

[0030] The aging life prediction initial model is trained using qualified lifespan time data to obtain the aging life prediction model.

[0031] The aging lifetime prediction results of SiC MOSFETs are output based on real-time aging impact factor data and aging lifetime prediction model.

[0032] In this embodiment, a combination of experimental design and response surface methodology is used to cover the main effects and interaction terms of aging influencing factors through a quadratic polynomial model. This allows for experiments on both single and interactive effects of aging influencing factors with fewer experimental groups, improving experimental efficiency and reducing R&D costs. Simultaneously, residual analysis of the lifetime prediction values ​​output by the response surface model and the initial lifetime time data is used to screen out qualified lifetime time data that meet the suitability and assumptions, eliminating biased data and preventing training distortion of the aging lifetime prediction model. Furthermore, an initial aging lifetime prediction model is constructed jointly based on the coupling effect and the response surface model, ensuring the model is adapted to the actual application of SiC MOSFETs (silicon carbide field-effect transistors) with multiple factors acting together, improving the accuracy of SiC MOSFET aging lifetime prediction and reducing the difficulty of constructing the aging lifetime prediction model.

[0033] Specifically, the aging factors affecting the lifespan of SiC MOSFETs include at least voltage, stress, and temperature. Through experiments investigating the interaction of voltage, stress, and temperature, the thermal expansion and stress coupling of the epoxy insulating material caused by SiC MOSFETs operating at high junction temperatures, high stress, and high voltage levels are obtained. This allows for the prediction of the aging lifespan of the SiC MOSFET packaging insulating material under the influence of voltage, stress, and temperature.

[0034] The experimental point combinations for selecting aging influencing factors based on experimental design methods include:

[0035] Based on the experimental design method, several experimental parameters of voltage, stress and temperature are selected according to an arithmetic sequence, and the experimental parameters are horizontally coded to obtain the experimental point combination.

[0036] In this embodiment, aging influencing factors =3, and three experimental parameters for each aging influencing factor are taken according to an arithmetic sequence, denoted as ( , , ; ∈1,2,3), where, Indicates the first Each voltage experimental parameter, Indicates the first One stress experiment parameter, Indicates the first Each temperature experimental parameter was used. The experimental parameter levels of the aging effect factors were coded and labeled from low to high as (-1, 0, 1). , , Each value represents one experimental group, and n groups are set up, where n is greater than or equal to 10:

[0037] ;

[0038] SiC MOSFET packaging material was fabricated into a thin film and placed on an electro-thermal-mechanical coupling accelerated aging test platform. Based on the combination of experimental points, the lifetime of the epoxy insulation material sample was tested to obtain the initial lifetime time data set under accelerated aging.

[0039] The process of performing lifetime testing on SiC MOSFETs based on experimental point combinations, and obtaining the initial lifetime time data set corresponding to each experimental point combination, also includes:

[0040] Lifetime tests were performed on SiC MOSFETs based on a preset number of experiments and combinations of experimental points to obtain the first set of experiments corresponding to each combination of experimental points; wherein the preset number of experiments was greater than or equal to two.

[0041] The lifetime distribution of the first experimental set is fitted using a Weibull distribution, and the initial lifetime time data set is obtained based on the lifetime distribution fitting results and the preset failure probability.

[0042] By employing repeated accelerated testing with multiple experimental points, multiple sets of repeated test results are obtained, avoiding the randomness of single tests and ensuring the representativeness and reliability of the data. Since semiconductor device failure follows a bathtub curve, the cumulative distribution function of the Weibull distribution is used to describe the three stages of semiconductor device failure: early failure due to manufacturing defects, random failure caused by chance factors, and wear-out failure due to aging. This approach can adapt to the failure patterns throughout the entire lifecycle of semiconductor devices, enabling the acquisition of continuous lifetime distributions through statistical fitting with less experimental data, thus improving the accuracy and reliability of analysis with limited experimental data.

[0043] In this embodiment, the preset number of experiments is 8. The lifetime distribution of the first experimental set is fitted using a Weibull distribution as follows:

[0044] ;

[0045] It can be represented as a linear expression:

[0046] ;

[0047] in, Represents the cumulative distribution function; This indicates the lifetime of the SiC MOSFET, and this indicates the failure time of the SiC MOSFET. Represents the scale parameter, when = When the failure probability is ; Represents shape parameters, when When, corresponding to early failure, when When, corresponding to random failure, when At that time, it corresponds to the loss of efficiency; This indicates the lifespan of the SiC MOSFET. , represents the shape parameter; , representing the scale parameter, is used to convert a nonlinear Weibull distribution into a linear form.

[0048] In this embodiment, the scale parameters and shape parameters are obtained by fitting using the least squares method. The preset failure probability is set to 63.2%, and the lifetime time corresponding to the failure probability of 63.2% is extracted as the characterization value to eliminate experimental randomness and accidental errors and obtain the initial lifetime time data set.

[0049] The failure probability at different time points is quantified by using the Weibull distribution, and the lifetime time data corresponding to the preset failure probability is extracted to construct an initial lifetime time data set.

[0050] Response surface methodology is used to construct response surface models relating lifespan to aging influencing factors based on initial lifespan time datasets, including:

[0051] A model incorporating quadratic polynomials is constructed based on voltage, stress, and temperature.

[0052] The initial lifetime time data set is substituted into a quadratic polynomial model using the least squares method to obtain the response surface model.

[0053] Based on voltage, stress, and temperature, a quadratic polynomial model is constructed as follows:

[0054] ;

[0055] in, Represents a constant; Indicates voltage; Indicates stress; Indicates temperature; Represents the coefficient of the first term of voltage; The one-time factor representing stress; The coefficient of the first-order term representing temperature; Represents the interaction coefficient between voltage and stress; Represents the interaction coefficient between voltage and temperature; Represents the interaction coefficient between stress and temperature; Represents the coefficient of the quadratic term of voltage; The coefficients of the quadratic term of stress; The coefficient of the quadratic term representing temperature.

[0056] In this embodiment, a quadratic polynomial model is constructed based on voltage, stress, and temperature. The coefficients of the first term in the quadratic polynomial model represent the linear influence of a single factor on lifespan, the coefficients of the quadratic term in the quadratic polynomial model represent the quadratic nonlinear influence of a single factor on lifespan, and the coefficients of the interaction term in the quadratic polynomial model represent the quadratic nonlinear influence of the combined effect of two factors on lifespan, thereby enabling the analysis of the influence of single factors and the combined effect of multiple factors.

[0057] It is understandable that the initial lifetime time data set includes at least horizontally coded voltage data, stress data, temperature data, and actual lifetime. The least squares method is used to fit the quadratic polynomial model based on the initial lifetime time data set to obtain the coefficient values ​​and thus obtain the response surface model.

[0058] Furthermore, residual analysis is performed on the lifetime predictions output by the response surface model and the initial lifetime time dataset, including:

[0059] The voltage, stress, and temperature corresponding to all experimental point combinations are used as inputs to the response surface model, and the lifetime prediction value is output.

[0060] Based on the actual lifetime corresponding to the experimental point combination in the initial lifetime time data set, calculate the residual between the actual lifetime and the lifetime prediction value;

[0061] Based on the distribution of residuals and the experimental order, construct a normal probability plot, a residual-fit value plot, and a residual order plot, respectively, and perform residual analysis based on the normal probability plot, the residual-fit value plot, and the residual order plot.

[0062] In this embodiment, a normal probability plot, a residual-fit value plot, and a residual order plot are constructed based on the distribution of the residuals and the experimental sequence to verify the adaptability of the response surface model, so as to ensure the accuracy of the subsequent aging lifetime prediction model construction.

[0063] Specifically, based on the distribution of the residuals and the experimental order, normal probability plots, residual-fit value plots, and residual order plots are constructed, including:

[0064] Calculate the theoretical quantile corresponding to each residual based on the residual size, and construct a normal probability graph;

[0065] Based on the predicted lifetime values ​​and residuals, a residual-fit value plot is constructed according to the experimental sequence.

[0066] Construct a residual sequence diagram based on the residuals and the corresponding experimental sequence.

[0067] The residual is calculated based on the actual lifespan and the predicted lifespan:

[0068] ;

[0069] in, Indicates the first The residuals of the group experiments Indicates the first The actual lifespan of the group experiment Indicates the first Lifetime prediction values ​​for the group experiment.

[0070] Assume there is The residuals from the group of experiments were sorted in ascending order as follows:

[0071] ;

[0072] The theoretical quantile for each residual is calculated as follows:

[0073] ;

[0074] in, Indicates the first The theoretical quantiles corresponding to the residuals of the group experiments. It represents the inverse function of the standard normal distribution.

[0075] A normal probability plot is constructed by using the horizontal axis as the theoretical quantile and the vertical axis as the residual, and adding a least-squares fitting line.

[0076] Based on the experimental sequence, a residual-fit value plot is constructed with the predicted lifetime value on the horizontal axis and the residual on the vertical axis.

[0077] Based on the experimental sequence, a residual sequence diagram is constructed with the horizontal axis representing the experimental sequence and the vertical axis representing the residuals.

[0078] Furthermore, residual analysis based on the normal probability plot, residual-fit value plot, and residual order plot includes:

[0079] Determine whether the normality assumption is satisfied based on the degree of deviation between the distribution of the residuals on the normal probability plot and the fitted line;

[0080] Determine whether the homogeneity of variance assumption is satisfied based on the degree of dispersion of the residual-fit value data points in the residual-fit value graph;

[0081] The degree of dispersion of the residual-ordered data points in the residual-fit value plot is used to determine whether the independence assumption is satisfied.

[0082] The presence of systematic bias is examined using normal probability plots, residual-fit value plots, and residual order plots to determine model fit and compliance with assumptions. If the residuals in the normal probability plot approximate a linear distribution, the normality assumption is satisfied; if the data points in the residual-fit value plot are randomly distributed on both sides of the zero line without a specific pattern, the homogeneity of variance assumption is satisfied; and if the residuals in the residual-order plot are randomly scattered around the center line, the independence assumption is satisfied. In this embodiment, preset deviation levels, a first preset distribution dispersion level corresponding to the residual-fit value data points, and a second preset distribution dispersion level corresponding to the residual-order data points can be pre-set. Whether the assumptions are satisfied is determined by whether the actual deviation level, the actual distribution dispersion level of the actual residual-fit value data points, and the actual distribution dispersion level of the actual residual-order data points respectively meet the preset deviation level, the first preset distribution dispersion level, and the second preset distribution dispersion level. It is understood that if all assumptions are satisfied, the response surface model is considered to be well-fitted. In other cases, a residual threshold can also be set to determine the suitability of the response surface model.

[0083] In this embodiment, residual analysis is performed on the lifetime prediction values ​​output by the response surface model and the initial lifetime time data set to obtain qualified lifetime time data whose residual analysis results satisfy model fit and assumption compliance.

[0084] If the residual analysis results simultaneously meet the assumptions of normality, homogeneity of variance, and independence, then the corresponding initial lifetime time data will be output as the qualified lifetime time data.

[0085] If the residual analysis results show that any of the assumptions are not met, an experimental error message will be output.

[0086] The reliability of experimental results is determined by whether the residuals satisfy the assumptions of normality, homogeneity of variance, and independence. If any of the assumptions are not met, the experiment is considered to have an error, and an experimental error message is output, prompting the experimenters to re-optimize the experimental design or supplement the experiment to ensure that the data used to train the aging life prediction model are all reliable experimental data, thereby improving the accuracy of the aging life prediction model.

[0087] Based on the coupling effect among aging influencing factors, and combined with a response surface methodology, an initial model for predicting aging lifetime is constructed, including:

[0088] An initial model for predicting aging life is constructed based on the dynamic coupling effect of voltage and stress, the dynamic coupling effect of stress and temperature, and the response surface model.

[0089] Specifically, the initial model for predicting basic aging lifetime, constructed based on the response surface methodology, is as follows:

[0090] ;

[0091] in, Indicates voltage; Indicates stress; Indicates voltage as Lifespan at that time This represents the critical voltage threshold at which voltage aging is negligible. Indicates the voltage aging factor. This represents the critical stress threshold at which stress aging is negligible. Indicates the stress aging coefficient. This indicates the critical temperature threshold at which temperature aging is negligible. This represents the temperature aging coefficient.

[0092] By correcting the dynamic coupling effects of voltage and stress, and stress and temperature, the initial model for predicting aging life is obtained as follows:

[0093] ;

[0094] in, This represents the fundamental coefficient for voltage-stress coupling. Correction factor for voltage-stress coupling. The fundamental coefficient representing the coupling between temperature and stress. This represents the correction factor for temperature and stress.

[0095] In this embodiment, the effect of stress on life loss caused by voltage or temperature is reflected by the product of the stress change magnitude and the correction coefficient. When the stress is greater, the voltage-induced electric shock breakthrough is more significant, and the temperature-induced thermal expansion is more significant. This reflects the superimposed effect of stress dynamic change on voltage and temperature, making the initial model for aging life prediction adaptable to life prediction under multiple operating conditions.

[0096] Specifically, the basic coefficients and correction coefficients in the initial state are set according to the coefficients of the first term, the second term, and the interaction term in the response surface model, which reduces the number of iterations in the aging lifetime prediction model and improves the model construction efficiency.

[0097] Furthermore, the coefficients of the initial aging life prediction model are fitted using a nonlinear regression method. The initial aging life prediction model is then trained using qualified lifespan data to obtain the aging life prediction model. The model is then trained using data that has been verified for reliability to ensure the accuracy of the initial aging life prediction model.

[0098] In some cases, methods for predicting the aging lifetime of SiC MOSFETs also include:

[0099] Select a set of reserved sample data and randomly divide it into a training set and a validation set;

[0100] The validation determination coefficients of the aging life prediction model are calculated using validation set data, and the fitted values ​​of the validation determination coefficients are obtained.

[0101] The training determination coefficient of the aging life prediction model is calculated using the training set data, and the fitted value of the training determination coefficient is obtained.

[0102] The results of verifying the fitted values ​​of the coefficient of determination, comparing the fitted values ​​of the trained coefficient of determination with the preset fitted values, and outputting the parameter fitting reliability evaluation results of the aging life prediction model are used.

[0103] Perform residual analysis based on the residuals between the validation set data and the training set data, and output the accuracy evaluation results of the aging life prediction model;

[0104] The decision to rebuild the aging life prediction model is based on the reliability and accuracy assessment results of the parameter fitting.

[0105] It is understood that the retained sample data should include at least actual lifetime, temperature, stress, and voltage data. The retained samples can be external samples or a set of experimental data retained during the experiment. The retained sample data is divided into a training set and a validation set in an 8:2 ratio.

[0106] The coefficient of determination is calculated as follows:

[0107] ;

[0108] in, Indicates the first The actual lifetime of the retained sample data. Indicates the first Predicted lifetime of retained sample data This represents the average predicted lifetime of all retained sample data.

[0109] In this embodiment, the preset fitting value is 0.5. If the fitting value of the verification coefficient of determination and the fitting value of the training coefficient of determination are both greater than 0.95, the parameter fitting of the aging life prediction model is considered reliable.

[0110] Simultaneously, a normal probability plot and an actual-predictive plot are plotted based on the residuals between the validation set data and the training set data. The actual-predictive plot is plotted with the horizontal axis representing the actual lifespan and the vertical axis representing the predicted lifespan. The accuracy evaluation results of the aging lifespan prediction model are output based on whether the normal probability plot satisfies a normal distribution and whether the slope of the fitted line of the actual-predictive plot is close to 1.

[0111] As a second embodiment of this application, a specific embodiment is provided for illustration. Assume that the voltage, stress, and temperature parameter settings and the corresponding Welbeck statistical lifetime characteristic values ​​are as shown in Table 1:

[0112] Table 1 Experimental parameters and results

[0113]

[0114] The n horizontally coded Vi, Si, Ti are shown in Table 2 below:

[0115] Table 2 Experimental Parameter Codes

[0116]

[0117] According to the least squares method, the coefficients of the linear term, quadratic term, and interaction term are respectively... =471.0333、 =-416.4125、 =-850.1625、 =-287.4750、 =326.1500、 =113.7750、 =301.9250、 =-8.0667、 =575.1333、 =49.4083.

[0118] The residuals for 15 sets of data were obtained: 41.1750, -140.7000, 140.7000, -41.1750, -173.0375, 8.8375, -8.8375, 173.0375, 131.8625, 32.3375, -32.3375, -131.8625, 1.4667, 3.0667, -4.5333. The residuals were plotted as follows: Figure 2 The normal probability graph shown is as follows: Figure 3 The residual-fit value plot shown and as follows Figure 4 The residual sequence diagram is shown.

[0119] by =3601 (s) =2.5 (kV), =10 (MPa) =20 + 273.15 (℃). An aging life prediction model was constructed, and the parameters of the aging life prediction model were obtained using nonlinear regression, as shown in Table 3.

[0120] Table 3 Parameters of Aging Life Prediction Model

[0121]

[0122] As a fourth embodiment of this application, an aging lifetime prediction system for SiC MOSFETs, connected to an electro-thermal-mechanical coupling accelerated aging test platform, includes:

[0123] The experimental unit is used to determine the aging factors that affect the lifetime of SiC MOSFETs. Based on the experimental design method, the experimental point combination of the aging factors is selected, and the lifetime of SiC MOSFETs is tested according to the experimental point combination to obtain the initial lifetime time data set corresponding to each experimental point combination.

[0124] The model building unit is used to construct a response surface model related to lifespan and aging influencing factors based on the initial lifespan time data set using the response surface analysis method. It performs residual analysis on the lifespan prediction value output by the response surface model and the initial lifespan time data set to obtain qualified lifespan time data that meet the model fit and assumption conditions. Based on the coupling effect between aging influencing factors, it constructs an initial aging lifespan prediction model in combination with the response surface model. The initial aging lifespan prediction model is trained with qualified lifespan time data to obtain the aging lifespan prediction model.

[0125] The lifetime prediction unit is used to output the aging lifetime prediction results of SiCMOSFET based on real-time aging influencing factor data and aging lifetime prediction model.

[0126] In this embodiment, the experimental unit is connected to the electro-thermal-mechanical coupling accelerated aging test platform, the model building unit is connected to the experimental unit, and the lifetime prediction unit is connected to the model building unit.

[0127] The specific embodiments described above are preferred embodiments of the aging lifetime prediction method and system for SiC MOSFETs in this application, and are not intended to limit the specific scope of this application. The scope of this application includes but is not limited to the specific embodiments described above. All equivalent changes made in accordance with the shape and structure of this application are within the protection scope of this application.

Claims

1. A method for predicting the aging lifetime of SiC MOSFETs, characterized in that: Includes the following steps: Identify the aging factors that affect the lifespan of SiC MOSFETs; Based on the experimental design method, experimental point combinations of aging influencing factors are selected, and life tests are performed on SiC MOSFETs according to the experimental point combinations to obtain the initial life time data set corresponding to each experimental point combination. A response surface model relating lifetime to aging influencing factors was constructed using the initial lifetime time data set based on response surface analysis. Residual analysis was performed on the lifetime prediction values ​​output by the response surface model and the initial lifetime time data set to obtain qualified lifetime time data that meet the model fit and assumption conditions. Based on the coupling effect among aging influencing factors, an initial model for aging lifetime prediction is constructed by combining a response surface model. The aging life prediction initial model is trained using qualified lifespan time data to obtain the aging life prediction model. The aging lifetime prediction results of SiC MOSFETs are output based on real-time aging influencing factor data and aging lifetime prediction model. The residual analysis is performed on the lifetime prediction values ​​output by the response surface model and the initial lifetime time data set to obtain qualified lifetime time data that satisfy the model fit and assumption conditions. The voltage, stress, and temperature corresponding to all experimental point combinations are used as inputs to the response surface model, and the lifetime prediction value is output. Based on the actual lifetime corresponding to the experimental point combination in the initial lifetime time data set, calculate the residual between the actual lifetime and the lifetime prediction value; Based on the distribution of residuals and the experimental order, construct a normal probability plot, a residual-fit value plot, and a residual order plot respectively, and perform residual analysis based on the normal probability plot, the residual-fit value plot, and the residual order plot; If the residual analysis results simultaneously meet the assumptions of normality, homogeneity of variance, and independence, then the corresponding initial lifetime time data will be output as the qualified lifetime time data. If the residual analysis results show that any of the assumptions are not met, an experimental error message will be output. The initial model for predicting aging lifespan is: ； in, Indicates the lifespan of SiC MOSFETs. Indicates voltage as Lifespan at that time Indicates the critical voltage threshold. Indicates voltage. This represents the fundamental coefficient for voltage-stress coupling. Correction factor for voltage-stress coupling. Indicates stress, Indicates the critical stress threshold. Indicates the stress aging coefficient. The fundamental coefficient representing the coupling between temperature and stress. This represents the correction factor for temperature and stress. This represents the critical temperature threshold. Indicates temperature.

2. The aging lifetime prediction method for SiC MOSFETs as described in claim 1, characterized in that: The experimental point combinations for selecting aging influencing factors based on experimental design include: Based on the experimental design method, several experimental parameters of voltage, stress and temperature are selected according to an arithmetic sequence, and the experimental parameters are horizontally coded to obtain the experimental point combination.

3. The aging lifetime prediction method for SiC MOSFETs as described in claim 1, characterized in that: The step of performing lifetime testing on SiC MOSFETs based on experimental point combinations and obtaining the initial lifetime time data set corresponding to each experimental point combination further includes: Lifetime tests were performed on SiC MOSFETs based on a preset number of experiments and combinations of experimental points to obtain the first set of experiments corresponding to each combination of experimental points; wherein the preset number of experiments was greater than or equal to two. The lifetime distribution of the first experimental set is fitted using a Weibull distribution, and the initial lifetime time data is obtained based on the lifetime distribution fitting results and the preset failure probability.

4. The aging lifetime prediction method for SiC MOSFETs as described in claim 1, characterized in that: The method of constructing a response surface model related to lifespan and aging influencing factors based on the initial lifespan time data set using response surface analysis includes: A quadratic polynomial model is constructed based on voltage, stress, and temperature. The initial lifetime time data set is substituted into a quadratic polynomial model using the least squares method to obtain the response surface model.

5. The aging lifetime prediction method for SiC MOSFETs as described in claim 1, characterized in that: The construction of the normal probability plot, residual-fit value plot, and residual order plot based on the distribution of residuals and the experimental order includes: Calculate the theoretical quantile corresponding to each residual based on the residual size, and construct a normal probability graph; Construct a residual-fit value plot based on the predicted lifetime value and the residual; Construct a residual sequence diagram based on the residuals and the corresponding experimental sequence.

6. The aging lifetime prediction method for SiC MOSFETs as described in claim 5, characterized in that: The residual analysis based on the normal probability plot, residual-fit value plot, and residual order plot includes: Determine whether the normality assumption is satisfied based on the degree of deviation between the distribution of the residuals on the normal probability plot and the fitted line; Determine whether the homogeneity of variance assumption is satisfied based on the degree of dispersion of the residual-fit value data points in the residual-fit value graph; The degree of dispersion of the residual-ordered data points in the residual-fit value plot is used to determine whether the independence assumption is satisfied.

7. The aging lifetime prediction method for SiC MOSFETs as described in claim 1, characterized in that: The initial model for predicting aging lifetime, based on the coupling effect between aging influencing factors and combined with a response surface model, includes: An initial model for predicting aging life is constructed based on the dynamic coupling effect of voltage and stress, the dynamic coupling effect of stress and temperature, and the response surface model.

8. The aging lifetime prediction method for SiC MOSFETs as described in claim 1, characterized in that: Also includes: Select a set of reserved sample data and randomly divide it into a training set and a validation set; The validation determination coefficients of the aging life prediction model are calculated using validation set data, and the fitted values ​​of the validation determination coefficients are obtained. The training determination coefficient of the aging life prediction model is calculated using the training set data, and the fitted value of the training determination coefficient is obtained. The results of verifying the fitted values ​​of the coefficient of determination, comparing the fitted values ​​of the trained coefficient of determination with the preset fitted values, and outputting the parameter fitting reliability evaluation results of the aging life prediction model are used. Perform residual analysis based on the residuals between the validation set data and the training set data, and output the accuracy evaluation results of the aging life prediction model; The decision to rebuild the aging life prediction model is based on the reliability and accuracy assessment results of the parameter fitting.

9. A aging lifetime prediction system for SiC MOSFETs, for implementing the method as described in any one of claims 1 to 8, characterized in that: include: The experimental unit is used to determine the aging factors that affect the lifetime of SiC MOSFETs. Based on the experimental design method, the experimental point combination of the aging factors is selected, and the lifetime of SiC MOSFETs is tested according to the experimental point combination to obtain the initial lifetime time data set corresponding to each experimental point combination. The model building unit is used to construct a response surface model related to lifespan and aging influencing factors based on the initial lifespan time data set using the response surface analysis method. It performs residual analysis on the lifespan prediction value output by the response surface model and the initial lifespan time data set to obtain qualified lifespan time data that meet the model fit and assumption conditions. Based on the coupling effect between aging influencing factors, it constructs an initial aging lifespan prediction model in combination with the response surface model. The initial aging lifespan prediction model is trained with qualified lifespan time data to obtain the aging lifespan prediction model. The lifetime prediction unit is used to output the aging lifetime prediction results of SiCMOSFET based on real-time aging influencing factor data and aging lifetime prediction model.

Images