A method and system for detecting reliability of a SiC power device package

Abstract

The application relates to the technical field of semiconductor device detection, and provides a SiC power device package reliability detection method and system, the detection method comprising the following steps: constructing a multi-physical field coupling finite element model, running the finite element model to obtain initial simulation data, performing parameter sensitivity analysis on the initial simulation data to determine a key stress parameter combination of a typical failure mode; collecting SiC power device working parameters, performing real-time calibration and denoising to obtain measured data, and generating a key region temperature gradient atlas; performing analysis on the initial simulation data and the measured data through a deep reinforcement learning algorithm, and iteratively optimizing the key stress parameter combination and the key region parameters; setting a failure criterion threshold, performing trend prediction, comparing the trend prediction result with the failure criterion threshold, and executing a multi-level response strategy according to a comparison result. The application can realize accurate detection of SiC power device package reliability in a multi-physical field coupling scene.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of semiconductor device detection, in particular to a SiC power device package reliability detection method and system. BACKGROUND

[0002] SiC (silicon carbide) power devices are a kind of semiconductor power devices, which have been widely used in electric vehicles, smart grids, aerospace, etc. due to their excellent high-temperature resistance, high frequency and high power density. The packaging of SiC power devices is the basis for ensuring their safety and fully exerting their excellent performance, so it is crucial to detect the packaging reliability of SiC power devices. There are also methods for detecting the packaging reliability of SiC power devices in the prior art, such as packaging thermal resistance analysis based on infrared temperature measurement or failure trend inference through electrical performance degradation curve. Although these existing detection methods can detect the packaging performance of SiC power devices to some extent, all detection processes rely on single physical parameter measurement, such as simple temperature monitoring or electrical performance testing, which cannot comprehensively and accurately reflect the packaging reliability of SiC power devices in complex application scenarios. SUMMARY

[0003] The application aims to solve the technical problem that the existing SiC power device packaging reliability detection method only relies on single physical parameter measurement during detection, which cannot comprehensively and accurately reflect the packaging reliability of SiC power devices in complex application scenarios. A SiC power device packaging reliability detection method and system are proposed to realize accurate detection of SiC power device packaging reliability in a multi-physical field coupling scenario.

[0004] To achieve the above-mentioned purpose, the application adopts the following technical solutions:

[0005] A SiC power device packaging reliability detection method, comprising the following steps:

[0006] Constructing a multi-physical field coupling finite element model of the SiC power device;

[0007] Running the multi-physical field coupling finite element model to obtain initial simulation data, and performing parameter sensitivity analysis on the initial simulation data to determine the key stress parameter combination of the typical failure mode;

[0008] Collecting the working parameters of the SiC power device in the running state, obtaining the measured data after real-time calibration and denoising of the working parameters; during the process of collecting the working parameters of the SiC power device in the running state, the key areas of the SiC power device are located by dynamic region tracking technology, and the key area temperature gradient map of the SiC power device is generated according to the measured data;

[0009] The initial simulation data is compared with the measured data by a deep reinforcement learning algorithm, and the key stress parameter combination and the key region parameter of the multi-physical field coupling finite element model are iteratively optimized based on the comparison result until the error between the initial simulation data and the measured data is within a set range, so as to obtain the optimized key stress parameter combination and the key region parameter of the multi-physical field coupling finite element model.

[0010] The final simulation data is obtained by simulation according to the optimized key stress parameter combination and the key region parameter of the multi-physical field coupling finite element model, and a failure criterion threshold is set according to the final simulation data.

[0011] The key region temperature gradient map is input into a trend prediction model for trend prediction and to obtain prediction data, the prediction data is compared with the failure criterion threshold, and a multi-level response strategy is executed according to the comparison result.

[0012] Further, the process of constructing the multi-physical field coupling finite element model of the SiC power device includes: constructing or importing a geometric model of the SiC power device, setting SiC power device parameters, setting boundary conditions, loads and typical failure modes of the model, and performing mesh division and solution setting to obtain the multi-physical field coupling finite element model.

[0013] Further, the setting of the SiC power device parameters includes setting the thermal conductivity, thermal expansion coefficient, elastic modulus and Poisson's ratio of the packaging material at different temperatures; the thermal conductivity of the packaging material at different temperatures is obtained by laser flash method.

[0014] Further, the typical failure modes include solder layer cracks, thermal resistance degradation, and bond wire shedding, the failure of the solder layer crack is determined by obtaining the strain concentration area in the thermal-mechanical stress simulation, the failure of the thermal resistance degradation is determined by obtaining the simulation thermal resistance parameter, and the failure of the bond wire shedding is determined by obtaining the stress analysis of the bonding area in the electro-thermal coupling simulation.

[0015] Further, the process of performing parameter sensitivity analysis on the initial simulation data to determine the key stress parameter combination of the typical failure mode is: a stress parameter sampling matrix is designed by Morris global sensitivity analysis method; finite element simulation is performed on each group of stress parameter combinations, and failure indicators are output; the normalized sensitivity coefficient of each stress parameter to failure time is calculated to determine the key stress parameter combination of the typical failure mode.

[0016] Further, the real-time calibration process is: the temperature parameter in the working parameter is calibrated in real time by using an environmental radiation compensation algorithm.

[0017] Further, the process of real-time calibration of the temperature parameter in the working parameter by using the environmental radiation compensation algorithm is: measuring the radiation intensity under different wavelengths in the process of collecting the temperature parameter of the SiC power device, and calculating and calibrating the measured temperature according to the blackbody radiation law.

[0018] Further, the process of denoising is: time alignment processing of the working parameter and the simulation data, and then using a wavelet threshold denoising algorithm to eliminate environmental noise in the working parameter.

[0019] Further, the failure criterion threshold includes a pre-warning threshold and a dangerous threshold, the multi-level response strategy is controlled by using a fuzzy logic control method, and the membership function used in the control process of the fuzzy logic control method is:

[0020] ;

[0021] In the formula, is the failure severity membership degree, the value range is ; x is the prediction data; a is the pre-warning threshold; and b is the dangerous threshold.

[0022] The detection system based on the above-mentioned SiC power device packaging reliability detection method and the same inventive concept comprises:

[0023] A model construction module is configured to construct a multi-physical field coupled finite element model of the SiC power device.

[0024] An initial simulation and parameter analysis module is configured to run the multi-physical field coupled finite element model to obtain initial simulation data, and perform parameter sensitivity analysis on the initial simulation data to determine a key stress parameter combination of a typical failure mode.

[0025] A real measurement data collection module comprises a working parameter collection unit, a data calibration and denoising unit, a dynamic region tracking unit, and a temperature gradient atlas generation unit.

[0026] The working parameter collection unit is configured to collect working parameters of the SiC power device in a running state.

[0027] The data calibration and denoising unit is configured to obtain real measurement data by real-time calibration and denoising of the working parameters.

[0028] The dynamic region tracking unit is configured to locate a key region of the SiC power device by using a dynamic region tracking technology.

[0029] The temperature gradient atlas generation unit is configured to generate a temperature gradient atlas of the key region of the SiC power device according to the real measurement data.

[0030] The comparative analysis and iterative optimization module is configured to compare the simulation data and the measured data by using a deep reinforcement learning algorithm, and iteratively optimize the key stress parameter combination and the key region parameter of the multi-physical field coupling finite element model based on the comparison result until the error between the initial simulation data and the measured data is within a set range, so as to obtain the optimized key stress parameter combination and the key region parameter of the multi-physical field coupling finite element model.

[0031] The final simulation and failure criterion setting module is configured to simulate to obtain final simulation data according to the optimized key stress parameter combination and the key region parameter of the multi-physical field coupling finite element model, and set a failure criterion threshold according to the final simulation data.

[0032] The trend prediction and response execution module includes a trend prediction unit and a comparative response unit.

[0033] The trend prediction unit is configured to input the key region temperature gradient map into a trend prediction model to perform trend prediction and obtain prediction data.

[0034] The comparative response unit is configured to compare the prediction data with the failure criterion threshold, and execute a multi-level response strategy according to the comparison result.

[0035] The application has the following beneficial effects:

[0036] The application firstly constructs a multi-physical field coupling finite element model to simulate and detect SiC power devices, which can fully consider the packaging reliability of SiC power devices in various complex application scenarios, and more accurately evaluate the thermal-mechanical stress distribution of SiC power devices in actual work, thereby providing strong support for the optimization design and reliability improvement of the devices. Meanwhile, the application iteratively optimizes and updates the simulation data by using the measured data, which can effectively improve the accuracy of the final detection result. Moreover, the measured data is calibrated by an environmental radiation compensation algorithm, which can effectively avoid the interference of environmental temperature on the measured data, ensure the reliability of the measured data, and further improve the accuracy of the final detection result.

[0037] The application can effectively ensure the consistency and comparability of the two groups of data by time alignment processing of the simulation data and the measured data, and then eliminate the environmental noise in the data by using a wavelet threshold denoising algorithm to effectively improve the signal-to-noise ratio of the data, thereby improving the accuracy of subsequent data analysis. BRIEF DESCRIPTION OF DRAWINGS

[0038] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the description of the embodiments will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0039] Figure 1 A flowchart of a SiC power device packaging reliability detection method provided by an embodiment of the present application is shown in the figure.

[0040] Figure 2 A schematic structural block diagram of a SiC power device packaging reliability detection system provided by an embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0041] The SiC power device packaging reliability detection method and system provided by the embodiments of the present application can simulate and detect the SiC power device by using the COMSOL Multiphysics software to construct a multi-physical field coupling finite element model, calibrate the temperature parameter in the working parameter by using the environmental radiation compensation algorithm, obtain the measured data by using the wavelet threshold method to denoise the working parameter, then analyze and process the measured data and the initial simulation data in real time by using the deep reinforcement learning algorithm, and realize the iterative optimization of each parameter, then obtain the final simulation data according to the optimized parameters, set the failure criterion threshold value by using the final simulation data, predict the trend of the future value of the measured data by using the prediction model to obtain the prediction data, and finally compare the prediction data with the failure criterion threshold value, and execute the multi-level response strategy according to the comparison result, so as to realize the failure prediction of the SiC power device. When used, the packaging reliability of the SiC power device in various complex application scenarios can be fully considered to realize the accurate detection of the packaging reliability of the SiC power device in the multi-physical field coupling scenario. Thus, the technical problem that the packaging reliability detection method of the SiC power device in the prior art only relies on a single physical parameter for measurement when detecting, so that it is difficult to comprehensively and accurately reflect the packaging reliability of the SiC power device in the complex application scenario is solved.

[0042] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the present application.

[0043] As Figure 1As shown, the embodiment provides a SiC power device package reliability detection method, applied to a SiC power device package reliability detection system, and the SiC power device can be SiC MOSFET / diode, etc. The detection method specifically includes the following steps:

[0044] A COMSOL Multiphysics software is used to build or import a geometric model of the SiC power device, set SiC power device parameters, set boundary conditions, loads and typical failure modes of the model, and perform mesh division and solution settings to obtain a multi-physical field coupling finite element model.

[0045] In the above steps, the setting of the SiC power device parameters can include setting the thermal conductivity, thermal expansion coefficient, elastic modulus, and Poisson's ratio of the packaging material at different temperatures, and the corresponding SiC power device parameters can be set according to actual needs by those skilled in the art during actual application; the thermal conductivity of the packaging material at different temperatures is obtained by laser flash method.

[0046] In some embodiments, the specific setting of the SiC power device parameters is as follows:

[0047]

[0048] In the above steps, the boundary conditions can be thermal boundary, electrical boundary, and mechanical boundary; the boundary conditions can be set according to actual needs by those skilled in the art during actual application.

[0049] In some embodiments, the boundary conditions and loads can be set as follows:

[0050] Thermal boundary: forced air cooling or liquid cooling is used, the convective heat transfer coefficient is set to 50-200 W / (m²·K), and the environmental temperature is set to 25-85℃;

[0051] Electrical boundary: for SiC MOSFET, the drain-source voltage Vds is set to 0-1200V, and the gate-source voltage Vgs is set to -5-20V;

[0052] Mechanical boundary: the substrate fixed edge displacement constraint is 0, and the power terminal is subjected to periodic vibration load (frequency 10-200Hz, acceleration 2-5g);

[0053] Load step: thermal cycling is performed, the temperature range is -40℃-150℃, the temperature rising rate is 10℃ / min, electromigration test is performed, direct current or pulse current is used, and the duty cycle is 50%-100%.

[0054] The typical failure modes include solder layer crack, thermal resistance degradation, and wire pull-off. The failure of the solder layer crack is determined by obtaining a strain concentration area in a thermal-mechanical stress simulation, the failure of the thermal resistance degradation is determined by obtaining a simulation thermal resistance parameter, and the failure of the wire pull-off is determined by obtaining a wire area stress analysis in an electro-thermal coupling simulation.

[0055] An initial simulation data is obtained by running a multi-physics coupling finite element model, and a parameter sensitivity analysis is performed on the initial simulation data to determine a key stress parameter combination of a typical failure mode.

[0056] The simulation analysis of obtaining the initial simulation data by running the multi-physics coupling finite element model is achieved by simultaneously solving the following equation groups:

[0057] (1);

[0058] (2);

[0059] (3);

[0060] In formula (1), is a stress tensor; is a strain tensor; T is a temperature field; in formula (2), is a temperature-dependent thermal conductivity; is a gradient temperature; is a Joule heat source term; is a thermoelastic heat source term; in formula (3), J is a current density; is an electrical conductivity; E is an electric field intensity; v is a carrier drift velocity; and B is a magnetic induction intensity.

[0061] Equation describes the deformation of the packaging material under stress, wherein the relationship between the stress tensor and the strain tensor reflects the elastic correlation of the material, and provides basic data of structural deformation for subsequent thermal and electrical physical field analysis.

[0062] Equation describes the heat conduction process in the material, the temperature-dependent thermal conductivity represents the change of the thermal conductivity with temperature, and the gradient temperature represents the spatial change rate of the temperature. The temperature-dependent thermal conductivity and the gradient temperature can accurately simulate the heat transfer in different temperature regions, and the Joule heat source term and the thermoelastic heat source term introduce the effects of the electric and force fields on the thermal field.

[0063] Equation factors such as electric field, carrier drift velocity, and magnetic induction intensity are combined to obtain the current density. The Joule heat source term is calculated by the current density and the electric field intensity, which reflects the heat generated when the current passes through the material.

[0064] In the process of current passing through the device, the current density can be calculated by formula (3), and the Joule heat source term Q joule can be calculated by the current density and the electric field intensity. joule The obtained Joule heat source term Q piezo is fed back to formula (2) and drives formula (2) to cause the change of the temperature gradient, and the change of the temperature gradient causes the change of the temperature field T, and the change of the temperature field drives the change of the stress tensor and the strain tensor in formula (1), and then causes the thermal stress generated by thermal expansion and contraction of the material, so that the thermoelastic heat source term Q piezo is calculated and obtained. piezo For most packaging materials, under adiabatic or high-speed deformation conditions, when the material is rapidly compressed, the temperature will rise; when it is rapidly stretched, the temperature will decrease. This part of the thermal effect caused by the volume change is the thermoelastic effect. The thermoelastic heat source term Q piezo is obtained and fed back to formula (2). Because the conductivity is a function of temperature, when the temperature field T changes in the above process, it will also act on the conductivity in formula (3), so that the current density in formula (3) changes again. As can be seen, by solving the above three equations, the cyclic coupling process of “electricity generates heat→heat-induced force→force generates heat” is completely covered. Instead of isolated analysis of a single physical field, the synchronous iterative calculation of multiple physical fields is realized, which can better fit the actual situation of mutual influence of multiple fields, improve the accuracy of model prediction, and provide an important reference for the design and optimization of SiC power devices.

[0065] The initial simulation data obtained by solving the above equation set for multi-physical field coupling finite element simulation is the direct input for subsequent parameter sensitivity analysis. The initial simulation data includes the physical quantity distribution parameters of the entire SiC power device packaging structure in space directly output by the coupled simulation and the parameters calculated to quantify the failure risk, and the two types of parameters are specifically:

[0066] 1. The physical quantity distribution parameters of the entire SiC power device packaging structure in space directly output by the coupled simulation mainly include:

[0067] Temperature field (T): solved by the above heat conduction equation (2), used to identify overheating areas and calculate temperature gradients;

[0068] Stress field (σ) and strain field (ε): solved by the above mechanical equilibrium equation (1), used to locate the mechanical stress / strain concentration area caused by thermal expansion mismatch;

[0069] Current density field (J): solved by the above current field equation (3), used to identify current crowding effect and high Joule heat area;

[0070] 2、 The calculated parameters for quantifying failure risk are based on the above-mentioned physical field distribution parameters and are determined / calculated according to the determination criteria of typical failure modes. These parameters include solder layer cracks, thermal resistance degradation, and wire disconnection, which are direct target response quantities for parameter sensitivity analysis. The determination / calculating process of each parameter for quantifying failure risk is as follows:

[0071] For solder layer cracks: the maximum equivalent strain of the solder layer is extracted from the strain field (ε) to determine the strain concentration area;

[0072] For thermal resistance degradation: the difference between the chip junction temperature and the shell temperature is calculated according to the temperature field (T) data, and combined with the input power to obtain the simulated thermal resistance value, i.e., the simulated thermal resistance parameter;

[0073] For wire disconnection: the maximum principal stress or shear stress is extracted from the stress field (σ) in the bonding area to analyze the stress in the bonding area;

[0074] After obtaining the above initial simulation data, the failure reference state under a certain set of fixed input parameters is obtained. Subsequently, parameter sensitivity analysis can be performed on the initial simulation data to determine the key stress parameter combination of the typical failure mode. The specific process is as follows: a stress parameter sampling matrix is designed by the Morris global sensitivity analysis method. The stress parameters include physical variables that affect the failure time of the SiC power device, including temperature-related parameters such as environmental temperature, junction temperature, and shell temperature, current-related parameters such as device operating current and current density, voltage-related parameters such as drain-source voltage and gate-source voltage, thermal cycle number, mechanical stress caused by packaging layer stress mismatch, and the like. Those skilled in the art can design the stress parameter sampling matrix according to actual needs. In some embodiments, the temperature can be 80℃ / 120℃ / 150℃, the current can be 10A / 20A / 30A, and the thermal cycle can be 100 times / 500 times / 1000 times. After the design is completed, the failure time data corresponding to each set of stress parameters is obtained by performing a coupled simulation once by changing one stress parameter in the stress parameter sampling matrix. Then, the normalized sensitivity coefficient of the failure time with respect to each changed stress parameter is calculated to determine the key stress parameter combination of the typical failure mode. The formula for calculating the sensitivity coefficient is as follows:

[0075] ;

[0076] wherein, is the sensitivity coefficient; is the failure time variation; Y is the reference failure time; is the variation of the i-th stress parameter; is the reference value of the i-th stress parameter;

[0077] Firstly, the reference value corresponding to the i-th stress parameter needs to be set according to the actual application scenario or the specification of the SiC power device and the stress parameter variation ; for example, the temperature reference value can be set to 25℃, and the current reference value to 10A. The stress parameter variation is a percentage or absolute value of the reference value, for example, the temperature variation is ±50℃, and the current variation is ±5A. When calculating the sensitivity coefficient, the reference failure time is obtained by finite element simulation, then a single stress parameter is changed, the failure time variation is obtained by re-simulation, and the corresponding value is brought into the formula to obtain the normalized sensitivity coefficient. When , it is determined that the stress parameter is a key stress parameter. The Morris global sensitivity analysis method can efficiently identify the stress parameter that has the greatest impact on the failure time, providing key guidance for optimization design and improving reliability, which helps to shorten the development cycle of SiC power devices and reduce costs. By determining the key stress parameter combination, targeted monitoring and protection of these areas can be strengthened to reduce the risk of failure and improve the service life and reliability of the device.

[0078] The above-mentioned Morris global sensitivity analysis method is a sensitivity analysis method based on "one factor at a time" variation, which evaluates the importance of parameters by calculating the normalized sensitivity coefficient of each stress parameter to the failure time. In this method, each stress parameter varies within a certain range while keeping other parameters constant, and the degree of influence of the parameter on the failure time is determined by observing the variation of the failure time. The greater the sensitivity coefficient, the more significant the impact of the parameter on the failure time, and thus it is determined as a key stress parameter.

[0079] In some embodiments, the key stress parameter combination of the typical failure mode of the SiC power device needs to cover the thermal-electric-mechanical multi-physical field coupling effect, specifically including at least the following parameters and typical combinations:

[0080] Single parameter: temperature, junction temperature, shell temperature, current density, device operating current, area, voltage, drain-source voltage Vds, gate-source voltage Vgs, number of thermal cycles, number of temperature fluctuation cycles, mechanical stress, inter-package stress caused by thermal expansion coefficient mismatch.

[0081] Combined parameters:

[0082] Temperature-current interaction: high temperature + high current to accelerate electromigration failure;

[0083] Temperature-thermal cycle interaction: temperature fluctuation + high temperature to cause solder layer fatigue;

[0084] Current-mechanical stress interaction: joule heat under high current density aggravates thermal stress.

[0085] The working parameters of the SiC power device in the running state are collected, the working parameters are calibrated in real time, then the working parameters calibrated in real time are time-aligned with the simulation data, the environmental noise in the working parameters is eliminated by using a wavelet threshold denoising algorithm to obtain measured data; in the process of collecting the working parameters of the SiC power device, the key area of the SiC power device is located by using a dynamic area tracking technology, and a temperature gradient map of the key area of the SiC power device is generated according to the measured data.

[0086] In the above steps, the process of collecting the working parameters of the SiC power device includes arranging temperature sensors, current sensors and voltage sensors at key positions of the SiC power device to collect the working temperature, current and voltage of the SiC power device in real time and obtain the working parameters. The key positions are positions with significant thermal-mechanical coupling and the highest failure risk obtained from the previous multi-physical field coupling simulation and parameter sensitivity analysis, including the solder layer between the chip and the substrate due to CTE mismatch, the power terminal bonding area due to high current density, the center of the chip surface with a power density of >500 W / cm2, the package edge corner with mechanical stress concentration and the gate driving position with voltage spikes due to high frequency switching.

[0087] Since the electrical parameters such as working current and voltage of the SiC power device are generally less affected by the outside during the working process, and the temperature parameter is greatly affected by the environment, the real-time calibration of the working parameters is generally the real-time calibration of the temperature parameter in the working parameters, and the real-time calibration method used is an environmental radiation compensation algorithm for real-time calibration of the temperature in the working parameters. The specific process is as follows: in the process of collecting the temperature parameters of the SiC power device, a dual-color infrared thermometer is integrated to measure the radiation intensity at different wavelengths, and then the blackbody radiation law is used to calculate and calibrate the measured temperature, so as to realize the real-time calibration of the measured temperature of the SiC power device to eliminate the influence of the environmental temperature. The real-time calibration of the measured temperature is realized by the following formula:

[0088] ;

[0089] In the formula, is the corrected true temperature; is the second radiation constant; , is the central wavelength of the dual-wavelength filter; , is the emissivity of the packaging material corresponding to the central wavelength; , is the blackbody radiation spectral density; T is the temperature.

[0090] In the above correction formula, the corrected true temperature is calculated by the center wavelength of the dual-wavelength filter, the emissivity of the packaging material at the corresponding wavelength, the blackbody radiation spectral density, and the second radiation constant. The dual-color infrared temperature measurement method measures the radiation intensity at two wavelengths simultaneously and uses the radiation characteristics at these two wavelengths to eliminate the influence of environmental radiation on the temperature measurement results, thereby improving the accuracy of temperature measurement and ensuring the reliability of temperature data. This algorithm is suitable for temperature measurement in various complex environments and provides a strong guarantee for the reliable operation of SiC power devices.

[0091] In the above steps, the key area of the SiC power device is located by dynamic area tracking technology, and the process of generating the key area temperature gradient map of the SiC power device according to the measured data is as follows: an infrared thermometer scans several areas (e.g. 100 areas) per second, combined with image recognition to locate temperature changes, and generates a key area temperature gradient map with a temperature difference ≥ 10 ℃ / mm. The key area is the part with significant thermal-mechanical coupling and the highest failure risk obtained from the previous multi-physical field coupling simulation and parameter sensitivity analysis process, including the solder layer between the chip and the substrate due to CTE mismatch, the power terminal bonding area due to high current density, the center of the chip surface with a power density > 500 W / cm2, the packaging edge corner with mechanical stress concentration, and the gate drive area with voltage spikes due to high-frequency switching. The key area and the key position described above are actually related and progressive. The key area represents a range of areas, while the key position is a specific point position in the key area. The temperature gradient map is constructed by the following two-dimensional heat flux density formula:

[0092] ;

[0093] where, is the heat flux vector at coordinates (x, y); is the temperature-dependent thermal conductivity; , are the components of the temperature gradient in the x and y directions, respectively; i and j are unit vectors of the coordinate axes. The two-dimensional heat flux density formula can intuitively display the temperature distribution and heat flow direction in the packaging structure of the SiC power device, providing an important basis for analyzing the heat conduction path and hot spot location, and helping to optimize the thermal design of the SiC power device. By constructing the temperature gradient map, potential thermal problem areas can be identified in a timely manner, and appropriate heat dissipation measures can be taken to improve the thermal stability and reliability of the device.

[0094] The initial simulation data and the measured data are compared and analyzed by a deep reinforcement learning algorithm, and based on the comparison and analysis results, the key stress parameter combination and the key region parameters of the multi-physical field coupling finite element model are iteratively optimized until the error between the initial simulation data and the measured data is within a set range, thereby obtaining the optimized key stress parameter combination and the key region parameters of the multi-physical field coupling finite element model.

[0095] The deep reinforcement learning algorithm adopts an improved DDPG algorithm, and the action value function is updated by the following Bellman equation:

[0096]

[0097]

[0098] In the formula, The mean square error of the current Q value and the target Q value is used to measure the mean square error of the current Q value and the target Q value. The network parameters are used to measure the mean square error of the current Q value and the target Q value. The Q value of the action a selected by the current network in the state s is used to measure the Q value of the action a selected by the current network in the state s. The maximum Q value of the next state is used to calculate the target Q value. The sample in the experience replay buffer D is used to calculate the average. The feedback obtained from the environment after performing the action a is used to calculate the instant reward. The discount factor is used to weigh the importance of future rewards. The next state is used to calculate the next state. The Q value of the target network is used to calculate the Q value of the target network. The action output by the target policy network is used to calculate the action output by the target policy network. If it is a discrete action space, the action that maximizes the Q value is taken. If it is a continuous action space, the action is directly given by the policy network.

[0099] In the improved DDPG algorithm, the action value function is updated by the Bellman equation to measure the mean square error of the current Q value and the target Q value, and the experience replay buffer is used to optimize the learning process. The algorithm dynamically adjusts the key stress parameter combination and the key region parameters through continuous trial and error and optimization to improve the adaptive ability and detection efficiency of the system. The improved DDPG algorithm dynamically adjusts the key stress parameter combination and the key region parameters through continuous trial and error and optimization, improves the adaptive ability of the system, and makes the system better adapt to the detection requirements under different working conditions. The application of the Bellman equation enables the algorithm to consider the importance of future rewards, thereby making more long-term decisions and improving the overall performance and stability of the system.

[0100] ​​When performing iterative optimization of temperature parameters, deviation comparison analysis needs to be performed, and the specific process is as follows: the temperature parameters in the measured data are taken as the heat flow boundary conditions of the multi-physical field coupling finite element model (for example, the temperature of the temperature measuring point is assigned to the corresponding grid node to replace the calculation of convective heat transfer in simulation); the deviation (for example, the root mean square error RMSE) between the measured temperature and the simulation temperature is compared, and if the deviation exceeds 5%, the thermal conductivity or convective heat transfer coefficient in the multi-physical field coupling finite element model is adjusted until the final deviation is less than 5% before performing iterative optimization of the parameters.

[0101] According to the optimized key stress parameter combination and the key region parameters of the multi-physical field coupling finite element model, final simulation data is obtained by simulation, and a failure criterion threshold is set according to the final simulation data; the set failure criterion threshold includes a warning threshold and a danger threshold. The key region temperature gradient map is input into a trend prediction model to perform trend prediction of the measured data and obtain prediction data, the prediction data is compared with the failure criterion threshold, and a multi-level response strategy is executed according to the comparison result. The multi-level response strategy includes a danger response strategy, a warning response strategy and a normal response strategy.

[0102] The specific process of comparing the prediction data with the failure criterion threshold and executing the multi-level response strategy according to the comparison result is as follows: the prediction data is compared with the danger threshold, and when the prediction data is greater than or equal to the danger threshold, the corresponding danger response strategy in the multi-level response strategy is triggered; when the prediction data is less than the danger threshold, the prediction data is compared with the warning threshold, and when the prediction data is greater than or equal to the warning threshold, the corresponding warning response strategy in the multi-level response strategy is triggered, and when the prediction data is less than the warning threshold, the corresponding normal response strategy in the multi-level response strategy is triggered, and the system is normal at this time.

[0103] The trend prediction is realized by using a sliding window time series analysis, and the state of the SiC power device is predicted by the following autoregressive moving average model:

[0104] ;

[0105] In the above formula, is the detection parameter of the time series at time t; is the coefficient of autoregression; is the historical observation value of the time series at time t-m; is the coefficient of moving average term; is the white noise error term; is the detection parameter of the white noise error term at time t-n; p and q are model orders, which are determined by minimizing the Akaike information criterion (AIC).

[0106] The sliding window time series analysis is a method for predicting future values of a time series by using an autoregressive moving average model (ARMA). In this method, the local characteristics of the time series are captured by a sliding window mechanism, and the autoregressive coefficients and moving average term coefficients are used to establish a prediction model. The white noise error term reflects the part of the variation that cannot be explained by the model, and the prediction performance of the model can be optimized by adjusting the model order (p, q). This analysis method can capture the trend of the SiC power device state in real time, providing strong support for subsequent timely warning and intervention, and helping to prevent the occurrence and development of faults. The application of the autoregressive moving average model improves the accuracy of state prediction, helps to discover potential faults in advance and take measures to avoid their occurrence, and improves the reliability and safety of the system.

[0107] The multi-level response strategy adopts a fuzzy logic control method for control, and the membership function used in the control process of the fuzzy logic control method is:

[0108] ;

[0109] In the formula, is the failure severity membership degree, the value range is ; x is the prediction data; a is the warning threshold; and b is the danger threshold.

[0110] The membership function is used to define the fuzzy set of failure severity, and the severity of the SiC power device failure is determined by comparing the prediction data with the warning threshold and the danger threshold. The membership function usually adopts a trapezoidal or triangular distribution form to reflect the influence of the prediction data on the failure severity in different intervals. The application of the membership function enables the system to more accurately evaluate the severity of the failure and make reasonable decisions accordingly, which helps to optimize the performance and reliability of the system. The fuzzy logic control method adopted in this embodiment can take corresponding response measures according to different levels of failure severity, realizing the refinement and intelligentization of fault handling, and improving the response speed and accuracy of the system.

[0111] Based on the same inventive concept as the above-mentioned SiC power device packaging reliability detection method, the embodiments of the present application also provide a SiC power device packaging reliability detection system, as shown in Figure 2 , which comprises:

[0112] A model construction module is configured to construct a multi-physical field coupled finite element model of the SiC power device.

[0113] An initial simulation and parameter analysis module is connected to the model construction module and is configured to run the multi-physical field coupled finite element model to obtain initial simulation data, and perform parameter sensitivity analysis on the initial simulation data to determine the key stress parameter combination of the typical failure mode.

[0114] The real data acquisition module comprises a working parameter acquisition unit, a data calibration and denoising unit, a dynamic region tracking unit and a temperature gradient atlas generation unit.

[0115] The working parameter acquisition unit is configured to acquire working parameters of the SiC power device in a running state.

[0116] The data calibration and denoising unit is connected with the working parameter acquisition unit and is configured to obtain real measurement data by real-time calibration and denoising of the working parameters.

[0117] The dynamic region tracking unit is connected with the working parameter acquisition unit and is configured to locate key regions of the SiC power device by a dynamic region tracking technology.

[0118] The temperature gradient atlas generation unit is connected with the data calibration and denoising unit and the dynamic region tracking unit and is configured to generate a key region temperature gradient atlas of the SiC power device according to the real measurement data.

[0119] The comparative analysis and iterative optimization module is connected with the initial simulation and parameter analysis module and the real data acquisition module and is configured to perform comparative analysis on the simulation data and the real measurement data by a deep reinforcement learning algorithm, to perform iterative optimization on the key stress parameter combination and the key region parameters of the multi-physical field coupled finite element model based on the comparative analysis result, and to obtain the optimized key stress parameter combination and the key region parameters of the multi-physical field coupled finite element model until the error between the initial simulation data and the real measurement data is within a set range.

[0120] The final simulation and failure criterion setting module is connected with the comparative analysis and iterative optimization module and is configured to perform simulation to obtain final simulation data according to the optimized key stress parameter combination and the key region parameters of the multi-physical field coupled finite element model, and to set a failure criterion threshold according to the final simulation data.

[0121] The trend prediction and response execution module comprises a trend prediction unit and a comparative response unit.

[0122] The trend prediction unit is connected with the temperature gradient atlas generation unit and is configured to input the key region temperature gradient atlas into a trend prediction model to perform trend prediction and obtain prediction data.

[0123] The comparative response unit is connected with the final simulation and failure criterion setting module and the trend prediction unit and is configured to compare the prediction data with the failure criterion threshold and to execute a multi-level response strategy according to the comparison result.

[0124] In the embodiments of the present application provided above, it should be understood that the disclosed system and method can also be implemented in other manners. The embodiments described above are merely exemplary. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architectural, functional and operational logic of systems, methods according to embodiments of the present application. In this regard, each block in the flowcharts or block diagrams can represent a module, a segment or a portion of code which comprises one or more executable instructions for implementing the specified logic function. It should also be noted that in some alternative implementations, the functions noted in the blocks can occur out of the order noted in the accompanying drawings. For example, two blocks shown in succession may, in fact, be executed substantially concurrently or in the reverse order, depending on the functionality involved. It should also be noted that each block in the block diagrams and / or flowcharts and combinations of blocks in the block diagrams and / or flowcharts can be implemented by dedicated hardware-based systems which perform the specified functions or acts or by combinations of dedicated hardware and computer instructions.

[0125] In addition, the various functional modules in the embodiments of the present application can be integrated together or exist separately, or two or more modules can be integrated to form an independent part.

[0126] If the functions are implemented in the form of software function modules and sold or used as an independent product, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application can be embodied in the form of a software product, and the computer software product is stored in a computer readable storage medium, and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods in the embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various other media which can store program codes.

[0127] The preferred embodiments of the present application have been described above with the aid of a number of exemplary drawings. The present application is not limited to the preferred embodiments described above. Various modifications and changes can be made by those skilled in the art without departing from the spirit and scope of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

[0128] It will be apparent to those skilled in the art that the application is not limited to the details of the above-exemplified embodiments and that the present application can be implemented in other particular forms without departing from the spirit or essential characteristics thereof. The embodiments should be considered in all respects as illustrative and not restrictive, the scope of the application being indicated by the appended claims rather than by the above description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. No reference herein to any prior art is to be taken as an admission that the application is not entitled to antedate such prior art by virtue of prior application. Any reference to the use of a term in the singular herein shall be understood in the context to describe a particular example or embodiment of the application and should not be construed as limiting the scope of the application to that particular example or embodiment. Any reference to use of terms in the plural herein shall be understood as describing a particular example or embodiment of the application and should not be construed as limiting the scope of the application to that particular example or embodiment.

Claims

1. A SiC power device package reliability detection method, characterized in that, The method comprises the following steps: constructing a multi-physics field coupling finite element model of the SiC power device; running the multi-physics field coupling finite element model to obtain initial simulation data, performing parameter sensitivity analysis on the initial simulation data to determine a key stress parameter combination of a typical failure mode; collecting working parameters of the SiC power device in a running state, obtaining measured data after real-time calibration and denoising of the working parameters; in the process of collecting the working parameters of the SiC power device in the running state, the key regions of the SiC power device are located through dynamic region tracking technology, and a key region temperature gradient map of the SiC power device is generated according to the measured data; comparing and analyzing the simulation data and the measured data through a deep reinforcement learning algorithm, iteratively optimizing the key stress parameter combination and the key region parameters of the multi-physics field coupling finite element model based on the comparison and analysis results until the error between the initial simulation data and the measured data is within a set range, thereby obtaining an optimized key stress parameter combination and key region parameters of the multi-physics field coupling finite element model; performing simulation according to the optimized key stress parameter combination and the key region parameters of the multi-physics field coupling finite element model to obtain final simulation data, and setting a failure criterion threshold according to the final simulation data; inputting the key region temperature gradient map into a trend prediction model to perform trend prediction and obtain prediction data, comparing the prediction data with the failure criterion threshold, and executing a multi-level response strategy according to the comparison result; the process of performing parameter sensitivity analysis on the initial simulation data to determine the key stress parameter combination of the typical failure mode comprises: designing a stress parameter sampling matrix through a Morris global sensitivity analysis method; performing finite element simulation on each group of stress parameter combinations, and outputting a failure index; calculating the normalized sensitivity coefficients of each stress parameter on the failure time to determine the key stress parameter combination of the typical failure mode.

2. The SiC power device package reliability detection method of claim 1, wherein: The process of constructing the multi-physics field coupling finite element model of the SiC power device comprises: constructing or importing a geometric model of the SiC power device, setting parameters of the SiC power device, setting boundary conditions, loads and typical failure modes of the model, and performing mesh division and solution setting to obtain the multi-physics field coupling finite element model.

3. The SiC power device package reliability detection method of claim 2, wherein: The setting of the parameters of the SiC power device comprises setting the thermal conductivity, thermal expansion coefficient, elastic modulus and Poisson's ratio of the packaging material of the SiC power device at different temperatures; the thermal conductivity of the packaging material at different temperatures is obtained by laser flash method.

4. The SiC power device package reliability detection method of claim 1, wherein: The typical failure modes include solder layer cracks, thermal resistance degradation and wire bonding line shedding; the failure of the solder layer cracks is determined by obtaining the strain concentration area in the thermal-mechanical stress simulation, the failure of the thermal resistance degradation is determined by obtaining the simulation thermal resistance parameter, and the failure of the wire bonding line shedding is determined by obtaining the stress analysis of the bonding area in the electro-thermal coupling simulation.

5. The SiC power device package reliability detection method of claim 1, wherein: The real-time calibration process comprises: using an environmental radiation compensation algorithm to perform real-time calibration on the temperature parameter in the working parameter.

6. The SiC power device package reliability detection method of claim 5, wherein: The process of real-time calibration of the temperature parameter in the working parameter by using the environmental radiation compensation algorithm is: measuring the radiation intensity under different wavelengths in the process of collecting the temperature parameter of the SiC power device, and then calculating and calibrating the measured temperature according to the blackbody radiation law.

7. The SiC power device package reliability detection method of claim 1, wherein: The process of denoising is: time alignment processing of the working parameter and the simulation data, and then eliminating the environmental noise in the working parameter by using the wavelet threshold denoising algorithm.

8. The SiC power device package reliability detection method of claim 1, wherein: The failure criterion threshold includes a pre-warning threshold and a danger threshold, the multi-level response strategy is controlled by using a fuzzy logic control method, and the membership function used in the control process of the fuzzy logic control method is: ； In the formula, is the failure severity degree membership, with a value range of 【0, 1】; x is the prediction data; a is the early warning threshold; and b is the danger threshold.

9. A SiC power device package reliability detection system, characterized by, The SiC power device packaging reliability detection method comprises: a model construction module, configured to construct a multi-physical field coupled finite element model of the SiC power device; an initial simulation and parameter analysis module, configured to run the multi-physical field coupled finite element model to obtain initial simulation data, and perform parameter sensitivity analysis on the initial simulation data to determine a key stress parameter combination of a typical failure mode; a real measurement data collection module, configured to collect working parameters of the SiC power device in a running state, and obtain real measurement data after real-time calibration and denoising of the working parameters; and locate a key region of the SiC power device by using a dynamic region tracking technology and generate a key region temperature gradient map; a comparative analysis and iterative optimization module, configured to perform comparative analysis on the simulation data and the real measurement data by using a deep reinforcement learning algorithm, and perform iterative optimization on the key stress parameter combination and key region parameters of the multi-physical field coupled finite element model based on a comparative analysis result, so as to obtain an optimized key stress parameter combination and key region parameters of the multi-physical field coupled finite element model; a final simulation and failure criterion setting module, configured to perform simulation according to the optimized key stress parameter combination and key region parameters of the multi-physical field coupled finite element model to obtain final simulation data, and set a failure criterion threshold according to the final simulation data; a trend prediction and response execution module, configured to input the key region temperature gradient map into a trend prediction model to perform trend prediction and obtain prediction data, compare the prediction data with the failure criterion threshold, and execute a multi-level response strategy according to a comparison result.

Images