Method and device for controlling packaging and sintering of power semiconductor device

Abstract

The invention provides a packaging and sintering control method and device for a power semiconductor device, and the method comprises the steps: achieving the monitoring of a dynamic stress field of a chip and a substrate interface through an embedded pressure sensing array, and obtaining the real-time stress distribution data. And in combination with a material characteristic database, multivariable collaborative optimization is performed on sintering process parameters, so that dynamic adjustment of the parameters is realized. The temperature and pressure curves of all the areas of the sintering platform are dynamically adjusted according to the optimized technological parameters by means of the micro induction heating element and the piezoelectric actuator, and refined self-adaptive multi-area sintering control is achieved. The sintering interface is subjected to real-time nondestructive testing, the interface quality index is obtained and serves as a feedback parameter to be input into the optimization process, closed-loop control is formed, and the accuracy of the sintering process is improved. The problems of uneven stress distribution and inconsistent bonding strength in the prior art are solved through a comprehensive real-time monitoring and dynamic regulation and control method, and the packaging quality and reliability of the power semiconductor device are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of encapsulation sintering, and particularly to a method and device for controlling the encapsulation sintering of power semiconductor devices. Background Art

[0002] In the encapsulation technology of power semiconductor devices, sintering is a key connection method, especially for high-power devices. With the development of power semiconductor devices towards high power density and high reliability, traditional sintering control methods have been difficult to meet the increasingly strict encapsulation requirements.

[0003] Currently, the sintering control methods commonly used in the industry mainly have a significant technical defect: the lack of real-time monitoring and dynamic regulation capabilities for the stress distribution during the sintering process. This results in the inability to timely adjust the sintering parameters to cope with local stress concentration problems, and further leads to quality problems such as uneven stress distribution and inconsistent interface bonding strength during the encapsulation of power semiconductor devices, ultimately affecting the reliability and service life of the devices. Summary of the Invention

[0004] The main purpose of the present invention is to solve the technical problem of the lack of real-time monitoring and dynamic regulation capabilities for stress distribution during the encapsulation sintering of existing power semiconductor devices; The first aspect of the present invention provides a method for controlling the encapsulation sintering of power semiconductor devices, and the method for controlling the encapsulation sintering of power semiconductor devices includes: Dynamically monitoring the stress field at the interface between the power semiconductor device chip and the substrate through a preset embedded pressure sensing array to obtain real-time stress distribution data; According to the real-time stress distribution data, combined with a pre-established material property database, multi-variable collaborative optimization is performed on the sintering process parameters to obtain optimized process parameters; Dynamically adjusting the temperature and pressure curves of each area of the sintering platform according to the process parameters through a preset micro induction heating element and piezoelectric actuator to achieve adaptive multi-area sintering; Performing real-time non-destructive detection on the sintering interface to obtain an interface quality index; and inputting the interface quality index as a feedback parameter into the multi-variable collaborative optimization to form a closed-loop control until the encapsulation sintering of the power semiconductor device is completed.

[0005] Optionally, in the first implementation manner of the first aspect of the present invention, the dynamically monitoring the stress field at the interface between the power semiconductor device chip and the substrate through a preset embedded pressure sensing array to obtain real-time stress distribution data includes: Arranging the pressure sensing array according to the structural characteristics of the interface between the power semiconductor device chip and the substrate to obtain a sensor layout scheme; Calibrate each sensor in the pressure sensing array under high-temperature environment to obtain a calibration coefficient matrix; Use the calibration coefficient matrix to correct the original interface data collected by the pressure sensing array to obtain corrected interface pressure data, and perform time-domain and spatial-domain filtering processing on the corrected interface pressure data to obtain filtered interface pressure data; According to the filtered interface pressure data and the sensor layout scheme through an interpolation algorithm, construct a continuous interface stress field distribution of the power semiconductor device and the substrate to obtain real-time stress distribution data.

[0006] Optionally, in the second implementation manner of the first aspect of the present invention, the multi-variable collaborative optimization of the sintering process parameters according to the real-time stress distribution data, combined with a pre-established material property database, to obtain optimized process parameters includes: According to the real-time stress distribution data, calculate the stress gradient of each region of the power semiconductor device chip to obtain a stress gradient matrix; Combined with the stress-strain relationship in the material property database, perform non-linear mapping on the stress gradient matrix to obtain a local pressure gradient adjustment factor; Take the current temperature data, pressure data, time data and local pressure gradient adjustment factor as input variables, construct a multi-dimensional optimization objective function, and use an improved particle swarm optimization algorithm to iteratively solve the multi-dimensional optimization objective function to obtain a preliminary optimization result; Input the preliminary optimization result into a pre-trained deep reinforcement learning model to obtain optimized process parameters.

[0007] Optionally, in the third implementation manner of the first aspect of the present invention, the combined stress-strain relationship in the material property database performs non-linear mapping on the stress gradient matrix to obtain a local pressure gradient adjustment factor, including: Perform singular value decomposition on the stress gradient matrix to obtain the principal stress direction and principal stress amplitude, and construct a non-linear mapping function according to the stress-strain curve in the material property database; Use the non-linear mapping function to transform the principal stress amplitude to obtain a local strain estimation value, and calculate the anisotropic pressure adjustment coefficient based on the local strain estimation value and the principal stress direction; Perform spatial smoothing processing on the anisotropic pressure adjustment coefficient to obtain a local pressure gradient adjustment factor.

[0008] Optionally, in the fourth implementation manner of the first aspect of the present invention, the dynamic adjustment of the temperature and pressure curves of each region of the sintering platform according to the process parameters by a preset micro induction heating element and a piezoelectric actuator to achieve adaptive multi-region sintering includes: According to the optimized process parameters, the sintering platform is meshed to obtain multiple independent control regions; For each independent control region, based on a preset silver particle sintering kinetic model, time series curves of temperature and pressure are generated; The time series curves are subjected to piecewise linearization processing to obtain a piecewise control instruction sequence, and the piecewise control instruction sequence is converted into a current modulation signal of a micro induction heating element and a voltage modulation signal of a piezoelectric actuator; According to the current modulation signal and the voltage modulation signal, the temperature and pressure of each independent control region are adjusted in real time to complete adaptive multi-region sintering.

[0009] Optionally, in the fifth implementation manner of the first aspect of the present invention, the generating time series curves of temperature and pressure for each independent control region based on a preset silver particle sintering kinetic model includes: According to the initial size distribution of silver particles required for sintering and the sintering temperature, a particle growth rate function in the silver particle sintering kinetic model is constructed; Using the random distribution function and the particle growth rate function in the silver particle sintering kinetic model, the initial distribution and contact state of silver particles are simulated to obtain an initial sintering structure; Based on the molecular dynamics principle, the initial sintering structure and the particle growth rate function in the silver particle sintering kinetic model, an interaction potential energy function between silver particles is constructed; Performing numerical integration on the interaction potential energy function, combining the initial sintering structure and the particle growth rate function, to obtain a neck growth curve predicted by the silver particle sintering kinetic model; According to the neck growth curve, the initial sintering structure and the densification sub-model in the silver particle sintering kinetic model, the volume shrinkage rate during the sintering process is calculated; Combining the volume shrinkage rate, the initial sintering structure, the particle growth rate function and the thermal expansion effect model in the silver particle sintering kinetic model, a coupling relationship equation of temperature and pressure is generated; Performing numerical solution on the coupling relationship equation to obtain time series curves of temperature and pressure.

[0010] Optionally, in the sixth implementation manner of the first aspect of the present invention, the real-time non-destructive detection of the sintering interface to obtain an interface quality index; and inputting the interface quality index as a feedback parameter into multi-variable collaborative optimization to form a closed-loop control until the sintering of the power semiconductor device package is completed includes: Using ultrasonic scanning technology to perform real-time scanning on the sintering interface to obtain an acoustic image, and using thermal reflection imaging technology to measure the thermal field distribution of the sintering interface to obtain a thermal reflection intensity map; Perform image fusion processing on the acoustic image and the thermal reflection intensity map to obtain a fused image, and use a pre-trained convolutional neural network to extract features and classify the fused image to obtain an interface defect distribution map; Calculate the interface quality index according to the interface defect distribution map, and add the interface quality index as a new input variable to the multivariable collaborative optimization to form a closed-loop control until the sintering of the power semiconductor device package is completed.

[0011] The second aspect of the present invention provides a device for controlling the sintering of a power semiconductor device, and the device for controlling the sintering of the power semiconductor device includes: A monitoring module for dynamically monitoring the stress field at the interface between the power semiconductor device chip and the substrate through a preset embedded pressure sensing array to obtain real-time stress distribution data; An optimization module for performing multivariable collaborative optimization on the sintering process parameters according to the real-time stress distribution data in combination with a pre-established material property database to obtain optimized process parameters; A regulation module for dynamically adjusting the temperature and pressure curves of each area of the sintering platform according to the process parameters through a preset micro induction heating element and a piezoelectric actuator to achieve adaptive multi-area sintering; An evaluation module for performing real-time non-destructive detection on the sintering interface to obtain an interface quality index; and inputting the interface quality index as a feedback parameter into the multivariable collaborative optimization to form a closed-loop control until the sintering of the power semiconductor device package is completed.

[0012] The above-mentioned method and device for controlling the sintering of a power semiconductor device realize dynamic stress field monitoring of the interface between the chip and the substrate through an embedded pressure sensing array to obtain real-time stress distribution data. In combination with the material property database, multivariable collaborative optimization is performed on the sintering process parameters to achieve dynamic adjustment of the parameters. Using a micro induction heating element and a piezoelectric actuator, the temperature and pressure curves of each area of the sintering platform are dynamically adjusted according to the optimized process parameters to achieve refined adaptive multi-area sintering control. By performing real-time non-destructive detection on the sintering interface, an interface quality index is obtained and input as a feedback parameter into the optimization process to form a closed-loop control, improving the accuracy of the sintering process. By comprehensively real-time monitoring and dynamic regulation methods, the problems of uneven stress distribution and inconsistent bonding strength in the prior art are solved, significantly improving the packaging quality and reliability of power semiconductor devices.

[0013] Other features and advantages of the present invention will be described in the following specification, and, in part, will be obvious from the specification, or will be understood by implementing the present invention. The objectives and other advantages of the present invention are achieved and obtained by the structures specifically pointed out in the specification, claims, and drawings.

[0014] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following provides preferred embodiments in conjunction with the accompanying drawings and describes them in detail as follows. Description of the Drawings

[0015] Figure 1 Schematic diagram of the first embodiment of the packaging sintering control method for a power semiconductor device in an embodiment of the present invention; Figure 2 Schematic diagram of an embodiment of the packaging sintering control device for a power semiconductor device in an embodiment of the present invention. Detailed Embodiments

[0016] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0017] The terms "including" and "having" and any variations thereof mentioned in the embodiments of the present invention are intended to cover non-exclusive inclusion. For example, a process, method, device, product, or equipment that includes a series of steps or units is not limited to the listed steps or units, but may optionally further include other unlisted steps or units, or may optionally further include other steps or units inherent to these processes, methods, products, or equipment.

[0018] For ease of understanding of this embodiment, first, a packaging sintering control method for a power semiconductor device disclosed in the embodiments of the present invention will be introduced in detail. As Figure 1 shown, this method includes the following steps: 101. Dynamically monitor the stress field at the interface between the power semiconductor device chip and the substrate through a preset embedded pressure sensing array to obtain real-time stress distribution data; In an embodiment of the present invention, the dynamic stress field monitoring of the interface between the power semiconductor device chip and the substrate by means of a preset embedded pressure sensing array to obtain real-time stress distribution data includes: arranging the pressure sensing array according to the structural characteristics of the interface between the power semiconductor device chip and the substrate to obtain a sensor layout scheme; calibrating each sensor in the pressure sensing array under a high-temperature environment to obtain a calibration coefficient matrix; using the calibration coefficient matrix to correct the original interface data collected by the pressure sensing array to obtain corrected interface pressure data, and performing time-domain and spatial-domain filtering processing on the corrected interface pressure data to obtain filtered interface pressure data; constructing a continuous stress field distribution of the power semiconductor device and the substrate interface according to the filtered interface pressure data and the sensor layout scheme through an interpolation algorithm to obtain real-time stress distribution data.

[0019] Specifically, during the packaging sintering process of power semiconductor devices, the realization of dynamic stress field monitoring requires the reasonable arrangement of embedded pressure sensors to obtain high-precision stress data. Since the packaged chips usually use large-size IGBTs or MOSFETs, the temperature and pressure gradient changes are obvious during the sintering process, and the stress distribution at the interface is extremely uneven. Therefore, the arrangement of sensors must consider the stress differences in different regions. The distribution of the stress field is affected by multiple factors, including chip size, silver paste flow characteristics, and thermal expansion effects during the sintering process. To optimize the sensor layout scheme, it is necessary to use finite element simulation to calculate the stress field distribution of the chip and the substrate, and construct an optimization objective function to make the sensor density higher in the regions where the stress changes violently and appropriately reduce it in the regions where the stress is relatively stable, so as to improve the data sampling efficiency and ensure the measurement accuracy. The optimization objective can be characterized by the Laplace operator of the stress gradient, that is, the optimal sensor layout should minimize the second derivative of the interface stress gradient, while considering the uniformity constraint of the sensor spacing. By solving the constrained optimization problem, the optimal sensor layout matrix can be obtained, and additional measurement points can be added at the chip edge and stress concentration regions to improve the data accuracy.

[0020] After determining the sensor layout, since the packaging sintering is usually carried out in a high-temperature environment of 200°C to 300°C, the measured values of the sensors will be affected by temperature drift. Therefore, local temperature calibration is required to eliminate the non-linear influence of temperature on the measured data. Since the piezoelectric response of the sensor shows non-linear drift with temperature change, simple linear calibration cannot provide sufficient accuracy. Therefore, a method of high-order polynomial fitting needs to be used for correction. The calibration process needs to construct a temperature-pressure non-linear compensation model. Assuming that the original output signal of the sensor is , the ambient temperature is , and the calibrated measured value is , its mathematical expression can be represented as: ; Among them, is the calibration coefficient obtained by fitting experimental data, and the exponential term is used to correct the exponential decay effect of thermal drift, term and term are used to compensate for the high-order nonlinear response of temperature, while the Fourier series term is used to correct the periodic drift of temperature to ensure that the calibration accuracy is high enough. The parameters of this model can be optimized by the least squares fitting or neural network regression method, so that the temperature compensation error is less than . After calibration, the output signals of all sensors can maintain high stability and high accuracy throughout the temperature range, thus ensuring the reliability of subsequent stress field calculations. After the data acquisition is completed, the measured signals usually contain high-frequency noise and discrete numerical errors. Therefore, it is necessary to perform filtering processing in the time domain and spatial domain to improve the signal quality and remove non-physical fluctuations. In terms of time domain filtering, the Adaptive Kalman Filter is used to dynamically adjust the noise covariance matrix, so as to minimize the measurement error and remove high-frequency noise without affecting the signal change trend. At the same time, the spatial domain filtering adopts the Bessel interpolation method to optimize the interpolation calculation between sensors, so that the final stress distribution can be reconstructed more smoothly. In the calculation process of Bessel interpolation, the goal is to construct a smooth pressure field function , and its mathematical expression is as follows: ; Among them, is the interpolation weight, is the Bessel surface coefficient, and are the Bessel basis functions respectively, satisfying the constraint condition , ensuring that the interpolation result satisfies continuity and smoothness. Through this method, the data of discrete measurement points can be reconstructed in high order in space, so that the finally obtained stress field distribution can not only accurately reflect the change trend of actual measurement data, but also provide relatively accurate estimated values in the unmeasured area. Finally, the stress distribution data after filtering and interpolation can be used for real-time stress field monitoring and provide accurate feedback for the optimization of sintering parameters.

[0021] 102. According to the real-time stress distribution data, combined with the pre-established material property database, perform multi-variable collaborative optimization on the sintering process parameters to obtain the optimized process parameters; In an embodiment of the present invention, based on the real-time stress distribution data and in combination with a pre-established material property database, multi-variable collaborative optimization is performed on the sintering process parameters. The optimized process parameters obtained include: calculating the stress gradient of each region of the power semiconductor device chip according to the real-time stress distribution data to obtain a stress gradient matrix; performing a non-linear mapping on the stress gradient matrix in combination with the stress-strain relationship in the material property database to obtain a local pressure gradient adjustment factor; using the current temperature data, pressure data, time data, and local pressure gradient adjustment factor as input variables to construct a multi-dimensional optimization objective function, and using an improved particle swarm optimization algorithm to iteratively solve the multi-dimensional optimization objective function to obtain a preliminary optimization result; and inputting the preliminary optimization result into a pre-trained deep reinforcement learning model to obtain the optimized process parameters.

[0022] Specifically, in the sintering process of power semiconductor device packaging, the first step of multi-variable collaborative optimization is to calculate the stress gradient of each region of the chip based on the real-time stress distribution data to obtain a stress gradient matrix. The real-time stress distribution data is provided by a pressure sensing array. After being processed by time-domain and spatial-domain filtering, a stress field distribution with high resolution is formed. The calculation of the stress gradient requires the use of a numerical difference method to perform spatial analysis on the stress change in the entire packaging area. Let the stress distribution function be , the stress gradient of the chip region can be represented by the gradient operator , that is: ; where respectively represent the change rates of stress in the x direction and the y direction. To avoid error accumulation caused by numerical differences, a high-order finite difference method is used for calculation. The calculation expression of the stress gradient matrix is: where is the grid step size, is located at The local stress gradient at [location], compared with the traditional second-order difference method, can reduce truncation errors and improve the accuracy of gradient calculation. The obtained stress gradient matrix reflects the stress variation of the encapsulation interface in different regions, and can identify high-gradient regions, that is, regions where insufficient bonding strength or microcrack initiation may occur, thus providing data support for subsequent optimization. After obtaining the stress gradient matrix, it is necessary to combine the stress-strain relationship in the material property database to perform a non-linear mapping on the stress gradient matrix to calculate the local pressure gradient adjustment factor. The material property database stores the stress-strain curves of the silver sintered layer under different temperatures and different sintering pressures. The data is measured through experiments and stored in a parameterized manner for quick call during the optimization process. The stress-strain relationship usually exhibits non-linear characteristics, so the stress gradient matrix needs to be transformed through a non-linear mapping function to obtain the local pressure adjustment factor . The form of the non-linear mapping function is: ; where, is the yield stress of the silver sintered layer, is a high-order polynomial or neural network model fitted based on experimental data. This function describes how the stress gradient affects the sintering bonding strength. Through this non-linear mapping, the local pressure adjustment factor of each region can be obtained to ensure that a more appropriate pressure is applied in the high stress gradient region to reduce the stress concentration phenomenon and improve the overall reliability of the encapsulation.

[0023] After calculating the local pressure adjustment factor. The next step is to construct a multi-dimensional optimization objective function, taking the current temperature data, pressure data, time data and local pressure gradient adjustment factor as input variables, and using an improved particle swarm optimization algorithm to solve. The optimization objectives include minimizing the residual stress at the sintering interface, maximizing the bonding strength, and controlling the local temperature gradient. The optimization objective function can be expressed as: ; where, and are the optimization variables, namely the sintering temperature field and pressure field, is the residual stress, is the bonding strength, is the local temperature gradient, is the weight factor used to balance the influence between various optimization objectives. The optimization process uses an improved particle swarm optimization (PSO) algorithm, introducing an adaptive inertia weight and a local convergence enhancement factor, so that the particle swarm can converge to the optimal solution faster during the multi-objective optimization process. The particle update formula is: ; where, is the velocity of the particle at each step, which is dynamically adjusted according to the optimization progress to maintain a large exploration ability in the early stage of the search and enhance the local convergence ability in the later stage of the search, is the historical optimal position of the particle itself, is the global optimal position of the population, is the learning factor, is a random number. This improved method can improve the search efficiency of the optimization algorithm on complex multi-objective functions and ensure convergence to the optimal solution.

[0024] Furthermore, the nonlinear mapping of the stress gradient matrix by combining the stress-strain relationship in the material property database to obtain the local pressure gradient adjustment factor includes: performing singular value decomposition on the stress gradient matrix to obtain the principal stress direction and principal stress amplitude, and constructing a nonlinear mapping function according to the stress-strain curve in the material property database; using the nonlinear mapping function to transform the principal stress amplitude to obtain the local strain estimate value, and calculating the anisotropic pressure adjustment coefficient based on the local strain estimate value and the principal stress direction; performing spatial smoothing processing on the anisotropic pressure adjustment coefficient to obtain the local pressure gradient adjustment factor.

[0025] Specifically, in the process of performing nonlinear mapping by combining the stress-strain relationship in the material property database, it is first necessary to perform singular value decomposition on the stress gradient matrix to extract the principal stress direction and principal stress amplitude. The stress gradient matrix is obtained by spatial differentiation of the stress field data measured by the sensor. It contains the stress change rate in each region of the interface and shows anisotropic distribution characteristics. Due to the influence of material properties and thermal stress during the sintering process on the encapsulation interface, there may be significant directional differences in the stress gradients in different regions. Therefore, directly using the scalar gradient value cannot accurately describe the local stress change situation. To obtain the principal direction information of the stress, it is first necessary to perform singular value decomposition on the local stress gradient matrix. The decomposed matrix contains a set of orthogonal bases, where the eigenvector corresponding to the largest singular value is the principal stress direction, and the corresponding singular value size represents the magnitude of the principal stress. Through this method, the main stress action direction at each position of the interface can be clearly identified, and the stress concentration area can be determined for subsequent nonlinear mapping calculations.

[0026] After obtaining the principal stress direction and principal stress amplitude, it is necessary to construct a non-linear mapping function in combination with the stress-strain relationship in the material property database to describe the deformation behavior of the material under different stress levels. The silver sintered layer in the power semiconductor package exhibits significant non-linear stress-strain characteristics during the sintering process. Especially under high stress conditions, its strain response does not conform to the linear elastic model. Therefore, it is necessary to use the experimentally measured stress-strain data for parametric modeling. The database stores the stress-strain curve data under different temperature conditions and is described by high-order polynomial fitting or neural network models, so that the mapping relationship between stress and strain can be obtained through numerical calculation. For the principal stress amplitude of each measurement point, it is necessary to query the material curve at the corresponding temperature in the database to calculate the local strain estimate at this stress level. Due to the strong plastic deformation characteristics of the material, the mapping function usually contains high-order non-linear terms to ensure the accuracy of strain estimation. In addition, due to the strain hardening effect during the sintering process of the material, the calculation of the stress-strain relationship needs to consider not only the current principal stress magnitude but also the stress changes in the surrounding area to obtain a more reasonable strain distribution.

[0027] After calculating the local strain estimate, it is necessary to calculate the anisotropic pressure adjustment coefficient in combination with the principal stress direction. The stress anisotropy at the package interface is mainly caused by the different mechanical properties and thermal expansion mismatches of the materials, and the pressure applied during the sintering process needs to be dynamically adjusted according to these strain distribution characteristics. The local pressure adjustment should follow the directional change law of the strain so that the applied external pressure can effectively offset the stress concentration phenomenon. The calculation of the pressure adjustment coefficient needs to consider not only the magnitude of the local strain but also the distribution information of the stress direction to ensure that the optimized pressure can act uniformly on the entire interface. For each measurement point, by calculating the angle between the principal stress direction and the local strain direction, the component of the strain along the principal stress direction is determined, and the magnitude of the externally applied pressure is adjusted accordingly. Since the stress states in different regions during the packaging process are different, more precise pressure adjustment is often required in high stress gradient regions, while the adjustment amplitude can be appropriately reduced in low stress regions to ensure overall stress uniformity.

[0028] After obtaining the anisotropic pressure regulation coefficient, it is necessary to perform spatial smoothing on it to reduce the volatility of local data and ensure the continuity of pressure regulation. Since the pressure regulation coefficient is calculated based on discrete measurement points, its spatial distribution may have numerical oscillations. This instability may lead to over-adjustment or under-adjustment of the pressure in local areas during the sintering process, thereby affecting the overall quality of the packaging interface. Therefore, a high-order bidirectional filtering method needs to be adopted to smooth the regulation coefficient so that its distribution in the entire packaging area is more uniform. The filtering process uses a spatial smoothing algorithm based on weighted averaging. By calculating the weighted average of the neighborhood around each point, the data is made smoother in space while avoiding interference from high-frequency noise. The finally obtained local pressure gradient regulation factor is used to guide the pressure adjustment in each area during the sintering process to ensure a uniform stress distribution at the packaging interface throughout the sintering cycle, thereby improving the long-term stability and reliability of the power semiconductor packaging.

[0029] 103. Dynamically adjust the temperature and pressure curves of each area of the sintering platform according to process parameters through preset micro induction heating elements and piezoelectric actuators to achieve adaptive multi-area sintering; In an embodiment of the present invention, the dynamically adjusting the temperature and pressure curves of each area of the sintering platform according to the process parameters through preset micro induction heating elements and piezoelectric actuators to achieve adaptive multi-area sintering includes: dividing the sintering platform into grids according to the optimized process parameters to obtain multiple independent control areas; for each independent control area, generating time series curves of temperature and pressure based on a preset silver particle sintering kinetic model; performing piecewise linearization on the time series curves to obtain a piecewise control instruction sequence, and converting the piecewise control instruction sequence into a current modulation signal of the micro induction heating element and a voltage modulation signal of the piezoelectric actuator; and according to the current modulation signal and voltage modulation signal, adjusting the temperature and pressure of each independent control area in real time to complete adaptive multi-area sintering.

[0030] Specifically, in the adaptive multi-region sintering process, it is first necessary to divide the sintering platform into grids according to the optimized process parameters to determine multiple independent control regions. Since the sintering process of power semiconductor device packaging involves a complex thermodynamic process of large-size chips and multi-layer materials, the temperature and pressure distributions in each region need to be precisely controlled based on real-time stress data and material properties. The grid division method needs to combine the finite element simulation results so that each independent control region can cover a region with a relatively uniform stress gradient and minimize the thermal interference and stress coupling between different regions. In the actual implementation process, the grid division algorithm can be based on the Voronoi diagram or the adaptive quadtree division method to ensure the control accuracy of high-stress change regions while reducing the control complexity of low-stress change regions. Let the control domain of the entire sintering platform be , where each grid region satisfies: ; where, is the stress gradient, n is the normal vector of the region boundary, is the differential area, and the grid division needs to ensure that the stress gradient within each is as small as possible to ensure the stress uniformity within the region and improve the stability of the sintering process. After completing the grid division, for each independent control region, based on the preset silver particle sintering kinetic model, time series curves of temperature and pressure need to be generated. The sintering kinetic process of silver particles involves multiple stages such as particle rearrangement, neck growth, densification, and grain growth. The kinetic characteristics of each stage are different, so the time curves of temperature and pressure need to precisely match these kinetic changes. The core driving force for silver particle sintering is the reduction of surface energy. The classical formula describing sintering kinetics is: ; where, is the neck radius between particles, is the surface energy, is the atomic volume, is the Boltzmann constant, is the sintering temperature, is the initial particle radius. This equation describes the relationship between the neck growth rate and temperature during the particle sintering process, indicating that the higher the temperature, the faster the sintering rate between particles. However, if the temperature is too high, the material may undergo over-sintering, resulting in abnormal grain growth and the formation of pores. Therefore, the temperature and pressure curves need to be dynamically adjusted according to the particle size distribution and stress state in different regions to ensure an appropriate densification degree of silver particles, thereby improving the bonding strength. To make the generated time series curve convenient for the control system to execute, it is necessary to perform piecewise linearization on the time series to obtain a piecewise control instruction sequence, and convert the instruction sequence into a current modulation signal for the micro induction heating element and a voltage modulation signal for the piezoelectric actuator. Since the changes in temperature and pressure during the sintering process are usually non-linear, directly using the original time series curve for control may cause the actuator to respond lag or over-adjust. Therefore, a method of piecewise linear approximation is required to decompose the non-linear change curve into several linear intervals, and the control parameters within each interval can be approximated as constant temperature and pressure increments. Let the original control curve of temperature be , after piecewise linearization, the control instruction of temperature can be expressed as: ; where is the time step, is the temperature change amount within each time step. Similarly, the control signal of pressure can also be piecewise linearized using the same method, and the piecewise linear control signal is converted into a current modulation signal for the micro induction heating element and a voltage modulation signal for the piezoelectric actuator through digital signal processing (DSP), enabling the control system to adjust the temperature and pressure in real time.

[0031] Finally, according to the generated current modulation signal and voltage modulation signal, the temperature and pressure in each independent control region are adjusted in real time to complete adaptive multi-region sintering. The temperature control of the micro induction heating element adopts the method of high-frequency electromagnetic induction heating, enabling the temperature to respond within milliseconds, while the pressure control of the piezoelectric actuator achieves high-precision pressure adjustment through the inverse piezoelectric effect. During the temperature adjustment process, the real-time feedback system will monitor the actual temperature distribution in each region and adjust the induction heating power through closed-loop control to compensate for the temperature drift caused by heat diffusion or environmental factors. Similarly, the pressure adjustment system will adjust the driving voltage of the piezoelectric actuator based on the real-time stress monitoring data to keep the pressure in each region at the target value obtained by optimized calculation. Finally, during the entire sintering process, the temperature and pressure curves in each region can achieve dynamic adaptive adjustment, thereby ensuring a uniform stress distribution throughout the power semiconductor package, improving the interface bonding quality, and optimizing the long-term reliability of the device.

[0032] Further, for each independent control region, generating the time series curves of temperature and pressure based on the preset silver particle sintering kinetic model includes: constructing the particle growth rate function in the silver particle sintering kinetic model according to the initial size distribution of silver particles required for sintering and the sintering temperature; using the random distribution function and the particle growth rate function in the silver particle sintering kinetic model to simulate the initial distribution and contact state of silver particles, and obtaining the initial sintering structure; constructing the interaction potential energy function between silver particles based on the molecular dynamics principle, the initial sintering structure and the particle growth rate function in the silver particle sintering kinetic model; performing numerical integration on the interaction potential energy function, combining the initial sintering structure and the particle growth rate function, and obtaining the neck growth curve predicted by the silver particle sintering kinetic model; calculating the volume shrinkage rate during the sintering process according to the neck growth curve, the initial sintering structure and the densification sub-model in the silver particle sintering kinetic model; combining the volume shrinkage rate, the initial sintering structure, the particle growth rate function and the thermal expansion effect model in the silver particle sintering kinetic model to generate the coupling relationship equation of temperature and pressure; and performing numerical solution on the coupling relationship equation to obtain the time series curves of temperature and pressure.

[0033] Specifically, when generating the time series curves of temperature and pressure for each independent control region, it is first necessary to construct the particle growth rate function in the silver particle sintering kinetic model according to the initial size distribution of silver particles required for sintering and the sintering temperature. The initial size distribution of silver particles directly affects the neck growth and densification rates during sintering. Therefore, it is necessary to determine its particle size distribution function through experimental measurement or statistical modeling. The particle growth rate is controlled by the diffusion mechanism. At higher temperatures, surface diffusion and volume diffusion dominate the sintering kinetics, while at lower temperatures, interface diffusion contributes more. The calculation of the growth rate needs to consider factors such as particle size, surface energy, sintering driving force and environmental temperature to determine the neck growth rate and densification rate of particles under different temperature and pressure conditions, so that the model can accurately describe the evolution law of silver particles during sintering.

[0034] After obtaining the particle growth rate function, it is necessary to use the random distribution function and the particle growth rate function in the sintering kinetics model to simulate the initial distribution and contact state of silver particles in order to obtain a reasonable initial sintering structure. The arrangement state of silver particles affects the contact area between particles and the formation of sintering necks. Therefore, the Monte Carlo method or the random sphere packing algorithm needs to be used to simulate the initial arrangement of particles on the substrate to make it conform to the packing density and contact rate observed in experiments. During the generation of the random distribution, the influence of the sintering temperature on the particle arrangement needs to be introduced, and the particle displacement caused by thermal expansion and surface diffusion needs to be considered to make the generated initial sintering structure more in line with the actual situation. At the same time, the contact points between particles need to be adjusted according to the particle size distribution and the pressure acting conditions to ensure that the model can accurately predict the subsequent neck growth and densification processes.

[0035] After determining the initial sintering structure, it is necessary to construct the interaction potential energy function between silver particles based on the principles of molecular dynamics, the initial sintering structure, and the particle growth rate function. Silver particles are subject to van der Waals forces, capillary forces, and diffusion-driven forces during the sintering process, which gradually enhance the binding force between particles. Therefore, these factors need to be considered in the interaction potential energy to accurately describe the binding state between particles. The potential energy function can adopt the Lennard-Jones potential function or the modified Morse potential function to characterize the interaction strength between particles, and combined with the influence of the sintering temperature and time, calculate the change in the relative position of particles during the sintering process. When constructing the potential energy function, the influence of particle morphology changes on the potential energy also needs to be considered to ensure that the potential energy calculation can reasonably reflect the movement trend of particles and the sintering kinetics behavior.

[0036] After obtaining the interaction potential energy function, it is necessary to perform numerical integration on it. Combining the initial sintering structure and the particle growth rate function, the neck growth curve predicted by the silver particle sintering kinetics model is obtained. The process of numerical integration needs to use molecular dynamics simulation or the finite element method to minimize the potential energy of the particle system to calculate the relative movement trajectory between particles, and calculate the neck growth rate based on the diffusion control equation. The calculation of the neck growth curve needs to consider the influence of different diffusion mechanisms. At lower temperatures, interface diffusion dominates neck growth, while at higher temperatures, volume diffusion contributes more. Therefore, the diffusion coefficient needs to be dynamically adjusted during the calculation to ensure that the simulation results conform to the neck growth data measured in experiments. In addition, the neck growth curve also needs to be corrected in combination with the action of pressure to ensure that the model can accurately describe the particle binding situation under different pressure conditions.

[0037] After obtaining the neck growth curve, it is necessary to calculate the volume shrinkage rate during the sintering process based on the initial sintered structure, the particle growth rate function, and the densification sub-model in the sintering kinetics model. The calculation of the volume shrinkage rate needs to consider the process of particle transformation from a loose packing state to a dense structure during sintering. Usually, empirical models or numerical simulation methods based on particle arrangement are used to calculate the volume change. Since there are the closure of sintering pores and grain rearrangement during the densification process of silver particles, the calculation of the volume shrinkage rate not only needs to consider the relative movement between particles but also needs to combine the pore shrinkage model to accurately predict the change trend of the sintered volume. At the same time, the shrinkage rate calculation also needs to consider the influence of temperature and pressure on sintering kinetics. At higher temperatures and greater pressures, the densification rate between particles accelerates. Therefore, the shrinkage rate calculation needs to combine experimental data for parameter correction to ensure the accuracy of the calculation results.

[0038] After calculating the volume shrinkage rate, it is necessary to combine the initial sintered structure, the particle growth rate function, and the thermal expansion effect model in the sintering kinetics model to generate a coupled relationship equation for temperature and pressure. The influence of temperature and pressure on the sintering process is mutually coupled. Higher temperatures accelerate the diffusion process and increase the neck growth rate, while greater pressures increase the contact area between particles and improve the densification efficiency. Therefore, the coupled equation for temperature and pressure needs to consider these factors simultaneously to optimize the sintering process. The coupled relationship equation usually combines the diffusion control equation and the rheological model to describe the relationship between temperature, pressure, and the densification of silver particles, and parameter fitting is performed through experimental data to improve the prediction ability of the equation.

[0039] After obtaining the coupled relationship equation, it is necessary to perform numerical solution on it to obtain the time series curves of temperature and pressure. Numerical solution usually adopts the finite difference method or the finite element method to ensure the stability and accuracy of the calculation results. During the numerical calculation process, it is necessary to combine the sintering kinetics model to dynamically adjust the calculation step size to adapt to the sintering rate changes under different temperature and pressure conditions. After the numerical solution is completed, the finally obtained time series curves of temperature and pressure can be used in the sintering control system to guide the temperature and pressure regulation during the sintering process, thereby achieving precise process optimization and improving the quality and long-term stability of the power semiconductor package.

[0040] 104. Perform real-time non-destructive detection on the sintering interface to obtain the interface quality index, and input the interface quality index as a feedback parameter into the multi-variable collaborative optimization to form a closed-loop control until the sintering of the power semiconductor device package is completed.

[0041] In one embodiment of the present invention, the real-time non-destructive detection of the sintering interface is carried out to obtain an interface quality index; and the interface quality index is input as a feedback parameter into the multi-variable collaborative optimization to form a closed-loop control until the sintering of the power semiconductor device package is completed, including: using ultrasonic scanning technology to perform real-time scanning on the sintering interface to obtain an acoustic image, and using thermal reflection imaging technology to measure the thermal field distribution of the sintering interface to obtain a thermal reflection intensity map; performing image fusion processing on the acoustic image and the thermal reflection intensity map to obtain a fused image, and using a pre-trained convolutional neural network to extract features and classify the fused image to obtain an interface defect distribution map; calculating the interface quality index according to the interface defect distribution map, and adding the interface quality index as a new input variable to the multi-variable collaborative optimization to form a closed-loop control until the sintering of the power semiconductor device package is completed.

[0042] Specifically, in the process of real-time non-destructive detection of the sintering interface, it is first necessary to use ultrasonic scanning technology to perform real-time scanning on the sintering interface to obtain an acoustic image. Ultrasonic scanning relies on the propagation characteristics of high-frequency ultrasonic waves on the material interface. The change in acoustic impedance in different regions will cause the reflection, transmission, or attenuation of ultrasonic waves. Therefore, by receiving the reflected wave signal and performing signal processing, an acoustic characteristic distribution map of the interface can be constructed. During the sintering process, defects such as uneven interface bonding, voids, and microcracks will change the propagation path of the sound wave, resulting in obvious differences in the echo signals of different regions. To improve the imaging accuracy, a high-frequency focused ultrasonic transducer needs to be used, and the angle and focus of the ultrasonic beam are adjusted through phased array technology to enhance the resolution ability of interface defects. After the signal is received, through time-domain denoising, envelope detection, and high-resolution reconstruction, an acoustic image is finally formed, which can clearly reflect the interface bonding quality and can be used for further analysis of the spatial distribution characteristics of interface defects.

[0043] While obtaining the acoustic image, it is also necessary to use thermal reflection imaging technology to measure the thermal field distribution of the sintering interface to obtain a thermal reflection intensity map. Thermal reflection imaging is based on the difference in thermal conductivity of different material regions, combined with the dynamic change of temperature distribution during the sintering process. By capturing the thermal radiation signal of the interface with a high-precision infrared camera and using the fast sampling ability of the thermal imager, a thermal reflection intensity distribution map of different regions of the interface is constructed. Since the difference in interface bonding quality during the sintering process will affect the local thermal conduction characteristics, thermal reflection imaging can effectively detect regions with poor bonding, such as high thermal resistance regions caused by insufficient sintering or local thermal short-circuit regions caused by over-sintering. After data acquisition, noise is removed through Fourier transform, and the image contrast is enhanced using adaptive histogram equalization to improve the clarity of the thermal reflection map and make the difference in interface bonding states more significant.

[0044] After obtaining the acoustic image and the thermal reflection intensity map, it is necessary to fuse these two types of image data to comprehensively utilize the acoustic and thermal information and improve the detection accuracy of the interface quality. Since the acoustic image mainly reflects the density distribution of the material, while the thermal reflection image provides information on the thermal conduction characteristics, these two types of data are complementary. During the image fusion process, spatial registration needs to be performed first to ensure the corresponding relationship of different modality data at the pixel level, and then the multi-scale wavelet transform method is used to decompose the image into low-frequency components and high-frequency detail components, and different components are processed separately during the fusion process. The low-frequency components are fused using the mean value to retain the overall structural information, while the high-frequency components are weighted using the Laplacian operator to highlight the characteristics of the interface defect area. The fused image not only contains the high-resolution density information of the acoustic image but also combines the temperature field distribution characteristics of the thermal reflection image, making the comprehensive evaluation of the interface quality more accurate.

[0045] After obtaining the fused image, it is necessary to use a pre-trained convolutional neural network (CNN) to extract features and classify the image to generate an interface defect distribution map. The convolutional neural network can automatically learn the features in the image and classify different types of interface defects through multiple layers of convolution, pooling, and fully connected operations. During the feature extraction stage, the convolutional layer extracts the edges, textures, and high-order features of the local area, and the pooling layer reduces the data dimension and enhances the invariance. During the classification stage, the Softmax layer is used to estimate the probabilities of the defect categories. During the training process, the network performs supervised learning based on a large number of sintering interface samples to optimize the network weights so that it can accurately identify different types of interface defects, such as microcracks, insufficient sintering, and over-sintering. The classified interface defect distribution map is visually displayed through pixel-level annotation to show the bonding quality of different regions of the interface and can be used to further calculate the interface quality index.

[0046] Based on the interface defect distribution map, it is necessary to calculate the interface quality index to quantitatively evaluate the overall quality of the sintering interface. The calculation of the interface quality index needs to combine the statistical characteristics of the defect distribution, including the area, location, density, and type weights of the defects. By calculating the proportion of defect pixels in the interface area and combining the weight factors of the defect categories, a comprehensive quality scoring function can be constructed to evaluate the bonding state of the entire packaging interface. The calculation of the interface quality index not only needs to consider the absolute number of defects but also needs to combine the impact of the defects on the overall structural reliability. For example, small defects in the edge area may have a greater impact on the thermal expansion adaptability, while poor bonding in the central area may have a greater impact on the electrical performance. Therefore, when calculating the interface quality index, it is necessary to adaptively adjust the weights of different regions to ensure the rationality of the quality evaluation results.

[0047] After obtaining the interface quality index, it is necessary to add it as a new input variable to the multi-variable collaborative optimization to form a closed-loop control, so as to adjust the sintering parameters in real time and optimize the encapsulation process. The core idea of the closed-loop control is to dynamically adjust the process parameters such as sintering temperature, pressure and time based on the change trend of the interface quality index, so as to compensate for the non-uniformity in the sintering process and improve the interface bonding quality. During the optimization process, the interface quality index is used to update the optimization objective function of the process parameters, and combined with the reinforcement learning algorithm, the optimization strategy is continuously adjusted through the trial-and-error mechanism, so that the final sintering result meets the expected reliability requirements. During the optimization iteration process, the system will automatically adjust the gradient distribution of the sintering temperature, the spatial distribution of the pressure and the sintering time to reduce the interface defects and improve the long-term stability of the overall encapsulation. Finally, with the continuous operation of the closed-loop optimization, the interface quality index gradually converges to a stable range until the encapsulation sintering process of the power semiconductor device is completed and reaches the quality standard set by the optimization target.

[0048] In this embodiment, the dynamic stress field of the chip and substrate interface is monitored through an embedded pressure sensing array to obtain real-time stress distribution data. Combining with the material property database, multi-variable collaborative optimization of the sintering process parameters is carried out to realize the dynamic adjustment of the parameters. Using a micro induction heating element and a piezoelectric actuator, the temperature and pressure curves of each area of the sintering platform are dynamically adjusted according to the optimized process parameters, realizing refined adaptive multi-area sintering control. By performing real-time non-destructive detection on the sintering interface, the interface quality index is obtained and input as a feedback parameter into the optimization process to form a closed-loop control, improving the accuracy of the sintering process. By comprehensively real-time monitoring and dynamic regulation methods, the problems of uneven stress distribution and inconsistent bonding strength in the prior art are solved, significantly improving the encapsulation quality and reliability of power semiconductor devices.

[0049] The above describes the encapsulation sintering control method of the power semiconductor device in the embodiment of the present invention. Next, the encapsulation sintering control device of the power semiconductor device in the embodiment of the present invention will be described. Please refer to Figure 2 , an embodiment of the encapsulation sintering control device of the power semiconductor device in the embodiment of the present invention includes: A monitoring module 201, configured to monitor the dynamic stress field of the chip and substrate interface of the power semiconductor device through a preset embedded pressure sensing array to obtain real-time stress distribution data; An optimization module 202, configured to perform multi-variable collaborative optimization on the sintering process parameters according to the real-time stress distribution data in combination with a pre-established material property database to obtain optimized process parameters; A regulation module 203, configured to dynamically adjust the temperature and pressure curves of each area of the sintering platform according to the process parameters through a preset micro induction heating element and a piezoelectric actuator to realize adaptive multi-area sintering; An evaluation module 204 is configured to perform real-time non-destructive detection on the sintering interface to obtain an interface quality index; and input the interface quality index as a feedback parameter into multi-variable collaborative optimization to form a closed-loop control until the sintering of the power semiconductor device package is completed.

[0050] In an embodiment of the present invention, the power semiconductor device package sintering control device runs the above-mentioned power semiconductor device package sintering control method. The power semiconductor device package sintering control device realizes the monitoring of the dynamic stress field at the interface between the chip and the substrate through an embedded pressure sensing array, and obtains real-time stress distribution data. Combining with the material property database, multi-variable collaborative optimization of the sintering process parameters is carried out to realize the dynamic adjustment of the parameters. By using a micro induction heating element and a piezoelectric actuator, the temperature and pressure curves of each area of the sintering platform are dynamically adjusted according to the optimized process parameters, realizing refined adaptive multi-area sintering control. By performing real-time non-destructive detection on the sintering interface, an interface quality index is obtained and input as a feedback parameter into the optimization process to form a closed-loop control, improving the accuracy of the sintering process. By means of comprehensive real-time monitoring and dynamic regulation methods, the problems of uneven stress distribution and inconsistent bonding strength in the prior art are solved, significantly improving the packaging quality and reliability of power semiconductor devices.

[0051] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described systems, devices, or units can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0052] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The foregoing storage medium includes: various media such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disc that can store program codes.

[0053] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for controlling the packaging and sintering of a power semiconductor device, characterized in that: The packaging and sintering control method of the power semiconductor device comprises: The dynamic stress field of the interface between the power semiconductor device chip and the substrate is monitored through a preset embedded pressure sensor array to obtain real-time stress distribution data; According to the real-time stress distribution data, combined with a pre-established material property database, multivariable collaborative optimization is performed on the sintering process parameters to obtain optimized process parameters; Dynamically adjust the temperature and pressure curves of each area of ​​the sintering platform according to the process parameters through preset micro-induction heating elements and piezoelectric actuators to achieve adaptive multi-area sintering; The sintering interface is subjected to real-time nondestructive testing to obtain the interface quality index, which is then input as a feedback parameter into the multivariable collaborative optimization to form a closed-loop control until the sintering of the power semiconductor device package is completed.

2. The method for controlling the packaging and sintering of a power semiconductor device according to claim 1, characterized in that: The dynamic stress field monitoring of the interface between the power semiconductor device chip and the substrate by the preset embedded pressure sensor array to obtain real-time stress distribution data includes: According to the structural characteristics of the interface between the power semiconductor device chip and the substrate, the pressure sensor array is arranged to obtain a sensor layout scheme; Calibrate each sensor in the pressure sensing array in a high temperature environment to obtain a calibration coefficient matrix; Using the calibration coefficient matrix to correct the original interface data collected by the pressure sensor array to obtain corrected interface pressure data, and performing time domain and space domain filtering processing on the corrected interface pressure data to obtain filtered interface pressure data; By using an interpolation algorithm, a continuous stress field distribution of the interface between the power semiconductor device and the substrate is constructed according to the filtered interface pressure data and the sensor layout scheme, and real-time stress distribution data is obtained.

3. The method for controlling the packaging and sintering of a power semiconductor device according to claim 1, characterized in that: The multivariable collaborative optimization of the sintering process parameters is performed based on the real-time stress distribution data and combined with a pre-established material property database, and the optimized process parameters include: Calculating the stress gradients of various regions of the power semiconductor device chip according to the real-time stress distribution data to obtain a stress gradient matrix; Combined with the stress-strain relationship in the material property database, the stress gradient matrix is ​​nonlinearly mapped to obtain the local pressure gradient adjustment factor; The current temperature data, pressure data, time data and local pressure gradient adjustment factor are used as input variables to construct a multidimensional optimization objective function, and the improved particle swarm optimization algorithm is used to iteratively solve the multidimensional optimization objective function to obtain a preliminary optimization result; The preliminary optimization results are input into the pre-trained deep reinforcement learning model to obtain the optimized process parameters.

4. The method for controlling the packaging and sintering of a power semiconductor device according to claim 3, characterized in that: The stress gradient matrix is ​​nonlinearly mapped in combination with the stress-strain relationship in the material property database to obtain the local pressure gradient adjustment factor, which includes: Performing singular value decomposition on the stress gradient matrix to obtain the principal stress direction and principal stress amplitude, and constructing a nonlinear mapping function according to the stress-strain curve in the material property database; The principal stress amplitude is transformed by using the nonlinear mapping function to obtain a local strain estimation value, and an anisotropic pressure adjustment coefficient is calculated based on the local strain estimation value and the principal stress direction; The anisotropic pressure adjustment coefficient is spatially smoothed to obtain a local pressure gradient adjustment factor.

5. The method for controlling the packaging and sintering of a power semiconductor device according to claim 1, characterized in that: The method of dynamically adjusting the temperature and pressure curves of each region of the sintering platform according to the process parameters by using the preset micro induction heating element and the piezoelectric actuator to achieve adaptive multi-region sintering includes: According to the optimized process parameters, the sintering platform is meshed to obtain multiple independent control areas; For each independent control area, a time series curve of temperature and pressure is generated based on a preset silver particle sintering kinetic model; Performing piecewise linearization processing on the time series curve to obtain a piecewise control instruction sequence, and converting the piecewise control instruction sequence into a current modulation signal of the micro induction heating element and a voltage modulation signal of the piezoelectric actuator; According to the current modulation signal and the voltage modulation signal, the temperature and pressure of each independently controlled area are adjusted in real time to complete the adaptive multi-area sintering.

6. The method for controlling the packaging and sintering of a power semiconductor device according to claim 5, characterized in that: For each independent control area, based on the preset silver particle sintering kinetics model, the time series curve of temperature and pressure is generated, including: According to the initial size distribution of silver particles and the sintering temperature required for sintering, a particle growth rate function in the silver particle sintering kinetic model is constructed; Using the random distribution function and the particle growth rate function in the silver particle sintering kinetics model, the initial distribution and contact state of the silver particles are simulated to obtain an initial sintering structure; Based on the molecular dynamics principle, initial sintering structure and particle growth rate function in the silver particle sintering kinetics model, constructing the interaction potential energy function between the silver particles; Numerically integrating the interaction potential energy function, combining the initial sintering structure and the particle growth rate function, and obtaining a neck growth curve predicted by a silver particle sintering kinetic model; Calculating the volume shrinkage during the sintering process according to the neck growth curve, the initial sintering structure and the densification sub-model in the silver particle sintering kinetics model; Combining the volume shrinkage, the initial sintering structure, the particle growth rate function and the thermal expansion effect model in the silver particle sintering kinetic model, a coupled relationship equation of temperature and pressure is generated; The coupling relationship equation is numerically solved to obtain a time series curve of temperature and pressure.

7. The method for controlling the packaging and sintering of a power semiconductor device according to claim 1, characterized in that: The method of performing real-time nondestructive testing on the sintering interface to obtain an interface quality index and inputting the interface quality index as a feedback parameter into the multivariable collaborative optimization to form a closed-loop control until the power semiconductor device packaging sintering is completed includes: The sintering interface is scanned in real time using ultrasonic scanning technology to obtain an acoustic image, and the thermal reflection imaging technology is used to measure the thermal field distribution of the sintering interface to obtain a thermal reflection intensity map; Performing image fusion processing on the acoustic image and the thermal reflection intensity map to obtain a fused image, and performing feature extraction and classification on the fused image using a pre-trained convolutional neural network to obtain an interface defect distribution map; The interface quality index is calculated according to the interface defect distribution map, and the interface quality index is added as a new input variable into the multivariable collaborative optimization to form a closed-loop control until the power semiconductor device packaging sintering is completed.

8. A packaging and sintering control device for a power semiconductor device, characterized in that: The packaging and sintering control device of the power semiconductor device comprises: A monitoring module is used to monitor the dynamic stress field of the interface between the power semiconductor device chip and the substrate through a preset embedded pressure sensor array to obtain real-time stress distribution data; An optimization module, used to perform multivariable collaborative optimization of sintering process parameters according to the real-time stress distribution data and in combination with a pre-established material property database, to obtain optimized process parameters; A control module, used to dynamically adjust the temperature and pressure curves of each area of ​​the sintering platform according to the process parameters through a preset micro-induction heating element and a piezoelectric actuator to achieve adaptive multi-area sintering; The evaluation module is used to perform real-time nondestructive testing on the sintering interface to obtain the interface quality index, and input the interface quality index as a feedback parameter into the multivariable collaborative optimization to form a closed-loop control until the sintering of the power semiconductor device package is completed.

Images