High-efficiency semiconductor device thermal management method and system

Abstract

The invention relates to the technical field of thermal management of semiconductor devices, and provides a high-efficiency semiconductor device thermal management method and system, and the method comprises the steps: arranging a three-dimensional temperature sensor array to obtain multi-dimensional temperature data, and generating a continuous temperature distribution matrix through spatial interpolation compensation; analyzing matrix characteristics through a dynamic weight fusion algorithm, and generating a heat conduction abnormal coefficient; a self-adaptive heat dissipation topology network containing active and passive heat dissipation paths is constructed in this way; dynamically distributing heat dissipation power and cooling liquid flow velocity according to a hierarchical control strategy; fusing data of the multi-source sensor to calculate a heat dissipation efficiency deviation; a multi-objective optimization model is established based on thermodynamic constraints and device performance, control parameters are iterated, and feature extraction rules are updated. The performance and reliability of the semiconductor device can be improved, and the service life of the semiconductor device is prolonged.

Description

Technical Field

[0001] The present invention relates to the technical field of thermal management of semiconductor devices, and more particularly, to a high-efficiency thermal management method and system for semiconductor devices. Background Art

[0002] During the operation of semiconductor devices, as power density continues to increase, effective heat management becomes a key factor restricting their performance and reliability. Existing thermal management technologies mainly rely on traditional heat dissipation methods, such as passive heat dissipation, air cooling, and liquid cooling. Although these methods can meet the heat dissipation needs to a certain extent, they have some limitations. For example, passive heat dissipation is less efficient and cannot meet the heat dissipation needs of high power density; air cooling is greatly affected by environmental conditions and has poor heat dissipation effect in areas with high heat flux density; although liquid cooling has high heat dissipation efficiency, the system is complex, the cost is high, and there are risks such as coolant leakage. In addition, most existing heat dissipation systems adopt fixed heat dissipation strategies and cannot be dynamically adjusted according to the actual operating status of semiconductor devices, resulting in low heat dissipation efficiency and high energy consumption.

[0003] In the process of implementing the embodiments of the present invention, the inventors found that there are at least the following problems or defects in the prior art: it is difficult to strike a balance between heat dissipation efficiency and energy consumption, the heat dissipation system is insufficiently adaptable to the operating state of semiconductor devices, and the reliability and stability of the heat dissipation system need to be improved. Summary of the Invention

[0004] The present invention provides a high-efficiency semiconductor device thermal management method and system.

[0005] In a first aspect of the present invention, a high-efficiency semiconductor device thermal management method is provided, comprising: S1. Arrange a three-dimensional temperature sensor array on the surface and key internal nodes of the semiconductor device to obtain multi-dimensional temperature data at a preset sampling frequency; S2. Perform spatial interpolation compensation on the multi-dimensional temperature data to generate a continuous temperature distribution matrix with a time stamp; S3. Analyze the gradient characteristics and thermal stress coupling characteristics of the temperature distribution matrix based on a dynamic weight fusion algorithm to generate a heat conduction anomaly coefficient; S4. Constructing an adaptive heat dissipation topology network based on the thermal conductivity anomaly coefficient, wherein the network includes an active heat dissipation path and a passive heat dissipation path; S5. Implement dynamic heat flow distribution in the heat dissipation topology network and use a hierarchical control strategy to generate radiator power instructions and coolant flow rate instructions; S6. Use multi-source sensor fusion technology to collect radiator operating status parameters in real time and calculate the actual heat dissipation efficiency deviation value; S7. Based on thermodynamic constraints and device performance parameters, a multi-objective optimization model is established to perform online iteration of control parameters and simultaneously update the feature extraction rules of the dynamic weight fusion algorithm.

[0006] Furthermore, the dynamic weight fusion algorithm in step S3 includes: S31. Extracting the transverse temperature gradient from the temperature distribution matrix and longitudinal temperature gradient , where T represents temperature, x and y represent the horizontal and vertical coordinate axis directions respectively; S32. Calculation of thermal stress coupling factors , where k is the transverse thermal expansion coefficient, is the longitudinal thermal expansion coefficient, represents the integral area element; S33. Constructing dynamic weight function ,in is the spatial weight coefficient, is the time weight coefficient, is the attenuation factor, is the baseline thermal stress value, t represents time; the first term reflects the relative strength of the spatial gradient and the baseline thermal stress, and the second term combines the time attenuation factor Characterize the dynamic trend of thermal stress; is the baseline thermal stress value; S34. Inputting the weighted temperature gradient data into a convolutional neural network, wherein the network comprises 5 layers of convolution kernels and 3 layers of attention mechanism; S35. Output of heat conduction anomaly coefficient through neural network , The active cooling path priority allocation is triggered when

[0007] Furthermore, the adaptive heat dissipation topology network construction in step S4 includes: S41. Divide the heat conduction sub-regions according to the semiconductor device packaging structure and establish a thermal resistance matrix between the sub-regions , where i and j represent different sub-regions; S42. Use the minimum spanning tree algorithm to determine the main heat dissipation path. The path selection criteria are: Minimum value, where represents the heat conduction anomaly coefficient of the jth sub-region; S43. Dynamically configure the backup cooling path. The backup path activation condition meets the following conditions: the main path thermal resistance change rate >10% or Sustained growth for more than 3 sampling periods; S44. Establish a path switching buffer mechanism to keep the heat dissipation power fluctuation within ±5% during path switching.

[0008] Furthermore, the hierarchical control strategy in step S5 includes: S51. Level 1 control: When When <0.5, only the passive cooling path is enabled. Adjust the heat sink power, where is the passive heat dissipation coefficient, Indicates the maximum temperature, Indicates the ambient temperature, is the heat sink power; S52. Secondary control: When 0.5≤ <0.8, start the micro fan array, the speed in, is the fan base speed, is the speed adjustment coefficient, is the thermal conductivity anomaly coefficient; S53. Level 3 control: When ≥0.8, activate the liquid cooling circulation system, the flow rate ,in is the pressure difference, is the coolant density, is the pipeline loss coefficient; S54. Smooth transition of control instructions to ensure that the power change rate when switching between adjacent control levels is ≤10% / s.

[0009] Furthermore, the multi-objective optimization model in step S7 includes: S71. Define the optimization goal set: minimize heat dissipation energy consumption , maximize heat dissipation efficiency , minimize temperature fluctuations ; S72. Establish physical constraints: 、 , Liquid cooling system pressure difference ,in, Represents junction temperature, Indicates the maximum junction temperature, is the maximum allowable value of the pressure difference of the liquid cooling system; S73. Construct the objective function as shown in the following formula; ,in, is the dynamically adjusted weight factor, is the baseline energy consumption, is the benchmark heat dissipation efficiency, is the base temperature fluctuation; S74. Use a hybrid optimization algorithm to solve the problem, combining the local search capability of the gradient descent method with the global search capability of the genetic algorithm; S75. The optimization result is fed back to the dynamic weight function to update the weight, as shown in the following formula; , .

[0010] Furthermore, the spatial interpolation compensation in step S2 includes: S21. Detect abnormal sensor data. When the temperature difference between adjacent sensors Start data repair when S22. Using the improved Kriging interpolation algorithm, the variation function is adjusted to ,in, is the heat source term; S23. Perform thermodynamic verification on the interpolated temperature field to ensure that the energy conservation equation is satisfied. ,in, represents the heat flux density, represents the material density, represents specific heat capacity, t represents time; S24. Generate a temperature distribution matrix with confidence ratings. Areas with confidence ratings below 90% are marked as high-risk monitoring areas.

[0011] Furthermore, the actual heat dissipation efficiency deviation calculation in step S6 includes: S61. Synchronously collect radiator inlet temperature , outlet temperature and flow Q; S62. Calculate the actual heat dissipation power as shown in the following formula; ,in, Indicates the coolant density, Indicates the specific heat capacity of the coolant; is the actual heat dissipation power; S63. Obtain theoretical heat dissipation power, as shown in the following formula; , where U is the driving voltage, I is the driving current, To drive efficiency; is the theoretical heat dissipation power; S64. The generation efficiency deviation coefficient is shown in the following formula: ;in, is the efficiency deviation coefficient; S65. When , triggers the cooling system health status diagnostic program.

[0012] Furthermore, the hybrid optimization algorithm in step S75 specifically includes: S751. When initializing the population, generate the initial particle swarm by combining the historical optimal solution and the current operating parameters; S752. Use adaptive step size control in the gradient descent stage: ; in, , is the initial step length, is the time constant, is the minimum value; S753. The genetic algorithm mutation operation uses an asymmetric mutation strategy, implementing small-scale perturbations on dominant genes and large-scale mutations on disadvantaged genes. S754. Establish an optimization process traceability mechanism to record the parameter adjustment trajectory and constraint violations of each iteration.

[0013] Furthermore, the control instruction smooth transition process in step S54 includes: S541. Set a transition time window before and after the control instruction switching point ; in, is the system time constant, is the transition time window; S542. Use S-curve for power ramp control: ; Where k is the curve steepness coefficient, is the transition midpoint time, is the power at the start of the transition, is the power at the end of the transition, is the instantaneous power during the transition process; S543. Real-time monitoring of the temperature change rate during the transition process. When the safety threshold is exceeded, the transition is interrupted and emergency cooling is enabled; S544. After the transition is completed, verify the control effect to ensure that the matching degree between the actual temperature distribution and the prediction model is ≥95%.

[0014] In a second aspect of the present invention, a high-efficiency semiconductor device thermal management system is provided, comprising: A three-dimensional sensing array module is used to arrange a three-dimensional temperature sensor array on the surface and key internal nodes of semiconductor devices to obtain multi-dimensional temperature data at a preset sampling frequency; Interpolation compensation module, used to perform spatial interpolation compensation on multi-dimensional temperature data and generate a continuous temperature distribution matrix with time stamps; The coefficient generation module is used to analyze the gradient characteristics and thermal stress coupling characteristics of the temperature distribution matrix based on the dynamic weight fusion algorithm to generate the heat conduction anomaly coefficient; A network construction module, configured to construct an adaptive heat dissipation topology network according to a heat conduction anomaly coefficient, wherein the network includes an active heat dissipation path and a passive heat dissipation path; Dynamic allocation module, used to implement dynamic heat flow allocation in the heat dissipation topology network, and generate radiator power instructions and coolant flow rate instructions using a hierarchical control strategy; Deviation calculation module, used to collect radiator working state parameters in real time through multi-source sensor fusion technology and calculate the actual heat dissipation efficiency deviation value; The feature extraction module is used to establish a multi-objective optimization model based on thermodynamic constraints and device performance parameters to perform online iteration of control parameters and synchronously update the feature extraction rules of the dynamic weight fusion algorithm.

[0015] According to the above-mentioned embodiments of the present invention, there are at least the following beneficial effects: the thermal management method and system of the present invention can achieve efficient heat dissipation and energy consumption optimization of semiconductor devices. By arranging a three-dimensional temperature sensor array on the surface of the semiconductor device and at key internal nodes, multi-dimensional temperature data can be obtained in real time to provide data support for accurate thermal management. Combined with the dynamic weight fusion algorithm and the adaptive heat dissipation topology network, the heat dissipation path and heat dissipation power can be flexibly adjusted to ensure the efficient operation of the heat dissipation system while reducing energy consumption. In addition, the application of hierarchical control strategies and multi-source sensor fusion technology can further improve the adaptability and reliability of the heat dissipation system, enabling it to be dynamically adjusted according to different operating conditions and heat dissipation requirements, thereby improving the performance and service life of semiconductor devices.

[0016] The present invention can also achieve online iteration of control parameters through a multi-objective optimization model, synchronously update the feature extraction rules of the dynamic weight fusion algorithm, and further enhance the intelligence level and adaptability of the heat dissipation system. The application of technologies such as spatial interpolation compensation and actual heat dissipation efficiency deviation calculation can effectively improve the accuracy of temperature data and the evaluation accuracy of heat dissipation efficiency, providing a more reliable basis for the optimization and control of the heat dissipation system. The comprehensive application of these technologies makes the thermal management system of the present invention have significant advantages in the field of heat dissipation of high-power density semiconductor devices, and can effectively solve the problems of low heat dissipation efficiency, high energy consumption, and poor adaptability existing in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily apparent by reading the following detailed description with reference to the accompanying drawings, in which several embodiments of the present invention are shown by way of example and not limitation, in which: Figure 1A schematic flow chart of a high-efficiency semiconductor device thermal management method according to an embodiment of the present invention; Figure 2 A schematic structural diagram of a high-efficiency semiconductor device thermal management system provided by one embodiment of the present invention; Figure 3 The figure schematically shows the structure of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0018] The principles and spirit of the present invention will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are provided solely to enable those skilled in the art to better understand and implement the present invention, and are not intended to limit the scope of the present invention in any way. Rather, these embodiments are provided to make the present invention more thorough and complete, and to fully convey the scope of the present invention to those skilled in the art.

[0019] Those skilled in the art will appreciate that the embodiments of the present invention may be implemented as a system, apparatus, device, method, or computer program product. Therefore, the present invention may be implemented in the following forms: entirely in hardware, entirely in software (including firmware, resident software, microcode, etc.), or in a combination of hardware and software.

[0020] It should be noted that any number of elements in the drawings is for illustration only and not for limitation, and any naming is only for distinction and does not have any limiting meaning.

[0021] Reference below Figure 1 , Figure 1 Schematic diagram of a high-performance semiconductor device thermal management method according to an embodiment of the present invention. Figure 1 As shown, a high-efficiency semiconductor device thermal management method 100 includes: S1. Arrange a three-dimensional temperature sensor array on the surface and key internal nodes of the semiconductor device to obtain multi-dimensional temperature data at a preset sampling frequency; S2. Perform spatial interpolation compensation on the multi-dimensional temperature data to generate a continuous temperature distribution matrix with a time stamp; S3. Analyze the gradient characteristics and thermal stress coupling characteristics of the temperature distribution matrix based on a dynamic weight fusion algorithm to generate a heat conduction anomaly coefficient; S4. Constructing an adaptive heat dissipation topology network based on the thermal conductivity anomaly coefficient, wherein the network includes an active heat dissipation path and a passive heat dissipation path; S5. Implement dynamic heat flow distribution in the heat dissipation topology network and use a hierarchical control strategy to generate radiator power instructions and coolant flow rate instructions; S6. Use multi-source sensor fusion technology to collect radiator operating status parameters in real time and calculate the actual heat dissipation efficiency deviation value; S7. Based on thermodynamic constraints and device performance parameters, a multi-objective optimization model is established to perform online iteration of control parameters and simultaneously update the feature extraction rules of the dynamic weight fusion algorithm.

[0022] It should be noted that in the thermal management of semiconductor devices, it is first necessary to arrange a three-dimensional temperature sensor array on the surface and key internal nodes of the semiconductor device to obtain multi-dimensional temperature data at a preset sampling frequency. The three-dimensional temperature sensor array here refers to the arrangement of multiple temperature sensors at different locations of the semiconductor device, including the key heat source areas on the surface and inside, to form a three-dimensional spatial distribution. These sensors can monitor temperature changes at different locations in real time, thereby providing basic data for subsequent thermal management. The preset sampling frequency refers to setting a suitable sampling frequency based on the operating characteristics and heat dissipation requirements of the semiconductor device, so as to obtain temperature data in a timely manner and ensure the response speed and accuracy of the thermal management system.

[0023] Specifically, the three-dimensional temperature sensor array can be optimized according to the packaging structure and power distribution of the semiconductor device. For example, the density of sensors can be increased in high-power areas and heat-sensitive areas to more accurately monitor temperature changes. The preset sampling frequency can be adjusted according to the dynamic characteristics of the device, usually between 10Hz and 100Hz to ensure that rapid temperature changes can be captured. The multi-dimensional nature of temperature data means that temperature information can be obtained from different directions and depths, thereby more comprehensively reflecting the thermal state of the semiconductor device. These temperature sensors can be of different types such as thermistors, thermocouples or semiconductor temperature sensors, and can be selected according to the specific application scenario and accuracy requirements.

[0024] Preferably, for the arrangement of the three-dimensional temperature sensor array, sensors can be set at key locations such as the chip surface, packaging layer, and heat sink contact surface of the semiconductor device to fully cover the heat conduction path. For example, micro-nano processing technology can be used to integrate micro temperature sensors on the chip surface, while patch sensors can be used on the packaging layer and heat sink contact surface. The sampling frequency can be set dynamically according to the operating mode of the semiconductor device. For example, the sampling frequency can be increased during high-load operation and reduced during low-load or standby mode to save energy and improve the response efficiency of the system. In addition, the sensor data can be pre-processed by intelligent algorithms, such as filtering and denoising, to improve the accuracy and reliability of the data.

[0025] In some embodiments, the dynamic weight fusion algorithm in step S3 includes: S31. Extract the transverse temperature gradient from the temperature distribution matrix and longitudinal temperature gradient , where T represents temperature, x and y represent the horizontal and vertical coordinate axis directions respectively; S32. Calculation of thermal stress coupling factors , where k is the transverse thermal expansion coefficient, is the longitudinal thermal expansion coefficient, represents the integral area element; S33. Constructing dynamic weight function ,in is the spatial weight coefficient, is the time weight coefficient, is the attenuation factor, is the baseline thermal stress value, t represents time; the first term reflects the relative strength of the spatial gradient and the baseline thermal stress, and the second term combines the time attenuation factor Characterize the dynamic trend of thermal stress; is the baseline thermal stress value; S34. Inputting the weighted temperature gradient data into a convolutional neural network, wherein the network comprises 5 layers of convolution kernels and 3 layers of attention mechanism; S35. Output of heat conduction anomaly coefficient through neural network , The active cooling path priority allocation is triggered when

[0026] It should be noted that the dynamic weight fusion algorithm is a key technical approach used in this invention to analyze the temperature distribution matrix. This algorithm extracts and weights the temperature gradient and thermal stress coupling features in the temperature distribution matrix to generate a thermal conduction anomaly coefficient. The temperature distribution matrix here refers to the temperature data matrix acquired through a three-dimensional temperature sensor array and processed through spatial interpolation compensation. It can intuitively reflect the temperature distribution within the semiconductor device. The temperature gradient refers to the rate of change of temperature in space, including transverse and longitudinal temperature gradients, which respectively reflect the temperature variation in different directions. The thermal stress coupling factor comprehensively considers the effects of thermal expansion and temperature gradients and is used to assess anomalies in the heat conduction process. The dynamic weight function weights the temperature gradient data based on spatial and temporal factors to highlight important features and suppress noise. A convolutional neural network is a deep learning model used to process image or matrix data. It extracts features and outputs a thermal conduction anomaly coefficient using convolution kernels and an attention mechanism.

[0027] Specifically, the dynamic weight fusion algorithm first extracts the transverse temperature gradient and longitudinal temperature gradient in the temperature distribution matrix. The transverse temperature gradient represents the change in temperature in the horizontal direction, while the longitudinal temperature gradient represents the change in temperature in the vertical direction. The calculation of the thermal stress coupling factor involves the transverse and longitudinal thermal expansion coefficients, which reflect the thermal expansion characteristics of the material in different directions. The spatial weight coefficient and the time weight coefficient in the dynamic weight function are used to adjust the importance of the temperature gradient data, and the attenuation factor is used to control the influence of the time factor. The convolutional neural network contains multiple layers of convolution kernels and an attention mechanism to extract complex features from the temperature gradient data. The output of the network is the thermal conduction anomaly coefficient, whose value ranges from 0 to 1. When the coefficient is greater than 0.8, the active heat dissipation path is preferentially allocated, which indicates that the degree of thermal conduction anomaly is high and more active heat dissipation measures need to be taken.

[0028] Preferably, in the dynamic weight fusion algorithm, the lateral and longitudinal thermal expansion coefficients can be accurately measured based on the material properties of the semiconductor device or obtained through experimental data. For example, for common silicon-based semiconductor materials, the lateral thermal expansion coefficient is approximately 2.6×10⁻ 6 / ℃, the longitudinal thermal expansion coefficient is about 3.0×10⁻ 6 / °C. The spatial weight coefficient in the dynamic weight function can be adjusted based on the sensor layout density and location importance, for example, assigning higher weights to high-power areas. The temporal weight coefficient can be set based on the response time of the cooling system and is typically between 0.1 and 1. The attenuation factor can be adjusted based on the dynamic characteristics of temperature changes, for example, between 0.01 and 0.1. Furthermore, the structure of the convolutional neural network can be optimized based on the characteristics of temperature data, for example, by increasing the number of convolution kernels or adjusting the number of attention mechanism layers to improve model accuracy and robustness.

[0029] In some embodiments, the adaptive heat dissipation topology network construction in step S4 includes: S41. Divide the heat conduction sub-regions according to the semiconductor device packaging structure and establish a thermal resistance matrix between the sub-regions , where i and j represent different sub-regions; S42. Use the minimum spanning tree algorithm to determine the main heat dissipation path. The path selection criteria are: Minimum value, where represents the heat conduction anomaly coefficient of the jth sub-region; S43. Dynamically configure the backup cooling path. The backup path activation condition meets the following conditions: the main path thermal resistance change rate >10% or Sustained growth for more than 3 sampling periods; S44. Establish a path switching buffer mechanism to keep the heat dissipation power fluctuation within ±5% during path switching.

[0030] It should be noted that the construction of the adaptive heat dissipation topology network is based on the division of heat conduction sub-regions based on the semiconductor device packaging structure. The primary heat dissipation path is determined through a thermal resistance matrix and a minimum spanning tree algorithm. At the same time, backup heat dissipation paths are dynamically configured and a path switching buffer mechanism is established. The heat conduction sub-regions here refer to areas with different heat conduction characteristics divided according to the semiconductor device packaging structure. The thermal resistance matrix is used to describe the thermal resistance relationship between sub-regions. The minimum spanning tree algorithm is an algorithm for determining the optimal path. This algorithm can be used to determine the primary heat dissipation path. The backup heat dissipation path is activated when an anomaly occurs in the primary path. The path switching buffer mechanism is used to maintain stable heat dissipation power during path switching.

[0031] Specifically, the construction of the adaptive heat dissipation topology network first requires dividing the heat conduction sub-regions according to the packaging structure of the semiconductor device. For example, for a multi-chip packaged semiconductor device, each chip and its surrounding area can be divided into a sub-region. Then, a thermal resistance matrix is established between the sub-regions. The elements in the thermal resistance matrix represent the thermal resistance between the sub-regions, which can be obtained through experimental measurement or thermal simulation. Next, the minimum spanning tree algorithm is used to determine the main heat dissipation path. The goal of this algorithm is to select the path with the smallest thermal resistance to ensure the highest heat dissipation efficiency. The activation condition of the backup heat dissipation path can be set according to the thermal resistance change rate or temperature change rate of the main path. For example, when the thermal resistance change rate of the main path exceeds 10% or the temperature continues to rise for more than 3 sampling cycles, the backup heat dissipation path is activated. The path switching buffer mechanism is used to maintain the stability of the heat dissipation power during path switching. For example, by gradually adjusting the heat dissipation power during the switching process, excessive power fluctuations are avoided.

[0032] Preferably, for the construction of the adaptive heat dissipation topology network, some operating steps can be further refined. For example, when dividing the heat conduction sub-areas, a more detailed division can be made according to the power density and thermal sensitivity of the chip, and the high power density area can be divided into smaller sub-areas to more accurately control the heat dissipation. When establishing the thermal resistance matrix, the differences in thermal conductivity characteristics between different materials can be considered. For example, for the thermal resistance between metal and non-metallic materials, it is necessary to accurately calculate according to the thermal conductivity of the actual material. When determining the main heat dissipation path, in addition to considering the minimum thermal resistance, the temperature distribution can also be combined to select a path with a lower temperature as the main heat dissipation path to improve the heat dissipation efficiency. For the activation conditions of the backup heat dissipation path, more parameters can be introduced for comprehensive judgment. For example, parameters such as temperature gradient and thermal stress coupling factor can be combined to more accurately judge whether the main path is abnormal.

[0033] In some embodiments, the hierarchical control strategy in step S5 includes: S51. Level 1 control: When When <0.5, only the passive cooling path is enabled. Adjust the heat sink power, where is the passive heat dissipation coefficient, Indicates the maximum temperature, Indicates the ambient temperature, is the heat sink power; S52. Secondary control: When 0.5≤ <0.8, start the micro fan array, the speed in, is the fan base speed, is the speed adjustment coefficient, is the thermal conductivity anomaly coefficient; S53. Level 3 control: When ≥0.8, activate the liquid cooling circulation system, the flow rate ,in is the pressure difference, is the coolant density, is the pipeline loss coefficient; S54. Smooth transition of control instructions to ensure that the power change rate when switching between adjacent control levels is ≤10% / s.

[0034] It should be noted that the hierarchical control strategy is to enable the passive cooling path, micro fan array and liquid cooling circulation system respectively according to the different ranges of the thermal conductivity anomaly coefficient, so as to achieve dynamic adjustment of the radiator power and coolant flow rate. The thermal conductivity anomaly coefficient here is an indicator to measure the heat conduction state and is used to judge the urgency of the heat dissipation demand. The passive cooling path refers to the method of dissipating heat through natural convection and heat sinks. The micro fan array is an active cooling method that improves the heat dissipation efficiency by increasing air flow. The liquid cooling circulation system is an efficient heat dissipation method that removes heat through the circulation of coolant. The smooth transition processing of control instructions is to ensure that the power of the cooling system changes smoothly when switching between different control levels to avoid thermal shock to semiconductor devices.

[0035] Specifically, the hierarchical control strategy adopts different heat dissipation measures according to different ranges of the thermal conductivity anomaly coefficient. When the thermal conductivity anomaly coefficient is less than 0.5, only the passive heat dissipation path is enabled, and the heat dissipation effect is controlled by adjusting the power of the heat sink. The passive heat dissipation coefficient is a parameter related to the material and structure of the heat sink and can be determined through experiments. When the thermal conductivity anomaly coefficient is between 0.5 and 0.8, the micro-fan array is started, and its speed can be calculated by a function related to the thermal conductivity anomaly coefficient. The speed gain coefficient is a parameter set according to the fan performance and heat dissipation requirements. When the thermal conductivity anomaly coefficient is greater than or equal to 0.8, the liquid cooling circulation system is activated, and the flow rate of the coolant can be calculated by a function related to the pressure difference, coolant density and pipeline loss coefficient. The smooth transition processing of control instructions is to ensure that the power change rate of the heat dissipation system does not exceed 10% / s when switching between different control levels, so as to avoid thermal shock to semiconductor devices.

[0036] Preferably, in the hierarchical control strategy, some operating steps can be further refined. For example, when the passive heat dissipation path is enabled, the power of the heat sink can be dynamically adjusted according to the operating status of the semiconductor device instead of being fixed at a value. For the speed control of the micro fan array, a temperature feedback mechanism can be introduced to dynamically adjust the speed gain coefficient according to the actual temperature change to improve the heat dissipation effect. When activating the liquid cooling circulation system, the dynamic adjustment of the temperature and flow of the coolant can be considered to ensure the heat dissipation efficiency. Smooth transition processing of control instructions can be achieved by setting a transition time window. For example, a suitable transition time window is set according to the response time of the system to ensure a smooth transition of the heat dissipation power during the switching process. In addition, an emergency cooling mechanism can be introduced. When the temperature exceeds the safety threshold, emergency cooling measures are immediately initiated to protect the semiconductor device.

[0037] In some embodiments, the multi-objective optimization model in step S7 includes: S71. Define the optimization goal set: minimize heat dissipation energy consumption , maximize heat dissipation efficiency , minimize temperature fluctuations ; S72. Establish physical constraints: 、 , Liquid cooling system pressure difference ,in, Represents junction temperature, Indicates the maximum junction temperature, is the maximum allowable value of the pressure difference of the liquid cooling system; S73. Construct the objective function as shown in the following formula; ,in, is the dynamically adjusted weight factor, is the baseline energy consumption, is the benchmark heat dissipation efficiency, is the base temperature fluctuation; S74. Use a hybrid optimization algorithm to solve the problem, combining the local search capability of the gradient descent method with the global search capability of the genetic algorithm; S75. Feedback the optimization results to the dynamic weight function and update , .

[0038] It should be noted that the multi-objective optimization model is based on thermodynamic constraints and device performance parameters. By defining a set of optimization objectives, establishing physical constraints, and constructing an objective function, it is combined with a hybrid optimization algorithm to achieve dynamic optimization of the heat dissipation system. The optimization objectives here include minimizing heat dissipation energy consumption, maximizing heat dissipation efficiency, and minimizing temperature fluctuations. These objectives reflect the comprehensive performance of the heat dissipation system in terms of energy consumption, efficiency, and stability. Physical constraints are set based on thermodynamic principles and the actual operating parameters of the device to ensure the safety and reliability of the heat dissipation system. The objective function is a mathematical model that comprehensively considers the optimization objectives and constraints and guides the solution process of the optimization algorithm. The hybrid optimization algorithm combines the advantages of gradient descent and genetic algorithms, leveraging both local and global search capabilities to quickly find the optimal solution.

[0039] Specifically, constructing a multi-objective optimization model first requires defining a set of optimization objectives. Minimizing heat dissipation energy consumption refers to minimizing the energy consumed by the cooling system while meeting cooling requirements. Maximizing heat dissipation efficiency refers to improving the cooling system's ability to transfer heat from semiconductor devices to the environment. Minimizing temperature fluctuations refers to maintaining stable semiconductor device temperatures to prevent performance degradation due to excessive temperature fluctuations. Physical constraints, such as the upper limit on junction temperature, the upper limit on the rate of temperature change, and the upper limit on the pressure differential of the liquid cooling system, ensure that the cooling system operates within a safe range. The objective function quantifies the optimization objectives through dynamically adjusted weighting factors, which are adjusted based on different application scenarios and priorities. The hybrid optimization algorithm combines historical optimal solutions and current operating parameters to generate the initial particle swarm during population initialization. Adaptive step-size control is used in the gradient descent phase, and an asymmetric mutation strategy is employed in the genetic algorithm mutation operation. These methods improve the efficiency and accuracy of the optimization algorithm.

[0040] Preferably, during the implementation of the multi-objective optimization model, some operational steps can be further refined. For example, when defining the optimization target set, the weights of different targets can be adjusted according to the specific application scenario. If the application scenario is more sensitive to energy consumption, the weight of minimizing heat dissipation energy consumption can be increased; if higher requirements are placed on temperature stability, the weight of minimizing temperature fluctuations can be increased. When establishing physical constraints, the parameters of the constraints can be dynamically adjusted based on the actual operating data of the semiconductor device. For example, if it is found that the junction temperature upper limit is too high, the parameter can be appropriately lowered based on experimental data to improve the safety of the system. In the hybrid optimization algorithm, more optimization strategies, such as simulated annealing algorithm, can be introduced to further improve the global search capability of the optimization algorithm. In addition, an optimization process traceability mechanism can be established to record the parameter adjustment trajectory and constraint violations of each iteration so as to analyze and improve the optimization process.

[0041] In some embodiments, the spatial interpolation compensation in step S2 includes: S21. Detect abnormal sensor data. When the temperature difference between adjacent sensors Start data repair when S22. Using the improved Kriging interpolation algorithm, the variation function is adjusted to ,in, is the heat source term; S23. Perform thermodynamic verification on the interpolated temperature field to ensure that the energy conservation equation is satisfied. ,in, represents the heat flux density, represents the material density, represents specific heat capacity, t represents time; S24. Generate a temperature distribution matrix with confidence ratings. Areas with confidence ratings below 90% are marked as high-risk monitoring areas.

[0042] It should be noted that spatial interpolation compensation is the process of processing multi-dimensional temperature data, filling in the blank areas between sensor data, and generating a continuous temperature distribution matrix with timestamps. This method can effectively address the problems of uneven sensor distribution or missing data, and improve the integrity and accuracy of temperature data. The improved Kriging interpolation algorithm is a statistically based interpolation method that uses a variogram to describe the spatial correlation of data, thereby performing more accurate interpolation. The anisotropy factor and noise compensation term are parameters in the variogram used to adjust the accuracy and adaptability of the interpolation. Thermodynamic verification uses the energy conservation equation to ensure that the interpolated temperature field conforms to physical laws, thereby improving data reliability. Confidence rating assesses the credibility of the interpolation results, and marking high-risk monitoring areas facilitates subsequent key monitoring.

[0043] Specifically, spatial interpolation compensation first detects abnormal sensor data. When the temperature difference between adjacent sensors exceeds a set threshold (e.g., 20°C), data repair is initiated. The improved Kriging interpolation algorithm describes the spatial correlation of the data using a specific variogram, which dynamically adjusts the interpolation accuracy based on the actual distribution characteristics of the data. The anisotropy factor describes the differences in data correlation in different directions, and the noise compensation term reduces the impact of random noise in the data. After interpolation, thermodynamic verification is performed to ensure that the interpolation results conform to the energy conservation equation, which accounts for the relationship between factors such as heat flux, material density, specific heat capacity, and time. Finally, a temperature distribution matrix with confidence ratings is generated. Areas with confidence ratings below 90% are marked as high-risk monitoring areas for subsequent focused monitoring.

[0044] Preferably, some operational steps can be further refined during the spatial interpolation compensation process. For example, when detecting abnormal sensor data, time series analysis methods can be combined to use the sensor's historical data to determine whether the current data is abnormal, thereby improving the accuracy of anomaly detection. For the improved Kriging interpolation algorithm, the values of the anisotropy factor and noise compensation term can be dynamically adjusted according to the actual temperature distribution characteristics. For example, in areas with large temperature gradients, the value of the anisotropy factor can be appropriately increased to better reflect the directionality of temperature changes; in areas with large noise, the value of the noise compensation term can be increased to smooth the data. In thermodynamic verification, the interpolated temperature field can be verified in more detail in combination with finite element analysis methods to ensure that it conforms to physical laws in different areas. In addition, for confidence rating, a machine learning algorithm can be introduced to automatically adjust the confidence threshold based on historical data and interpolation results to improve the identification accuracy of high-risk monitoring areas.

[0045] In some embodiments, the actual heat dissipation efficiency deviation calculation in step S6 includes: S61. Synchronously collect radiator inlet temperature , outlet temperature and flow Q; S62. Calculate the actual heat dissipation power ,in, Indicates the coolant density, Indicates the specific heat capacity of the coolant; S63. Obtain theoretical heat dissipation power , where U is the driving voltage, I is the driving current, To drive efficiency; is the theoretical heat dissipation power; S64. Generation efficiency deviation coefficient ; S65. When , triggers the cooling system health status diagnostic program.

[0046] It should be noted that the actual heat dissipation efficiency deviation is calculated by synchronously collecting parameters such as the radiator inlet temperature, outlet temperature, and flow rate, and calculating the deviation between the actual heat dissipation power and the theoretical heat dissipation power, thereby evaluating the performance of the heat dissipation system. The radiator inlet temperature and outlet temperature here refer to the temperature of the coolant when it enters and leaves the radiator, and the flow rate refers to the flow rate of the coolant in the radiator. The actual heat dissipation power is calculated based on these parameters and reflects the heat dissipation capacity of the radiator in actual operation. The theoretical heat dissipation power is the ideal heat dissipation capacity calculated based on the design parameters and operating conditions of the radiator. The efficiency deviation coefficient is used to quantify the difference between the actual heat dissipation power and the theoretical heat dissipation power. When the deviation exceeds a certain threshold, the heat dissipation system health status diagnostic program is triggered to promptly detect and resolve possible problems with the heat dissipation system.

[0047] Specifically, the actual heat dissipation efficiency deviation calculation first synchronously collects the radiator inlet temperature, outlet temperature, and flow rate. These parameters can be obtained in real time through sensors installed on the radiator. The actual heat dissipation power is calculated based on the coolant's density, specific heat capacity, flow rate, and the inlet and outlet temperature difference. It reflects the heat dissipation effect of the radiator in actual operation. The theoretical heat dissipation power is calculated based on the radiator's design parameters, such as drive voltage, drive current, and drive efficiency. It represents the heat dissipation capacity of the radiator under ideal conditions. The efficiency deviation coefficient is calculated by taking the difference between the actual heat dissipation power and the theoretical heat dissipation power and the ratio of the theoretical heat dissipation power. It is used to evaluate the actual operating efficiency of the cooling system. When the absolute value of the efficiency deviation coefficient exceeds 15%, it indicates that the cooling system may have a fault or performance degradation. At this time, the cooling system health status diagnostic program is triggered to conduct a comprehensive inspection and diagnosis of the cooling system.

[0048] Preferably, in the process of calculating the actual heat dissipation efficiency deviation, some operating steps can be further refined. For example, when collecting the radiator inlet temperature, outlet temperature and flow rate, high-precision sensors can be used, and multi-point sampling can be performed to improve the accuracy of the data. For the calculation of the actual heat dissipation power, the parameter values of density and specific heat capacity can be adjusted according to different coolant types to ensure the accuracy of the calculation results. When calculating the theoretical heat dissipation power, parameters such as the driving voltage, driving current and driving efficiency can be dynamically adjusted according to the actual operating conditions of the radiator to be closer to the actual operating conditions. In addition, the threshold setting of the efficiency deviation coefficient can be adjusted according to the specific application scenario and reliability requirements of the heat dissipation system. For example, in situations where the reliability requirements of the heat dissipation system are high, the threshold can be set lower so that potential problems can be discovered more promptly.

[0049] In some embodiments, the hybrid optimization algorithm in step S75 specifically includes: S751. When initializing the population, generate the initial particle swarm by combining the historical optimal solution and the current operating parameters; S752. Use adaptive step size control in the gradient descent stage: ; in, , is the initial step size, is the time constant, The minimum value is used to prevent the denominator from being zero; S753. The genetic algorithm mutation operation uses an asymmetric mutation strategy, implementing small-scale perturbations on dominant genes and large-scale mutations on disadvantaged genes. S754. Establish an optimization process traceability mechanism to record the parameter adjustment trajectory and constraint violations of each iteration.

[0050] It should be noted that the hybrid optimization algorithm is an optimization method that combines the gradient descent method and the genetic algorithm, and is used to solve multi-objective optimization models. This algorithm can quickly find the optimal solution by utilizing the local search capability of the gradient descent method and the global search capability of the genetic algorithm. When initializing the population, the initial particle swarm is generated by combining the historical optimal solution and the current operating parameters in order to improve the convergence speed and accuracy of the algorithm. Adaptive step size control can dynamically adjust the step size according to the gradient information during the optimization process to avoid falling into the local optimal solution. The asymmetric mutation strategy is a mutation operation in a genetic algorithm that implements small-scale perturbations on dominant genes and large-span mutations on disadvantaged genes to maintain the diversity of the population. The optimization process traceability mechanism is used to record the parameter adjustment trajectory and constraint violations of each iteration in order to analyze and improve the optimization process.

[0051] Specifically, during the initialization phase, the hybrid optimization algorithm generates an initial particle swarm by combining historical optimal solutions and current operating parameters. The historical optimal solution refers to the best solution obtained during previous optimization processes and provides a good starting point for the algorithm. Current operating parameters refer to relevant parameters of the cooling system's current operating state, such as temperature and flow rate, which reflect the current operating environment. During the gradient descent phase, adaptive step size control is used to dynamically adjust the step size based on the magnitude of the gradient. When the gradient is large, the step size can be increased to accelerate convergence; when the gradient is small, the step size is reduced to avoid overshoot. An asymmetric mutation strategy mutates the population within the genetic algorithm. For well-performing genes, small perturbations are applied to fine-tune their performance; for underperforming genes, large mutations are applied to explore new solutions. An optimization process traceability mechanism records the parameter adjustment trajectory and constraint violations at each iteration. This facilitates analysis of optimization issues, such as frequent constraint violations, and provides a basis for algorithm improvement.

[0052] Preferably, during the implementation of the hybrid optimization algorithm, some operational steps can be further refined. For example, when initializing the population, initial particle swarms for high-load, medium-load, and low-load conditions can be generated respectively according to the different operating modes of the cooling system to improve the adaptability of the algorithm to different working conditions. In the adaptive step size control, a time decay factor can be introduced to gradually reduce the step size as the number of iterations increases, so as to ensure that the algorithm can converge stably in the later stage. For asymmetric mutation strategies, the mutation probability can be adjusted dynamically, and the intensity of the mutation can be determined according to the diversity of the population. When the population diversity is low, the mutation probability is increased to introduce new gene combinations; when the population diversity is high, the mutation probability is reduced to maintain the stability of the population. In addition, the optimization process tracing mechanism can be combined with visualization tools to display the parameter adjustment trajectory and constraint violations of each iteration in graphical form, which is convenient for intuitive analysis of the optimization process.

[0053] In some embodiments, the control instruction smooth transition process in step S54 includes: S541. Set a transition time window before and after the control instruction switching point ; in, is the system time constant; S542. Use S-curve for power ramp control: ; Where k is the curve steepness coefficient, is the transition midpoint time, is the power at the start of the transition, is the power at the end of the transition, is the instantaneous power during the transition process; S543. Real-time monitoring of the temperature change rate during the transition process. When the safety threshold is exceeded, the transition is interrupted and emergency cooling is enabled; S544. After the transition is completed, verify the control effect to ensure that the matching degree between the actual temperature distribution and the prediction model is ≥95%.

[0054] It should be noted that the smooth transition processing of control instructions is intended to ensure that the change in heat dissipation power is smooth when switching between different control levels of the heat dissipation system, thereby avoiding thermal shock to semiconductor devices. The control instruction switching point refers to the time point when the heat dissipation system switches from one control level to another. The transition time window is a time interval set before and after the switching point for smooth transition. The S-curve is a smooth mathematical curve used to control the rate of change of power to ensure that the power change rate does not exceed the set safety threshold. Real-time monitoring of the temperature change rate is to promptly detect abnormal conditions during the transition process. When the temperature change rate exceeds the safety threshold, the transition is interrupted and emergency cooling measures are activated to protect the safety of the semiconductor devices. After the transition is completed, the control effect is verified to ensure that the actual temperature distribution matches the prediction model to a high level, thereby ensuring the effectiveness and reliability of the heat dissipation system.

[0055] Specifically, the smooth transition of control instructions first sets a transition time window before and after the control instruction switching point. The size of this window can be adjusted based on the system's response time and cooling requirements. For example, for a cooling system with a short response time, a smaller transition time window can be set; for a system with a longer response time, a larger transition time window is required. An S-shaped curve is used to control power changes. Its steepness coefficient determines the curvature of the curve, thereby affecting the rate of power change. The steepness coefficient can be adjusted based on the actual needs of the cooling system to ensure that the power change rate does not exceed 10% / s. During the transition process, the temperature change rate is monitored in real time, obtaining real-time data from temperature sensors installed in the cooling system. If the temperature change rate exceeds a set safety threshold, the transition process is immediately interrupted and emergency cooling measures are initiated, such as increasing coolant flow or activating backup cooling equipment. After the transition is complete, the control effectiveness is verified by comparing the actual temperature distribution with the results of the prediction model. If the match is less than 95%, the control strategy needs to be adjusted.

[0056] Preferably, some operating steps can be further refined during the smooth transition processing of control instructions. For example, when setting the transition time window, the window size can be dynamically adjusted according to the specific working conditions of the cooling system. For cooling systems running at high loads, the transition time window can be appropriately increased to ensure a smooth transition of the cooling power. When using the S-curve for power ramp control, the steepness coefficient can be adjusted according to the characteristics of different cooling devices. For devices with higher cooling efficiency, the steepness coefficient can be appropriately reduced to speed up the transition speed; while for devices with lower cooling efficiency, the steepness coefficient needs to be increased to ensure the smoothness of the transition process. When monitoring the temperature change rate in real time, the temperature change trend can be predicted in combination with a machine learning algorithm to detect potential abnormalities in advance. In addition, in the control effect verification stage, in addition to comparing the actual temperature distribution with the results of the prediction model, other performance indicators such as cooling energy consumption and cooling efficiency can also be introduced to comprehensively evaluate the cooling system.

[0057] The above-described embodiments of the present invention have the following beneficial effects: The present invention can achieve efficient heat dissipation and energy consumption optimization for semiconductor devices. By arranging a three-dimensional temperature sensor array to acquire multi-dimensional temperature data and performing spatial interpolation compensation on the multi-dimensional temperature data, a continuous temperature distribution matrix with timestamps can be generated, providing accurate data support for heat dissipation control. A dynamic weight fusion algorithm is used to analyze the gradient characteristics and thermal stress coupling characteristics of the temperature distribution matrix to generate a heat conduction anomaly coefficient. This is then used to construct an adaptive heat dissipation topology network, comprising active and passive heat dissipation paths. This network can flexibly adjust the heat dissipation path based on the heat conduction anomaly coefficient, ensuring efficient operation of the heat dissipation system. Dynamic heat flow distribution is implemented within the heat dissipation topology network, and a hierarchical control strategy is employed to generate heat sink power commands and coolant flow rate commands, further optimizing heat dissipation and energy consumption. Multi-source sensor fusion technology is used to collect heat sink operating state parameters in real time and calculate the actual heat dissipation efficiency deviation, providing a basis for dynamic adjustment of the heat dissipation system. A multi-objective optimization model is established based on thermodynamic constraints and device performance parameters to perform online iteration of control parameters. The feature extraction rules of the dynamic weight fusion algorithm are simultaneously updated, further enhancing the intelligence and adaptability of the heat dissipation system.

[0058] In addition, the present invention can also perform thermodynamic verification on the interpolated temperature field through an improved Kriging interpolation algorithm to ensure that the energy conservation equation is satisfied, generate a temperature distribution matrix with confidence rating, and improve the accuracy and reliability of the temperature data. When switching the heat dissipation path, by establishing a path switching buffer mechanism, the heat dissipation power fluctuation is kept within ±5%, ensuring the stable operation of the heat dissipation system. The smooth transition processing of control instructions in the hierarchical control strategy can ensure that the power change rate when switching between adjacent control levels is ≤10% / s, avoiding problems such as excessive power fluctuations in the heat dissipation system during the switching process. The hybrid optimization algorithm combines the local search capability of the gradient descent method and the global search capability of the genetic algorithm, and can quickly and accurately solve the multi-objective optimization model, providing a more reliable solution for the optimization control of the heat dissipation system.

[0059] like Figure 2 As shown, in some embodiments, a high-efficiency semiconductor device thermal management system 200 includes: The three-dimensional sensor array module 201 is used to arrange a three-dimensional temperature sensor array on the surface and key nodes inside the semiconductor device to obtain multi-dimensional temperature data at a preset sampling frequency; The interpolation compensation module 202 is used to perform spatial interpolation compensation on the multi-dimensional temperature data to generate a continuous temperature distribution matrix with a time stamp; The coefficient generation module 203 is used to analyze the gradient characteristics and thermal stress coupling characteristics of the temperature distribution matrix based on a dynamic weight fusion algorithm to generate a heat conduction anomaly coefficient; A network construction module 204 is used to construct an adaptive heat dissipation topology network according to the heat conduction anomaly coefficient, wherein the network includes an active heat dissipation path and a passive heat dissipation path; Dynamic allocation module 205, for implementing dynamic heat flow allocation in the heat dissipation topology network, and generating radiator power instructions and coolant flow rate instructions using a hierarchical control strategy; Deviation calculation module 206, used to collect radiator working state parameters in real time through multi-source sensor fusion technology and calculate the actual heat dissipation efficiency deviation value; The feature extraction module 207 is used to establish a multi-objective optimization model based on thermodynamic constraints and device performance parameters to perform online iteration of control parameters and synchronously update the feature extraction rules of the dynamic weight fusion algorithm.

[0060] It is understood that the modules described in the high-performance semiconductor device thermal management system 200 are similar to those described in the reference Figure 1 Therefore, the operations, features, and beneficial effects described above for the high-efficiency semiconductor device thermal management method are also applicable to the high-efficiency semiconductor device thermal management system 200 and the modules included therein, and will not be repeated here.

[0061] Reference below Figure 3 , which shows a schematic structural diagram of an electronic device 300 suitable for implementing some embodiments of the present invention. The electronic devices in some embodiments of the present invention may include, but are not limited to, mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 3 The terminal device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0062] like Figure 3 As shown, electronic device 300 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 301, which can perform various appropriate actions and processes based on programs stored in a read-only memory (ROM) 302 or programs loaded from a storage device 308 into a random access memory (RAM) 303. RAM 303 also stores various programs and data required for the operation of electronic device 300. Processing device 301, ROM 302, and RAM 303 are interconnected via a bus 304. An input / output (I / O) interface 305 is also connected to bus 304.

[0063] Typically, the following devices may be connected to the I / O interface 305: an input device 306 including, for example, a touch screen, a touchpad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 307 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 308 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 309. The communication device 309 may allow the electronic device 300 to communicate with other devices wirelessly or by wire to exchange data. Figure 3 The electronic device 300 is shown with various devices, but it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed instead. Figure 3 Each block shown in the figure may represent one device, or may represent multiple devices as needed.

[0064] Furthermore, the storage medium of the embodiment of the present application stores program instructions that can implement all the above methods, wherein the program instructions can be stored in the above storage medium in the form of a software product, including a number of instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) or a processor to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, 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 disk, or a terminal device such as a computer, server, mobile phone, or tablet.

[0065] The above descriptions merely illustrate some preferred embodiments of the present invention and the underlying technical principles. Those skilled in the art should understand that the scope of the invention encompassed by the embodiments of the present invention is not limited to technical solutions formed by specific combinations of the aforementioned technical features. It also encompasses other technical solutions formed by any combination of the aforementioned technical features or their equivalents, without departing from the aforementioned inventive concept. For example, a technical solution formed by replacing the aforementioned features with (but not limited to) technical features with similar functions disclosed in the embodiments of the present invention.

Claims

1. A high-performance semiconductor device thermal management method, characterized in that: The following steps are involved: S1. Arrange a three-dimensional temperature sensor array on the surface and key internal nodes of the semiconductor device to obtain multi-dimensional temperature data at a preset sampling frequency; S2. Perform spatial interpolation compensation on the multi-dimensional temperature data to generate a continuous temperature distribution matrix with a time stamp; S3. Analyze the gradient characteristics and thermal stress coupling characteristics of the temperature distribution matrix based on a dynamic weight fusion algorithm to generate a heat conduction anomaly coefficient; S4. Constructing an adaptive heat dissipation topology network based on the thermal conductivity anomaly coefficient, wherein the network includes an active heat dissipation path and a passive heat dissipation path; S5. Implement dynamic heat flow distribution in the heat dissipation topology network and use a hierarchical control strategy to generate radiator power instructions and coolant flow rate instructions; S6. Use multi-source sensor fusion technology to collect radiator operating status parameters in real time and calculate the actual heat dissipation efficiency deviation value; S7. Based on thermodynamic constraints and device performance parameters, a multi-objective optimization model is established to perform online iteration of control parameters and simultaneously update the feature extraction rules of the dynamic weight fusion algorithm.

2. The method according to claim 1, characterized in that The dynamic weight fusion algorithm in step S3 includes: S31. Extract the transverse temperature gradient and longitudinal temperature gradient from the temperature distribution matrix. The transverse temperature gradient is: ; The longitudinal temperature gradient is: ; Where T represents temperature, x and y represent the horizontal and vertical coordinate axes respectively; S32. Calculate the thermal stress coupling factor as shown in the following formula; , where k is the transverse thermal expansion coefficient, is the longitudinal thermal expansion coefficient, represents the integral area element; S33. Construct a dynamic weight function as shown in the following formula; ,in, is the spatial weight coefficient, is the time weight coefficient, is the attenuation factor, is the baseline thermal stress value, t represents time; the first term reflects the relative strength of the spatial gradient and the baseline thermal stress, and the second term combines the time attenuation factor Characterize the dynamic trend of thermal stress; is the baseline thermal stress value; S34. Inputting the weighted temperature gradient data into a convolutional neural network, wherein the network comprises 5 layers of convolution kernels and 3 layers of attention mechanism; S35. Output of heat conduction anomaly coefficient through neural network ,when The active cooling path priority allocation is triggered when 3. The method according to claim 2, characterized in that The adaptive heat dissipation topology network construction in step S4 includes: S41. Divide the heat conduction sub-regions according to the semiconductor device packaging structure and establish a thermal resistance matrix between the sub-regions , where i and j represent different sub-regions; S42. Use the minimum spanning tree algorithm to determine the main heat dissipation path. The path selection criteria are: Minimum value, where represents the heat conduction anomaly coefficient of the jth sub-region; S43. Dynamically configure the backup cooling path. The backup path activation condition meets the following conditions: the main path thermal resistance change rate >10% or Sustained growth for more than 3 sampling periods; S44. Establish a path switching buffer mechanism to keep the heat dissipation power fluctuation within ±5% during path switching.

4. The method according to claim 3, characterized in that The hierarchical control strategy in step S5 includes: S51. Level 1 control: When When <0.5, only the passive cooling path is enabled. Adjust the heat sink power, where is the passive heat dissipation coefficient, Indicates the maximum temperature, Indicates the ambient temperature, is the heat sink power; S52. Secondary control: When 0.5≤ <0.8, start the micro fan array, the speed ;in, is the fan base speed, is the speed adjustment coefficient, is the thermal conductivity anomaly coefficient; S53. Level 3 control: When ≥0.8, activate the liquid cooling circulation system, the flow rate ,in is the pressure difference, is the coolant density, is the pipeline loss coefficient; S54. Smooth transition of control instructions to ensure that the power change rate when switching between adjacent control levels is ≤10% / s.

5. The method according to claim 4, characterized in that The multi-objective optimization model in step S7 includes: S71. Define the optimization goal set: minimize heat dissipation energy consumption , maximize heat dissipation efficiency , minimize temperature fluctuations ; S72. Establish physical constraints: 、 , Liquid cooling system pressure difference ,in, Represents junction temperature, Indicates the maximum junction temperature, is the maximum allowable value of the pressure difference of the liquid cooling system; S73. Construct the objective function as shown in the following formula; ,in, is the dynamically adjusted weight factor, is the baseline energy consumption, is the benchmark heat dissipation efficiency, is the base temperature fluctuation; S74. Use a hybrid optimization algorithm to solve the problem, combining the local search capability of the gradient descent method with the global search capability of the genetic algorithm; S75. The optimization result is fed back to the dynamic weight function to update the weight, as shown in the following formula; ， 。 6. The method according to claim 5, characterized in that The spatial interpolation compensation in step S2 includes: S21. Detect abnormal sensor data. When the temperature difference between adjacent sensors Start data repair when S22. Using the improved Kriging interpolation algorithm, the variation function is adjusted to ,in, is the heat source term; S23. Perform thermodynamic verification on the interpolated temperature field to ensure that the energy conservation equation is satisfied. ,in, represents the heat flux density, represents the material density, represents specific heat capacity, t represents time; S24. Generate a temperature distribution matrix with confidence ratings. Areas with confidence ratings below 90% are marked as high-risk monitoring areas.

7. The method according to claim 6, characterized in that The actual heat dissipation efficiency deviation calculation in step S6 includes: S61. Synchronously collect radiator inlet temperature , outlet temperature and flow Q; S62. Calculate the actual heat dissipation power as shown in the following formula; ,in, Indicates the coolant density, Indicates the specific heat capacity of the coolant; is the actual heat dissipation power; S63. Obtain theoretical heat dissipation power, as shown in the following formula; , where U is the driving voltage, I is the driving current, To drive efficiency; is the theoretical heat dissipation power; S64. The generation efficiency deviation coefficient is shown in the following formula: ;in, is the efficiency deviation coefficient; S65. When , triggers the cooling system health status diagnostic program.

8. The method according to claim 7, characterized in that The hybrid optimization algorithm in step S75 specifically includes: S751. When initializing the population, generate the initial particle swarm by combining the historical optimal solution and the current operating parameters; S752. Use adaptive step size control in the gradient descent stage: ； in, , is the initial step size, is the time constant, is the minimum value; S753. The genetic algorithm mutation operation uses an asymmetric mutation strategy, implementing small-scale perturbations on dominant genes and large-scale mutations on disadvantaged genes. S754. Establish an optimization process traceability mechanism to record the parameter adjustment trajectory and constraint violations of each iteration.

9. The method according to claim 8, characterized in that The control instruction smooth transition process in step S54 includes: S541. Set a transition time window before and after the control instruction switching point, as shown in the following formula; ； in, is the system time constant, is the transition time window; S542. Use S-curve for power ramp control: ； Where k is the curve steepness coefficient, is the transition midpoint time, is the power at the start of the transition, is the power at the end of the transition, is the instantaneous power during the transition process; S543 real-time monitoring of the temperature change rate during the transition process, when the temperature change rate exceeds the safety threshold when the transition is interrupted and enable emergency cooling; S544. After the transition is completed, verify the control effect to ensure that the matching degree between the actual temperature distribution and the prediction model is ≥95%.

10. A high-performance semiconductor device thermal management system, characterized in that: include: A three-dimensional sensing array module is used to arrange a three-dimensional temperature sensor array on the surface and key internal nodes of semiconductor devices to obtain multi-dimensional temperature data at a preset sampling frequency; Interpolation compensation module, used to perform spatial interpolation compensation on multi-dimensional temperature data and generate a continuous temperature distribution matrix with time stamps; The coefficient generation module is used to analyze the gradient characteristics and thermal stress coupling characteristics of the temperature distribution matrix based on the dynamic weight fusion algorithm to generate the heat conduction anomaly coefficient; A network construction module, configured to construct an adaptive heat dissipation topology network according to a heat conduction anomaly coefficient, wherein the network includes an active heat dissipation path and a passive heat dissipation path; Dynamic allocation module, used to implement dynamic heat flow allocation in the heat dissipation topology network, and generate radiator power instructions and coolant flow rate instructions using a hierarchical control strategy; Deviation calculation module, used to collect radiator working state parameters in real time through multi-source sensor fusion technology and calculate the actual heat dissipation efficiency deviation value; The feature extraction module is used to establish a multi-objective optimization model based on thermodynamic constraints and device performance parameters to perform online iteration of control parameters and synchronously update the feature extraction rules of the dynamic weight fusion algorithm.

Images