Chip aging test board and intelligent heat dissipation method and system thereof

Abstract

The present application relates to the technical field of chip testing, and particularly relates to a chip aging test board and an intelligent heat dissipation method and system thereof, which comprises a substrate provided with sockets for mounting a plurality of chips to be tested; a support plate fixedly installed on the bottom surface of the substrate; a plurality of screws locked into threaded holes at the four corners of the sockets by penetrating the support plate and the substrate; and a plurality of air guide inclined surfaces arranged on the support plate and corresponding to the positions of the sockets. The present application forms a stable mechanical connection by arranging a support plate on the bottom surface of the substrate and locking the screws into the threaded holes of the sockets by penetrating the support plate and the substrate, so as to resist the downward impact force during chip mounting and dismounting and the thermal stress generated by high-low temperature cycles. In addition, the plurality of air guide inclined surfaces arranged on the support plate help to improve the heat exchange between the air and the chips, and the intelligent scheme taking into account temperature control and vibration management solves the problems of response lag and insufficient temperature regulation capacity of the traditional scheme.

Description

Technical Field

[0001] This invention relates to the field of chip testing technology, and in particular to a chip aging test board and its intelligent heat dissipation method and system. Background Technology

[0002] In a chip aging test system, when a high-density pin chip is subjected to aging test, the chip needs to be placed in a socket. The socket is connected to the aging test board (i.e., BIB board). Power supply and test signals are transmitted to the chip through the socket via the aging test board. Due to the high pin density between the socket and the aging test board, the existing technology usually uses a nut to fix the connection between the two.

[0003] However, throughout the testing process, the thin, weak, and rigid aging test board is susceptible to deformation due to the thermal stress generated by high and low temperature cycling and the downward impact force during chip loading and unloading. This deformation can lead to loosening of the connecting nuts, resulting in poor contact between the socket and the aging test board, ultimately causing errors in the test results and severely affecting the reliability and accuracy of chip aging tests. Furthermore, existing aging test boards use a PID-controlled fan strategy during testing, which is not only slow to respond but also unable to handle the dynamic requirements of different chip layouts and testing stages. Summary of the Invention

[0004] The main objective of this invention is to provide a chip aging test board and its intelligent heat dissipation method and system, aiming to solve the technical problems mentioned in the background art.

[0005] This invention proposes a chip aging test board, comprising:

[0006] A substrate having sockets for mounting multiple chips to be tested;

[0007] A support plate, which is fixedly installed on the bottom surface of the substrate, is used to provide structural strength to prevent substrate deformation;

[0008] Multiple screws pass through the support plate and the base plate and are locked into the threaded holes at the four corners of the socket to achieve mechanical fixation;

[0009] Multiple air guide slopes are disposed on the support plate and corresponding to the position of the socket to guide the cooling airflow toward the chip;

[0010] A temperature sensor network, distributed and integrated on the substrate, is used to collect surface temperature data of the chip in real time.

[0011] The tilt angle, surface curvature, and orientation relative to the chip of the air guide slope are definable to collaboratively shape differentiated flow fields for different chip regions.

[0012] Preferably, an insulating and heat-conducting layer is provided at the contact point between the support plate and the substrate to enhance heat transfer and improve temperature uniformity.

[0013] Preferably, the support plate is a perforated grid structure.

[0014] Preferably, it also includes an independently controlled multi-fan array;

[0015] Each fan in the multi-fan array is independently aligned with one or more of the air guide slope inlets;

[0016] Each fan is controlled by an independent signal to achieve independent adjustment of the inlet velocity of each guide slope;

[0017] The fan array, the air guide slope, and the chip layout on the substrate together constitute a software-definable dynamic flow field shaping hardware foundation.

[0018] This invention also provides an intelligent heat dissipation method for a chip aging test board, comprising the following steps:

[0019] A computational fluid dynamics BIB parameterized model is established based on the geometric parameters of the substrate, support plate, and air guide slope.

[0020] Based on the BIB parameterized model, a dataset of inclined plane flow velocity-temperature rise-vibration mapping relationship is generated through CFD simulation.

[0021] A lightweight neural network prediction model is trained using the slope velocity-temperature rise-vibration mapping dataset. The input of the neural network prediction model is the current inlet wind speed of the wind guide slope, and the output is the predicted temperature distribution field of all chips and the vibration prediction value of all substrates.

[0022] The real-time temperature distribution field of all chips is obtained through a temperature sensor network, and the real-time vibration value of the substrate is obtained through a vibration sensor.

[0023] Based on the real-time temperature distribution field and real-time vibration values, the lightweight neural network prediction model is invoked, and the optimal target wind speed combination is solved through a multi-objective optimization algorithm.

[0024] The optimal target wind speed combination is converted into PWM control signals for each wind turbine in the multi-turbine array and then driven to execute.

[0025] Preferably, the step of training a lightweight neural network prediction model using the inclined plane flow velocity-temperature rise-vibration mapping dataset includes:

[0026] The slope velocity-temperature rise-vibration mapping dataset is preprocessed to obtain a standardized dataset.

[0027] A neural network with multiple output branches is constructed based on the input and output dimensions of the standardized dataset. The number of neurons in the input layer is set to correspond to the number of wind guide slopes, and the number of neurons in the two output layers is set to correspond to the number of monitored chips and the number of substrates, respectively.

[0028] Based on the standardized dataset, the mean squared error is used as the loss function, and the neural network is trained through the backpropagation algorithm until its prediction accuracy on the preset validation set reaches a preset threshold to obtain a lightweight neural network prediction model.

[0029] Preferably, the step of calling the lightweight neural network prediction model based on the real-time temperature distribution field and real-time vibration values, and solving for the optimal target wind speed combination through a multi-objective optimization algorithm, includes:

[0030] The temperature prediction function and vibration prediction function are obtained based on the lightweight neural network prediction model.

[0031] A comprehensive performance index is constructed based on the temperature prediction function and the vibration prediction function.

[0032] Obtain chip temperature safety constraints and fan speed feasible region constraints, aim to minimize the comprehensive performance index, and construct a constrained optimization problem under the condition of satisfying chip temperature safety constraints and fan speed feasible region constraints;

[0033] A fast quadratic programming method based on the sensitivity matrix is ​​used to iteratively solve the constrained optimization problem and output the optimal target wind speed combination.

[0034] Preferably, the step of constructing a comprehensive performance index based on the temperature prediction function and the vibration prediction function includes:

[0035] The priority of temperature control requirements and the priority of vibration suppression requirements are determined based on the real-time temperature distribution and the real-time vibration value.

[0036] The weighting coefficients in the overall system performance index are dynamically adjusted according to the priority of the temperature control requirements and the priority of the vibration suppression requirements.

[0037] Based on the weighting coefficients, and combined with the temperature prediction function and vibration prediction function, a comprehensive performance index is constructed.

[0038] Preferably, the step of iteratively solving the constrained optimization problem using a fast quadratic programming method based on the sensitivity matrix and outputting the optimal target wind speed combination includes:

[0039] The real-time wind speed combination is obtained, and the sensitivity matrix of temperature distribution field and vibration level to wind speed is obtained by automatic differentiation algorithm based on the lightweight neural network prediction model and the real-time wind speed combination.

[0040] Based on the sensitivity matrix, the lightweight neural network prediction model is expanded using a first-order Taylor series in real-time wind speed combinations to construct a linear prediction model.

[0041] A quadratic programming problem is constructed based on the linearized predicted temperature distribution field and vibration prediction values ​​of the linearized prediction model.

[0042] The optimal target wind speed combination is obtained by solving the quadratic programming problem using the effective set method.

[0043] The present invention also provides an intelligent heat dissipation system for a chip aging test board, comprising multiple modules, which are used to implement the steps of an intelligent heat dissipation method for a chip aging test board.

[0044] The beneficial effects of this invention are as follows: By setting a support plate on the bottom surface of the substrate and using multiple screws to pass through the support plate and the substrate and lock them to the threaded holes of the socket, a stable mechanical connection structure is formed. This effectively resists the downward impact force during chip loading and unloading, as well as the thermal stress generated by high and low temperature cycles, thus preventing substrate deformation and solving the problem of easy loosening of traditional nut connections. Furthermore, the multiple air guide slopes set on the support plate can accurately guide the cooling airflow to the chip surface inside the socket, which helps to improve the heat exchange between the air and the chip. In addition, the intelligent solution that takes into account both temperature control and vibration management solves the problems of slow response and insufficient temperature control capability of traditional solutions. Attached Figure Description

[0045] Figure 1 This is a front view of an embodiment of this application.

[0046] Figure 2 This is a bottom view of an embodiment of this application;

[0047] Figure 3 This is a schematic diagram of an embodiment of the present application showing how an airflow direction is guided by a guide slope.

[0048] Figure 4 This is a schematic diagram of a method flow according to an embodiment of this application;

[0049] Figure 5 This is a schematic diagram of the system structure according to an embodiment of this application.

[0050] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings.

[0051] Reference numerals: 1. Connector protective plate; 2. Base plate; 3. Socket; 4. Positioning pin seat; 5. Connector; 6. Support plate; 7. Slide rail; 8. Washer; 9. Screw. Detailed Implementation

[0052] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0053] like Figures 1-3 As shown, this application provides a chip aging test board, comprising:

[0054] The substrate 2 has a socket 3 for mounting multiple chips to be tested; wherein, the substrate 2 is a PCB, and one end of the substrate 2 is also provided with a connector 5 and a connector protection plate 1, and the substrate 2 is also provided with a positioning pin seat 4 for positioning the test board, and the side of the substrate 2 is also provided with a slide 7 for guiding the installation of the test board.

[0055] Support plate 6 is fixedly installed on the bottom surface of the substrate 2 to provide structural strength and prevent deformation of the substrate 2;

[0056] Multiple screws 9 pass through the support plate 6 and the base plate 2 and are locked into the threaded holes at the four corners of the socket 3 to achieve mechanical fixation;

[0057] Multiple air guide slopes are set on the support plate 6 and correspond to the position of the socket 3 to guide the cooling airflow to the chip;

[0058] A temperature sensor network, distributed and integrated on the substrate 2, is used to collect the surface temperature data of the chip in real time. The sensors in the temperature sensor network include the following arrangement methods: First, when manufacturing the socket 3, a miniature thermocouple junction or thermistor is pre-embedded inside it near the top surface of the chip package. When the chip is inserted, its top contacts the sensor of the socket 3 through a tiny gap or an extremely thin thermal pad; Second, on the substrate 2, a temperature sensor is installed directly below or adjacent to each socket 3. A thermally conductive pillar (such as a copper pillar) or an elastic thermally conductive block is set above the sensor, with its top slightly protruding from the surface of the substrate 2. When the chip is inserted into the socket 3, its bottom (or the heat conducted up through the package) will contact the thermally conductive pillar, and the heat will be quickly conducted to the sensor below.

[0059] The tilt angle, surface curvature, and orientation relative to the chip of the air guide slope are definable to collaboratively shape differentiated flow fields for different chip regions.

[0060] It should be noted that conventional socket 3 and substrate 2 are connected and fixed by solder paste. When the pin density is too high, soldering can easily cause solder bridging, leading to abnormalities. This invention provides a support plate 6 on the bottom surface of substrate 2, and uses multiple screws 9 to pass through the support plate 6 and substrate 2 and lock them to the threaded holes at the four corners of socket 3, forming a stable mechanical connection structure. Due to the mechanical connection, a gasket 8 can be placed between substrate 2 and support plate 6 to isolate the electrical connection between the metal parts and substrate 2. Furthermore, the support plate 6 is made of metal materials such as AL6061 or S304, which can provide sufficient strength and rigidity to effectively resist the downward impact during chip loading and unloading, as well as high and low temperature cycling. The thermal stress generated prevents deformation of the substrate 2, thus solving the problem of easy loosening of traditional nut connections. This further prevents abnormal signal connection of the socket 3, ensuring that the socket 3 and substrate 2 maintain good contact at all times. In the aging test system, the aging test boards are arranged in a shelf format, with multiple boards placed at once. A high and low temperature test chamber provides the required temperature. Temperature uniformity of all chips in the test chamber is a key indicator affecting the test results. Air at a certain temperature blows over the aging test boards, but because the socket 3 is recessed, the air cannot effectively exchange heat with the chips. This invention simultaneously provides a structural form that helps improve heat exchange between the air and the chips. Figure 3 As shown, when the air flows from left to right, a portion of it blows onto the guide slope on the support plate 6. Guided by the guide slope, it blows directly downwards onto the chip surface, generating effective heat exchange, further improving the chip temperature uniformity and enhancing the test results.

[0061] It's important to further clarify that chip aging test facilities need to test chips of different sizes, power consumptions, and package types. A fixed airflow ramp design might be optimal for chip A, but ineffective for chip B. Since the chip's power consumption is the primary factor determining heat dissipation requirements, the airflow ramp needs to be optimized based on the amount of heat dissipation. In high-power server chips, which generate significant heat and require robust cooling capabilities, a steeper ramp angle and a larger airflow area are preferred. A steeper angle allows airflow to impact the chip surface more vertically and at a higher speed, quickly removing a large amount of heat and preventing overheating. Temperature rise deviations can occur because low-power IoT communication chips generate relatively little heat. Excessive airflow might actually cause the chip temperature to drop too low or produce unnecessary temperature fluctuations. Therefore, a gentler tilt angle and smoother curvature are needed. This design allows for a more even and gentle distribution of airflow to the chip surface and surrounding area, achieving precise temperature control and ensuring a uniform temperature across the entire chip surface. Furthermore, different high and low temperature test chambers vary in fan power, duct design, outlet location, and uniformity. To adapt to the airflow characteristics of the test chamber itself, different guide slopes need to be designed. For example, in a high-speed circulating test chamber with high wind speed and pressure, directly blowing air onto the chip might... Turbulence and rebound can occur, affecting heat exchange efficiency. Therefore, the curvature design of the guide slope becomes crucial. A fluid dynamics-optimized surface can rectify the flow, transforming high-speed turbulence into a smooth, uniform downwash airflow, ensuring effective heat exchange with the chip. For test chambers with weak wind speeds or poor uniformity, the guide slope needs to more efficiently capture and utilize limited airflow. In this case, a larger tilt angle and a smoother surface can reduce wind resistance, directing as much airflow as possible to the chip and compensating for insufficient system airflow. Furthermore, to match different substrate 2 layouts and BIB designs, the layout of the substrate 2 under test and the overall stacking density of the BIB directly affect the smoothness of airflow. When multiple BIBs are closely arranged on the shelf, airflow will be obstructed between the boards, which can easily create dead corners. Therefore, it is necessary to design special composite angles or asymmetrical curved surfaces for each support board 6. For example, the slope not only guides the airflow downwards, but also has a slight lateral angle to guide the airflow to the gap between adjacent BIBs, improve the overall airflow distribution, and enhance the temperature uniformity of all chips on the entire shelf. For tall components on the substrate 2, if there are tall capacitors, heat sinks, or other components around the chip, they will block the airflow. The design of the airflow guide slope needs to avoid these components, or generate vortices through a specific curvature to throw the airflow to the back of the components, ensuring that all parts of the chip can be blown.In some tests that require rapid heating and cooling or reaching extreme temperatures, the heat exchange efficiency is extremely high. In order to achieve an extremely fast temperature change rate, it is necessary to maximize the heat exchange efficiency. In this case, a design with the maximum tilt angle is adopted, combined with the measure of filling the space between the support plate 6 and the substrate 2 with thermally conductive adhesive. The thermally conductive adhesive can quickly conduct the heat of the chip to the entire support plate 6, while the optimized air guide slope can efficiently cool or heat the support plate 6 and the chip, forming an efficient active thermal management system.

[0062] This invention utilizes a definable airflow guide slope (definable here means that different specifications of airflow guide slopes can be defined and replaced for different test tasks; if the support plate 6 and the airflow guide slope are integrated, then the support plate 6 can be replaced; alternatively, the airflow guide slope can be designed as an independent replaceable module). This allows the support plate 6 to be optimally adapted to the heat dissipation requirements of different products by designing different airflow guide slopes or designing replaceable and adjustable airflow guide slopes, thus achieving hardware flexibility and reconfigurability. In contrast, a fixed, homogeneous design would limit the optimization space of the algorithm to wind speed distribution. The definable solution allows the optimization space of the algorithm and its underlying design method to be expanded to the level of flow field morphology, achieving hardware and software synergy.

[0063] In one embodiment of the present invention, an insulating and thermally conductive layer is provided at the contact point between the support plate and the substrate to enhance heat transfer and improve temperature uniformity.

[0064] It should be noted that the insulating and thermally conductive layer is actually a thermally conductive adhesive, which can enhance heat transfer and promote temperature uniformity.

[0065] In one embodiment of the present invention, the support plate is a perforated grid structure.

[0066] It should be noted that the perforated grid structure can reduce the difficulty of processing and reduce weight, further reducing the corresponding load.

[0067] In one embodiment of the present invention, a multi-fan array under independent control is also included;

[0068] Each fan in the multi-fan array is independently aligned with one or more of the air guide slope inlets;

[0069] Each fan is controlled by an independent signal to achieve independent adjustment of the inlet velocity of each guide slope;

[0070] The fan array, the air guide slope, and the chip layout on the substrate together constitute a software-definable dynamic flow field shaping hardware foundation.

[0071] It should be noted that the above solution allows the present invention to flexibly adjust the flow field according to requirements, so as to better adapt to the heat dissipation requirements in chip aging test scenarios.

[0072] like Figure 4 As shown, an intelligent heat dissipation method for a chip aging test board, applied to a chip aging test board, includes the following steps:

[0073] S1. Establish a computational fluid dynamics BIB parameterized model based on the geometric parameters of the substrate, support plate and air guide slope;

[0074] S2, Based on the BIB parameterized model, a dataset of inclined plane flow velocity-temperature rise-vibration mapping relationship is generated through CFD simulation;

[0075] S3. A lightweight neural network prediction model is trained using the slope velocity-temperature rise-vibration mapping relationship dataset. The input of the neural network prediction model is the current inlet wind speed of the wind guide slope, and the output is the predicted temperature distribution field of all chips and the vibration prediction value of all substrates.

[0076] S4 obtains the real-time temperature distribution field of all chips through a temperature sensor network and obtains the real-time vibration value of the substrate through a vibration sensor.

[0077] S5. Based on the real-time temperature distribution field and real-time vibration value, the lightweight neural network prediction model is invoked, and the optimal target wind speed combination is solved through a multi-objective optimization algorithm.

[0078] S6, convert the optimal target wind speed combination into PWM control signals for each wind turbine in the multi-wind turbine array, and drive its execution.

[0079] As described in steps S1-S6 above, in the actual scenario of chip aging test, the tested chips have different properties and will generate different temperatures, resulting in temperature differences in the chip temperature distribution field. Moreover, when testing chips of different sizes, power consumptions and packaging forms, the fluctuations in temperature distribution are more complex. Therefore, this invention uses an intelligent solution that takes into account both temperature control and vibration management to achieve more precise airflow regulation to balance chip temperature, ensure the accuracy of aging test, and at the same time avoid excessive vibration caused by fan operation, thereby improving the long-term operational reliability of the equipment.

[0080] In existing technologies, to address the problem of dynamically changing relative hot and cold spots caused by the complex thermal field distribution of chips on the test board due to individual power consumption differences and positional effects during chip aging testing, a strategy of unified PID control of the fan is adopted. This strategy not only has a lag in response but also struggles to cope with the dynamic requirements of different chip layouts and different testing stages. Furthermore, it does not consider the impact of vibration on the test system, which may lead to problems such as local chip overheating and excessive equipment vibration during testing, affecting the accuracy of test results and equipment lifespan. In contrast, this invention combines CFD simulation modeling, lightweight neural network prediction, and multi-objective optimization algorithms to accurately match hardware characteristics to optimize the flow field and respond to temperature and vibration changes in real time. This solves the problems of lag in response and insufficient temperature control capability of traditional solutions.

[0081] Specifically, firstly, a computational fluid dynamics (CFD) parametric model is established based on the geometric parameters of the substrate, support plate, and guide slope. The geometric parameters of the substrate, support plate, and guide slope are obtained from the design drawings of the actual test board, including the dimensions of the substrate, the perforated grid structure parameters of the support plate, and the inclination angle and surface curvature of the guide slope. By converting these physical parameters into model parameters recognizable by CFD simulation, the airflow and heat exchange characteristics of the test board can be accurately reflected. Prior to this step, the following steps are also included:

[0082] A1, obtain the layout of the chip particles on the test board and the expected power consumption distribution;

[0083] A2, based on the layout and expected power consumption distribution, with the goal of improving the temperature uniformity of chip particles, the geometric parameters of each air guide slope on the support plate are optimized and calculated. The geometric parameters include at least the tilt angle, surface curvature radius and azimuth angle.

[0084] A3. Based on the geometric parameters obtained from the optimization calculation, determine or manufacture the corresponding air guide slope.

[0085] By optimizing the hardware configuration through the above steps to lay the foundation for the optimal flow field, and then optimizing the control parameters online through this solution to achieve precise dynamic adjustment, a deep collaboration between hardware and software at the design level, not just the control level, is achieved. This allows the solution to optimize the flow field configuration through algorithms, resulting in a more uniform temperature distribution during the test.

[0086] Then, based on the BIB parametric model, a mapping dataset of inclined plane flow velocity-temperature rise-vibration relationship is generated through CFD simulation. In the CFD simulation process, different inlet wind speed conditions of the wind guide inclined plane are first set. Under each wind speed condition, the temperature rise data of the corresponding chip (i.e., the difference between the chip temperature and the ambient temperature) and the vibration data of the substrate (obtained through the vibration transmission characteristics generated by the fan operation in the simulation) are calculated by the simulation software. The inclined plane flow velocity, chip temperature rise and substrate vibration data under different wind speeds are correlated to form a mapping dataset. Through a limited number of CFD simulations of typical conditions, data covering multiple operating states can be quickly obtained without a large number of physical experiments.

[0087] Next, a lightweight neural network prediction model is trained using the aforementioned mapping dataset. The model input is the current inlet wind speed of the guide slope, and the output is the predicted temperature distribution field of all chips and the vibration prediction value of the substrate. The lightweight neural network prediction model can reduce computational overhead while ensuring prediction accuracy, adapt to rapid prediction in subsequent real-time testing, and improve overall control efficiency.

[0088] Next, in the real-time data acquisition stage, the real-time temperature distribution field of all chips is obtained through a temperature sensor network. The temperature sensor network adopts a distributed integration method, with at least one temperature sensor placed near each chip. The sensors transmit the acquired temperature data to the data processing module through the I2C or SPI communication protocol. At the same time, the real-time vibration value of the substrate is obtained through vibration sensors. The vibration sensors are installed on the edge of the substrate or at key positions on the support plate to collect the vibration acceleration data of the substrate during the test.

[0089] Next, based on the real-time temperature distribution field and real-time vibration values, the trained lightweight neural network prediction model is invoked, and the optimal target wind speed combination is solved through a multi-objective optimization algorithm. This step can take into account both temperature and vibration objectives, avoiding the imbalance caused by single-objective optimization. For example, during the chip heating stage, temperature uniformity is prioritized, allowing vibration to fluctuate within a low range. During the temperature stabilization stage, the focus is on reducing vibration to ensure long-term stable operation of the equipment.

[0090] Finally, the optimal target wind speed combination is converted into PWM control signals for each fan in the multi-fan array and driven to execute. In specific implementation, a mapping relationship between wind speed and PWM signal is first established. Based on this mapping relationship, each wind speed value in the optimal wind speed combination is converted into a corresponding PWM duty cycle. Then, the PWM signal is sent to each fan through the drive circuit to control the fan to run at the target wind speed. After the fan runs, the cooling airflow in the corresponding area is increased, thereby reducing the temperature of the corresponding chip, realizing intelligent heat dissipation, and making the temperature distribution more uniform during the test.

[0091] In one embodiment of the present invention, the step of training a lightweight neural network prediction model using the inclined plane flow velocity-temperature rise-vibration mapping dataset includes:

[0092] S31, Perform standardization preprocessing on the inclined surface flow velocity-temperature rise-vibration mapping relationship dataset to obtain a standardized dataset;

[0093] S32, construct a neural network with multiple output branches according to the input and output dimensions of the standardized dataset, set the number of neurons in the input layer to correspond to the number of wind guide slopes, and set the number of neurons in the two output layers to correspond to the number of monitored chips and the number of substrates, respectively.

[0094] S33. Based on the standardized dataset, using mean squared error as the loss function, the neural network is trained through backpropagation until its prediction accuracy on the preset validation set reaches a preset threshold to obtain a lightweight neural network prediction model.

[0095] As described in steps S31-S33 above, the present invention processes and learns the obtained inclined plane flow velocity-temperature rise-vibration mapping relationship dataset to obtain a lightweight neural network prediction model that specifically predicts the chip temperature distribution field and substrate vibration values.

[0096] The input and output dimensions of the standardized dataset are the number of wind-guiding slopes. For example, if there are 6 wind-guiding slopes on the test board, the input layer needs to be set with 6 neurons, each neuron corresponding to the standardized wind speed data of one wind-guiding slope. The output dimension is divided into two branches: one branch corresponds to the number of monitored chips, used to output the standardized predicted temperature of each chip; the other branch corresponds to the number of substrates, used to output the standardized predicted vibration value of the substrates. The hidden layer of the neural network is set to two layers. The number of neurons in the first hidden layer can be set to 2-3 times the number of neurons in the input layer, and the number of neurons in the second hidden layer is set to 1 / 2 of the number in the first layer. The hidden layers all use the ReLU activation function to enhance the model's ability to fit nonlinear relationships and avoid the gradient vanishing problem. The output layer uses a linear activation function for the temperature prediction branch and also uses a linear activation function for the vibration prediction branch.

[0097] In the step of training a lightweight neural network prediction model by using the mean squared error as the loss function based on a standardized dataset and training the neural network structure through the backpropagation algorithm until its prediction accuracy on the preset validation set reaches a preset threshold, the specific form of the loss function is the mean squared error. For the temperature prediction branch, the sum of the squares of the differences between the predicted temperature and the actual temperature of each chip is calculated, and then the average value is taken. For the vibration prediction branch, the square of the difference between the predicted vibration value and the actual vibration value (standardized vibration data from CFD simulation) is calculated. Finally, the mean squared errors of the two branches are added together with a 1:1 weight to obtain the total loss function.

[0098] In one embodiment of the present invention, the step of calling the lightweight neural network prediction model based on the real-time temperature distribution field and real-time vibration values, and solving for the optimal target wind speed combination through a multi-objective optimization algorithm includes:

[0099] S51, Obtain the temperature prediction function and vibration prediction function according to the lightweight neural network prediction model;

[0100] S52, Construct a comprehensive performance index based on the temperature prediction function and the vibration prediction function;

[0101] S53, obtain the chip temperature safety constraint and the fan speed feasible region constraint, with the goal of minimizing the comprehensive performance index and the condition of satisfying the chip temperature safety constraint and the fan speed feasible region constraint, construct a constrained optimization problem;

[0102] S54 employs a fast quadratic programming method based on the sensitivity matrix to iteratively solve the constrained optimization problem and output the optimal target wind speed combination.

[0103] As described in steps S51-S54 above, due to the coupling relationship between temperature and vibration in chip aging tests, increasing the fan speed can enhance heat dissipation to balance the chip temperature difference, but it may lead to increased substrate vibration. Decreasing the fan speed can reduce vibration, but it may cause local chip overheating. Moreover, the priority of temperature and vibration requirements differs in different test stages. For example, in the heating stage and the stabilization stage, the heating stage needs to prioritize ensuring that the temperature reaches the target value quickly to shorten the test time, while the stabilization stage needs to focus on suppressing vibration to ensure the long-term reliability of the equipment. Therefore, this invention is based on a lightweight neural network prediction model, combined with real-time collected temperature and vibration data, to construct and solve a constrained optimization problem that takes into account both temperature uniformity and vibration suppression. Finally, it outputs the optimal target fan speed combination that is suitable for the current test conditions. Compared with the existing optimization strategy that only targets temperature, which is difficult to adapt to dynamic conditions, this invention can ensure that the temperature field is stable and the system vibration is within a reasonable range during chip aging tests.

[0104] The temperature prediction function refers to the function obtained by inputting the current inlet wind speed of the guide slope into the lightweight neural network prediction model, and then outputting the branch corresponding to "predicted temperature distribution field of all chips" at the model's output. The predicted temperature of each chip is a function of wind speed. The vibration prediction function refers to the function of the predicted vibration value of the substrate and wind speed obtained by the branch corresponding to the "predicted vibration value of the substrate" at the model output. The specific steps to obtain these two functions are: by calling the trained lightweight neural network prediction model, the forward propagation process of the model is encapsulated into an independent function interface at the code level, which returns the predicted temperature distribution field of all chips (corresponding to the output of the temperature prediction function) and the predicted vibration value of the substrate (corresponding to the output of the vibration prediction function). After the prediction ability of the lightweight neural network prediction model is transformed into a mathematical function, the sensitivity of the function to the inlet wind speed of the guide slope can be calculated by the automatic differentiation algorithm, that is, the degree of influence of wind speed change on chip temperature and substrate vibration.

[0105] In the process of obtaining chip temperature safety constraints and fan speed feasible region constraints, and constructing a constrained optimization problem with the goal of minimizing the overall performance index while satisfying the above two types of constraints, the core of the chip temperature safety constraint is to avoid chip overheating and damage. Specifically, the predicted temperature of all monitored chips should not exceed their critical safety temperature, which is determined according to the chip datasheet. The core of the fan speed feasible region constraint is to avoid heat dissipation failure in the multi-fan array due to fan overload operation or no-wind conditions. Specifically, the inlet velocity of each guide slope should not be less than 0 (the lower limit of fan shutdown) and not greater than the rated maximum velocity of the fan, which is determined according to the fan model. The parameters of the above constraints are obtained by combining hardware characteristics and testing standards. For example, the chip critical safety temperature is obtained by consulting the technical documentation of the chip under test, and the rated maximum velocity of the fan is determined by the technical specifications provided by the fan supplier. The specific form of constructing the optimization problem is: "the inlet velocity of each guide slope ≥ 0 and ≤ the rated maximum velocity of the fan" and "the predicted temperature of all monitored chips ≤ the chip critical safety temperature". Under the constraints, solve for the inlet wind speed combination of the wind guide slope to minimize the comprehensive performance index.

[0106] In one embodiment of the present invention, the step of constructing a comprehensive performance index based on the temperature prediction function and the vibration prediction function includes:

[0107] S521, determine the priority of temperature control requirements and the priority of vibration suppression requirements based on the real-time temperature distribution and the real-time vibration value;

[0108] S522, dynamically adjust the weighting coefficients in the overall system performance index according to the priority of the temperature control requirements and the priority of the vibration suppression requirements;

[0109] S523, Based on the weighting coefficients, and combining the temperature prediction function and the vibration prediction function, a comprehensive performance index is constructed.

[0110] As described in steps S521-S523 above, since the operating conditions are not constant throughout the chip aging test process, and the temperature changes rapidly, it is necessary to control the temperature of all chips to the target temperature as soon as possible. At this time, even if the substrate vibration is slightly high, it is acceptable as long as it is within the safe range. The priority of temperature control is significantly higher than that of vibration suppression. In the stable stage, the chip temperature has been maintained near the target temperature. If high wind speed is maintained continuously, it will not only waste energy, but also cause the substrate to be in a high vibration level for a long time, accelerating mechanical fatigue. At this time, the priority of vibration suppression is correspondingly increased. Therefore, this invention dynamically determines the priority of temperature control and vibration suppression by collecting real-time temperature distribution and real-time vibration values, and then adjusts the weight coefficient of temperature uniformity and vibration suppression in the comprehensive performance index, so as to adapt to the comprehensive performance index of the current test conditions.

[0111] Specifically, the priority of temperature control and vibration suppression needs is first determined based on real-time temperature distribution and real-time vibration values. This step quantifies real-time operating data to determine whether the primary need at the current stage is temperature control or vibration suppression. The specific logic is as follows: First, the deviation between the real-time temperature distribution and the target temperature is calculated. The "average absolute deviation between the real-time temperatures of all chips and the corresponding target temperatures" can be used as a metric. If the metric is greater than the deviation threshold, it is determined that the current stage is a critical heating phase, and the priority of temperature control needs is higher than that of vibration suppression needs. If the metric is less than the deviation threshold, the real-time vibration values ​​are further considered. If the real-time vibration values ​​are greater than the safety threshold, the priority of vibration suppression needs is higher than that of temperature control needs. If the real-time vibration values ​​are less than the safety threshold and the metric is less than the deviation threshold, the priority of vibration suppression needs is higher.

[0112] Then, the weighting coefficients in the system's overall performance index are dynamically adjusted based on the priority of temperature control and vibration suppression requirements. By adjusting the values ​​of the temperature uniformity weighting coefficient and the vibration suppression weighting coefficient, priority differences are transformed into weight differences. The weighting coefficients are adjusted according to the principle that higher priority targets correspond to higher weights. Specifically, when the priority of temperature control is higher than that of vibration suppression, the temperature deviation item has a higher proportion in the overall performance index; when the priority of vibration suppression is higher than that of temperature control, the vibration item has a higher proportion in the overall performance index; when the priorities of the two are equal, they are considered in a balanced manner. It should be noted that the specific values ​​of the weighting coefficients need to be calibrated based on actual test results. For example, in actual chip testing, if it is found that the heating rate is still not up to standard when the temperature uniformity weighting coefficient is 0.8, the temperature uniformity weighting coefficient can be adjusted to 0.9 to further increase the weighting of temperature control, thereby making the optimization direction more in line with actual needs.

[0113] Finally, based on the weighting coefficients and combining the temperature prediction function and vibration prediction function, a comprehensive performance index is constructed, as shown in the formula:

[0114] ;

[0115] In the formula, Indicates comprehensive performance indicators, The temperature uniformity weighting coefficient represents the degree of importance attached to the temperature control target and is dynamically adjusted according to the testing phase: a larger value is set during the heating phase and a smaller value is set during the stabilization phase. This represents the vibration suppression weighting coefficient (which dynamically adjusts according to the testing phase, setting a small value during the heating phase and increasing it during the stabilization phase). This indicates the number of chips being monitored. Indicates the first Predicted temperature of each chip, Indicates the first The target temperature of each chip The predicted vibration value of the substrate is used to construct comprehensive performance indicators, and the temperature component needs to be considered. and vibration terms Normalize the data to convert it into a dimensionless value.

[0116] Through the above scheme, the present invention can simultaneously optimize the temperature and vibration conditions during the testing process, and by incorporating the dynamic weighting coefficient into the comprehensive performance index, it can more accurately reflect the core requirements of the current working condition in real time, making the guidance of the comprehensive performance index more precise.

[0117] In one embodiment of the present invention, the step of iteratively solving the constrained optimization problem using a fast quadratic programming method based on the sensitivity matrix and outputting the optimal target wind speed combination includes:

[0118] S541, Obtain the real-time wind speed combination. Based on the lightweight neural network prediction model and the real-time wind speed combination, obtain the sensitivity matrix of temperature distribution field and vibration level to wind speed through automatic differentiation algorithm.

[0119] S542, Based on the sensitivity matrix, the lightweight neural network prediction model is expanded using a first-order Taylor series in real-time wind speed combination to construct a linearized prediction model.

[0120] S543, construct a quadratic programming problem based on the linearized predicted temperature distribution field and vibration prediction values ​​of the linearized prediction model;

[0121] S544, the effective set method is used to solve the quadratic programming problem to obtain the optimal target wind speed combination.

[0122] Temperature linearization prediction formula for linearized prediction model: ;

[0123] In the formula, This represents the linearized predicted temperature distribution field. Represents the real-time temperature distribution field. The sensitivity matrix to temperature and wind speed is represented (obtained through automatic differentiation of a lightweight neural network prediction model). This represents the optimal target wind speed combination. Indicates real-time wind speed combinations;

[0124] Linearized prediction model vibration linearization prediction formula: ;

[0125] In the formula, This represents the linearized vibration prediction value. Indicates the real-time vibration value. The sensitivity matrix of vibration to wind speed is represented (obtained through automatic differentiation of a lightweight neural network prediction model). This represents the optimal target wind speed combination. 1 indicates the real-time wind speed combination;

[0126] The objective function of the quadratic programming problem is: ;

[0127] In the formula, This represents the optimal target wind speed combination. Represents the coefficient matrix of the quadratic term. Let T denote the coefficient vector of the linear term, and T denote the transpose operation.

[0128] Quadratic coefficient matrix: ;

[0129] In the formula, Represents the coefficient matrix of the quadratic term. This represents the weighting coefficient for temperature uniformity. This represents the sensitivity matrix of temperature to wind speed. This represents the sensitivity matrix of vibration to wind speed. The value represents the weighting coefficient for vibration suppression, and T represents the transpose operation.

[0130] As described in steps S541-S544 above, the nonlinear mapping relationship of the lightweight neural network prediction model leads to high difficulty in solving the optimization problem based on the model and long computation time, making it difficult to meet the millisecond-level response requirements of real-time control. However, near the current real-time wind speed combination, the small changes in wind speed can have an approximate linear relationship with temperature and vibration. Therefore, this invention quantifies this linear effect through a sensitivity matrix, which simplifies the complex nonlinear model into a linear model near the operating point, thereby reducing the difficulty of solving the optimization problem and enabling the multi-objective optimization process to quickly respond to changes in operating conditions in the real-time control scenario of chip aging test.

[0131] Specifically, the current real-time wind speed combination is first determined as the linearization operating point. The current real-time wind speed combination is the real-time inlet wind speed of each guide slope in the multi-fan array collected by the wind speed sensor. The significance of selecting the current real-time wind speed combination as the operating point is that in chip aging tests, wind speed adjustment is usually a small correction based on the current state to avoid drastic system fluctuations. Therefore, in the local range near the current wind speed, the linear approximation has high accuracy and can effectively reflect the actual impact of wind speed changes on temperature and vibration.

[0132] Then, based on the lightweight neural network prediction model, an automatic differentiation algorithm is used to calculate the sensitivity matrix of the temperature distribution field to wind speed and the sensitivity matrix of the vibration level to wind speed. This step can solve the derivative of the nonlinear model at the operating point using mathematical methods to obtain the influence coefficients of wind speed changes on temperature and vibration. The lightweight neural network prediction model is derived from the training results of the aforementioned steps and has been encapsulated as a function interface that can calculate derivatives. The automatic differentiation algorithm in the neural network model can be implemented through the automatic differentiation tool built into the framework. The calculation process is as follows: the current real-time wind speed combination is input into the lightweight neural network prediction model, and the automatic differentiation tool is used to calculate the partial derivatives of the temperature distribution field and vibration values ​​output by the model with respect to the wind speed of each wind guide slope. The obtained partial derivative matrix is ​​the sensitivity matrix of the temperature distribution field to wind speed and the sensitivity matrix of the vibration level to wind speed. The sensitivity matrix can be used to convert how wind speed changes affect temperature and vibration into a quantified coefficient matrix.

[0133] Next, based on the sensitivity matrix and the linearized operating point, a first-order Taylor expansion is used to construct the linearized prediction formulas for temperature and vibration. This step can improve optimization efficiency, simplifying the complex neural network forward propagation calculation into matrix multiplication and addition. While reducing the time required for a single prediction, sufficient prediction accuracy can be maintained.

[0134] Next, a quadratic programming problem is constructed based on the linearized predicted temperature distribution field and vibration prediction values ​​from the linearized prediction model. This quadratic programming problem mainly includes an objective function and constraints. The constraints inherit the chip temperature safety constraint and the fan speed feasible region constraint. Specifically, in the linearized predicted temperature distribution field, the temperature of all chips does not exceed the critical safety temperature, and in the optimal target wind speed combination, each wind speed is between 0 and the fan's rated maximum wind speed. The objective function is constructed based on the linearized prediction results and comprehensive performance indicators. The quadratic term coefficient matrix quantifies the quadratic impact of different wind speed adjustments on the comprehensive performance indicators. This allows the objective function to reflect the nonlinear relationship between wind speed and performance indicators (even if the model has been linearized, the objective function still needs to retain its quadratic form to ensure the convexity of the optimization); the coefficient vector of the first-order term is constructed based on the linearized prediction bias. Its core is to transform the target deviation of temperature and vibration into a linear constraint on wind speed adjustment, guiding the optimization direction towards the direction where the temperature approaches the target and the vibration approaches the minimum. Thus, through the standardized quadratic programming form, the multi-objective optimization problem can be transformed into a format that can be directly processed by the mathematical solver. At the same time, matrix operations are used to make the objective function more accurately reflect the demand priority of real-time operating conditions.

[0135] Finally, the effective set method is used to solve the quadratic programming problem, obtaining the optimal target wind speed combination. This method can quickly locate the wind speed combination that minimizes the overall performance index while satisfying the chip temperature safety and fan speed feasible region requirements. Therefore, the optimization result is both adapted to real-time operating conditions and conforms to the hardware operation safety boundary. In specific implementation, the effective constraint set is first initialized. The initial feasible solution can be selected from the real-time wind speed combination. Since the real-time wind speed combination is the actual wind speed of the current hardware operation, it must satisfy the fan speed feasible region constraint. Furthermore, through prior control, the real-time temperature distribution field has been ensured to meet safety requirements. Based on the linear model, the real-time wind speed can be inferred. The linearized predicted temperature corresponding to the speed combination also meets the safety constraints. The initial effective constraint set by default includes all "non-loose constraints" (such as some wind speeds approaching the rated maximum wind speed of the wind turbine, or some chip temperatures approaching the critical safety temperature, which need to be included in constraint management first). Then, the iterative solution stage begins. The first step is to solve the equality-constrained quadratic programming subproblem under the current effective constraint set. At this time, the inequality constraints in the effective constraint set need to be transformed into equality constraints. Combining the quadratic coefficient matrix and the linear coefficient vector of the objective function, the optimal trial solution is solved using the Lagrange multiplier method. The solution process will automatically "reduce temperature deviation and suppress vibration". The direction of the wind speed is tilted. If the temperature uniformity weight coefficient is high, the proportion of the term corresponding to temperature sensitivity in the quadratic term coefficient matrix will be higher, and the trial solution will prioritize adjusting the wind speed that has a significant impact on temperature. If the vibration suppression weight coefficient is high, it will guide the trial solution to reduce the adjustment of wind speed that has a large impact on vibration. The second step is to verify whether the trial solution satisfies all constraints (including relaxed constraints not included in the effective constraint set). If the trial solution satisfies all constraints, it means that the optimal solution under the current effective constraint set is the global optimal solution, which can be output as the optimal target wind speed combination. If the trial solution violates a relaxed constraint (such as a wind speed trial value exceeding the wind turbine's rated value), it will be considered a failure. If the maximum wind speed is set, or the predicted temperature of a chip after linearization exceeds the critical safety temperature, the violated constraint needs to be added to the effective constraint set, the subproblem needs to be solved again, and finally the iteration terminates. When the trial solution satisfies all constraints, and all constraints in the effective constraint set are "necessary constraints" (i.e. the optimal solution still satisfies the constraint after removing any constraint), the iteration terminates and the current trial solution is output as the optimal target wind speed combination. In this way, not only can the temperature distribution field and vibration value that meet the requirements be predicted through the linearization model, but also the wind speed can be kept within the hardware feasible range, thereby realizing the real-time control requirements of chip aging test.

[0136] like Figure 5 As shown, this invention also discloses an intelligent heat dissipation system for a chip aging test board, comprising:

[0137] The modeling module is used to establish a computational fluid dynamics BIB parametric model based on the geometric parameters of the substrate, support plate and air guide slope.

[0138] The CFD simulation module is used to generate a dataset of inclined flow velocity-temperature rise-vibration mapping relationship based on the BIB parametric model through CFD simulation.

[0139] The model training module is used to train a lightweight neural network prediction model using the inclined plane velocity-temperature rise-vibration mapping relationship dataset. The input of the neural network prediction model is the current inlet wind speed of the wind guide inclined plane, and the output is the predicted temperature distribution field of all chips and the vibration prediction value of all substrates.

[0140] The real-time acquisition module is used to acquire the real-time temperature distribution field of all chips through a temperature sensor network and to acquire the real-time vibration value of the substrate through a vibration sensor.

[0141] The optimization solution module is used to call the lightweight neural network prediction model based on the real-time temperature distribution field and real-time vibration value, and solve for the optimal target wind speed combination through a multi-objective optimization algorithm;

[0142] The drive control module is used to convert the optimal target wind speed combination into PWM control signals for each wind turbine in the multi-wind turbine array and drive its execution.

[0143] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, apparatus, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0144] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A chip aging test board, characterized in that, include: A substrate having sockets for mounting multiple chips to be tested; A support plate is fixedly installed on the bottom surface of the substrate to provide structural strength to prevent substrate deformation. An insulating and heat-conducting layer is provided at the contact point between the support plate and the substrate to enhance heat transfer and improve temperature uniformity. Multiple screws pass through the support plate and the base plate and are locked into the threaded holes at the four corners of the socket to achieve mechanical fixation; Multiple air guide slopes are disposed on the support plate and corresponding to the position of the socket to guide the cooling airflow toward the chip; A temperature sensor network, distributed and integrated on the substrate, is used to collect surface temperature data of the chip in real time. as well as An independently controlled multi-fan array, wherein each fan in the multi-fan array is independently aligned with one or more of the air guide slope inlets, and each fan is controlled by an independent signal to achieve independent adjustment of the air velocity at each air guide slope inlet; The tilt angle, surface curvature, and orientation relative to the chip of the air guide slope are definable, so as to collaboratively shape differentiated flow fields for different chip regions. The fan array, the air guide slope, and the chip layout on the substrate together constitute a software-definable dynamic flow field shaping hardware foundation for working in conjunction with an intelligent heat dissipation method for a chip aging test board.

2. The chip aging test board according to claim 1, characterized in that, The support plate is a perforated grid structure.

3. A smart heat dissipation method for a chip aging test board, characterized in that, The method applied to a chip aging test board according to any one of claims 1 includes the following steps: A computational fluid dynamics BIB parameterized model is established based on the geometric parameters of the substrate, support plate, and air guide slope. Based on the BIB parameterized model, a dataset of inclined plane flow velocity-temperature rise-vibration mapping relationship is generated through CFD simulation. A lightweight neural network prediction model is trained using the slope velocity-temperature rise-vibration mapping dataset. The input of the neural network prediction model is the current inlet wind speed of the wind guide slope, and the output is the predicted temperature distribution field of all chips and the vibration prediction value of all substrates. The real-time temperature distribution field of all chips is obtained through a temperature sensor network, and the real-time vibration value of the substrate is obtained through a vibration sensor. Based on the real-time temperature distribution field and real-time vibration values, the lightweight neural network prediction model is invoked, and the optimal target wind speed combination is solved through a multi-objective optimization algorithm. The optimal target wind speed combination is converted into PWM control signals for each wind turbine in the multi-turbine array and then driven to execute.

4. The intelligent heat dissipation method for a chip aging test board according to claim 3, characterized in that, The steps for training a lightweight neural network prediction model using the inclined plane flow velocity-temperature rise-vibration mapping dataset include: The slope velocity-temperature rise-vibration mapping dataset is preprocessed to obtain a standardized dataset. A neural network with multiple output branches is constructed based on the input and output dimensions of the standardized dataset. The number of neurons in the input layer is set to correspond to the number of wind guide slopes, and the number of neurons in the two output layers is set to correspond to the number of monitored chips and the number of substrates, respectively. Based on the standardized dataset, the mean squared error is used as the loss function, and the neural network is trained through the backpropagation algorithm until its prediction accuracy on the preset validation set reaches a preset threshold to obtain a lightweight neural network prediction model.

5. The intelligent heat dissipation method for a chip aging test board according to claim 3, characterized in that, The step of calling the lightweight neural network prediction model based on the real-time temperature distribution field and real-time vibration values, and solving for the optimal target wind speed combination through a multi-objective optimization algorithm includes: The temperature prediction function and vibration prediction function are obtained based on the lightweight neural network prediction model. A comprehensive performance index is constructed based on the temperature prediction function and the vibration prediction function. Obtain chip temperature safety constraints and fan speed feasible region constraints, aim to minimize the comprehensive performance index, and construct a constrained optimization problem under the condition of satisfying chip temperature safety constraints and fan speed feasible region constraints; A fast quadratic programming method based on the sensitivity matrix is ​​used to iteratively solve the constrained optimization problem and output the optimal target wind speed combination.

6. The intelligent heat dissipation method for a chip aging test board according to claim 5, characterized in that, The steps for constructing a comprehensive performance index based on the temperature prediction function and the vibration prediction function include: The priority of temperature control requirements and the priority of vibration suppression requirements are determined based on the real-time temperature distribution and the real-time vibration value. The weighting coefficients in the overall system performance index are dynamically adjusted according to the priority of the temperature control requirements and the priority of the vibration suppression requirements. Based on the weighting coefficients, and combined with the temperature prediction function and vibration prediction function, a comprehensive performance index is constructed.

7. The intelligent heat dissipation method for a chip aging test board according to claim 5, characterized in that, The steps of iteratively solving the constrained optimization problem using a fast quadratic programming method based on the sensitivity matrix and outputting the optimal target wind speed combination include: The real-time wind speed combination is obtained, and the sensitivity matrix of temperature distribution field and vibration level to wind speed is obtained by automatic differentiation algorithm based on the lightweight neural network prediction model and the real-time wind speed combination. Based on the sensitivity matrix, the lightweight neural network prediction model is expanded using a first-order Taylor series in real-time wind speed combinations to construct a linear prediction model. A quadratic programming problem is constructed based on the linearized predicted temperature distribution field and vibration prediction values ​​of the linearized prediction model. The optimal target wind speed combination is obtained by solving the quadratic programming problem using the effective set method.

8. An intelligent heat dissipation system for a chip aging test board, characterized in that, It includes multiple modules for implementing the steps of the method according to any one of claims 3 to 7.

Images