Intelligent temperature control method and device of multi-chip stacking structure and computer equipment

Abstract

The invention relates to the technical field of chip stacking, and discloses an intelligent temperature control method and device of a multi-chip stacking structure and computer equipment. The method comprises the following steps: acquiring target temperature data of each layer of chip in a multi-chip stacking structure, and establishing a nonlinear temperature response model; constructing a hierarchical control framework, and inputting the target temperature data into an agent auxiliary temperature control algorithm for processing to obtain temperature distribution characteristic data; performing optimal sampling analysis to obtain local temperature regulation and control parameters; constructing a temperature-flow multi-stage coordination control model, and generating a cooling liquid flow distribution scheme and a flow direction control instruction; the flow and temperature of cooling liquid are controlled, gradient cooling treatment is conducted on the temperature abnormal area, and temperature adjusting process data are obtained; and carrying out multistable distribution analysis to generate a temperature control optimization strategy. According to the invention, the problem of temperature balance in a multi-chip stacking structure is effectively solved.

Description

Technical Field

[0001] The present invention relates to the field of chip stacking technology, and in particular to an intelligent temperature control method, device and computer equipment for a multi-chip stacking structure. Background Art

[0002] Multi-chip stacking structures are widely used in modern electronic devices due to their advantages such as high integration, low power consumption and miniaturization. However, the three-dimensional stacking of multiple chips leads to a sharp increase in the heat density per unit volume, making it difficult to effectively dissipate the heat, and easily forming local hot spots inside the chip, which seriously affects the stability and reliability of the system.

[0003] Traditional multi-chip stacking heat dissipation methods mainly rely on fixed heat dissipation structures and simple temperature threshold control. They cannot accurately adjust the temperature change characteristics of different regions, and lack the overall grasp and dynamic response capabilities of the temperature field distribution. Especially in high-performance computing scenarios, the dynamic changes in chip loads lead to diverse and uncertain temperature distributions, and traditional temperature control methods are difficult to meet the needs of precise temperature control. Summary of the invention

[0004] The present invention provides an intelligent temperature control method, device and computer equipment for a multi-chip stacking structure, which effectively solves the temperature balance problem in the multi-chip stacking structure.

[0005] In a first aspect, the present invention provides an intelligent temperature control method for a multi-chip stacking structure, the intelligent temperature control method for a multi-chip stacking structure comprising: Collect the target temperature data of each layer of chips in the multi-chip stacking structure and establish a nonlinear temperature response model; Building a hierarchical control framework based on the nonlinear temperature response model, and inputting the target temperature data into the agent-assisted temperature control algorithm for processing to obtain temperature distribution characteristic data; Performing optimal sampling analysis on the temperature distribution characteristic data to obtain local temperature control parameters; Building a temperature-flow multi-level coordinated control model based on the local temperature control parameters, and generating a coolant flow distribution plan and flow direction control instructions; Based on the coolant flow distribution scheme and the flow direction control instruction, the coolant flow and temperature are controlled, and a gradient cooling process is performed on the temperature abnormal area to obtain temperature adjustment process data; A multi-stable distribution analysis is performed on the temperature regulation process data, the region identification parameters of the agent-assisted temperature control algorithm and the coordination parameters of the hierarchical control framework are updated, and a temperature control optimization strategy is generated.

[0006] In a second aspect, the present invention provides an intelligent temperature control device of a multi-chip stacking structure, the intelligent temperature control device of the multi-chip stacking structure comprising: An acquisition module is used to acquire target temperature data of each layer of chips in a multi-chip stacking structure and establish a nonlinear temperature response model; A construction module is used to construct a hierarchical control framework based on the nonlinear temperature response model, and input the target temperature data into the agent-assisted temperature control algorithm for processing to obtain temperature distribution characteristic data; An analysis module, used for performing optimal sampling analysis on the temperature distribution characteristic data to obtain local temperature control parameters; A generation module, used to construct a temperature-flow multi-level coordinated control model based on the local temperature control parameters, and generate a coolant flow distribution plan and a flow direction control instruction; A processing module, used to control the coolant flow rate and temperature based on the coolant flow distribution scheme and the flow direction control instruction, and to perform gradient cooling on the temperature abnormal area to obtain temperature adjustment process data; An updating module is used to perform multi-stable distribution analysis on the temperature adjustment process data, update the region identification parameters of the agent-assisted temperature control algorithm and the coordination parameters of the hierarchical control framework, and generate a temperature control optimization strategy.

[0007] The third aspect of the present invention provides a computer device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the computer device executes the above-mentioned intelligent temperature control method for a multi-chip stacking structure.

[0008] A fourth aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned intelligent temperature control method for a multi-chip stacking structure.

[0009] In the technical solution provided by the present invention, by setting temperature sensors with different characteristics, using sinusoidal characteristic sensors to capture periodic temperature changes, and using hyperbolic tangent characteristic sensors to monitor sudden temperature anomalies, the comprehensiveness and accuracy of temperature monitoring are significantly improved. A hierarchical control framework combined with an agent-assisted temperature control algorithm is adopted to achieve the coordinated cooperation of global temperature monitoring and local precise adjustment, and effectively solve the temperature balance problem in the multi-chip stacking structure. Minkowski distance calculation and K-means clustering analysis are introduced to accurately locate and grade the temperature anomaly area, providing a reliable data basis for subsequent precise temperature control. A multi-level coordinated control model based on heat dissipation efficiency is established, and differentiated control of different temperature areas is achieved by adaptively adjusting the coolant flow and flow direction. A step-type flow control strategy is adopted to avoid drastic fluctuations in the coolant flow and ensure the stability of the temperature adjustment process. Through multi-stable distribution analysis and dynamic parameter optimization, the adaptive update of control parameters is achieved, and the system's adaptability to complex working conditions is improved. A complete closed-loop optimization mechanism for temperature control is constructed, and the long-term stability and reliability of system control are guaranteed through real-time monitoring and parameter adjustment. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0011] Figure 1 A schematic diagram of a flow chart of an intelligent temperature control method for a multi-chip stacking structure provided in an embodiment of the present application; Figure 2 A schematic block diagram of the structure of an intelligent temperature control device with a multi-chip stacking structure provided in an embodiment of the present application; Figure 3 A schematic block diagram of the structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0012] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0013] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change based on actual conditions.

[0014] It should also be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0015] It should be further understood that the term “and / or” used in the specification and appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0016] In conjunction with the accompanying drawings, some embodiments of the present application are described in detail below. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0017] See also Figure 1 , Figure 1 A schematic diagram of a flow chart of an intelligent temperature control method for a multi-chip stacking structure provided in an embodiment of the present application, such as Figure 1 As shown, the intelligent temperature control method of a multi-chip stacking structure provided in an embodiment of the present application includes steps S100 to S600.

[0018] Step S100, collecting target temperature data of each layer of chips in a multi-chip stacking structure, and establishing a nonlinear temperature response model; It is understandable that the execution subject of the present invention may be a multi-chip stacking structure intelligent temperature control device, or a terminal or a server, which is not limited here. The embodiment of the present invention is described by taking a server as the execution subject as an example.

[0019] Specifically, each layer of the multi-chip stacking structure is grid-divided, and multiple sampling points are defined on each layer of the chip according to the division rule to provide spatial distribution information. In the arrangement of sampling points, in order to improve the accuracy of temperature acquisition and the sensitivity of response, temperature sensors with different characteristics are used. The sine characteristic temperature sensor is arranged in the central area with a spacing of D between adjacent sampling points to sample and monitor the overall temperature change trend of the chip layer. At the same time, in the edge area of ​​each layer of the chip, a hyperbolic tangent characteristic temperature sensor is arranged, and the spacing between adjacent sampling points is D / 2, which is denser than the central area, so as to better capture the subtle changes in the temperature of the edge area and form a complete temperature sensing network topology. In this temperature sensing network topology, the sine characteristic temperature sensor obtains the initial temperature data by temperature periodic sampling. These data are used to reflect the temperature change characteristics of each area of ​​the chip over time. The hyperbolic tangent characteristic temperature sensor is used for mutation detection configuration, which obtains mutation detection data reflecting the abnormal temperature state by capturing the drastic change of temperature. The initial temperature data and mutation detection data are sorted and classified and labeled to obtain a preprocessed temperature sequence. The preprocessed temperature sequence is decomposed by wavelet multiscale to extract the temperature information at different scales in the multi-chip stacking structure. Wavelet multiscale decomposition is an effective signal processing technology that decomposes the temperature signal at different time scales to reveal the details and overall trend of temperature changes and obtain the target temperature data. The target temperature data is subjected to principal component analysis to reduce the dimension of the data and extract key temperature features to obtain the target temperature data. The temperature feature vector is mapped to a nonlinear function, and the nonlinear temperature response model is constructed using the sine function and the hyperbolic tangent function respectively. The sine function can describe the periodic change characteristics of temperature, while the hyperbolic tangent function can describe the asymptotic change and mutation behavior of temperature. The combination of the two functions can more comprehensively reflect the nonlinear response characteristics of temperature in the chip stacking structure.

[0020] Step S200, constructing a hierarchical control framework based on a nonlinear temperature response model, and inputting target temperature data into an agent-assisted temperature control algorithm for processing to obtain temperature distribution characteristic data; Specifically, a hierarchical control framework is established based on the nonlinear temperature response model, which includes a global temperature monitoring unit and a local control unit. The global temperature monitoring unit, as the master control layer, performs global temperature perception and monitoring of the entire multi-chip stacking structure, while the local control unit, as the slave control layer, mainly fine-tunes and responds to the temperature of local hot spots and abnormal areas. The master-slave hierarchical structure enables global and local control to work together, thereby achieving precise control of the temperature of the multi-chip stack. In order to effectively coordinate the operations between the master control layer and the slave control layer, the control parameters of the hierarchical control framework are configured, and the response time of the global temperature monitoring unit is specifically set to T1, and the response time of the local control unit is set to T2, to ensure the orderly coordination of global monitoring and local control in time. The configuration of the response time matches the complexity of the tasks at each level. Global monitoring involves a large range of temperature data collection and analysis, and a longer response time T1 is set, while local control requires a rapid response to sudden temperature anomalies, and a shorter response time T2 is set. Through the setting, the coordination parameters between the layers are obtained to ensure that the temperature data can be effectively transmitted and the control linkage can be carried out between different levels. The target temperature data is segmented and normalized, and the temperature data from different sources and different scales are standardized to eliminate unnecessary dimensional effects and obtain standardized temperature data, so that the data has better consistency and comparability in subsequent algorithm processing and analysis. The standardized temperature data is input into the agent-assisted temperature control algorithm, which constructs a temperature state vector based on the time series characteristics of the temperature data, thereby reflecting the dynamic characteristics of temperature changes over time. Based on this state vector, the temperature evolution trend is calculated using the state transition matrix to obtain a temperature prediction model. Based on the temperature prediction model, a sampling strategy is generated, and the temperature field is optimized for sampling to determine the distribution of each target temperature point. Under the guidance of the sampling strategy, it is ensured that the sampling points can cover the areas with large temperature gradients and hot spots in the system, and a reasonable distribution of the target temperature points is obtained. Based on the distribution of the target temperature points, the boundaries of the temperature regions are divided to obtain the division results of the temperature regions. According to the regional division results, the temperature mean vector and temperature gradient vector of each temperature region are calculated. The temperature mean vector reflects the average temperature level of each region, while the temperature gradient vector is used to describe the change of temperature within each region. These feature vectors are combined with the time series characteristics of the temperature data to construct a complete regional temperature feature, which can fully reflect the temperature status and its changing trend in different regions. Based on the regional temperature characteristics and inter-layer coordination parameters, the hierarchical control is optimized to obtain the temperature distribution characteristic data. The hierarchical control optimization method combines the integrity of global monitoring with the flexibility of local regulation. Through collaboration between different levels and differentiated control of different temperature zones, the temperature of the entire multi-chip stacking structure can be kept within a reasonable range.

[0021] Step S300, performing optimal sampling analysis on temperature distribution characteristic data to obtain local temperature control parameters; Specifically, the temperature spatial distribution data is constructed based on the temperature distribution characteristic data, and the temperature spatial distribution data is input into the Minkowski distance calculation unit. The distance order p is set to 2, and the distance value between each sampling point and its surrounding 8 neighboring points is calculated to generate the temperature gradient vector describing the temperature change between each sampling point. Through these gradient vectors, a temperature field distance matrix is ​​formed, which intuitively represents the temperature change amplitude and spatial relationship between each sampling point. All distance values ​​in the temperature field distance matrix are sorted according to the size of the amplitude, and the distance values ​​are divided into high gradient area, medium gradient area and low gradient area by using the K-means clustering method. This division method helps to accurately identify the temperature change characteristics in the multi-chip stacking structure. Connected domain analysis is performed on the high gradient area to obtain abnormal temperature areas, that is, those areas with local overheating risks. Hot spots are graded according to the temperature mean Tm in the abnormal temperature area. When the temperature in the area is greater than 1.2 times Tm, the area is defined as a first-level hotspot; when the temperature is between 1.0Tm and 1.2Tm, it is a second-level hotspot; and when the temperature is between 0.8Tm and 1.0Tm, it is a third-level hotspot. Different degrees of temperature anomalies are distinguished to ensure that cooling resources can be allocated more targeted and obtain corresponding hotspot level data. Based on the hotspot level data, the coolant flow demand of each hotspot area is calculated. The baseline coolant flow of the first-level hotspot is set to Q. These areas have the highest temperature and the heat needs to be taken away quickly and effectively, so the flow demand is the largest. For the second-level hotspot, its coolant flow demand is set to 0.8Q to reduce the waste of coolant but still maintain appropriate cooling capacity; for the third-level hotspot, the flow is set to 0.6Q to achieve the purpose of maintaining temperature stability. The flow distribution parameters are formed based on the level-based data flow distribution method. The spatial distribution of the flow distribution parameters is optimized. By calculating the temperature gradient vector of the hot spot area, determining the flow direction of the coolant, and obtaining the corresponding flow control parameters, it is ensured that the coolant can flow through each hot spot area along the optimal path, thereby achieving efficient heat removal and avoiding the continuous deterioration of local overheating. Based on the flow distribution parameters and flow control parameters, combined with the maximum temperature bearing capacity Tmax of the chip, the corresponding control constraints are set to obtain the basic control parameters. The maximum temperature bearing capacity of the chip is an important constraint to ensure that the entire system is not damaged under normal working conditions, so the basic control parameters must meet this condition. In order to improve the flexibility and adaptability of temperature control, the basic control parameters are dynamically compensated and calculated. The compensation coefficient is set according to the temperature change rate of the hot spot area, so that the basic control parameters can be compensated and adjusted in time to ensure the accuracy and real-time responsiveness of temperature control and obtain local temperature control parameters.

[0022] Step S400: constructing a temperature-flow multi-level coordinated control model based on local temperature control parameters, and generating a coolant flow distribution plan and flow direction control instructions; Specifically, based on the local temperature control parameters, a thermodynamic parameter model including cooling power, fluid resistance and temperature drop coefficient is established. This model is used to describe the heat transfer characteristics in the chip stacking structure and the interaction between the coolant and the chip. Through this thermodynamic model, a temperature-flow mapping relationship is generated, that is, the required coolant flow and corresponding cooling conditions under different temperature changes. The temperature-flow mapping relationship is processed by region, and the entire chip area is divided into high temperature control area, medium temperature control area and low temperature control area based on the hotspot level data to meet the heat dissipation requirements of different areas respectively. For these areas, the control target temperatures are set to F1, F2 and F3 respectively to ensure that each area can be maintained within the appropriate temperature range and obtain the corresponding partition control parameters. Based on the partition control parameters, a three-stage series control loop is constructed to meet the dynamic change requirements of different temperature areas. The main purpose of the three-stage series control loop is to form a global and local multi-level regulation, so that the temperature control targets at different levels can cooperate with each other to ensure the overall thermal balance of the system. After completing the construction of the three-level control loop, the basic flow configuration scheme is obtained. In order to verify and optimize the scheme, the cooling effect is simulated and calculated to obtain the flow threshold data. The simulation calculation helps evaluate the actual cooling effect of the basic flow configuration scheme under different working conditions. Based on the flow threshold data, the ratio of the actual temperature drop ΔT to the theoretical temperature drop ΔT0 is defined as the heat dissipation efficiency η. The heat dissipation efficiency value is calculated for each control area to obtain the efficiency distribution data, which reflects the heat dissipation performance of each area. The efficiency distribution data is normalized, and the area with the highest heat dissipation efficiency is defined as the reference area. Then, the efficiency ratio β of other areas relative to the reference area is calculated to obtain the flow correction coefficient. Based on the flow correction coefficient, the coolant flow is redistributed. The reference flow of each control area is multiplied by the correction coefficient β to obtain the final flow distribution scheme of the coolant. The corrected flow distribution effectively improves the utilization efficiency of the coolant, so that the cooling resources are more concentrated in the areas that need heat dissipation, avoiding the waste of coolant. At the same time, according to the coolant flow distribution scheme, the fluid pressure drop ΔP of each control area is calculated. In order to ensure that the coolant can effectively flow through each area that needs to be cooled, the flow direction of the coolant is determined based on the minimum pressure drop principle, and the flow direction control instruction is finally obtained. The minimum pressure drop principle ensures that the coolant can reduce resistance as much as possible during the flow process, thereby improving the cooling efficiency and flow stability.

[0023] Step S500, controlling the coolant flow rate and temperature based on the coolant flow distribution scheme and the flow direction control instruction, and performing gradient cooling on the temperature abnormal area to obtain temperature adjustment process data; Specifically, based on the coolant flow distribution scheme and the flow control instruction, the coolant reference flow is converted into a specific control signal to generate an execution instruction sequence. The execution instruction sequence specifically includes the coolant flow required for each area and its corresponding regulation requirements. The execution instruction sequence is prioritized according to the order of temperature anomaly levels, and the cooling control instruction of the first-level hot spot is set to the highest priority to ensure that it can get the coolant resource first to quickly reduce the temperature and prevent the chip from being damaged due to overheating. Secondly, the control instructions of the second-level hot spot and the third-level hot spot are executed in sequence to ensure that each temperature abnormal area can be cooled in a timely and effective manner. Through the priority sorting mechanism, an execution timing scheme is obtained to ensure that cooling resources can be used in the most reasonable way to achieve dynamic response to temperature control requirements in different areas. Based on the execution timing scheme, the cooling execution unit is started, and the flow of the coolant is gradually increased from the initial value Q0 to the target value Qt according to the step size ΔQ, forming a gradually increasing flow control process. In each control cycle, the flow is kept at a stable output of the set value to obtain a step-by-step control curve. The step control curve helps to achieve progressive cooling of the temperature abnormal area, avoiding problems such as rapid temperature changes or uneven cooling caused by sudden changes in the coolant flow, thereby improving the cooling efficiency and system stability. Based on the step control curve, each temperature abnormal area is accurately cooled and controlled, and the temperature response data is obtained by measuring the temperature change in real time. The temperature response data is compared and analyzed with the steady-state temperature set of the system to evaluate the cooling effect. In the comparative analysis, the temperature change rate of each control area and the temperature difference between adjacent areas are calculated to obtain the temperature distribution data in the entire multi-chip stacking structure. These data reflect the current cooling effect and reveal the temperature coupling relationship between different areas and the diffusion path of cooling. Deviation analysis is performed based on the temperature distribution data to determine the gap between the cooling effect and the expected target, and the flow correction value is obtained. The flow correction value reflects whether the coolant flow in each area needs to be adjusted to ensure the continuous optimization of the cooling effect. The flow correction value is input into the cooling controller, and the controller performs online compensation for the coolant flow based on it to obtain the corrected control data. The online compensation mechanism ensures that the system can adapt to the dynamic needs of temperature changes in real time, improve the flexibility and response speed of cooling, and enable the temperature abnormal area to return to normal as soon as possible. The corrected control data is stored and sorted, and various data in the entire temperature adjustment process are recorded, including the temperature change curve, flow change curve and control parameter sequence, to obtain the temperature adjustment process data.

[0024] Step S600: Perform multi-stable distribution analysis on the temperature adjustment process data, update the region identification parameters of the agent-assisted temperature control algorithm and the coordination parameters of the hierarchical control framework, and generate a temperature control optimization strategy.

[0025] Specifically, the temperature regulation process data is identified into the heating segment, stable segment and fluctuation segment to obtain accurate time series feature data. The heating segment describes the process of the system gradually heating up from the initial state, the stable segment reflects the stage when the system tends to thermal equilibrium, and the fluctuation segment represents the temperature fluctuation caused by load or environmental changes during the cooling process. Through this classification, the temperature change characteristics of different stages are mastered. The stable segment in the time series feature data is analyzed to obtain a steady-state temperature set. The steady-state temperature set contains all temperature points of the system in a relatively stable state. In order to effectively classify and analyze the steady-state points, the density clustering algorithm is used to process the steady-state temperature set, classify different steady-state points, and obtain steady-state distribution parameters. Based on the steady-state distribution parameters, the entire temperature field is spatially divided to obtain the structural characteristics of the temperature field. Through the division, different areas in the system are clearly marked, and the temperature characteristics of each area are described by steady-state parameters. The structural characteristics of the temperature field are dynamically optimized to identify the parts that need to be improved and calculate the parameter corrections. These corrections are used to adjust the various key parameters of temperature control to better adapt to the complex and changeable thermal environment in the system. According to the parameter correction, the temperature partition parameters and hotspot discrimination parameters of the agent-assisted temperature control algorithm are updated. This update process can improve the agent-assisted algorithm's ability to identify different temperature zones, so that it can detect hotspots more sensitively and take timely measures when facing complex temperature fields. At the same time, the monitoring unit sensitivity coefficient and the control unit response coefficient in the hierarchical control framework are corrected to improve the monitoring unit's perception of temperature changes and the control unit's response speed to cooling requirements, and the updated control parameters are obtained. The updated control parameters and heat dissipation efficiency values ​​are substituted into the temperature field evolution model, and the weights of each control parameter are adjusted according to the distribution of heat dissipation efficiency to generate an optimized control scheme. The heat dissipation efficiency is a key indicator for measuring the cooling effect of each area. The control weight is dynamically adjusted by the distribution of heat dissipation efficiency to ensure that the cooling resources are reasonably allocated, so that the thermal management strategy can be optimized at any time to meet the needs of the current system. After generating the optimized control scheme, the heat dissipation efficiency η is used as the evaluation index of the control effect, and the control objective function J=f(η,δ,c) is constructed, where δ represents the control deviation and c is the control cost. The control objective function J comprehensively considers the cooling effect (i.e., heat dissipation efficiency η), the accuracy of the control process (i.e., control deviation δ), and the control cost (i.e., c) such as energy consumption and cooling resource consumption. By solving the objective function, the optimal strategy for temperature control is finally obtained. This strategy can minimize energy consumption and cooling resource waste while ensuring the heat dissipation effect of the system, thus achieving economical and efficient temperature management.

[0026] In the embodiment of the present invention, by setting temperature sensors with different characteristics, using sinusoidal characteristic sensors to capture periodic temperature changes, and using hyperbolic tangent characteristic sensors to monitor sudden temperature anomalies, the comprehensiveness and accuracy of temperature monitoring are significantly improved. A hierarchical control framework combined with an agent-assisted temperature control algorithm is adopted to achieve the coordinated cooperation of global temperature monitoring and local precise adjustment, and effectively solve the temperature balance problem in the multi-chip stacking structure. Minkowski distance calculation and K-means clustering analysis are introduced to accurately locate and grade the temperature anomaly area, providing a reliable data basis for subsequent precise temperature control. A multi-level coordinated control model based on heat dissipation efficiency is established, and differentiated control of different temperature areas is achieved by adaptively adjusting the coolant flow and flow direction. A step-type flow control strategy is adopted to avoid drastic fluctuations in the coolant flow and ensure the stability of the temperature adjustment process. Through multi-stable distribution analysis and dynamic parameter optimization, the adaptive update of control parameters is achieved, and the system's adaptability to complex working conditions is improved. A complete closed-loop optimization mechanism for temperature control is constructed, and the long-term stability and reliability of system control are guaranteed through real-time monitoring and parameter adjustment.

[0027] In a specific embodiment, the process of executing step S100 may specifically include the following steps: For each layer of chips in the multi-chip stacking structure, the sampling points are divided into grids, the sinusoidal temperature sensors are arranged in the central area with a spacing of D between adjacent sampling points, and the hyperbolic tangent temperature sensors are arranged in the edge area with a spacing of D / 2 between adjacent sampling points, and the temperature sensing network topology is obtained; Perform temperature periodic sampling on the sinusoidal characteristic temperature sensor to obtain initial temperature data, and perform mutation detection configuration on the hyperbolic tangent characteristic temperature sensor to obtain mutation detection data; The initial temperature data and mutation detection data are sorted and classified to obtain a preprocessed temperature sequence, and the preprocessed temperature sequence is decomposed by wavelet multi-scale to obtain the target temperature data; Principal component analysis is performed on the target temperature data to obtain the temperature eigenvector, which is then mapped using nonlinear functions. The nonlinear temperature response model is constructed using sine function and hyperbolic tangent function.

[0028] Specifically, each layer of the multi-chip stack structure is divided into sampling points according to a grid. The sampling points are selected based on the characteristics and uniformity of heat conduction in the chip to ensure accurate monitoring of the temperature of the entire chip area. In this grid, the sinusoidal temperature sensors are arranged at adjacent sampling points with a spacing of In the central area of ​​the grid, the hyperbolic tangent characteristic temperature sensors are arranged at the edge of the grid, and the spacing between adjacent sampling points is , forming a heterogeneous temperature sensing network, in which the sinusoidal temperature sensor monitors relatively semi-smooth and uniform temperature changes, while the hyperbolic tangent temperature sensor focuses on areas with more drastic temperature changes, resulting in a temperature sensing network topology that can monitor the overall trend and capture local anomalies. After the sensors are arranged, data is collected from the sensors. For the sinusoidal temperature sensor, the temperature is periodically sampled to obtain the initial temperature data. The formula for temperature periodic sampling is: ; in, Indicates a moment The temperature value, is the amplitude of temperature change, is the sampling frequency, Φ is the phase constant, is the basic temperature. Through this formula, the sinusoidal temperature sensor can capture the temperature change of the chip in a stable state, which is suitable for describing areas with relatively uniform temperature fluctuations. For the hyperbolic tangent temperature sensor, a mutation detection configuration is performed to capture the temperature anomaly that occurs. The mutation detection formula for hyperbolic tangent temperature change is: ; in, Indicates the hyperbolic tangent characteristic sensor at time The temperature value, is the amplitude of temperature change, is the coefficient that controls the steepness of temperature change, is the time point when the mutation occurs. As the basic temperature. With this configuration, the hyperbolic tangent characteristic temperature sensor can keenly detect temperature mutations caused by local hot spots of the chip, such as high-load calculations in a short period of time or poor local heat dissipation. The initial temperature data and mutation detection data are sorted and classified to obtain the preprocessed temperature sequence. The preprocessed temperature sequence is subjected to wavelet multi-scale decomposition to decompose the complex temperature signal into components at different scales in order to reveal the characteristics of temperature changes at different time scales. The wavelet decomposition formula is: ; in, represents the original temperature signal, is the wavelet basis function, is the wavelet coefficient obtained by decomposition, represents the scale of decomposition, is the time translation factor. Through wavelet multi-scale decomposition, the target temperature data is obtained, which contains temperature change information at different scales. Principal component analysis is performed on the target temperature data to reduce the data dimension and extract key temperature features. Principal component analysis is a linear dimensionality reduction method that can project high-dimensional data into a low-dimensional space while retaining the main information of the data as much as possible. The main features are extracted by finding the direction with the largest covariance in the data. The formula for principal component analysis is as follows: ; in, is the principal component eigenvector, is the original temperature data matrix, is the eigenvector matrix. Through this process, the obtained temperature eigenvector reflects the main temperature change characteristics in the multi-chip stacking structure. The temperature eigenvector is mapped by nonlinear function. The sine function and hyperbolic tangent function are used to construct a nonlinear temperature response model to adapt to different types of temperature changes. The specific temperature response model is expressed as: ; in, represents the output of the nonlinear temperature response model, is the temperature eigenvector, and are the weights of the sine and hyperbolic tangent terms, respectively, and To control the frequency factor of temperature change, is the bias term. The sine term is used to describe periodic and relatively gentle temperature characteristics, while the hyperbolic tangent term is used to describe nonlinear temperature changes. This combination can comprehensively describe the temperature dynamic characteristics of different regions in the multi-chip stacking structure, ensuring that the model predicts and responds to temperature changes more accurately.

[0029] In a specific embodiment, the process of executing step S200 may specifically include the following steps: A hierarchical control framework is constructed based on the nonlinear temperature response model, with the global temperature monitoring unit as the master control layer and the local control unit as the slave control layer. Configure the control parameters of the hierarchical control framework, set the response time of the global temperature monitoring unit to T1, set the response time of the local control unit to T2, and obtain the inter-layer coordination parameters; The target temperature data is segmented and normalized to obtain standardized temperature data, and the standardized temperature data is input into the agent-assisted temperature control algorithm. The temperature state vector is constructed according to the time series characteristics, and the temperature evolution trend is calculated based on the state transition matrix to obtain the temperature prediction model. Generate a sampling strategy based on the temperature prediction model, and optimize the sampling of the temperature field to obtain the target temperature point distribution; The target temperature point distribution is divided into temperature region boundaries to obtain the region division results. Based on the region division results, the temperature mean vector and temperature gradient vector of each temperature region are calculated, and the regional temperature features are constructed in combination with the time series features. Based on the regional temperature characteristics and inter-layer coordination parameters, hierarchical control optimization is carried out to obtain temperature distribution characteristic data.

[0030] Specifically, a hierarchical control framework is constructed based on the nonlinear temperature response model to achieve global and local coordinated temperature control. In this control framework, a global temperature monitoring unit is set as the main control layer, and multiple local control units are set as slave control layers. The global temperature monitoring unit is mainly responsible for overall temperature perception and trend analysis. Its goal is to monitor the temperature changes of the entire multi-chip stacking structure and identify potential abnormal areas in the system. The local control unit focuses on the refined temperature control of specific areas, and quickly responds to hotspots found in the global layer and adjusts local heat dissipation. In order to coordinate the work between the main control layer and the slave control layer, the control parameters of the hierarchical control framework are configured. The response time of the global temperature monitoring unit is set to , because global monitoring involves the collection and processing of a large amount of data, the response time is relatively long to ensure sufficient accuracy. The response time of the local control unit is set to , Setting ratio The relationship between these two response times is expressed by the interlayer coordination parameter, denoted as ,in It is the proportional factor of inter-layer coordination, ensuring that the master control layer and the slave control layer can work in effective coordination. After determining the hierarchical control framework, the collected target temperature data is segmented and normalized. The temperature data is divided into the heating stage, the stabilization stage, and the cooling stage according to the time period, and the normalization process is to eliminate the dimensional differences between different sensors, so that the data has better consistency and comparability in the subsequent algorithm processing. The normalization formula is: ; in, represents the normalized temperature data, is the original temperature data, and are the minimum and maximum values ​​of the temperature in the entire data set, respectively. Through normalization processing, a set of standardized temperature data is obtained to eliminate the temperature measurement deviation between different chip layers. The standardized temperature data is input into the agent-assisted temperature control algorithm. The agent-assisted algorithm constructs a temperature state vector based on the time series characteristics of the temperature to predict the future temperature change trend. The construction formula of the temperature state vector is as follows: ; in, It is at the moment The temperature state vector contains the current and previous Temperature data at each moment. Based on the temperature state vector, the state transition matrix is ​​used to calculate the evolution trend of the temperature. State transition matrix The calculation formula is: ; in, Indicates the moment The predicted temperature value, It is a state transition matrix that describes the relationship between the changes in system temperature at different times. Through this process, a temperature prediction model is established to predict the temperature distribution of each chip layer in the future. Based on the temperature prediction model, an optimized sampling strategy is generated. The goal of this sampling strategy is to capture the temperature hotspots and cooling demand points in the chip as accurately as possible, so as to optimize the sampling of the temperature field and obtain a reasonable distribution of the target temperature points. The sampling strategy selects sampling points according to the predicted temperature gradient, so that the sampling points are more concentrated in areas with drastic temperature changes, so as to more accurately control the temperature of these areas. After optimizing the sampling of the temperature field, the distribution data of the target temperature points is obtained. The target temperature point distribution is divided into temperature area boundaries. By calculating the temperature difference between each sampling point and its surrounding neighboring points, the boundaries of different temperature areas are determined to obtain the area division results. According to the area division results, the temperature mean vector and temperature gradient vector of each temperature area are calculated. The temperature mean vector The formula used to describe the overall temperature level of each area is: ; in, Indicates The average temperature of the region, is the number of sampling points in the area, Indicates The temperature value of the sampling point. Temperature gradient vector It is used to describe the temperature variation trend in each region, and its calculation formula is: ; in, Indicates The temperature gradient vector of the region is are the directions of the spatial coordinate axes respectively. Combining these time series features, the temperature characteristics of each area are finally constructed, including information on temperature levels and changing trends. Based on the regional temperature characteristics and inter-layer coordination parameters, the hierarchical control is optimized. The global monitoring unit determines the heat load of each area based on the temperature prediction model and calculates the heat load of each area based on the inter-layer coordination parameters. Adjust the response speed and cooling strategy of the local control unit. The local control unit determines the flow rate and direction of the coolant according to the temperature mean and gradient information in the area to ensure that the hot spot area can be cooled in a timely and effective manner. After layered and coordinated control, the temperature distribution characteristic data of the entire multi-chip stacking structure is finally obtained.

[0031] In a specific embodiment, the process of executing step S300 may specifically include the following steps: The temperature spatial distribution data is constructed according to the temperature distribution characteristic data, and the temperature spatial distribution data is input into the Minkowski distance calculation unit, the distance order p is set to 2, the distance value between each sampling point and its surrounding 8 neighboring points is calculated, the temperature gradient vector is generated, and the temperature field distance matrix is ​​obtained; The distance values ​​in the temperature field distance matrix are sorted according to the magnitude, and the distance values ​​are divided into high gradient area, medium gradient area and low gradient area based on the K-means clustering method. The high gradient area is analyzed for connected domains to obtain abnormal temperature areas. According to the temperature mean Tm of the abnormal temperature area, the hot spots are divided into three levels: when the temperature is greater than 1.2Tm, it is a first-level hot spot; when the temperature is between 1.0Tm and 1.2Tm, it is a second-level hot spot; when the temperature is between 0.8Tm and 1.0Tm, it is a third-level hot spot, and the hot spot level data is obtained; The coolant flow demand is calculated based on the hotspot level data, the base flow of the first-level hotspot is set to Q, the flow of the second-level hotspot is set to 0.8Q, and the flow of the third-level hotspot is set to 0.6Q, and the flow distribution parameters are obtained; Optimize the spatial distribution of flow distribution parameters, calculate the temperature gradient vector of the hot spot area, determine the coolant flow direction, and obtain the flow control parameters; Based on the flow distribution parameters and flow direction control parameters, the control constraints are set in combination with the maximum temperature bearing capacity Tmax of the chip to obtain the basic control parameters. The basic control parameters are dynamically compensated and calculated. The compensation coefficient is set according to the temperature change rate of the hot spot area to obtain the local temperature control parameters.

[0032] Specifically, temperature spatial distribution data is constructed based on the temperature distribution characteristic data to describe the temperature conditions at each location of each chip. The temperature spatial distribution data is input into the Minkowski distance calculation unit, and the distance order is set to , that is, use the Euclidean distance to calculate the distance between each sampling point and its eight surrounding neighborhood points. The Minkowski distance calculation formula is as follows: ; in, Indicates The sampling point and The distance between neighboring points, and The sampling points and The temperature value, is the order of the distance, set to 2. Through this formula, the temperature difference between each sampling point and its surrounding neighborhood is obtained, and the temperature gradient vector is generated. The temperature gradient vector is used to describe the trend of temperature change in space, and its formula is: ; in, Indicates The temperature gradient at each sampling point is Indicates temperature, They are respectively the directions of the spatial coordinate axes. By calculating the temperature gradient for all sampling points, the temperature field distance matrix is ​​obtained, which describes the temperature differences between various positions in the entire chip stacking structure. The distance values ​​in the temperature field distance matrix are sorted and classified based on the magnitude of the amplitude. The K-means clustering method is used to divide the distance values ​​into high gradient areas, medium gradient areas, and low gradient areas to identify areas with drastic temperature changes. These areas usually represent areas with higher heat dissipation requirements. Connected domain analysis is performed on high gradient areas to find continuous areas with obvious temperature changes. These areas are usually potential abnormal temperature areas. The location and range of the abnormal temperature area are obtained through connected domain analysis. After determining the abnormal temperature area, based on its temperature mean Divide the hot spots. Greater than 1.2 When the temperature is above , the area is defined as a Level 1 hot spot, indicating that the temperature is very high and there is a serious risk of overheating; when the temperature is above to When the temperature is between to When the temperature is between , it is defined as a third-level hot spot, which means that the temperature is at a high level but still within the normal range. Through classification, the hot spot level data of each abnormal temperature area is obtained. Based on the hot spot level data, the coolant flow demand of each area is calculated. The baseline coolant flow of the first-level hot spot is set to , because these areas have the highest temperatures, they require a larger coolant flow rate to quickly reduce the temperature. For the secondary hotspot, its coolant flow rate requirement is set to , while meeting the cooling demand and reducing unnecessary energy consumption; for the third-level hotspot, the cooling night flow is set to , to ensure that the temperature of these areas can be kept within a reasonable range, and to obtain the flow distribution parameters of the coolant. Optimize the spatial distribution of the flow distribution parameters. Calculate the temperature gradient vector of each hot spot area and determine the flow direction of the coolant to remove the heat most effectively. The optimized distribution of flow needs to consider the direction and amplitude of the temperature gradient. Based on the temperature gradient vector, determine how the coolant should flow through each hot spot area to achieve the best heat dissipation effect. The flow direction control parameters are used to describe the flow direction of the coolant and the flow speed in each area to ensure that the coolant can preferentially flow through those areas with higher temperatures and more urgent heat dissipation needs. Based on the flow distribution parameters and flow direction control parameters, the control constraints are set in combination with the maximum temperature bearing capacity Tmax of the chip to obtain the basic control parameters. The maximum temperature bearing capacity of the chip is a key parameter to ensure that the entire control process will not cause thermal damage to the chip. When designing a cooling solution, ensure that the temperature of any area does not exceed The basic control parameters include the flow rate, flow direction and heat dissipation priority of each area of ​​the cooling night to ensure that the coolant can be distributed quickly and effectively under different heat load conditions. In order to improve the flexibility and adaptability of the temperature control system, the basic control parameters are dynamically compensated. Real-time adjustments are made according to the temperature change rate of the hot spot area to cope with changes in environmental conditions and workloads. Set the compensation coefficient , the compensation formula is: ; in, is the coolant flow after compensation, is the basic coolant flow rate, is the compensation coefficient, Indicates the rate of change of temperature over time. If the temperature change rate is large, it means that the temperature rises quickly and the flow rate of the coolant needs to be increased to quickly reduce the temperature. Through this formula, the cooling strategy is dynamically adjusted according to the actual temperature change, and the local temperature control parameters are finally obtained.

[0033] In a specific embodiment, the process of executing step S400 may specifically include the following steps: Based on the local temperature control parameters, a thermodynamic parameter model including cooling power, fluid resistance and temperature drop coefficient is established, and a temperature-flow mapping relationship is generated; The temperature-flow mapping relationship is processed by region, and the chip area is divided into high temperature control area, medium temperature control area and low temperature control area based on the hotspot level data. The control target temperatures are set to F1, F2 and F3 respectively, and the partition control parameters are obtained; A three-stage series control loop is constructed based on the partition control parameters to obtain a basic flow configuration scheme, and the cooling effect of the basic flow configuration scheme is simulated and calculated to obtain flow threshold data; Based on the flow threshold data, the ratio of the actual temperature drop ΔT to the theoretical temperature drop ΔT0 is defined as the heat dissipation efficiency η. The heat dissipation efficiency value is calculated for each control area to obtain the efficiency distribution data. The efficiency distribution data is normalized, the area with the highest heat dissipation efficiency is defined as the reference area, and the efficiency ratio β of other areas relative to the reference area is calculated to obtain the flow correction coefficient; The coolant flow is redistributed based on the flow correction coefficient, and the baseline flow of each area is multiplied by the correction coefficient β to obtain the coolant flow distribution plan. The fluid pressure drop ΔP of each control area is calculated according to the coolant flow distribution plan, and the coolant flow direction is determined based on the minimum pressure drop principle to obtain the flow direction control instruction.

[0034] Specifically, based on the local temperature control parameters, a thermodynamic parameter model including cooling power, fluid resistance and temperature drop coefficient is established to describe the flow of coolant in the multi-chip stacking structure and the effect of temperature control. It indicates the ability of the cooling system to remove heat from the chip. The calculation formula is: ; in, is the cooling power, is the mass flow rate of the coolant, is the specific heat capacity of the coolant, is the temperature difference between the inlet and outlet of the coolant. It is used to describe the resistance encountered by the coolant during its flow, and its formula is: ; in, is the fluid resistance, is the friction factor, is the length of the cooling pipe, is the pipe diameter, is the coolant density, is the flow rate. And the temperature drop coefficient It is used to describe the temperature reduction ability of the coolant, which is related to the coolant flow rate and heat transfer efficiency. Based on these parameters, a temperature-flow mapping relationship is established to describe the relationship between the coolant flow rate and the temperature control effect. The mapping relationship formula is: ; in, is the coolant outlet temperature, is the coolant inlet temperature, is the temperature drop coefficient, is the mass flow rate of the coolant. Through this mapping relationship, the temperature reduction effect that the system can achieve under different coolant flow conditions is determined. The temperature-flow mapping relationship is processed by region. Based on the hotspot level data, the chip area is divided into a high-temperature control area. The medium-temperature control area and the low-temperature control area are set to control the target temperature as , and . The high temperature control zone corresponds to the first-level hot spot area, which has a higher temperature and needs to be treated first; the temperature in the medium temperature control zone is relatively low, but still needs to be controlled; the temperature in the low temperature control zone is close to room temperature and only needs a small amount of adjustment. Based on the control target temperature, the partition control parameters are obtained to guide the cooling strategies of different temperature zones. Based on the partition control parameters, a three-stage series control loop is constructed. These three control loops correspond to the high temperature, medium temperature and low temperature zones, respectively. The series connection method is adopted so that the control process can be carried out continuously between different temperature zones to ensure the temperature regulation of the entire chip is coordinated and consistent. Through the series control loop, a basic flow configuration scheme is designed to allocate corresponding coolant flow to each zone. In order to verify the effectiveness of the basic flow configuration scheme, the cooling effect is simulated and calculated to obtain the flow threshold data for each zone. The heat dissipation efficiency of the cooling system is evaluated based on the flow threshold data. The actual temperature drop Theoretical temperature drop The ratio is defined as the heat dissipation efficiency , and its calculation formula is: ; in, is the actual measured temperature drop, is the theoretically calculated temperature drop, Indicates the heat dissipation efficiency, reflecting the gap between the actual performance of the cooling system and the theoretical optimal performance. By calculating the heat dissipation efficiency value for each control area, the efficiency distribution data is obtained, which is used to evaluate the cooling effect of different areas. The efficiency distribution data of the heat dissipation efficiency is normalized. The area with the highest heat dissipation efficiency is defined as the reference area, assuming that the efficiency value of this area is , and then calculate the efficiency ratio of other areas relative to the benchmark area , the formula is: ; in, For the The efficiency ratio of a region relative to the benchmark region, For the The heat dissipation efficiency of each area, is the heat dissipation efficiency of the reference area. Through this ratio, the degree of deviation of the cooling efficiency of each area relative to the reference area is determined. Based on the calculated flow correction coefficient , redistribute the coolant flow. With correction factor Multiply to get the corrected coolant flow , the formula is: ; By making corrections, the flow rate in areas with low cooling efficiency is increased to ensure that these areas can obtain sufficient coolant and improve the overall heat dissipation effect. According to the corrected coolant flow distribution plan, the fluid pressure drop in each control area is calculated. The formula for calculating fluid pressure drop is: ; in, is the pressure drop, is the fluid resistance, is the corrected coolant flow rate. The pressure drop describes the resistance encountered by the coolant when it flows through each area, and reflects the consumption of the coolant on different flow paths. In order to ensure that the coolant can flow through each control area in the optimal path, the minimum pressure drop principle is used to determine the flow direction of the coolant. The minimum pressure drop principle means that on the premise of meeting the heat dissipation requirements of each area, those paths with lower pressure drops are selected to reduce the flow energy consumption of the coolant and improve the overall efficiency of the cooling system. Based on this principle, the coolant flow control instruction is generated to guide the flow path of the coolant in each control area.

[0035] In a specific embodiment, the process of executing step S500 may specifically include the following steps: Based on the coolant flow distribution scheme and the flow direction control instruction, the coolant reference flow is converted into a control signal to obtain an execution instruction sequence; Prioritize the execution instruction sequence according to the order of temperature anomaly level, execute the cooling control instruction of the first-level hot spot first, and execute the control instructions of the second-level hot spot and the third-level hot spot in sequence to obtain the execution timing plan; Based on the execution timing scheme, the cooling execution unit is started, and the coolant flow rate is gradually increased from the initial value Q0 to the target value Qt according to the step length ΔQ, and a stable flow output is maintained in each control cycle to obtain a step control curve; According to the step control curve, each temperature abnormal area is cooled and controlled to obtain temperature response data, and the temperature response data is compared and analyzed with the steady-state temperature set to calculate the temperature change rate of each control area and the temperature difference of adjacent areas to obtain temperature distribution data; Deviation analysis is performed on the temperature distribution data to obtain the flow correction value, and the flow correction value is input into the cooling controller to perform online compensation on the coolant flow to obtain the corrected control data; The corrected control data is stored and sorted, and the temperature change curve, flow change curve and control parameter sequence in the complete temperature adjustment process are recorded to obtain the temperature adjustment process data.

[0036] Specifically, based on the coolant flow distribution scheme and flow control instructions, the coolant reference flow is converted into a specific control signal to obtain an executable instruction sequence. Cooling operations are performed on each hot spot area, and the coolant reference flow is converted from a theoretical scheme into a specific hardware control signal. Indicates the reference flow rate, which is converted into control signals such as voltage or current by the control module , the formula is: ; in, is the control signal, is the conversion coefficient, which reflects the proportional relationship between the flow rate and the control signal. Through the conversion, the controller drives the cooling pump to work according to the flow rate demand, and accurately adjusts the coolant supply to each hot spot area. Prioritize the execution instruction sequence. Since the degree of temperature abnormality in different areas is different, these areas are sorted in order of temperature abnormality level, and the cooling control instructions of the first-level hot spot are executed first, followed by the control instructions of the second-level and third-level hot spots, to obtain an orderly execution timing plan to ensure that the area with the highest temperature can get the coolant resources first, so as to minimize the risk of overheating. For the implementation of priority, set the control signal priority of the first-level hot spot area to the highest, that is, ,in , ,and Represents the priority of the first, second and third level hot spots respectively. Based on the execution timing plan, the cooling execution unit is started. In order to prevent the sudden temperature change from affecting the chip, a step-by-step increase method is adopted. At the initial moment, the flow rate of the coolant is set to , then follow the step size Gradually increase to target flow The formula for the incremental process is: ; in, Indicates at time Coolant flow at the time, The flow rate increment within each control cycle until the target flow rate is reached . Maintain the stability of flow output in each control cycle to avoid unstable cooling effect caused by drastic changes in flow, and obtain a step control curve to describe the gradual change of coolant flow over time. According to the generated step control curve, perform cooling control on each temperature abnormal area and record the temperature response data of these areas during the cooling process. The temperature response data includes the actual temperature value at each time point , these temperature values ​​are used to reflect the temperature changes of each area under the action of the coolant. The temperature response data is compared and analyzed with the steady-state temperature set of the system to evaluate the effect of the current cooling strategy. Steady-state temperature set It represents the temperature of each region when the system reaches the temperature equilibrium state. The comparison formula is: ; in, For in time The difference between the moment temperature and the steady-state temperature is calculated by calculating the temperature change rate of each control area and the temperature difference between adjacent areas , obtain temperature distribution data, reflecting the dynamic temperature change characteristics of the entire multi-chip stacking structure. According to the temperature distribution data, the cooling effect of each control area is analyzed for deviations to find out the difference between the actual cooling effect and the expected effect. By evaluating the temperature changes in each area, it is determined whether the coolant flow needs to be adjusted. Assuming the deviation is , the deviation calculation formula is: ; in, is the correction value of flow rate, is the proportional coefficient, which indicates the influence of temperature deviation on the flow rate. After obtaining the deviation, the flow rate correction value is input into the cooling controller to compensate the coolant flow rate online and obtain the corrected control data. The formula for online compensation is: ; The corrected flow rate is used for the next cycle of coolant supply to ensure that the temperature control of each area is always maintained at the optimal state. Through dynamic compensation, the cooling system can respond to temperature changes in real time and quickly adjust the cooling strategy to cope with different heat load conditions. The corrected control data is systematically stored to record the temperature change curve during the entire temperature adjustment process. , flow change curve , and control parameter sequence to obtain temperature regulation process data, including key parameters of the temperature control system and its dynamic response.

[0037] In a specific embodiment, the process of executing step S600 may specifically include the following steps: Identify the temperature rise segment, stable segment and fluctuation segment of the temperature regulation process data to obtain time series characteristic data; Perform state analysis on the stable segment in the time series characteristic data to obtain a steady-state temperature set, and use a density clustering algorithm to classify the steady-state points in the steady-state temperature set to obtain steady-state distribution parameters; Based on the steady-state distribution parameters, the temperature field space is divided to obtain the temperature field structure characteristics, and the temperature field structure characteristics are dynamically optimized to obtain the parameter correction amount; The temperature partition parameters and hot spot discrimination parameters of the agent-assisted temperature control algorithm are updated according to the parameter correction amount, and the sensitivity coefficient of the monitoring unit and the response coefficient of the control unit in the hierarchical control framework are corrected to obtain the updated control parameters; The updated control parameters and heat dissipation efficiency values ​​are substituted into the temperature field evolution model. The control parameter weights are adjusted according to the heat dissipation efficiency distribution to generate an optimized control scheme. Based on the optimized control scheme, the heat dissipation efficiency η is used as the control effect evaluation index, and the control objective function J=f(η,δ,c) is set, where δ is the control deviation and c is the control cost, to solve the temperature control optimization strategy.

[0038] Specifically, the temperature control process data is identified into heating segment, stable segment and fluctuation segment. Through segmentation, the temperature characteristics of different stages are identified, and targeted control optimization is performed. The heating segment mainly describes the process of the system gradually heating up from the initial cold state. This process generally occurs when the system is just started or the load suddenly increases; the stable segment describes the temperature maintenance process after the system reaches thermal equilibrium. At this time, the temperature fluctuates within a small range and is a relatively static state; the fluctuation segment reflects the dynamic response of the system when it is subject to external disturbances or load changes. By segmenting the temperature control process data, the time series feature data is extracted. State analysis is performed on the stable segment in the time series feature data to extract the steady-state temperature set. , these data reflect the temperature distribution characteristics of the system in equilibrium. A density clustering algorithm (such as DBSCAN) is used to classify the steady-state points in the steady-state temperature set to obtain the steady-state distribution parameters. Adjacent points with similar densities are grouped together to identify the distribution characteristics of temperature in the chip stacking structure. Assume that the steady-state temperature set is , the density clustering algorithm divides it into different categories, and the obtained steady-state distribution parameters are expressed as a set: ; in, Indicates Steady-state temperature clusters contain steady-state temperature points under similar conditions. Each steady-state cluster corresponds to a specific temperature region, and these clusters reflect the spatial distribution and local characteristics of temperature in different regions. Based on the obtained steady-state distribution parameters, the temperature field is spatially divided, and the entire chip area is divided into different control areas. Each control area has similar temperature characteristics, which is convenient for independent regulation. The result of spatial division is the structural characteristics of the temperature field, in which the temperature change trend and distribution law of each area are relatively consistent. According to the structural characteristics of the temperature field, it is dynamically optimized, and the control strategy of each area is adjusted according to the real-time temperature data and environmental changes. The goal of dynamic optimization is to calculate the parameter correction amount , the formula is: ; in, is the correction amount, is the correction factor, is the real-time measured temperature, is the target temperature. The correction is used to fine-tune the temperature distribution to ensure that the temperature of each area is closer to the expected target. The temperature partition parameters and hot spot discrimination parameters of the agent-assisted temperature control algorithm are updated according to the parameter correction. The agent-assisted control algorithm is used to achieve real-time control of the temperature of different areas. Its temperature partition parameters determine the control range of each area, while the hot spot discrimination parameters determine whether a certain area is marked as a hot spot area that needs to be cooled. The sensitivity coefficient of the monitoring unit in the hierarchical control framework is and the control unit response coefficient Calibrate. Monitoring unit sensitivity coefficient Reflects the system's ability to perceive temperature changes, and controls the unit response coefficient It indicates the speed at which the system responds to temperature changes. The corrected control parameters are expressed as: ; in, and The corrections are sensitivity and response coefficient, respectively. These corrections ensure that the system can accurately monitor temperature changes and respond quickly under different working conditions. Substitute it into the temperature field evolution model, use the heat dissipation efficiency distribution to adjust the weight of the control parameters, and generate an optimized control solution. Used to describe the effectiveness of the cooling system, which is defined as the ratio of the actual temperature drop to the theoretical temperature drop. In order to make the optimization control scheme more adaptable to the needs of different regions, different weights are assigned to the control parameters based on the heat dissipation efficiency. : ; in, For the The weight of the control area, is the heat dissipation efficiency of the area, is the total number of control areas. By weight allocation, more control resources are allocated to areas with low heat dissipation efficiency, thereby improving the cooling effect of these areas. Based on the optimized control scheme, the heat dissipation efficiency is As an evaluation index of the control effect, a control objective function is set , which is defined as follows: ; in, For heat dissipation efficiency, is the control deviation, which represents the difference between the current temperature and the target temperature. It is the control cost, which is used to describe the cost of coolant consumption, energy consumption, etc. The optimization goal is to minimize the control deviation and control cost while maximizing the heat dissipation efficiency. The objective function is solved to obtain the optimal temperature control strategy.

[0039] See also Figure 2 , Figure 2 A schematic block diagram of the structure of the intelligent temperature control device 200 with a multi-chip stacking structure provided in an embodiment of the present application, as shown in FIG. Figure 2 As shown, the intelligent temperature control device 200 of the multi-chip stacking structure includes: The acquisition module 210 is used to acquire the target temperature data of each layer of chips in the multi-chip stacking structure and establish a nonlinear temperature response model; A construction module 220 is used to construct a hierarchical control framework based on a nonlinear temperature response model, and input the target temperature data into an agent-assisted temperature control algorithm for processing to obtain temperature distribution characteristic data; An analysis module 230 is used to perform optimal sampling analysis on the temperature distribution characteristic data to obtain local temperature control parameters; A generation module 240 is used to construct a temperature-flow multi-level coordinated control model based on local temperature control parameters, and generate a coolant flow distribution plan and a flow direction control instruction; The processing module 250 is used to control the coolant flow and temperature based on the coolant flow distribution scheme and the flow direction control instruction, and to perform gradient cooling on the temperature abnormal area to obtain temperature adjustment process data; The updating module 260 is used to perform multi-stable distribution analysis on the temperature adjustment process data, update the region identification parameters of the agent-assisted temperature control algorithm and the coordination parameters of the hierarchical control framework, and generate a temperature control optimization strategy.

[0040] Through the cooperation of the above components, by setting temperature sensors with different characteristics, using sinusoidal characteristic sensors to capture periodic temperature changes, and using hyperbolic tangent characteristic sensors to monitor sudden temperature anomalies, the comprehensiveness and accuracy of temperature monitoring are significantly improved. The hierarchical control framework combined with the agent-assisted temperature control algorithm is adopted to achieve the coordinated cooperation of global temperature monitoring and local precise adjustment, and effectively solve the temperature balance problem in the multi-chip stacking structure. The Minkowski distance calculation and K-means clustering analysis are introduced to accurately locate and grade the temperature anomaly areas, providing a reliable data basis for subsequent precise temperature control. A multi-level coordinated control model based on heat dissipation efficiency is established, and differentiated control of different temperature areas is achieved by adaptively adjusting the coolant flow and flow direction. The step-type flow control strategy is adopted to avoid the drastic fluctuation of the coolant flow and ensure the stability of the temperature adjustment process. Through multi-stable distribution analysis and dynamic parameter optimization, the adaptive update of control parameters is achieved, and the system's adaptability to complex working conditions is improved. A complete closed-loop optimization mechanism for temperature control is constructed, and the long-term stability and reliability of system control are guaranteed through real-time monitoring and parameter adjustment.

[0041] See also Figure 3 , Figure 3 The computer device 300 provided in the embodiment of the present application is a schematic block diagram of the structure, wherein the computer device 300 includes a processor 301 and a memory 302, wherein the processor 301 and the memory 302 are connected via a device bus 303, wherein the memory 302 may include a non-volatile storage medium and an internal memory.

[0042] The non-volatile storage medium can store a computer program. The computer program includes program instructions, and when the program instructions are executed by the processor 301, the processor 301 can execute any of the above-mentioned intelligent temperature control methods for a multi-chip stacking structure.

[0043] The processor 301 is used to provide computing and control capabilities and support the operation of the entire computer device 300 .

[0044] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor 301, the processor 301 can execute any of the above-mentioned intelligent temperature control methods for the multi-chip stacking structure.

[0045] Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a part of the structure related to the present application scheme, and does not constitute a limitation on the computer device 300 involved in the present application scheme. The specific computer device 300 may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0046] It should be understood that the processor 301 may be a central processing unit (CPU), and the processor 301 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0047] It should be noted that technicians in the relevant field can clearly understand that for the convenience and simplicity of description, the specific working process of the computer device 300 described above can refer to the corresponding process of the intelligent temperature control method of the aforementioned multi-chip stacking structure, and will not be repeated here.

[0048] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by one or more processors, the one or more processors implement the intelligent temperature control method for a multi-chip stacking structure as provided in an embodiment of the present application.

[0049] The computer-readable storage medium may be an internal storage unit of the computer device 300 in the aforementioned embodiment, such as a hard disk or memory of the computer device 300. The computer-readable storage medium may also be an external storage device of the computer device 300, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped with the computer device 300.

[0050] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0051] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., various media that can store program codes.

[0052] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An intelligent temperature control method for a multi-chip stacking structure, characterized in that: include: Collect the target temperature data of each layer of chips in the multi-chip stacking structure and establish a nonlinear temperature response model; Building a hierarchical control framework based on the nonlinear temperature response model, and inputting the target temperature data into the agent-assisted temperature control algorithm for processing to obtain temperature distribution characteristic data; Performing optimal sampling analysis on the temperature distribution characteristic data to obtain local temperature control parameters; Building a temperature-flow multi-level coordinated control model based on the local temperature control parameters, and generating a coolant flow distribution plan and flow direction control instructions; Based on the coolant flow distribution scheme and the flow direction control instruction, the coolant flow and temperature are controlled, and a gradient cooling process is performed on the temperature abnormal area to obtain temperature adjustment process data; A multi-stable distribution analysis is performed on the temperature regulation process data, the region identification parameters of the agent-assisted temperature control algorithm and the coordination parameters of the hierarchical control framework are updated, and a temperature control optimization strategy is generated.

2. The intelligent temperature control method of a multi-chip stacking structure according to claim 1, characterized in that: The collecting of target temperature data of each layer of chips in the multi-chip stacking structure and establishing a nonlinear temperature response model includes: For each layer of chips in the multi-chip stacking structure, the sampling points are divided into grids, the sinusoidal temperature sensors are arranged in the central area with a spacing of D between adjacent sampling points, and the hyperbolic tangent temperature sensors are arranged in the edge area with a spacing of D / 2 between adjacent sampling points, and the temperature sensing network topology is obtained; Performing temperature periodic sampling on the sinusoidal characteristic temperature sensor to obtain initial temperature data, and performing mutation detection configuration on the hyperbolic tangent characteristic temperature sensor to obtain mutation detection data; Sorting and classifying the initial temperature data and the mutation detection data to obtain a preprocessed temperature sequence, and performing wavelet multi-scale decomposition on the preprocessed temperature sequence to obtain target temperature data; The target temperature data is subjected to principal component analysis to obtain a temperature characteristic vector, and the temperature characteristic vector is subjected to nonlinear function mapping, and a nonlinear temperature response model is constructed by using a sine function and a hyperbolic tangent function.

3. The intelligent temperature control method of a multi-chip stacking structure according to claim 2, characterized in that: The hierarchical control framework is constructed based on the nonlinear temperature response model, and the target temperature data is input into the agent-assisted temperature control algorithm for processing to obtain temperature distribution characteristic data, including: Building a hierarchical control framework based on the nonlinear temperature response model, setting a global temperature monitoring unit as a master control layer, and setting a local control unit as a slave control layer; Configure the control parameters of the hierarchical control framework, set the response time of the global temperature monitoring unit to T1, set the response time of the local control unit to T2, and obtain the inter-layer coordination parameters; The target temperature data is segmented and normalized to obtain standardized temperature data, and the standardized temperature data is input into the agent-assisted temperature control algorithm, a temperature state vector is constructed according to the time series characteristics, and the temperature evolution trend is calculated based on the state transition matrix to obtain a temperature prediction model; Generate a sampling strategy based on the temperature prediction model, and optimize the sampling of the temperature field to obtain the target temperature point distribution; Performing temperature region boundary division on the target temperature point distribution to obtain region division results, and calculating the temperature mean vector and temperature gradient vector of each temperature region according to the region division results, and constructing regional temperature features in combination with time series features; Based on the regional temperature characteristics and the inter-layer coordination parameters, layered control optimization is performed to obtain temperature distribution characteristic data.

4. The intelligent temperature control method of a multi-chip stacking structure according to claim 3, characterized in that: The optimal sampling analysis of the temperature distribution characteristic data to obtain the local temperature control parameters includes: Construct temperature spatial distribution data according to the temperature distribution characteristic data, and input the temperature spatial distribution data into a Minkowski distance calculation unit, set the distance order p=2, calculate the distance value between each sampling point and its surrounding 8 neighborhood points, generate a temperature gradient vector, and obtain a temperature field distance matrix; The distance values ​​in the temperature field distance matrix are sorted according to the magnitude, and the distance values ​​are divided into high gradient area, medium gradient area and low gradient area based on the K-means clustering method, and the high gradient area is analyzed for connected domains to obtain abnormal temperature areas; According to the temperature mean Tm of the abnormal temperature area, the hot spots are divided into three levels: when the temperature is greater than 1.2Tm, it is a first-level hot spot; when the temperature is between 1.0Tm and 1.2Tm, it is a second-level hot spot; when the temperature is between 0.8Tm and 1.0Tm, it is a third-level hot spot, and the hot spot level data is obtained; Calculate the coolant flow demand based on the hotspot level data, set the base flow of the first-level hotspot to Q, the flow of the second-level hotspot to 0.8Q, and the flow of the third-level hotspot to 0.6Q, and obtain the flow distribution parameters; Optimizing the spatial distribution of the flow distribution parameters, calculating the temperature gradient vector of the hot spot area, determining the flow direction of the coolant, and obtaining the flow direction control parameters; Based on the flow distribution parameters and the flow direction control parameters, control constraints are set in combination with the maximum temperature bearing capacity Tmax of the chip to obtain basic control parameters, and dynamic compensation calculation is performed on the basic control parameters. The compensation coefficient is set according to the temperature change rate of the hot spot area to obtain the local temperature control parameters.

5. The intelligent temperature control method of a multi-chip stacking structure according to claim 4, characterized in that: The temperature-flow multi-level coordinated control model is constructed based on the local temperature control parameters, and a coolant flow distribution scheme and flow direction control instructions are generated, including: Based on the local temperature control parameters, a thermodynamic parameter model including refrigeration power, fluid resistance and temperature drop coefficient is established, and a temperature-flow mapping relationship is generated; The temperature-flow mapping relationship is processed by region, and the chip area is divided into a high temperature control area, a medium temperature control area and a low temperature control area based on the hotspot level data, and the control target temperatures are set to F1, F2 and F3 respectively to obtain the partition control parameters; A three-stage series control loop is constructed based on the partition control parameters to obtain a basic flow configuration scheme, and a refrigeration effect of the basic flow configuration scheme is simulated and calculated to obtain flow threshold data; Based on the flow threshold data, the ratio of the actual temperature drop ΔT to the theoretical temperature drop ΔT0 is defined as the heat dissipation efficiency η, and the heat dissipation efficiency value is calculated for each control area to obtain efficiency distribution data; Normalizing the efficiency distribution data, defining the area with the highest heat dissipation efficiency as a reference area, calculating the efficiency ratio β of other areas relative to the reference area, and obtaining a flow correction coefficient; The coolant flow is redistributed based on the flow correction coefficient, the baseline flow of each area is multiplied by the correction coefficient β to obtain a coolant flow distribution plan, and the fluid pressure drop ΔP of each control area is calculated according to the coolant flow distribution plan, the coolant flow direction is determined based on the minimum pressure drop principle, and the flow direction control instruction is obtained.

6. The intelligent temperature control method of a multi-chip stacking structure according to claim 5, characterized in that: The method of controlling the coolant flow rate and temperature based on the coolant flow distribution scheme and the flow direction control instruction, and performing gradient cooling on the temperature abnormal area to obtain temperature adjustment process data includes: Based on the coolant flow distribution scheme and the flow direction control instruction, convert the coolant reference flow into a control signal to obtain an execution instruction sequence; Prioritizing the execution instruction sequence according to the order of temperature anomaly levels, giving priority to executing the cooling control instruction of the first-level hot spot, and executing the control instructions of the second-level hot spot and the third-level hot spot in sequence, to obtain an execution timing plan; Based on the execution timing scheme, the cooling execution unit is started, and the coolant flow rate is gradually increased from the initial value Q0 to the target value Qt according to the step length ΔQ, and a stable flow output is maintained in each control cycle to obtain a step control curve; Cooling control is performed on each temperature abnormality area according to the step control curve to obtain temperature response data, and the temperature response data is compared and analyzed with the steady-state temperature set to calculate the temperature change rate of each control area and the temperature difference of adjacent areas to obtain temperature distribution data; Performing deviation analysis on the temperature distribution data to obtain a flow correction value, and inputting the flow correction value into a cooling controller to perform online compensation on the coolant flow to obtain corrected control data; The corrected control data is stored and sorted, and the temperature change curve, flow change curve and control parameter sequence in the complete temperature adjustment process are recorded to obtain the temperature adjustment process data.

7. The intelligent temperature control method of a multi-chip stacking structure according to claim 6, characterized in that: The multi-stable distribution analysis of the temperature adjustment process data is performed, the region identification parameters of the agent-assisted temperature control algorithm and the coordination parameters of the hierarchical control framework are updated, and the temperature control optimization strategy is generated, including: Identify the temperature rising section, stable section and fluctuation section of the temperature adjustment process data to obtain time series characteristic data; Performing state analysis on the stable segments in the time series characteristic data to obtain a steady-state temperature set, and using a density clustering algorithm to classify the steady-state points in the steady-state temperature set to obtain steady-state distribution parameters; Based on the steady-state distribution parameters, the temperature field is spatially divided to obtain the temperature field structural characteristics, and the temperature field structural characteristics are dynamically optimized to obtain parameter correction values; According to the parameter correction amount, the temperature partition parameters and hot spot discrimination parameters of the agent-assisted temperature control algorithm are updated, and the sensitivity coefficient of the monitoring unit and the response coefficient of the control unit in the hierarchical control framework are corrected to obtain updated control parameters; The updated control parameters and heat dissipation efficiency values ​​are substituted into the temperature field evolution model, the control parameter weights are adjusted according to the heat dissipation efficiency distribution, an optimized control scheme is generated, and based on the optimized control scheme, the heat dissipation efficiency η is used as the control effect evaluation index, and the control objective function J=f(η,δ,c) is set, where δ is the control deviation and c is the control cost, to solve the temperature control optimization strategy.

8. An intelligent temperature control device with a multi-chip stacking structure, characterized in that: An intelligent temperature control method for executing a multi-chip stacking structure according to any one of claims 1 to 7, comprising: An acquisition module is used to acquire target temperature data of each layer of chips in a multi-chip stacking structure and establish a nonlinear temperature response model; A construction module is used to construct a hierarchical control framework based on the nonlinear temperature response model, and input the target temperature data into the agent-assisted temperature control algorithm for processing to obtain temperature distribution characteristic data; An analysis module, used for performing optimal sampling analysis on the temperature distribution characteristic data to obtain local temperature control parameters; A generation module, used to construct a temperature-flow multi-level coordinated control model based on the local temperature control parameters, and generate a coolant flow distribution plan and a flow direction control instruction; A processing module, used to control the coolant flow rate and temperature based on the coolant flow distribution scheme and the flow direction control instruction, and to perform gradient cooling on the temperature abnormal area to obtain temperature adjustment process data; An updating module is used to perform multi-stable distribution analysis on the temperature adjustment process data, update the region identification parameters of the agent-assisted temperature control algorithm and the coordination parameters of the hierarchical control framework, and generate a temperature control optimization strategy.

9. A computer device, characterized in that: The computer device comprises: a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instructions in the memory to enable the computer device to execute the intelligent temperature control method for a multi-chip stacking structure according to any one of claims 1 to 7.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the intelligent temperature control method for the multi-chip stacking structure according to any one of claims 1 to 7 is implemented.

Images