A high-computing-power chip partition liquid cooling control method and system

Abstract

The present application relates to the technical field of semiconductor device liquid cooling heat dissipation, and particularly relates to a high-computing-power chip partition liquid cooling control method and system. The method obtains regional temperature states of multiple heat dissipation regions and actual flow rates of multiple liquid cooling branches, and forms a regional cooling demand set; converts the regional cooling demand into branch candidate flow rates according to a regional-branch cooling influence relationship, and combines hydraulic coupling constraints of liquid cooling system operation to generate a branch target flow rate set; adjusts valve opening degrees and performs circulation pump compensation according to differences between the target flow rates and the actual flow rates, and obtains branch actual flow rates and regional temperature response results; forms flow rate execution residuals and regional temperature response residuals respectively, corrects hydraulic control parameters and the regional-branch cooling influence relationship, and generates parameter update results of a next control cycle. The present application matches regional cooling demand under cross cooling conditions with multi-branch flow rate distribution, and distinguishes hydraulic execution deviation from temperature response deviation.

Description

Technical Field

[0001] This invention relates to the field of semiconductor device heat dissipation and liquid cooling control technology, and in particular to a high-performance chip partitioned liquid cooling control method and system. Background Technology

[0002] In the field of liquid cooling technology for semiconductor devices, high-performance chips typically employ a liquid cooling system consisting of a liquid cooling plate, multiple liquid cooling branches, branch regulating valves, a circulating pump, temperature sensors, and flow sensors. The coolant flow rate is adjusted based on the chip area temperature, the actual flow rate of the branch, or the supply and return pressure difference. Some existing solutions divide the chip into multiple heat dissipation areas and adjust the target flow rate or the opening of the regulating valve for each liquid cooling branch according to a preset correspondence between the heat dissipation area and the liquid cooling branch.

[0003] Existing solutions often rely on a fixed correspondence between heat dissipation areas and liquid cooling branches for flow distribution. When a flow change in one liquid cooling branch simultaneously affects multiple heat dissipation areas, or when a heat dissipation area is cooled by multiple liquid cooling branches, the fixed correspondence fails to reflect the actual relationship between the flow change in the liquid cooling branch and the temperature changes in multiple heat dissipation areas. Furthermore, when multiple liquid cooling branches share a circulating pump, supply manifold, and return manifold, flow adjustment in one liquid cooling branch will cause changes in the supply and return pressure differential, affecting the actual flow of other liquid cooling branches. When each branch independently establishes its target flow rate, it is easy for the target flow rates of multiple branches to fail to be established simultaneously.

[0004] Existing feedback control typically adjusts branch control valves or circulating pumps based on the deviation between the actual and target flow rates of the branch, or on changes in the adjusted regional temperature. When the regional temperature does not change as expected, existing methods struggle to distinguish between situations where the target flow rate of the branch has not been actually established and situations where the actual flow rate of the branch has been established but the regional temperature response does not meet expectations. The cooling relationship between hydraulic control parameters and the cooling effect between the heat dissipation area and the liquid-cooled branch is easily corrected using the same feedback path. Therefore, it is necessary to address the cross-cooling relationship and hydraulic coupling state between multiple heat dissipation areas and multiple liquid-cooled branches of high-performance chips, and to solve the problems of jointly forming the target flow rate set of the branches and distinguishing between flow rate execution deviations and regional temperature response deviations. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a high-performance chip partitioned liquid cooling control method, comprising: S100: Obtain the regional temperature status of multiple heat dissipation areas and the actual flow rate of multiple liquid cooling branches; form a regional cooling demand set based on the regional temperature status; S200. Obtain the region-branch cooling influence relationship, which characterizes the change in liquid cooling branch flow rate and the temperature change of multiple heat dissipation areas; form branch candidate flow rates based on the region cooling demand set and the region-branch cooling influence relationship; apply hydraulic coupling constraints to the branch candidate flow rates according to the liquid cooling system operation constraints to generate a branch target flow rate set. S300. Based on the difference between the target flow rate set of the branch and the actual flow rate of the branch, generate and execute valve opening adjustment command and circulating pump compensation command, and obtain the adjusted actual flow rate of the branch and the area temperature response result. S400: Compare the target flow set of the branch with the adjusted actual flow of the branch to form a flow execution residual; determine the expected regional temperature response based on the regional-branch cooling influence relationship and the adjusted actual flow of the branch; compare the expected regional temperature response with the regional temperature response result to form a regional temperature response residual; correct the hydraulic control parameters based on the flow execution residual; correct the regional-branch cooling influence relationship based on the regional temperature response residual; and generate the parameter update result for the next control cycle.

[0006] Furthermore, the regional temperature state includes the regional temperature, the regional reference temperature, and the regional temperature change. The temperature of the region is compared with the reference temperature of the region to generate a regional temperature difference; The change in regional temperature is generated based on the difference between the regional temperature in the current control cycle and the regional temperature in the previous control cycle. The regional temperature rise rate is generated based on the ratio of the regional temperature change to the control cycle duration. The regional cooling requirements of each heat dissipation area are determined based on the regional temperature difference and the regional temperature rise rate, and the regional cooling requirement set is formed according to the regional identifier.

[0007] Furthermore, the region-branch cooling influence relationship includes region identifier, branch identifier, and cooling influence coefficient; Multiple area signs are associated with the same branch road sign, and multiple branch road signs are associated with the same area sign; The cooling influence coefficient characterizes the relationship between the unit flow rate change of the corresponding liquid cooling branch and the regional temperature change of the corresponding heat dissipation area.

[0008] Furthermore, the formation of the cooling influence coefficient includes: Adjust the branch regulating valve opening of the target liquid cooling branch according to the valve opening change limit of a single control cycle, and obtain the actual flow change of the target liquid cooling branch. Acquire the regional temperature changes of multiple heat dissipation areas within the regional temperature response time window; Multiple cooling influence coefficients are formed for the target liquid-cooled branch based on the actual flow rate changes of the branch and the regional temperature changes of each heat dissipation area. Multiple cooling influence coefficients are associated with branch identifiers and area identifiers to form the area-branch cooling influence relationship.

[0009] Furthermore, obtaining the actual flow of the branch includes: The actual flow rate of the liquid-cooled branch is obtained by a flow sensor installed in the liquid-cooled branch; When the flow sensor malfunctions, the inlet and outlet pressures of the liquid-cooled branch are acquired, the branch pressure difference is calculated, and the actual flow rate of the branch is calculated based on the branch pressure difference and the branch resistance parameters.

[0010] Furthermore, the hydraulic control parameters include branch resistance parameters, valve opening-flow correspondence parameters, and circulating pump operating status; The operating constraints of the liquid cooling system include the upper limit of total flow, the upper limit of branch flow, the lower limit of branch flow, the range of supply and return hydraulic pressure difference, the operating range of the circulating pump, and the limit of valve opening change in a single control cycle. The hydraulic coupling constraints include: The sum of the candidate flows of multiple branches is compared with the upper limit of the total flow, and the candidate flows of each branch are compared with the corresponding upper limit and lower limit of the branch flow. When the sum of multiple branch candidate flows exceeds the total flow limit, the corresponding branch candidate flows are adjusted according to the cooling requirements of each area. The effect of a change in the flow rate of one liquid-cooled branch on the actual flow rate of other liquid-cooled branches is corrected based on the hydraulic control parameters to form the target flow rate set of the branches.

[0011] Furthermore, generating and executing the valve opening adjustment command and the circulating pump compensation command includes: Based on the difference between the target flow rate and the actual flow rate of each branch and the valve opening-flow rate corresponding parameter, the valve opening adjustment amount of each branch regulating valve is generated. The valve opening adjustment command is executed according to the valve opening change limit of the single control cycle. Obtain the pressure of the supply manifold and the return manifold to form a supply and return hydraulic pressure difference; When the supply and return hydraulic pressure difference is lower than the lower limit of the supply and return hydraulic pressure difference range, a circulating pump speed adjustment command is generated and executed based on the total difference between the target flow rate of multiple branches and the actual flow rate of multiple branches.

[0012] Furthermore, the difference between the target flow rate of each branch and the actual flow rate of the corresponding branch after adjustment is used as the flow rate execution residual; Based on the actual flow rate changes of each liquid cooling branch before and after adjustment and the corresponding cooling influence coefficient, the expected regional temperature response of each heat dissipation area is determined. The difference between the predicted regional temperature response and the corresponding regional temperature response result is taken as the regional temperature response residual; When the flow execution residual exceeds the flow feedback deviation threshold, the branch resistance parameter and the valve opening-flow correspondence parameter are corrected. When the flow execution residual does not exceed the flow feedback deviation threshold and the regional temperature response residual exceeds the temperature response residual threshold, the corresponding cooling influence coefficient is corrected.

[0013] Furthermore, when the flow execution residual of the liquid cooling branch exceeds the flow feedback deviation threshold for multiple consecutive control cycles, and the opening of the corresponding branch regulating valve reaches the upper limit of the branch regulating valve opening, an abnormal branch mark is generated. Stop allocating branch candidate traffic to the abnormal branch; Based on the region-branch cooling influence relationship, other liquid-cooled branches associated with the heat dissipation area corresponding to the abnormal branch are obtained; According to the upper limit of the branch flow of the other liquid cooling branches, the regional cooling demand corresponding to the abnormal branch is redistributed to the other liquid cooling branches; When the sum of the upper limits of the branch flow of the other liquid cooling branches is less than the sum of the candidate flow of the redistributed branches, a power reduction request is output to the chip.

[0014] This invention also provides a high-performance chip partitioned liquid cooling control system, comprising: The status acquisition module is used to acquire the regional temperature status of multiple heat dissipation areas and the actual flow rate of multiple liquid cooling branches. A regional cooling demand determination module, connected to the status acquisition module, is used to form a regional cooling demand set based on the regional temperature status. The target flow generation module is connected to the regional cooling demand determination module and the status acquisition module. It is used to obtain the regional-branch cooling influence relationship, form branch candidate flow based on the regional cooling demand set and the regional-branch cooling influence relationship, and perform hydraulic coupling constraints on the branch candidate flow according to the liquid cooling system operation constraints to generate a branch target flow set. The valve pump execution control module is connected to the target flow generation module and the status acquisition module. It is used to generate and execute valve opening adjustment commands and circulating pump compensation commands based on the difference between the target flow set of the branch and the actual flow of the branch, and to obtain the adjusted actual flow of the branch and the area temperature response results. The feedback result determination module is connected to the valve pump execution control module and the target flow generation module. It is used to compare the target flow set of the branch with the adjusted actual flow of the branch to form a flow execution residual, and to determine the expected regional temperature response based on the regional-branch cooling influence relationship and the adjusted actual flow of the branch. It compares the expected regional temperature response with the regional temperature response result to form a regional temperature response residual. The parameter update module, connected to the feedback result determination module and the target flow generation module, is used to perform residual correction of hydraulic control parameters based on the flow rate, correct the regional-branch cooling influence relationship based on the regional temperature response residual, generate parameter update results for the next control cycle, and transmit the parameter update results for the next control cycle to the target flow generation module.

[0015] The key innovations of this invention include: (1) Establish the region-branch cooling influence relationship, associate the flow rate change of the liquid cooling branch with the temperature change of multiple heat dissipation areas, and characterize the cooling effect of a liquid cooling branch on multiple heat dissipation areas and multiple liquid cooling branches on the same heat dissipation area through the cooling influence coefficient. The region cooling demand is converted into branch candidate flow rate by the region-branch cooling influence relationship.

[0016] (2) The candidate flow rate of the branch is constrained together with the upper limit of the total flow rate, the upper limit of the branch flow rate, the lower limit of the branch flow rate, the range of the supply and return hydraulic pressure difference, the working range of the circulating pump and the hydraulic control parameters. The influence of the flow rate change of one liquid cooling branch on the actual flow rate of other liquid cooling branches is corrected, and the branch target flow rate set of multiple liquid cooling branches is formed.

[0017] (3) The flow execution residual is formed based on the target flow of the branch and the actual flow of the branch after adjustment, and the regional temperature response residual is formed based on the expected regional temperature response and the regional temperature response result. The flow execution residual is used to correct the hydraulic control parameters, and the regional temperature response residual is used to correct the regional-branch cooling influence relationship. The two types of correction results are respectively used to participate in the target flow generation and valve pump control of the next control cycle.

[0018] The following are its main beneficial effects: (1) By characterizing the actual relationship between the flow rate change of the liquid cooling branch and the temperature change of multiple heat dissipation areas through the regional-branch cooling influence relationship, the regional cooling demand is no longer allocated solely based on the fixed positional relationship between the heat dissipation area and the liquid cooling branch, thus reducing the mismatch between the branch flow rate demand and the regional cooling effect under cross-cooling conditions.

[0019] (2) By jointly forming a set of target flow rates for branches under the constraints of the liquid cooling system operation and the hydraulic coupling of the branches, the flow rate regulation of one liquid cooling branch can affect the flow rate of other liquid cooling branches in the calculation of target flow rates, thereby reducing the situation of total flow exceeding the limit, branch flow conflict and multiple target flow rates not being established at the same time caused by independent regulation of each branch.

[0020] (3) By using the residual flow rate and the residual flow rate of the area temperature response to determine the different states of the branch target flow rate not being actually established and the actual flow rate being established but the area temperature response deviating from the expected result, the branch resistance parameter, valve opening-flow rate corresponding parameter and cooling influence coefficient are respectively put into the corresponding correction path, so as to avoid uniformly treating control deviations of different reasons as continuing to increase the branch flow rate. Attached Figure Description

[0021] Figure 1 A flowchart illustrating a high-performance chip partitioning liquid cooling control method provided in this application embodiment; Figure 2 This is a structural block diagram of a high-performance chip partitioned liquid cooling control system provided in an embodiment of this application. Detailed Implementation

[0022] Example 1: Refer to Figure 1 This is a flowchart illustrating a high-performance chip partitioning liquid cooling control method provided in an embodiment of the present invention. The process may include at least steps S100-S400: S100: Obtain the regional temperature status of multiple heat dissipation areas and the actual flow rate of multiple liquid cooling branches; form a regional cooling demand set based on the regional temperature status; S200. Obtain the region-branch cooling influence relationship, which characterizes the change in liquid cooling branch flow rate and the temperature change of multiple heat dissipation areas; form branch candidate flow rates based on the region cooling demand set and the region-branch cooling influence relationship; apply hydraulic coupling constraints to the branch candidate flow rates according to the liquid cooling system operation constraints to generate a branch target flow rate set. S300. Based on the difference between the target flow rate set of the branch and the actual flow rate of the branch, generate and execute valve opening adjustment command and circulating pump compensation command, and obtain the adjusted actual flow rate of the branch and the area temperature response result. S400: Compare the target flow set of the branch with the adjusted actual flow of the branch to form a flow execution residual; determine the expected regional temperature response based on the regional-branch cooling influence relationship and the adjusted actual flow of the branch; compare the expected regional temperature response with the regional temperature response result to form a regional temperature response residual; correct the hydraulic control parameters based on the flow execution residual; correct the regional-branch cooling influence relationship based on the regional temperature response residual; and generate the parameter update result for the next control cycle.

[0023] S100: Obtain the regional temperature status of multiple heat dissipation areas and the actual flow rate of multiple liquid cooling branches; form a regional cooling demand set based on the regional temperature status: This embodiment applies to high-performance chips equipped with liquid-cooled cooling plates. The liquid-cooled cooling plate is thermally connected to the heat dissipation surface of the high-performance chip. Multiple liquid-cooling branches are internally located within the liquid-cooled cooling plate, each connected to a main supply pipe and a main return pipe. Each liquid-cooling branch is equipped with a branch regulating valve, and the main supply pipe is equipped with a circulation pump. The heat dissipation surface of the high-performance chip is divided into multiple heat dissipation zones according to the chip's functional area, heat source location, or the coverage area of ​​the liquid-cooled branches. Each heat dissipation zone is labeled. Temperature sensors are located on the chip packaging structure, the chip's heat dissipation surface, or the side of the liquid-cooled cooling plate closest to the chip, and are associated with the corresponding zone label. The liquid cooling controller reads the current status of the temperature sensor, flow sensor, pressure sensor, branch regulating valve, and circulation pump according to a control cycle, and executes the zoned liquid cooling control process described in this embodiment.

[0024] The regional temperature status is determined by the status acquisition module. This module first aligns the temperature values ​​from multiple temperature sensors according to their acquisition time, and then assigns the temperature values ​​to the corresponding heat dissipation areas based on the correspondence between the temperature sensors and area identifiers. When multiple temperature sensors are configured for a heat dissipation area, the status acquisition module compares the temperature values ​​of each sensor within the same area. When the differences between the temperature values ​​are within the configured range, the average of the multiple temperature values ​​is used as the regional temperature. When the difference between a temperature value and other temperature values ​​within the same area exceeds the configured range for multiple consecutive acquisition cycles, the corresponding temperature sensor is recorded as abnormal, and the regional temperature is formed using the temperature values ​​from the remaining sensors. When only one temperature sensor is configured for a heat dissipation area, the liquid cooling controller also reads the temperature changes of the chip's internal temperature detection unit or adjacent heat dissipation areas. When a single-cycle temperature change occurs in the region and the chip's input power consumption does not change accordingly, the valid regional temperature from the previous control cycle is retained for this control cycle and reread in the next acquisition cycle.

[0025] The regional reference temperature is selected based on the current operating status of the high-performance chip. In one embodiment, the chip power consumption control interface provides the current chip input power consumption and power consumption level to the liquid cooling controller. The liquid cooling controller reads the target temperature corresponding to the power consumption level as the regional reference temperature. When different heat dissipation areas have different target temperatures, the regional reference temperatures are configured separately according to the area identifier. In another embodiment, the regional reference temperature is formed by the regional temperatures of multiple heat dissipation areas. The liquid cooling controller excludes the heat dissipation area with the highest current temperature and the heat dissipation area where the temperature sensor is in an abnormal state, and then forms the regional reference temperature based on the regional temperatures of the remaining heat dissipation areas. The latter embodiment is suitable for the operating condition where different functional areas inside the chip alternately form local hot spots. The regional reference temperature changes with the overall temperature level of the chip, and a fixed temperature value is not directly used as the control basis for all heat dissipation areas.

[0026] The zone cooling demand determination module compares the zone temperature with the zone reference temperature to form a zone temperature difference. Simultaneously, it reads the zone temperature of the same zone in the current control cycle and the previous control cycle, calculates the zone temperature change based on the difference, and determines the zone temperature rise rate based on the ratio of the zone temperature change to the control cycle duration. The zone temperature difference reflects the degree of deviation of the current zone from the zone reference temperature, while the zone temperature rise rate reflects the direction of temperature change in the zone under the current operating state. When both the zone temperature difference and the zone temperature rise rate are positive, the zone cooling demand increases accordingly. When the zone temperature difference is positive but the zone temperature rise rate is not positive, the zone cooling demand is primarily based on the zone temperature difference. When the zone temperature difference is not positive but the zone temperature rise rate is positive, the liquid cooling controller retains the basic cooling demand to avoid missing rapid temperature rise states before the zone temperature exceeds the zone reference temperature. When both the zone temperature difference and the zone temperature rise rate are not positive, the zone cooling demand is taken from the basic cooling demand corresponding to that heat dissipation zone.

[0027] In one implementation, the regional cooling demand determination module configures conversion coefficients for the regional temperature difference and the regional temperature rise rate, respectively. The product of the regional temperature difference and its corresponding conversion coefficient, and the product of the regional temperature rise rate and its corresponding conversion coefficient are added together. The result, after being limited by a lower and upper limit, is used as the regional cooling demand. The lower limit corresponds to the basic cooling demand required to maintain coolant circulation in the heat dissipation area, while the upper limit corresponds to the allowable cooling demand of the heat dissipation area under the current chip input power consumption and liquid supply temperature. The conversion coefficients can be configured based on the cold plate thermal test results or corrected based on the regional temperature response in subsequent control cycles. This process does not directly provide the target flow rate of the liquid cooling branch; instead, it first converts the chip-side temperature state into the regional-side cooling demand, avoiding direct adjustment of a single branch without considering the combined effect of multiple liquid cooling branches.

[0028] The actual flow rate of multiple liquid-cooled branches is read by flow sensors installed in the corresponding liquid-cooled branches. The status acquisition module continuously checks the output values ​​of the flow sensors. If the flow sensor does not return a detection value, the detection value remains unchanged while the opening of the branch regulating valve changes, or the detection value is in the opposite direction to the flow change reflected by the branch pressure difference, the flow sensor is recorded as being in an abnormal state. At this time, the status acquisition module reads the inlet and outlet pressures of the liquid-cooled branch, calculates the branch pressure difference, and converts the actual flow rate of the branch based on the current branch regulating valve opening, branch resistance parameters, and branch pressure difference. After the flow sensor recovers, the liquid-cooled controller compares the directly detected actual flow rate of the branch with the actual flow rate of the branch converted from the branch pressure difference; once the difference returns to the configured range, the detection value of the flow sensor is restored.

[0029] The regional cooling demand determination module summarizes the regional cooling demands of each heat dissipation area according to the regional identifier, forming a regional cooling demand set. The regional cooling demand set includes the regional identifier and the corresponding regional cooling demand; when a temperature sensor is in an abnormal state, the valid status of the corresponding regional temperature status is also retained for subsequent steps to determine whether the regional cooling demand participates in flow allocation. The status acquisition module simultaneously retains the actual flow rate of the branch corresponding to each branch identifier, the branch pressure difference, the branch regulating valve opening, and the circulating pump speed. S200 receives the regional cooling demand set, the actual flow rate of multiple liquid cooling branches, and the currently valid regional-branch cooling influence relationship, and completes the conversion of regional cooling demand into liquid cooling branch flow demand accordingly.

[0030] S200. Obtain the region-branch cooling influence relationship, which characterizes the flow rate change of the liquid cooling branch and the temperature change of multiple heat dissipation areas; based on the region cooling demand set and the region-branch cooling influence relationship, form branch candidate flow rates; apply hydraulic coupling constraints to the branch candidate flow rates according to the liquid cooling system operation constraints to generate a branch target flow rate set: The region-branch cooling influence relationship is stored in the parameter storage unit of the liquid cooling controller, which includes region identifier, branch identifier, and cooling influence coefficient. The cooling influence coefficient characterizes the correlation between the unit flow rate change of the corresponding liquid cooling branch and the regional temperature change of the corresponding heat dissipation area within the region temperature response time window. Since the coolant transfers cooling energy to adjacent locations through the cold plate material and chip packaging structure as it flows inside the liquid cooling plate, a flow rate change in the same liquid cooling branch may cause temperature changes in multiple heat dissipation areas; the same heat dissipation area may also be cooled by multiple liquid cooling branches simultaneously. Therefore, the target flow generation module does not establish a fixed one-to-one correspondence between region and branch locations, but instead reads multiple branch identifiers associated with each region identifier and their corresponding cooling influence coefficients.

[0031] The region-branch cooling influence relationship can be established before the liquid cooling system is put into operation. During calibration, the chip input power consumption is kept within the configured range, the liquid supply temperature remains stable, and the liquid cooling controller selects a target liquid cooling branch and adjusts the branch regulating valve opening of the target liquid cooling branch according to the valve opening change limit of a single control cycle. After the actual flow rate of the branch enters a stable range, the liquid cooling controller records the change in the actual flow rate of the branch before and after adjustment, and records the regional temperature change of multiple heat dissipation areas within the regional temperature response time window. The target flow generation module generates multiple cooling influence coefficients corresponding to the target liquid cooling branch based on the actual flow rate change of the branch and the regional temperature change of each heat dissipation area. This process is then repeated for multiple liquid cooling branches to form the region-branch cooling influence relationship according to the branch identifier and the region identifier.

[0032] During calibration, if the chip input power consumption or liquid supply temperature changes beyond the configured range within the regional temperature response time window, this regional temperature change will not be used to form the cooling influence coefficient. When the actual flow rate of a branch does not reach a stable range, the liquid cooling controller extends the acquisition time. However, if the extended acquisition time reaches the configured upper limit and the flow rate still does not stabilize, the calibration ends, and the most recent valid cooling influence coefficient for that branch is retained. For liquid cooling branches that do not produce identifiable temperature changes for a certain heat dissipation area, the corresponding cooling influence coefficient is recorded as zero; this liquid cooling branch does not undertake the regional cooling demand proposed by that heat dissipation area in this control cycle.

[0033] After receiving the regional cooling demand set generated by S100, the target flow generation module first looks up the corresponding branch identifiers and cooling influence coefficients based on the regional identifier. For a heat dissipation area associated with multiple liquid cooling branches, the target flow generation module allocates the regional cooling demand for that heat dissipation area according to the cooling influence coefficient of each liquid cooling branch; liquid cooling branches with larger cooling influence coefficients bear larger flow demands. For a liquid cooling branch associated with multiple heat dissipation areas, the target flow generation module summarizes the flow demands borne by that liquid cooling branch for multiple heat dissipation areas, forming the branch candidate flow rate for that liquid cooling branch. The resulting branch candidate flow rate reflects the heat-side demand, but does not yet consider the circulating pump's liquid supply capacity, branch flow boundaries, and the hydraulic influence between parallel branches.

[0034] The operating constraints of the liquid cooling system are formed by the structural parameters of the liquid cooling plate, the operating range of the branch regulating valves, and the operating parameters of the circulating pump. These constraints include the upper limit of total flow, the upper limit of branch flow, the lower limit of branch flow, the range of supply and return hydraulic pressure difference, the operating range of the circulating pump, and the limit of valve opening variation in a single control cycle. The lower limit of branch flow corresponds to the flow boundary for the liquid cooling branch to maintain continuous circulation. The upper limit of branch flow is determined by the dimensions of the liquid cooling branch flow channel, the opening range of the branch regulating valve, and the allowable pressure difference. The upper limit of total flow is determined based on the current speed of the circulating pump, the supply and return hydraulic pressure difference, and the flow state of the supply and return main pipes. The nameplate flow rate of the circulating pump is not directly used as the upper limit of total flow for each operating condition.

[0035] The target flow generation module first compares the candidate flow of each branch with the corresponding upper and lower flow limits of the branch. Candidate flow below the lower flow limit is adjusted to the lower flow limit; flow above the upper flow limit is temporarily recorded as unallocated cooling demand, and other liquid-cooled branches with non-zero cooling impact coefficients on the same heat dissipation area are identified. When other liquid-cooled branches still have flow margin, the unallocated cooling demand is allocated according to the cooling impact coefficients and flow margins of the other liquid-cooled branches; when no available flow margin is available, the unallocated cooling demand is retained for subsequent exception handling or chip power reduction control.

[0036] After completing the branch boundary verification, the target flow generation module compares the sum of multiple branch candidate flows with the total flow limit. If the sum of multiple branch candidate flows does not exceed the total flow limit, the module continues to verify whether the candidate flows of each branch can be established simultaneously based on the hydraulic control parameters. If the sum of multiple branch candidate flows exceeds the total flow limit, the lower limit of the flow rate for each liquid-cooled branch is retained first, and then the remaining flow rate is allocated from high to low according to the regional cooling demand. When a liquid-cooled branch simultaneously undertakes the cooling demand of multiple regions, its flow allocation is calculated according to the cooling influence coefficient corresponding to that branch; when the same region is jointly undertaken by multiple liquid-cooled branches, further allocation is avoided to liquid-cooled branches that are already close to their flow limit.

[0037] Hydraulic control parameters include branch resistance parameters, valve opening-flow correspondence parameters, and circulating pump operating status. The target flow generation module calculates the impact of a flow change in a liquid-cooled branch on the supply-return hydraulic pressure difference based on the current branch regulating valve opening, branch pressure difference, and branch resistance parameters. It then calculates the impact on the actual flow of other parallel liquid-cooled branches based on the change in the supply-return hydraulic pressure difference. For example, if the regulating valve opening of the first liquid-cooled branch increases, the pressure in the main supply pipe decreases or the pressure in the main return pipe increases. Even if the valve opening of the second liquid-cooled branch remains unchanged, the actual flow rate may still decrease. The target flow generation module then corrects the candidate flow rates for the first and second liquid-cooled branches accordingly, ensuring that the corrected flow rates match the current circulating pump operating range and the supply-return hydraulic pressure difference range.

[0038] In one implementation, the target flow generation module calculates the branch flow changes caused by valve opening changes sequentially along the liquid-cooled branches, and updates the supply-return hydraulic pressure difference and the expected flow of the remaining liquid-cooled branches after each calculation, until the expected flow changes of multiple liquid-cooled branches are all within the configured range. In another implementation, the parameter storage unit stores the branch flow change relationships of multiple branch regulating valves under different opening combinations. The target flow generation module reads the corresponding hydraulic control parameters based on the current valve opening combination and the candidate branch flow. Both methods are constrained by the actual hydraulic state of the liquid-cooled branches and do not treat the candidate flow of each branch as an independent control variable.

[0039] After verifying the branch boundaries, total flow, and hydraulic coupling, the target flow generation module forms a branch target flow set according to the branch identifier. The branch target flow set includes the target flow of each liquid-cooled branch within the current control cycle. For liquid-cooled branches where the flow sensor is in an abnormal state but the branch pressure difference is valid, the branch target flow still participates in subsequent regulation, and its actual flow is calculated from the branch pressure difference and branch resistance parameters. For liquid-cooled branches where neither the actual flow nor the branch pressure difference has a valid detection value, its branch target flow is not increased in this control cycle, and the cooling demand of the corresponding area is transferred to other liquid-cooled branches with cooling influence. S300 generates control actions for the branch regulating valve and circulating pump based on the difference between the branch target flow set and the current branch actual flow.

[0040] S300. Based on the difference between the target flow rate set of the branch and the actual flow rate of the branch, generate and execute valve opening adjustment commands and circulating pump compensation commands to obtain the adjusted actual flow rate of the branch and the area temperature response results: The valve-pump execution control module receives the branch target flow set generated by S200, and simultaneously reads the current branch actual flow, branch regulating valve opening, circulating pump speed, supply main pressure, and return main pressure obtained by S100. For each liquid-cooled branch, the valve-pump execution control module calculates the difference between the branch target flow and the corresponding branch actual flow, and then calculates the valve opening adjustment amount of the branch regulating valve based on the valve opening-flow correspondence parameter. When the flow difference is positive, the valve opening adjustment direction is to increase; when the flow difference is negative, the valve opening adjustment direction is to decrease. When the flow difference is within the flow feedback deviation range, the current opening of the regulating valve of that branch is maintained in this control cycle.

[0041] The valve opening adjustment is constrained by the branch control valve opening range and the valve opening change limit for a single control cycle. When the calculated valve opening adjustment exceeds the valve opening change limit for a single control cycle, the valve pump execution control module intercepts the valve opening adjustment for the current control cycle according to the limit, and the remaining adjustment is executed in subsequent control cycles. This method keeps flow changes and regional temperature responses within a identifiable range and avoids sudden changes in supply and return hydraulic pressure difference caused by simultaneous large movements of multiple branch control valves. When the current opening of the branch control valve has reached its upper limit but the target flow rate of the branch is still greater than the actual flow rate of the branch, the valve pump execution control module retains the unfinished flow difference of the branch and records the valve opening limit status in the execution results.

[0042] When multiple liquid-cooled branches share a circulating pump, the valve-pump execution control module first executes the branch regulating valve control, and then determines the circulating pump operation based on the adjusted supply and return hydraulic pressure difference. Specifically, the liquid-cooled controller sends valve opening adjustment commands to multiple branch regulating valves. After receiving opening acknowledgments from the branch regulating valves, it reads the supply and return main pressures and calculates the supply and return hydraulic pressure difference. If some branch regulating valves do not return opening acknowledgments, the circulating pump compensation is not increased in this control cycle, and the status of the branch regulating valve is reread. If there is still no acknowledgment after rereading, the branch is kept at the current opening, and other branches continue to execute according to the confirmed valve opening.

[0043] When the supply and return hydraulic pressure difference is below the lower limit of the supply and return hydraulic pressure difference range, it indicates that the current operating state of the circulating pump is insufficient to support the target flow rate set of the branch. The valve-pump execution control module summarizes the total difference between the target flow rate of multiple branches and the actual flow rate of multiple branches, and generates a circulating pump speed adjustment amount based on the current speed and operating range of the circulating pump. The circulating pump speed adjustment amount is also limited according to the allowable variation range of a single control cycle, and then a circulating pump compensation command is generated. When the supply and return hydraulic pressure difference is within the configured range, the circulating pump speed is not increased to avoid replacing the zonal control of the branch regulating valve by increasing the overall flow of the circulating pump. When the supply and return hydraulic pressure difference is higher than the upper limit of the configured range, in one implementation, the circulating pump speed is reduced; in another implementation, the pressure difference change is absorbed by the main pipe regulating valve.

[0044] After the valve opening adjustment command and the circulating pump compensation command are executed, the liquid cooling controller enters the flow stabilization monitoring phase. Flow stabilization monitoring does not rely solely on a fixed waiting time; instead, it continuously reads the actual flow rate of each liquid cooling branch. When the flow rate change at adjacent acquisition times is within the configured range, and the supply and return hydraulic pressure difference does not continue to change unidirectionally, the current flow rate is recorded as the adjusted actual flow rate of the branch. If the liquid cooling branch continues to fluctuate after reaching the upper limit of the flow monitoring time, the flow execution result of that liquid cooling branch retains the fluctuating state. When generating the flow execution residual, S400 does not use a single instantaneous flow rate value as the sole basis, but instead uses the average actual flow rate of the branch within this monitoring phase.

[0045] For liquid-cooled branches where the flow sensor is in an abnormal state, the liquid-cooled controller reads the branch inlet and outlet pressures after the valve actuation is completed. Based on the adjusted branch differential pressure, branch resistance parameters, and the valve opening-flow correspondence parameter, it calculates the adjusted actual flow rate of the branch. If the branch regulating valve opening has increased, but the branch differential pressure and calculated flow rate have not changed accordingly, the valve actuation status, branch differential pressure change, and calculated flow rate of that branch are retained in the flow execution results, and the results are attributed to changes in branch resistance, valve jamming, or insufficient circulating pump capacity in the S400 area.

[0046] The acquisition of zone temperature response begins after the actual flow rate of the branch enters a stable range. The liquid cooling controller records the moment when the flow rate stabilizes as the starting point of the zone temperature response time window and reads the zone temperature of each heat dissipation area at the starting point. The length of the response time window is configured based on the chip packaging structure, the thermal capacity of the liquid cooling plate, and historical temperature response times. If the window is too short, the change in coolant flow rate has not yet been transmitted to the chip heat dissipation area; if the window is too long, changes in chip workload may interfere with the temperature response corresponding to changes in branch flow rate. Therefore, in one embodiment, the liquid cooling controller determines the current response time window based on the time from the change in flow rate in the previous control cycle to the start of continuous change in zone temperature.

[0047] After the response time window ends, the liquid cooling controller reads the temperature of each heat dissipation area, calculates the temperature change between the start and end points of the window, and generates a temperature response result according to the area identifier. If the chip input power consumption changes beyond the configured range within the response time window, the liquid cooling controller simultaneously records the power consumption change status; if the liquid supply temperature changes beyond the configured range, it also records the liquid supply temperature change status. These statuses are used in the S400 to verify the validity of the temperature response residual, preventing temperature changes caused by sudden changes in chip power consumption or liquid supply temperature from being directly attributed to liquid cooling branch flow regulation.

[0048] The flow execution results are recorded according to the branch identifier, including the branch target flow, the adjusted actual branch flow, the branch regulating valve opening before and after adjustment, and the circulating pump speed before and after adjustment. The area temperature response results are recorded according to the area identifier, including the start temperature, end temperature, and area temperature change within the response time window. Both types of results correspond to the same valve and pump action in time. Based on this, the S400 verifies whether the liquid cooling system has established the branch target flow and whether the established flow change has produced an area temperature response consistent with the area-branch cooling influence relationship.

[0049] S400: Compare the target flow rate set of the branch with the adjusted actual flow rate of the branch to form a flow execution residual; determine the expected regional temperature response based on the regional-branch cooling influence relationship and the adjusted actual flow rate of the branch; compare the expected regional temperature response with the regional temperature response result to form a regional temperature response residual; correct the hydraulic control parameters based on the flow execution residual; correct the regional-branch cooling influence relationship based on the regional temperature response residual; and generate the parameter update result for the next control cycle. The feedback result determination module first associates the target flow rate set of each branch with the adjusted actual flow rate of the branch obtained by S300, according to the branch identifier. The difference between the target flow rate and the adjusted actual flow rate of each liquid-cooled branch is used as the flow execution residual for that branch. The direction of the flow execution residual reflects the deviation of the actual flow rate from the target flow rate, and the absolute value of the flow execution residual reflects the degree of deviation. For liquid-cooled branches that experience continuous fluctuations during the flow stability monitoring phase, the average actual flow rate of the branch during the monitoring phase is used to form the flow execution residual, and the flow fluctuation state is retained until the parameter update process.

[0050] After generating the flow execution residual, the feedback result determination module calculates the expected regional temperature response for each heat dissipation area based on the actual flow changes of multiple liquid cooling branches before and after adjustment and the current region-branch cooling influence relationship. For a heat dissipation area associated with only one liquid cooling branch, the expected regional temperature response is formed by the actual flow change of that liquid cooling branch and the corresponding cooling influence coefficient. For a heat dissipation area associated with multiple liquid cooling branches, the temperature response corresponding to the flow change of each liquid cooling branch is calculated separately and then summarized. In the case where one liquid cooling branch is associated with multiple heat dissipation areas, the actual flow change of the same branch is calculated with the cooling influence coefficients corresponding to multiple areas to form multiple expected regional temperature responses.

[0051] The regional temperature response record reflects the actual observed regional temperature changes within the response time window. The feedback result determination module compares the predicted regional temperature response with the actual regional temperature response under the corresponding regional identifier, forming the regional temperature response residual. When the actual flow rate of the liquid cooling branch increases, the predicted regional temperature response follows the direction of regional temperature decrease; if the actual regional temperature decrease is less than the predicted decrease, the regional temperature response residual reflects an overestimation of the cooling effect of the current cooling influence coefficient on that branch. If the actual regional temperature decrease is greater than the predicted decrease, the regional temperature response residual reflects an underestimation of the cooling effect of the current cooling influence coefficient on that branch.

[0052] If the chip input power consumption or liquid supply temperature changes beyond the configured range within the response time window, the residual temperature response of the region in this control cycle is not directly used to correct the cooling influence coefficient. The liquid cooling controller retains the current region temperature response result and re-acquires it in the next control cycle. When changes in chip input power consumption and liquid cooling branch flow rate cause changes in region temperature in the same direction, the two are not simply subtracted and mixed; instead, the cooling influence relationship is re-verified after the chip input power consumption enters a stable range. When the temperature sensor exhibits an abnormal state within the response time window, the cooling influence coefficient of the corresponding heat dissipation region retains the most recent valid value.

[0053] The parameter update module selects different correction targets based on the flow execution residual and the regional temperature response residual. When the flow execution residual of a certain liquid cooling branch exceeds the flow feedback deviation threshold, it indicates that the actual flow of the liquid cooling branch has not been established according to the branch target flow. The parameter update module reads the opening change of the branch control valve, the branch differential pressure change, and the circulation pump speed change. When the branch control valve opening has changed but the actual flow change of the branch is insufficient, the branch resistance parameters are recalculated based on the actual branch differential pressure and actual flow, and the valve opening-flow correspondence parameters are re-formed based on the valve opening change and the actual flow change. When multiple liquid cooling branches simultaneously have flow execution residuals in the same direction and the supply and return pressure difference is lower than the configuration range, the parameter update module corrects the circulation pump operating status and does not attribute this situation to the local resistance changes of multiple liquid cooling branches.

[0054] When the flow execution residual does not exceed the flow feedback deviation threshold, but the area temperature response residual exceeds the temperature response residual threshold, it indicates that the branch target flow has been established in the liquid cooling system, but the actual area temperature change is inconsistent with the current area-branch cooling influence relationship. The parameter update module recalculates the cooling influence coefficients under the corresponding area and branch identifiers based on the actual flow changes of the branches and the area temperature response results in this control cycle. When a heat dissipation area is simultaneously affected by multiple liquid cooling branches, only the cooling influence coefficients corresponding to the liquid cooling branches whose actual flow changes in this control cycle are corrected; the original cooling influence coefficients of liquid cooling branches whose actual flow has not undergone identifiable changes are maintained. When the difference between the recalculated cooling influence coefficient and the original cooling influence coefficient exceeds the configuration range, the parameter update module does not replace the original coefficient temporarily, but instead arranges a small flow change in the same direction for verification in the next control cycle.

[0055] When neither the flow execution residual nor the area temperature response residual exceeds the corresponding threshold, the parameter update module retains the current hydraulic control parameters and the area-branch cooling influence relationship. When both the flow execution residual and the area temperature response residual exceed the corresponding threshold, the hydraulic control parameters are processed according to the flow execution residual first, without directly correcting the cooling influence coefficient based on the temperature response of the current cycle. This is because when the target flow rate has not been actually established, insufficient temperature response cannot reflect changes in the cooling influence relationship itself. After the hydraulic control parameters are corrected, the liquid-cooled controller re-executes branch flow regulation in the next control cycle, and then checks the area temperature response based on the new flow execution results.

[0056] When the flow execution residual of a certain liquid cooling branch exceeds the flow feedback deviation threshold for multiple consecutive control cycles, and the opening of the branch's regulating valve reaches its upper limit, the parameter update module generates an abnormal branch marker. This abnormal branch marker is associated with the branch identifier and transmitted to the target flow generation module for the next control cycle. The target flow generation module stops allocating candidate flow rates to this abnormal liquid cooling branch and then searches for other liquid cooling branches associated with the heat dissipation area covered by the abnormal liquid cooling branch based on the region-branch cooling influence relationship. If other liquid cooling branches still have flow margin, the regional cooling demand is reallocated according to the corresponding cooling influence coefficient and the branch's upper flow limit.

[0057] If the sum of the upper limits of the flow rates of other liquid cooling branches is less than the sum of the candidate flow rates of the redistributed branches, the liquid cooling controller outputs a chip power reduction request to the chip power control interface. The chip power reduction request includes the current insufficient cooling capacity status and the corresponding heat dissipation area identifier. After the chip power consumption is adjusted, S100 re-acquires the area temperature status of each heat dissipation area and forms a new set of area cooling requirements based on the adjusted area temperature difference and area temperature rise rate. When the abnormal liquid cooling branch restores the actual flow rate detection and valve action acknowledgment after maintenance, it does not immediately assume the original area cooling requirements, but first runs a control cycle near the lower limit of the branch flow rate; after its flow execution residual returns to the configuration range, it then participates in the subsequent branch candidate flow allocation.

[0058] The parameter update module ultimately generates updated hydraulic control parameters based on branch identifiers, and updated region-branch cooling influence relationships based on region and branch identifiers. It also appends the parameter validity status, constituting the parameter update result for the next control cycle. At the start of the next control cycle, S100 reads the updated parameter validity status; S200 calls the valid region-branch cooling influence relationships to generate candidate branch flow rates and executes hydraulic coupling constraints using the updated hydraulic control parameters. Thus, the region temperature status, actual branch flow rate, target flow rate execution status, and region temperature response form an interconnected liquid cooling control process within continuous control cycles.

[0059] Example 2: Figure 2 A structural block diagram of a high-performance chip partitioned liquid cooling control system according to an embodiment of the present invention is shown. Figure 2 As shown, the structure may include: The status acquisition module 01 is used to acquire the regional temperature status of multiple heat dissipation areas and the actual flow rate of multiple liquid cooling branches. This module is communicatively connected to temperature sensors installed on the high-performance chip, chip packaging structure, or the heat-conducting surface of the liquid cooling plate, as well as flow sensors installed on each liquid cooling branch. It reads temperature and flow detection values ​​according to a control cycle. Multiple temperature sensors are bound to corresponding regional identifiers, and multiple flow sensors are bound to corresponding branch identifiers. The status acquisition module performs time alignment on the temperature and flow detection values ​​according to the acquisition time, ensuring that the regional temperature status and actual branch flow rate within the same control cycle correspond to the same liquid cooling operating condition.

[0060] When multiple temperature sensors are set in the same heat dissipation area, the status acquisition module performs difference verification on the temperature detection values ​​in the same area. When the differences between multiple temperature detection values ​​are within the configured range, the area temperature of the heat dissipation area is formed based on the multiple temperature detection values. When a temperature detection value deviates from the other temperature detection values ​​in the same area for multiple consecutive acquisition cycles, the corresponding temperature sensor is set to an abnormal state, and the area temperature is formed by using the other valid temperature detection values ​​in the current control cycle. The area temperature status also includes an area reference temperature and an area temperature change. The area reference temperature is obtained from the target temperature corresponding to the current power consumption level of the high-performance chip, or formed by the area temperatures of multiple non-abnormal heat dissipation areas. The area temperature change is obtained by comparing the area temperature of the current control cycle with the area temperature of the previous control cycle.

[0061] If the flow sensor fails to return a detection value, the detection value remains unchanged while the branch regulating valve opening changes, or the direction of flow change reflected by the flow detection value and the branch pressure difference is inconsistent, the status acquisition module will set the corresponding flow sensor to an abnormal state. At this time, the status acquisition module reads the branch inlet pressure and branch outlet pressure of the corresponding liquid cooling branch to form a branch pressure difference. Then, based on the branch pressure difference, the current branch regulating valve opening, and the effective branch resistance parameters from the previous control cycle, the status acquisition module calculates the actual branch flow rate. The status acquisition module transmits the formed regional temperature status to the regional cooling demand determination module and transmits the actual branch flow rate to the target flow generation module and the valve-pump execution control module, respectively.

[0062] The regional cooling demand determination module 02, connected to the status acquisition module, is used to form a regional cooling demand set based on the regional temperature status. After receiving the regional temperature status arranged by regional identifiers, the module compares the regional temperature of each heat dissipation area with the corresponding regional reference temperature to form a regional temperature difference. It also forms a regional temperature rise rate based on the regional temperature change in adjacent control cycles and the control cycle. The regional temperature difference reflects the degree of deviation of the current temperature of the heat dissipation area from the regional reference temperature, while the regional temperature rise rate reflects the direction of temperature change in the heat dissipation area under the current chip operating state. Both serve as the basis for forming the regional cooling demand.

[0063] When the temperature difference and temperature rise rate of a region are both positive, the region cooling demand determination module determines the region cooling demand for that heat dissipation region based on the temperature difference and temperature rise rate. When the temperature difference is positive but the temperature rise rate is not positive, the region cooling demand is determined based on the temperature difference. When the temperature difference is not positive but the temperature rise rate is positive, the basic cooling demand for the corresponding heat dissipation region is retained. When both the temperature difference and the temperature rise rate are not positive, the region cooling demand for that heat dissipation region is limited to the range of the basic cooling demand. The basic cooling demand corresponds to the lower limit of the flow rate of the liquid cooling branch that maintains continuous flow, and the cooling demand of the corresponding region is not directly set to zero due to a decrease in region temperature within a single control cycle.

[0064] In one implementation, the regional temperature difference and the regional temperature rise rate are multiplied by configured conversion factors, and the results are summed and then limited by an upper limit and a lower limit of cooling demand to form the regional cooling demand. If the regional temperature state is abnormal, or if the chip input power consumption changes beyond the configured range in the current control cycle, the regional cooling demand determination module does not expand the corresponding regional cooling demand, but instead retains the valid value from the previous control cycle and waits for recalculation in the next control cycle. The regional cooling demand determination module aggregates multiple regional cooling demands according to regional identifiers to form a regional cooling demand set, and transmits the regional cooling demand set to the target flow generation module.

[0065] The target flow generation module 03, connected to the regional cooling demand determination module and the status acquisition module, is used to acquire the regional-branch cooling influence relationship, form branch candidate flow rates based on the regional cooling demand set and the regional-branch cooling influence relationship, and apply hydraulic coupling constraints to the branch candidate flow rates according to the liquid cooling system operation constraints to generate a set of branch target flow rates. The target flow generation module reads the currently valid regional-branch cooling influence relationship from the parameter storage area. The regional-branch cooling influence relationship includes a regional identifier, a branch identifier, and a cooling influence coefficient. The cooling influence coefficient characterizes the correlation between the unit flow rate change of the corresponding liquid cooling branch and the temperature change of the corresponding heat dissipation area within the regional temperature response time window. When the same branch identifier is associated with multiple regional identifiers, the flow rate change of the liquid cooling branch has a cooling effect on multiple heat dissipation areas; when the same regional identifier is associated with multiple branch identifiers, the cooling demand of the heat dissipation area is shared by multiple liquid cooling branches.

[0066] The target flow generation module reads the regional cooling demand according to the regional identifier and searches for multiple branch identifiers associated with that regional identifier. When a heat dissipation region is associated with multiple liquid cooling branches, the regional cooling demand of the heat dissipation region is allocated according to the cooling influence coefficients corresponding to the multiple liquid cooling branches; when a liquid cooling branch is associated with multiple heat dissipation regions, the multiple regional cooling demands undertaken by the liquid cooling branch are summarized to form the branch candidate flow of the liquid cooling branch. Liquid cooling branches with a cooling influence coefficient of zero do not undertake the regional cooling demand proposed by the corresponding heat dissipation region in the current control cycle; when the cooling influence relationship is missing or in an invalid state, the target flow generation module retains the valid cooling influence relationship of the previous control cycle and does not add branch candidate flow based on the missing relationship.

[0067] The operating constraints of the liquid cooling system include the upper limit of total flow, the upper limit of branch flow, the lower limit of branch flow, the range of supply and return hydraulic pressure difference, the operating range of the circulating pump, and the limit of valve opening change in a single control cycle. The target flow generation module first compares the candidate flow of each branch with the corresponding upper and lower limits of branch flow. The candidate flow of the branch below the lower limit is adjusted to the lower limit, and the portion above the upper limit is transferred to the liquid cooling branch that has a non-zero cooling influence coefficient for the same heat dissipation area and still has flow margin. When the sum of multiple candidate flow rates exceeds the upper limit of total flow, the lower limits of branch flow of multiple liquid cooling branches are first retained, and then the remaining flow is allocated according to the regional cooling demand and cooling influence coefficient.

[0068] After completing the flow boundary verification, the target flow generation module calls the hydraulic control parameters to correct the hydraulic coupling between parallel liquid-cooled branches. The hydraulic control parameters include branch resistance parameters, valve opening-flow correspondence parameters, and circulating pump operating status. When an increase in the flow rate of one liquid-cooled branch causes a change in the supply-return hydraulic pressure difference, the target flow generation module calculates the expected flow rate changes of other liquid-cooled branches based on the changed supply-return hydraulic pressure difference and adjusts the corresponding branch candidate flow rates. After continuous correction, the expected flow rates of multiple liquid-cooled branches are within the corresponding branch flow rate boundaries, and the sum of the flow rates of multiple branches is within the upper limit of the total flow rate corresponding to the current circulating pump operating status. The target flow generation module summarizes the corrected flow rate values ​​according to the branch identifier to form a branch target flow rate set, and transmits the branch target flow rate set to the valve-pump execution control module and the feedback result determination module.

[0069] The valve-pump execution control module 04, connected to the target flow generation module and the status acquisition module, is used to generate and execute valve opening adjustment commands and circulating pump compensation commands based on the difference between the target flow set of the branch and the actual flow of the branch, and to obtain the adjusted actual flow of the branch and the regional temperature response results. The valve-pump execution control module associates the target flow of each branch with the actual flow of the branch provided by the status acquisition module according to the branch identifier, and calculates the flow difference between the two. When the flow difference is positive, a valve opening adjustment amount is generated to increase the opening of the branch regulating valve based on the valve opening-flow correspondence parameter; when the flow difference is negative, a valve opening adjustment amount is generated to decrease the opening of the branch regulating valve; when the flow difference is within the flow feedback deviation range, the current opening of the branch regulating valve is maintained.

[0070] The valve-pump execution control module limits the valve opening adjustment amount according to the valve opening change limit in a single control cycle, and then issues a valve opening adjustment command to the corresponding branch control valve. Multiple branch control valves receive commands in the same control cycle, but the circulating pump compensation command is generated after the branch control valve's action acknowledgment is completed. If some branch control valves do not return an opening acknowledgment, the valve-pump execution control module rereads the status of the corresponding branch control valve; if no acknowledgment is obtained after rereading, the liquid-cooled branch maintains the branch control valve opening before adjustment, and the remaining liquid-cooled branches continue to operate according to the confirmed valve opening.

[0071] After the branch regulating valve completes its operation, the valve pump execution control module reads the pressure of the main supply pipe and the main return pipe to form a supply and return hydraulic pressure difference. When the supply and return hydraulic pressure difference is lower than the lower limit of the configuration range, a circulation pump compensation command is generated based on the total difference between the target flow rate of multiple branches and the actual flow rate of multiple branches, as well as the current speed and operating range of the circulation pump. When the supply and return hydraulic pressure difference is within the configuration range, the circulation pump speed is not increased. When the supply and return hydraulic pressure difference is higher than the upper limit of the configuration range, the circulation pump speed is reduced according to the operating range of the circulation pump or the current state of the main pipe regulating valve is maintained.

[0072] After the valve opening adjustment command and the circulating pump compensation command are executed, the valve-pump execution control module continuously reads the flow detection values ​​of multiple liquid-cooled branches. When the flow change at adjacent acquisition times is within the configured range and the supply and return pressure difference does not continue to change unidirectionally, the current detection value is taken as the actual flow of the adjusted branch. If flow fluctuation still exists after reaching the upper limit of the flow monitoring time, the average flow within the monitoring time is used as the actual flow of the adjusted branch, and the corresponding flow fluctuation state is retained. When the flow sensor is in an abnormal state, the actual flow of the adjusted branch is calculated based on the adjusted branch pressure difference, branch resistance parameters, and branch regulating valve opening.

[0073] After the adjusted branch flow rate enters a stable range, the valve-pump execution control module records the start time of the regional temperature response time window and reads the regional temperature of each heat dissipation area at the start and end times, forming a regional temperature response result. If the chip input power consumption or liquid supply temperature changes beyond the configured range within the response time window, the corresponding abnormal state is retained in the regional temperature response result. The feedback result determination module then verifies whether the regional temperature response participates in the cooling effect correction. The valve-pump execution control module transmits the adjusted branch flow rate and regional temperature response result to the feedback result determination module, and transmits the branch regulating valve's action receipt and the circulating pump's operating status to the status acquisition module for reading in the next control cycle.

[0074] The feedback result determination module 05, connected to the valve-pump execution control module and the target flow generation module, compares the target flow set of the branch with the adjusted actual flow of the branch to form a flow execution residual. It also determines the expected regional temperature response based on the region-branch cooling influence relationship and the adjusted actual flow of the branch, and compares the expected regional temperature response with the actual regional temperature response to form a regional temperature response residual. The feedback result determination module, according to the branch identifier, matches the target flow set of the branch provided by the target flow generation module with the adjusted actual flow of the branch provided by the valve-pump execution control module, and uses the difference between the target flow of each branch and the corresponding actual flow as the flow execution residual. For liquid-cooled branches where the flow detection value fluctuates, the flow execution residual is formed based on the average flow rate over the flow monitoring time, not a single instantaneous flow value.

[0075] After generating the flow execution residual, the feedback result determination module reads the current region-branch cooling influence relationship and calculates the expected regional temperature response of each heat dissipation area based on the actual flow changes of multiple liquid cooling branches before and after adjustment and the corresponding cooling influence coefficients. When a heat dissipation area is associated with multiple liquid cooling branches, the regional temperature response corresponding to the flow changes of multiple liquid cooling branches is calculated separately, and then summarized according to the same region identifier; when a liquid cooling branch is associated with multiple heat dissipation areas, the actual flow change of the same branch is calculated with the cooling influence coefficients corresponding to multiple regions respectively.

[0076] The feedback result determination module compares the predicted regional temperature response with the actual regional temperature response according to the regional identifier, forming a regional temperature response residual. If there are sudden changes in chip input power consumption, liquid supply temperature, or abnormal temperature sensor conditions in the regional temperature response result, the corresponding regional temperature response residual is set to an invalid state and will not participate in the cooling influence coefficient correction in this control cycle. Both the flow execution residual and the regional temperature response residual are accompanied by corresponding branch identifiers, regional identifiers, and valid states, and are transmitted to the parameter update module; specifically, the flow execution residual points to the correction path for the hydraulic control parameters, and the regional temperature response residual points to the correction path for the regional-branch cooling influence relationship.

[0077] The parameter update module 06, connected to the feedback result determination module and the target flow generation module, is used to correct hydraulic control parameters based on the flow rate execution residual, correct the regional-branch cooling influence relationship based on the regional temperature response residual, generate the parameter update result for the next control cycle, and transmit the parameter update result for the next control cycle to the target flow generation module. After receiving the two types of residuals, the parameter update module first checks the effective status and deviation degree of the flow execution residual. When the flow execution residual of a certain liquid-cooled branch exceeds the flow feedback deviation threshold, it reads the branch regulating valve opening, branch pressure difference, and circulating pump operating status before and after the adjustment of the liquid-cooled branch. When the branch regulating valve opening has changed but the actual flow rate change in the branch is insufficient, it corrects the branch resistance parameters based on the actual branch pressure difference and actual flow rate, and corrects the valve opening-flow rate corresponding parameters based on the valve opening change and flow rate change. When multiple liquid-cooled branches simultaneously exhibit flow execution residuals in the same direction and the supply and return pressure difference is lower than the configured range, it corrects the circulating pump operating status and does not attribute this status to the local resistance changes of multiple liquid-cooled branches.

[0078] When the flow execution residual does not exceed the flow feedback deviation threshold, but the effective area temperature response residual exceeds the temperature response residual threshold, the parameter update module corrects the cooling influence coefficient based on the actual flow change of the corresponding liquid cooling branch and the actual temperature change of the heat dissipation area. When a heat dissipation area is associated with multiple liquid cooling branches, only the cooling influence coefficient corresponding to the liquid cooling branch whose actual flow changes within the current control cycle is corrected; the original cooling influence coefficient is maintained for liquid cooling branches whose actual flow has not undergone identifiable changes. When the difference between the newly formed cooling influence coefficient and the original cooling influence coefficient exceeds the configuration range, the original cooling influence coefficient remains valid, and the newly formed cooling influence coefficient is set to a pending verification state, and is re-verified through small branch flow changes in subsequent control cycles.

[0079] When both the flow execution residual and the regional temperature response residual exceed their corresponding thresholds, the parameter update module first corrects the hydraulic control parameters and does not replace the cooling influence coefficient based on the regional temperature response residual of the current control cycle. If a liquid cooling branch experiences flow execution residual exceeding limits for multiple consecutive control cycles, and the branch regulating valve opening reaches its upper limit, an abnormal branch flag is generated and transmitted to the target flow generation module. In the next control cycle, the target flow generation module stops allocating branch candidate flow to the corresponding abnormal liquid cooling branch and allocates the corresponding regional cooling demand to the available liquid cooling branches based on the regional-branch cooling influence relationship.

[0080] The parameter update module generates the parameter update result for the next control cycle from the corrected branch resistance parameters, valve opening-flow correspondence parameters, circulating pump operating status, cooling influence coefficient, and corresponding effective status. After receiving the parameter update result for the next control cycle, the target flow generation module uses the effective cooling influence coefficient to generate candidate branch flow rates and executes hydraulic coupling constraints using the updated hydraulic control parameters. The status acquisition module re-reads the regional temperature status and actual branch flow rate in the next control cycle, thereby continuing to form a cyclical control relationship of regional cooling demand, target branch flow rate, valve and pump control, and feedback correction.

Claims

1. A method for partitioned liquid cooling control of a high-performance computing chip, characterized in that, include: S100: Obtain the regional temperature status of multiple heat dissipation areas and the actual flow rate of multiple liquid cooling branches; A set of regional cooling requirements is formed based on the temperature conditions of the region. S200. Obtain the region-branch cooling influence relationship, which characterizes the change in liquid cooling branch flow rate and the temperature change of multiple heat dissipation areas; form branch candidate flow rates based on the region cooling demand set and the region-branch cooling influence relationship; apply hydraulic coupling constraints to the branch candidate flow rates according to the liquid cooling system operation constraints to generate a branch target flow rate set. S300. Based on the difference between the target flow rate set of the branch and the actual flow rate of the branch, generate and execute valve opening adjustment command and circulating pump compensation command, and obtain the adjusted actual flow rate of the branch and the area temperature response result. S400: Compare the target flow set of the branch with the adjusted actual flow of the branch to form a flow execution residual; Based on the cooling influence relationship between the region and the branch and the actual flow rate of the adjusted branch, the expected regional temperature response is determined, and the expected regional temperature response is compared with the regional temperature response result to form the regional temperature response residual. Based on the flow rate, residual correction hydraulic control parameters are executed, and the regional-branch cooling influence relationship is corrected based on the regional temperature response residual, generating parameter update results for the next control cycle.

2. The method according to claim 1, characterized in that, The regional temperature status includes the regional temperature, the regional reference temperature, and the regional temperature change. The temperature of the region is compared with the reference temperature of the region to generate a regional temperature difference; The change in regional temperature is generated based on the difference between the regional temperature in the current control cycle and the regional temperature in the previous control cycle. The regional temperature rise rate is generated based on the ratio of the regional temperature change to the control cycle duration. The regional cooling requirements of each heat dissipation area are determined based on the regional temperature difference and the regional temperature rise rate, and the regional cooling requirement set is formed according to the regional identifier.

3. The method according to claim 1, characterized in that, The region-branch cooling influence relationship includes region identifier, branch identifier, and cooling influence coefficient; Multiple area signs are associated with the same branch road sign, and multiple branch road signs are associated with the same area sign; The cooling influence coefficient characterizes the relationship between the unit flow rate change of the corresponding liquid cooling branch and the regional temperature change of the corresponding heat dissipation area.

4. The method according to claim 3, characterized in that, The formation of the cooling effect coefficient includes: Adjust the branch regulating valve opening of the target liquid cooling branch according to the valve opening change limit of a single control cycle, and obtain the actual flow change of the target liquid cooling branch. Acquire the regional temperature changes of multiple heat dissipation areas within the regional temperature response time window; Multiple cooling influence coefficients are formed for the target liquid-cooled branch based on the actual flow rate changes of the branch and the regional temperature changes of each heat dissipation area. Multiple cooling influence coefficients are associated with branch identifiers and area identifiers to form the area-branch cooling influence relationship.

5. The method according to claim 1, characterized in that, Obtaining the actual flow of the branch includes: The actual flow rate of the liquid-cooled branch is obtained by a flow sensor installed in the liquid-cooled branch; When the flow sensor malfunctions, the inlet and outlet pressures of the liquid-cooled branch are acquired, the branch pressure difference is calculated, and the actual flow rate of the branch is calculated based on the branch pressure difference and the branch resistance parameters.

6. The method according to claim 1, characterized in that, The hydraulic control parameters include branch resistance parameters, valve opening-flow rate correspondence parameters, and circulating pump operating status. The operating constraints of the liquid cooling system include the upper limit of total flow, the upper limit of branch flow, the lower limit of branch flow, the range of supply and return hydraulic pressure difference, the operating range of the circulating pump, and the limit of valve opening change in a single control cycle. The hydraulic coupling constraints include: The sum of the candidate flows of multiple branches is compared with the upper limit of the total flow, and the candidate flows of each branch are compared with the corresponding upper limit and lower limit of the branch flow. When the sum of multiple branch candidate flows exceeds the total flow limit, the corresponding branch candidate flows are adjusted according to the cooling requirements of each area. The effect of a change in the flow rate of one liquid-cooled branch on the actual flow rate of other liquid-cooled branches is corrected based on the hydraulic control parameters to form the target flow rate set of the branches.

7. The method according to claim 6, characterized in that, Generating and executing the valve opening adjustment command and the circulating pump compensation command includes: Based on the difference between the target flow rate and the actual flow rate of each branch and the valve opening-flow rate corresponding parameter, the valve opening adjustment amount of each branch regulating valve is generated. The valve opening adjustment command is executed according to the valve opening change limit of the single control cycle. Obtain the pressure of the supply manifold and the return manifold to form a supply and return hydraulic pressure difference; When the supply and return hydraulic pressure difference is lower than the lower limit of the supply and return hydraulic pressure difference range, a circulating pump speed adjustment command is generated and executed based on the total difference between the target flow rate of multiple branches and the actual flow rate of multiple branches.

8. The method according to claim 6, characterized in that: The difference between the target flow rate of each branch and the actual flow rate of the corresponding branch after adjustment is used as the flow rate execution residual. Based on the actual flow rate changes of each liquid cooling branch before and after adjustment and the corresponding cooling influence coefficient, the expected regional temperature response of each heat dissipation area is determined. The difference between the predicted regional temperature response and the corresponding regional temperature response result is taken as the regional temperature response residual; When the flow execution residual exceeds the flow feedback deviation threshold, the branch resistance parameter and the valve opening-flow correspondence parameter are corrected. When the flow execution residual does not exceed the flow feedback deviation threshold and the regional temperature response residual exceeds the temperature response residual threshold, the corresponding cooling influence coefficient is corrected.

9. The method according to claim 8, characterized in that, When the flow execution residual of the liquid cooling branch exceeds the flow feedback deviation threshold for multiple consecutive control cycles, and the opening of the corresponding branch regulating valve reaches the upper limit of the branch regulating valve opening, an abnormal branch mark is generated. Stop allocating branch candidate traffic to the abnormal branch; Based on the region-branch cooling influence relationship, other liquid-cooled branches associated with the heat dissipation area corresponding to the abnormal branch are obtained; According to the upper limit of the branch flow of the other liquid cooling branches, the regional cooling demand corresponding to the abnormal branch is redistributed to the other liquid cooling branches; When the sum of the upper limits of the branch flow of the other liquid cooling branches is less than the sum of the candidate flow of the redistributed branches, a power reduction request is output to the chip.

10. A high-performance chip partitioned liquid cooling control system, applied to the method of any one of claims 1 to 9, characterized in that, include: The status acquisition module is used to acquire the regional temperature status of multiple heat dissipation areas and the actual flow rate of multiple liquid cooling branches. A regional cooling demand determination module, connected to the status acquisition module, is used to form a regional cooling demand set based on the regional temperature status. The target flow generation module is connected to the regional cooling demand determination module and the status acquisition module. It is used to obtain the regional-branch cooling influence relationship, form branch candidate flow based on the regional cooling demand set and the regional-branch cooling influence relationship, and perform hydraulic coupling constraints on the branch candidate flow according to the liquid cooling system operation constraints to generate a branch target flow set. The valve pump execution control module is connected to the target flow generation module and the status acquisition module. It is used to generate and execute valve opening adjustment commands and circulating pump compensation commands based on the difference between the target flow set of the branch and the actual flow of the branch, and to obtain the adjusted actual flow of the branch and the area temperature response results. The feedback result determination module is connected to the valve pump execution control module and the target flow generation module. It is used to compare the target flow set of the branch with the adjusted actual flow of the branch to form a flow execution residual, and to determine the expected regional temperature response based on the regional-branch cooling influence relationship and the adjusted actual flow of the branch. It compares the expected regional temperature response with the regional temperature response result to form a regional temperature response residual. The parameter update module, connected to the feedback result determination module and the target flow generation module, is used to perform residual correction of hydraulic control parameters based on the flow rate, correct the regional-branch cooling influence relationship based on the regional temperature response residual, generate parameter update results for the next control cycle, and transmit the parameter update results for the next control cycle to the target flow generation module.

Images