A method for optimizing heat dissipation performance of an integrated circuit module

Abstract

The application discloses a kind of integrated circuit module heat dissipation performance optimization method, comprising: obtaining the functional unit distribution information of integrated circuit module, physical adjacent relationship, temperature sampling information and to-be-executed task sequence;According to the thermal diffusion direction between functional units, thermal diffusion delay and thermal coupling strength, establish thermal diffusion directed graph;Predict the temperature rise curve of each functional unit in future time window, and determine thermal pulse source unit;Along the thermal diffusion direction towards module cooling boundary, adjacent functional units with thermal margin are selected to form thermal shadow chain;Before thermal pulse source unit enters predicted high heat load period, cold shadow reservation processing is executed to thermal shadow chain;In high heat load period, task is divided into source end reservation task fragment and chain end migration task fragment, and is distributed according to the order of end functional unit to head-end functional unit step-by-step backfilling.This method can form thermal buffer path in advance without changing hardware structure, reduce local temperature rise peak value, improve the heat dissipation regulation stability of integrated circuit module.

Description

Technical Field

[0001] This invention relates to the field of integrated circuits, and more specifically to a method for optimizing the heat dissipation performance of integrated circuit modules. Background Technology

[0002] During operation, integrated circuit modules typically experience localized temperature rises due to continuous computation by functional units, execution of instantaneous high-load tasks, and thermal coupling between adjacent functional units. When a functional unit performs a high-power task for a short period, its temperature rises rapidly and transfers to surrounding functional units. If the heat cannot dissipate in time or be properly dispersed, localized hot spots can easily form. Localized hot spots not only reduce the operational stability of the integrated circuit module but may also lead to frequency reduction, increased task latency, and in severe cases, affect the module's lifespan and reliability.

[0003] Existing technologies for optimizing the heat dissipation performance of integrated circuit modules mainly include adding heat dissipation structures, optimizing packaging materials, setting heat sinks or vapor chambers, adjusting chip layout, controlling frequency reduction based on temperature, and migrating tasks based on processor temperature. While these solutions can reduce module temperature to some extent, they still have certain shortcomings: Firstly, adding heat dissipation structures or changing packaging materials usually requires changes to the hardware structure, resulting in high implementation costs and poor adaptability to established integrated circuit modules; secondly, conventional task migration methods mostly select functional units with lower temperatures for task allocation based on the current or predicted temperature, focusing primarily on "avoiding high-temperature units," without fully considering the correlation between the heat diffusion path before high heat loads, heat margin reservation, and heat diffusion model calibration after high heat loads.

[0004] Especially within integrated circuit modules, heat transfer is not random but closely related to the physical adjacency between functional units, the strength of thermal coupling, the delay in thermal diffusion, and the location of the module's cooling boundary. If task migration only occurs after the temperature of a particular functional unit rises, the optimal adjustment window has often been missed; conversely, migrating tasks only to the functional unit with the lowest current temperature may create new hotspots or even disrupt the original thermal diffusion gradient. Therefore, current technologies lack a method for optimizing heat dissipation performance that can proactively create a cold shadow reserve area before high thermal loads arrive, utilize this reserve area to accommodate task migration during high thermal load phases, and then use the temperature drop process after the high thermal load ends to reverse-correct the thermal diffusion model. Summary of the Invention

[0005] To address the aforementioned problems, this invention provides a method for optimizing the heat dissipation performance of integrated circuit modules, which effectively overcomes the shortcomings of existing technologies.

[0006] This invention is achieved through the following technical solution: a method for optimizing the heat dissipation performance of an integrated circuit module, comprising the following steps: S1. Obtain the functional unit distribution information, physical adjacency relationship between functional units, temperature sampling information, and task sequence to be executed of the integrated circuit module. The integrated circuit module includes multiple functional units that can execute tasks independently or collaboratively. The functional unit distribution information includes at least the planar position of each functional unit in the integrated circuit module, its relative position with the module cooling boundary, and its corresponding allowable operating temperature. S2. Based on the physical adjacency relationship between the functional units, temperature sampling information, and task execution power consumption, a directed heat diffusion graph is established. The directed heat diffusion graph includes multiple unit nodes and heat diffusion edges connecting adjacent unit nodes. Each unit node corresponds to a functional unit. Each heat diffusion edge records the heat diffusion direction, heat diffusion delay, and heat coupling strength. The heat diffusion direction is from the heat-generating functional unit to the functional unit adjacent to it and close to the module cooling boundary. S3. Within the current control cycle, based on the sequence of tasks to be executed and the current temperature of each functional unit, predict the temperature rise curve of each functional unit in the future time window, and determine the functional unit whose predicted temperature rise curve exceeds the preset heat pulse condition as the heat pulse source unit. S4. Starting from the thermal pulse source unit, along the thermal diffusion direction toward the module cooling boundary in the thermal diffusion directional diagram, select at least two sequentially adjacent functional units to form a thermal shadow chain. Each functional unit in the thermal shadow chain has a current thermal margin higher than a preset thermal margin threshold, and the thermal coupling strength between adjacent functional units in the thermal shadow chain is higher than a preset coupling threshold. S5. Before the heat pulse source unit enters the predicted high heat load period, perform cold shadow reservation processing on the functional units in the heat shadow chain. The cold shadow reservation processing includes delaying, transferring or reducing the non-critical task load on the functional units in the heat shadow chain, so that the temperature of the first functional unit in the heat shadow chain near the heat pulse source unit is lower than the temperature of the heat pulse source unit, and forming a temperature gradient from the heat pulse source unit to the module cooling boundary. S6. When the heat pulse source unit enters the predicted high heat load period, the tasks to be executed assigned to the heat pulse source unit are divided into source end retention task segments and chain end migration task segments. The source end retention task segments continue to be executed by the heat pulse source unit, and the chain end migration task segments are assigned to the heat shadow chain for execution in the order of being filled back from the end functional unit to the beginning functional unit of the heat shadow chain. S7. During the process of the thermal pulse source unit executing the source end retention task segment and the thermal shadow chain executing the chain end migration task segment, the temperature gradient between the thermal pulse source unit and each functional unit in the thermal shadow chain is continuously monitored. If the temperature gradient is lower than a preset gradient threshold, the allocation of chain end migration task segments to the first functional unit of the thermal shadow chain is paused, and chain end migration task segments are allocated to the last functional unit of the thermal shadow chain or other thermal shadow chains in priority. S8. After the predicted high heat load period ends, the temperature drop curves of each functional unit in the thermal shadow chain are collected. Based on the actual power consumption of the chain-end migration task segment, the thermal diffusion response formed by the thermal pulse source unit is separated from the temperature drop curve. The thermal diffusion response is used to correct the thermal diffusion delay and thermal coupling strength in the thermal diffusion directed graph for the generation of the thermal shadow chain in the next control cycle.

[0007] As a preferred technical solution, when establishing the directed heat diffusion graph in step S2, adjacent unit pairs are first generated based on the planar position of each functional unit, and then the heat diffusion direction is determined based on the temperature response sequence of the adjacent unit pairs during the historical task execution process. The heat diffusion delay is determined based on the time required for adjacent functional units to reach the corresponding temperature rise ratio after a functional unit experiences a temperature rise.

[0008] As a preferred technical solution, the preset thermal pulse conditions in step S3 include: the predicted temperature rise value of the functional unit in the future time window exceeds the temperature rise threshold, and the predicted temperature peak arrival time of the functional unit is earlier than the predicted temperature peak arrival time of its adjacent functional units; the functional unit that meets the preset thermal pulse conditions is determined as a thermal pulse source unit.

[0009] As a preferred technical solution, when forming the thermal shadow chain in step S4, a functional unit that simultaneously meets the following conditions is selected from multiple adjacent functional units of the thermal pulse source unit as the first functional unit: First, the functional unit is located between the thermal pulse source unit and the module cooling boundary; second, the thermal coupling strength between the functional unit and the thermal pulse source unit is the largest or the thermal diffusion delay is the shortest; third, the current thermal margin of the functional unit is higher than the current thermal margin of the thermal pulse source unit.

[0010] As a preferred technical solution, the start time of the cold shadow reservation process in step S5 is determined based on the predicted temperature peak arrival time of the thermal pulse source unit, the thermal diffusion delay of the first functional unit in the thermal shadow chain relative to the thermal pulse source unit, and the preset safety advance amount, so that the thermal shadow chain completes the establishment of a low-load state before the temperature of the thermal pulse source unit rises rapidly.

[0011] As a preferred technical solution, in step S6, when dividing the task to be executed into source-end retained task segments and chain-end migration task segments, firstly identify the task segments in the task to be executed whose local data interaction frequency with the thermal pulse source unit is higher than a preset interaction threshold, and regard them as source-end retained task segments; then identify the task segments in the task to be executed whose calculation results can be delayed or can be executed in parallel, and regard them as chain-end migration task segments.

[0012] As a preferred technical solution, the step-by-step backfilling from the end functional unit to the beginning functional unit in step S6 means that the chain-end migration task segment is first assigned to the end functional unit that is farthest from the heat pulse source unit and close to the module cooling boundary. When the heat margin of the end functional unit is lower than the task acceptance threshold, the chain-end migration task segment is then assigned to the adjacent functional unit in the heat shadow chain that is closer to the heat pulse source unit.

[0013] As a preferred technical solution, in step S7, if the temperature difference between the thermal pulse source unit and the thermal shadow chain head-end functional unit is less than a preset temperature difference threshold, the execution frequency of the source-end retention task segment executed by the thermal pulse source unit is reduced, or the unexecuted chain-end migration task segment is transferred to another thermal shadow chain that does not share a head-end functional unit with the thermal pulse source unit.

[0014] As a preferred technical solution, when there are two or more thermal pulse source units in the same control cycle, and the thermal shadow chains corresponding to different thermal pulse source units have overlapping functional units, the overlapping functional units are marked as shared cold shadow units. The shared cold shadow units do not undertake chain end migration task segments during the corresponding predicted high heat load period, but are only used to maintain the thermal isolation margin between different thermal pulse source units.

[0015] As a preferred technical solution, after correcting the directed heat diffusion graph in step S8, the heat dissipation performance of the integrated circuit module is determined based on the change in thermal coupling strength before and after the correction. If the thermal coupling strength of the same heat diffusion edge decreases or the heat diffusion delay increases within multiple consecutive control cycles, the priority of selecting the thermal shadow chain associated with the heat diffusion edge is increased to form an early cold shadow reservation in the heat dissipation performance decay region.

[0016] The beneficial effects of this invention are: by establishing a directed heat diffusion graph and determining the heat pulse source unit and heat shadow chain based on the heat diffusion direction, heat diffusion delay and heat coupling strength, this invention enables heat dissipation optimization to no longer be passively scheduled based solely on the current temperature, but to plan the task distribution in advance by combining the actual heat transfer paths between functional units, which can more accurately suppress the formation of local hot spots.

[0017] This invention performs a cold shadow reservation process on the thermal shadow chain before the thermal pulse source unit enters the predicted high heat load period, so that the thermal shadow chain forms a cold shadow reservation area with thermal margin before the arrival of high heat load. This cold shadow reservation area serves as a heat diffusion buffer area in the early stage and as a receiving area for the chain end migration task segment in the high heat load stage. This enables the same method step to produce different technical effects at different stages, with strong linkage, and can reduce the temperature rise peak of the thermal pulse source unit.

[0018] This invention allocates the chain-end migration task segments in a stepwise manner from the end functional unit to the middle functional unit and then to the beginning functional unit. This allows the task load to be released preferentially to the side closer to the module cooling boundary, avoiding direct occupation of the beginning functional unit near the heat pulse source unit and weakening the temperature gradient. This creates a synergistic relationship between task execution and heat diffusion, improving the stability of heat dissipation control.

[0019] This invention acquires the temperature drop curve of the thermal shadow chain after the high thermal load ends, and separates the thermal diffusion response formed by the thermal pulse source unit conduction from it. This response is used to correct the thermal diffusion delay and thermal coupling strength in the directed thermal diffusion graph, so that the generation of the thermal shadow chain in the next control cycle is more consistent with the actual thermal state of the module. This allows for continuous optimization of heat dissipation performance without changing the hardware structure of the integrated circuit module. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is an overall flowchart of a method for optimizing the heat dissipation performance of an integrated circuit module according to the present invention; Figure 2 This is a schematic diagram illustrating the principle of establishing a directed thermal diffusion graph in this invention; Figure 3 This is a schematic diagram of the generation of thermal shadow chains and cold shadow reserved areas based on thermal pulse source units in this invention; Figure 4 This is a timing diagram illustrating the cold shadow reservation, chain-end migration task segment allocation, and temperature gradient changes in this invention. Figure 5 This is a flowchart of the hierarchical backfilling, allocation, and dynamic adjustment of chain-end migration task segments in this invention; Figure 6 This is a schematic diagram illustrating the formation of a shared cold shadow unit when multiple thermal pulse source units correspond to overlapping thermal shadow chains in this invention.

[0022] Explanation of reference numerals in the attached figures: 10, Integrated circuit module; 21, Thermal pulse source unit; 21a, First thermal pulse source unit; 21b, Second thermal pulse source unit; 31, First-end functional unit; 32, Middle functional unit; 33, End functional unit; 41, Cold shadow reserved area; 51, Shared cold shadow unit. Detailed Implementation

[0023] All features disclosed in this specification, or all steps in all disclosed methods or processes, may be combined in any way, except for mutually exclusive features and / or steps.

[0024] Any feature disclosed in this specification (including any appended claims, abstract, and drawings) may be replaced by other equivalent or similar features for a similar purpose, unless specifically stated otherwise. That is, unless specifically stated otherwise, each feature is merely one example of a series of equivalent or similar features.

[0025] like Figures 1 to 6 As shown, this embodiment provides a method for optimizing the heat dissipation performance of an integrated circuit module. This method is used to predict, schedule, and correct the heat distribution state of the integrated circuit module 10 during operation, so as to reduce local hot spots caused by local high-load tasks inside the integrated circuit module 10. The integrated circuit module 10 can be a chip module, packaged module, or board-level integrated module including multiple functional units. The multiple functional units can be processing cores, graphics processing units, storage control units, communication interface units, accelerated computing units, or other circuit functional areas that can independently or collaboratively perform tasks. Each functional unit has a defined planar position within the integrated circuit module 10, and there are physical adjacency and thermal coupling relationships between the functional units. The heat generated by a functional unit when performing a high-power task will gradually diffuse along the adjacent functional units and eventually be released towards the module cooling boundary.

[0026] In specific implementation, step S1 is first executed to obtain the functional unit distribution information, physical adjacency relationships between functional units, temperature sampling information, and a sequence of tasks to be executed from the integrated circuit module 10. The functional unit distribution information characterizes the arrangement position of each functional unit within the integrated circuit module 10, and may include the center coordinates, boundary coordinates, area size, adjacent functional unit identifiers, distance from the module's cooling boundary, and allowable operating temperature of each functional unit. The physical adjacency relationships between functional units determine whether there is a direct heat diffusion path between two functional units, such as whether the two functional units are adjacent, separated by a low thermal conductivity area, or located in the same heat diffusion direction. The temperature sampling information can be obtained from the temperature sensor built into the integrated circuit module 10, an on-chip sensor array, a thermistor circuit, or an external temperature detection unit, or it can be calculated from the operating power consumption and historical temperature rise model. The sequence of tasks to be executed includes the task type to be assigned to each functional unit, task start time, task duration, estimated task power consumption, task priority, and data dependencies between tasks.

[0027] After obtaining the above basic information, step S2 is executed to establish a directed heat diffusion graph based on the physical adjacency relationships between functional units, temperature sampling information, and task execution power consumption. The directed heat diffusion graph is as follows: Figure 2 As shown, it includes multiple unit nodes and heat diffusion edges connecting adjacent unit nodes. Each unit node corresponds to a functional unit in the integrated circuit module 10, and each heat diffusion edge corresponds to the heat transfer relationship between two adjacent functional units. The heat diffusion edge does not simply indicate that two functional units are adjacent, but simultaneously records the heat diffusion direction, heat diffusion delay, and thermal coupling strength. The heat diffusion direction indicates which functional unit the heat is mainly transferred from to which functional unit. In this embodiment, the heat diffusion direction preferentially points from the heat-generating functional unit to the adjacent functional unit close to the module's cooling boundary, thus enabling the directed heat diffusion graph to reflect the trend of heat transfer from inside the integrated circuit module 10 to the external heat dissipation path. The heat diffusion delay indicates the time required for an adjacent functional unit to exhibit a corresponding temperature response after a functional unit experiences a temperature rise. The thermal coupling strength indicates the strength of the thermal influence between two adjacent functional units; the higher the thermal coupling strength, the more significant the impact of a functional unit's temperature change on its adjacent functional units.

[0028] When establishing a directed heat diffusion graph, adjacent unit pairs can be generated based on the planar positions of each functional unit within the integrated circuit module 10. For each adjacent unit pair, temperature sampling information from historical task execution can be analyzed. When a functional unit experiences a temperature rise due to task execution, if an adjacent functional unit experiences a corresponding temperature rise after a certain period, the heat diffusion direction can be determined based on the order of their temperature responses. For example, if functional unit A's temperature rises first, followed by functional unit B, and functional unit B is located between functional unit A and the module's cooling boundary, the heat diffusion direction can be determined as from functional unit A to functional unit B. Furthermore, based on the time required for functional unit B to reach a preset temperature rise ratio after functional unit A's temperature rise, the heat diffusion delay between functional unit A and functional unit B is determined; based on the ratio of functional unit B's temperature rise to functional unit A's temperature rise, combined with the physical distance between them and the material's thermal conductivity, the thermal coupling strength of the corresponding heat diffusion edge is determined. The directed heat diffusion graph established in this way not only reflects the adjacency relationship between functional units but also reflects the direction and efficiency of heat transfer within the integrated circuit module 10.

[0029] After establishing the directed thermal diffusion graph, step S3 is executed. Within the current control cycle, based on the sequence of tasks to be executed and the current temperature of each functional unit, the temperature rise curve of each functional unit within the future time window is predicted. Functional units whose predicted temperature rise curve exceeds the preset thermal pulse condition are identified as thermal pulse source units 21. The current control cycle can be a time period for task scheduling and thermal management of the integrated circuit module 10, such as several milliseconds, several microseconds, or other time lengths set according to the system operating status. The future time window is a predicted time period extending backward from the current control cycle, used to determine whether a certain functional unit will generate a significant temperature rise in a short period of time. The predicted temperature rise curve can be calculated based on the power consumption of the tasks to be executed, the current temperature of the functional unit, the historical temperature change rate, the heat dissipation boundary conditions, and the thermal diffusion delay and thermal coupling strength in the directed thermal diffusion graph.

[0030] When identifying the heat pulse source unit 21, the function unit with the highest current temperature is not simply designated as the heat pulse source unit 21. Instead, it is determined whether a certain function unit will form a rapidly rising heat pulse within a future time window. Specifically, the preset heat pulse conditions may include: the predicted temperature rise value of the function unit within the future time window exceeds the temperature rise threshold, and the predicted temperature peak arrival time of the function unit is earlier than the predicted temperature peak arrival time of its adjacent function units. Function units that meet the above conditions can be identified as heat pulse source units 21. In this way, function units that are about to experience localized high heat loads can be identified in advance, rather than waiting until the function unit has become significantly overheated before processing, thus providing advance time for subsequent heat shadow chain generation and cold shadow reservation processing.

[0031] After determining the heat pulse source unit 21, step S4 is executed. Starting from the heat pulse source unit 21, at least two sequentially adjacent functional units are selected along the heat diffusion direction towards the module cooling boundary in the heat diffusion directional diagram to form a heat shadow chain. In this embodiment, for ease of description, the heat shadow chain may include a first functional unit 31 near the heat pulse source unit 21, an intermediate functional unit 32 located on the side of the first functional unit 31 away from the heat pulse source unit 21, and a last functional unit 33 near the module cooling boundary, such as... Figure 3 As shown. The first functional unit 31, the middle functional unit 32, and the last functional unit 33 are not limited to three fixed physical units. In practical applications, the thermal shadow chain can include two, three, or more sequentially adjacent functional units. When the thermal shadow chain includes more functional units, the functional units in the chain can be determined sequentially according to their distance from the thermal pulse source unit 21 from near to far, with the end closest to the thermal pulse source unit 21 as the first end and the end closest to the module cooling boundary as the last end.

[0032] When forming a thermal shadow chain, both thermal margin and thermal coupling conditions must be met. Each functional unit in the thermal shadow chain should have a current thermal margin higher than a preset thermal margin threshold. The current thermal margin can be understood as the difference between the allowable operating temperature of the functional unit and its current temperature, and can also be further corrected by considering short-term task load and heat dissipation capacity. If a functional unit has a low current temperature, a high allowable operating temperature, and is not occupied by critical tasks in a short period of time, then this functional unit has a high thermal margin and is suitable as a candidate unit in the thermal shadow chain. At the same time, the thermal coupling strength between adjacent functional units in the thermal shadow chain should be higher than a preset coupling threshold to ensure that the heat generated by the thermal pulse source unit 21 can be effectively diffused along the chain. By considering both thermal margin and thermal coupling strength, it is possible to avoid selecting functional units that have a low temperature but weak thermal coupling with the thermal pulse source unit 21, and also to avoid selecting functional units that are adjacent but close to thermal saturation.

[0033] In one specific selection method, a functional unit that simultaneously meets multiple conditions is selected from multiple adjacent functional units of the heat pulse source unit 21 as the first functional unit 31. First, this functional unit is located between the heat pulse source unit 21 and the module cooling boundary, and can spatially absorb the heat diffused outward from the heat pulse source unit 21. Second, the thermal coupling strength between this functional unit and the heat pulse source unit 21 is the greatest or the thermal diffusion delay is the shortest, and it can respond promptly to the temperature rise of the heat pulse source unit 21 in terms of heat transfer relationship. Third, the current thermal margin of this functional unit is higher than the current thermal margin of the heat pulse source unit 21, thereby playing a thermal buffering role in the high heat load stage. After determining the first functional unit 31, intermediate functional units 32 and end functional units 33 are selected along the thermal diffusion direction toward the module cooling boundary, ultimately forming a thermal shadow chain from the heat pulse source unit 21 to the module cooling boundary.

[0034] After the thermal shadow chain is formed, step S5 is executed. Before the thermal pulse source unit 21 enters the predicted high thermal load period, cold shadow reservation processing is performed on the functional units in the thermal shadow chain, forming a cold shadow reservation area 41 in the area where the thermal shadow chain is located. The cold shadow reservation processing does not simply stop the functional units from working, but rather delays, transfers, or reduces the non-critical task load on each functional unit in the thermal shadow chain according to its task status. For tasks that can be delayed, they can be delayed until after the high thermal load period; for tasks that can be migrated, they can be temporarily transferred to other functional units that do not affect the thermal margin of the thermal shadow chain; for tasks that can be executed with reduced power consumption, the execution frequency or parallelism can be temporarily reduced. In this way, the first functional unit 31, the middle functional unit 32, and the last functional unit 33 are kept in a low load state before the high thermal load arrives, and the temperature of the first functional unit 31 is lower than the temperature of the thermal pulse source unit 21, thereby forming a temperature gradient from the thermal pulse source unit 21 to the module cooling boundary.

[0035] The start time of the cold shadow reservation process can be determined based on the predicted peak temperature arrival time of the heat pulse source unit 21, the thermal diffusion delay of the first functional unit 31 in the heat shadow chain relative to the heat pulse source unit 21, and a preset safety advance amount. For example, if it is predicted that the heat pulse source unit 21 will reach its peak temperature at a future time T2, and there is a certain thermal diffusion delay in the heat diffusion from the heat pulse source unit 21 to the first functional unit 31, the cold shadow reservation process can be started a certain period of time before T2. Figure 4 As shown. By pre-setting a safety lead time, the thermal shadow chain can be established in a low-load state before the temperature of the thermal pulse source unit 21 rises rapidly. In this way, when the thermal pulse source unit 21 enters the high thermal load period, the thermal shadow chain is no longer an ordinary low-temperature idle area, but a cold shadow reserved area 41 arranged in advance along the heat diffusion direction, which can play a role in heat diffusion and task migration at the same time.

[0036] After the thermal pulse source unit 21 enters the predicted high heat load period, step S6 is executed, dividing the tasks to be executed assigned to the thermal pulse source unit 21 into source-end retained task segments and chain-end migration task segments. Source-end retained task segments refer to the parts of the tasks that need to continue to be executed by the thermal pulse source unit 21, such as task segments with high frequency of local data interaction with the thermal pulse source unit 21, sensitive to execution location, or with high migration costs. Chain-end migration task segments refer to the parts of the tasks that can be executed with the assistance of functional units in the thermal shadow chain, such as task segments where calculation results can be delayed, data dependencies are weak, parallel execution is possible, or migration costs are low. During task division, task segments with a frequency of local data interaction with the thermal pulse source unit 21 that is higher than a preset interaction threshold can be identified first and designated as source-end retained task segments; then, task segments where calculation results can be delayed or parallel execution is identified and designated as chain-end migration task segments. This division method can reduce the load on the thermal pulse source unit 21 while avoiding a significant increase in data interaction latency due to excessive migration.

[0037] The allocation of chain-end migration task segments follows a step-by-step backfilling sequence from the end functional unit 33 of the thermal shadow chain to the beginning functional unit 31. Specifically, the chain-end migration task segments are first allocated to the end functional unit 33, which is furthest from the thermal pulse source unit 21 and closest to the module cooling boundary. When the thermal margin of the end functional unit 33 is lower than the task acceptance threshold, the chain-end migration task segments are then allocated to adjacent functional units in the thermal shadow chain that are closer to the thermal pulse source unit 21, such as the intermediate functional unit 32. When the thermal margin of the intermediate functional unit 32 is also lower than the task acceptance threshold, a portion of the chain-end migration task segments are then considered for allocation to the beginning functional unit 31. This allocation method differs from the conventional "nearest migration" approach. Instead of prioritizing the migration of tasks to the beginning functional unit 31, which is closest to the thermal pulse source unit 21, it prioritizes utilizing the end functional units 33, which are closest to the module cooling boundary, to handle the task load, thus releasing the task load as much as possible towards the heat dissipation boundary. This prevents the first functional unit 31 from heating up too early and maintains the temperature gradient between the heat pulse source unit 21 and the first functional unit 31, so that the heat shadow chain can both undertake the migration task and maintain the heat diffusion buffering effect.

[0038] During the execution of source-end retention task segments and chain-end migration task segments, step S7 is executed to continuously monitor the temperature gradient between the heat pulse source unit 21 and each functional unit in the heat shadow chain. The temperature gradient can include the temperature difference between the heat pulse source unit 21 and the first functional unit 31, or the overall temperature decrease trend along the heat diffusion direction of the heat pulse source unit 21, the first functional unit 31, the intermediate functional unit 32, and the last functional unit 33. If the temperature gradient remains above a preset gradient threshold, it indicates that the heat shadow chain can still serve as an effective heat diffusion buffer zone and can continue to receive chain-end migration task segments. If the monitored temperature gradient is below the preset gradient threshold, it indicates that the temperature of the first functional unit 31 or the side of the heat shadow chain closest to the heat pulse source unit 21 has increased. Continuing to allocate chain-end migration task segments to this area may weaken the cold shadow retention effect or even lead to the formation of new hot spots. Therefore, in this case, the allocation of chain-end migration task segments to the first functional unit 31 of the heat shadow chain is suspended, and chain-end migration task segments are preferentially allocated to the last functional unit 33 or other heat shadow chains.

[0039] Furthermore, if the temperature difference between the heat pulse source unit 21 and the first-end functional unit 31 of the heat shadow chain is less than a preset temperature difference threshold, stronger dynamic adjustment measures can be taken. For example, the execution frequency of the source-end reserved task segments executed by the heat pulse source unit 21 can be reduced to reduce the heat generated by the heat pulse source unit 21; alternatively, unexecuted chain-end migration task segments can be transferred to another heat shadow chain that does not share the first-end functional unit 31 with the heat pulse source unit 21 for execution, so as to avoid multiple task segments being concentrated on the same heat diffusion path. Through the above dynamic adjustment methods, this embodiment can continuously maintain the thermal buffering capacity of the heat shadow chain during the high heat load stage, and prevent the cold shadow reserved area 41 from rapidly failing due to task acceptance.

[0040] After the predicted high heat load period ends, step S8 is executed to collect the temperature drop curves of each functional unit in the thermal shadow chain. Since the heat generated by the thermal pulse source unit 21 during the high heat load phase diffuses along the thermal shadow chain, the first functional unit 31, the middle functional unit 32, and the last functional unit 33 in the thermal shadow chain will exhibit a certain temperature drop process after the high heat load ends. This temperature drop curve includes not only the heat generated by the execution of the chain-end migration task segment itself, but also the thermal diffusion response formed by the thermal pulse source unit 21. Therefore, in this embodiment, based on the actual power consumption of the chain-end migration task segment, the thermal diffusion response formed by the thermal pulse source unit 21 is separated from the temperature drop curve. Specifically, based on the execution time, power consumption, and known thermal response characteristics of each chain-end migration task segment, the temperature rise component caused by the chain-end migration task segment itself can be estimated, and this temperature rise component can be subtracted from the measured temperature drop curve to obtain a thermal diffusion response that is closer to the contribution of the thermal pulse source unit 21.

[0041] After obtaining the thermal diffusion response, the thermal diffusion delay and thermal coupling strength in the directed thermal diffusion graph are corrected using this response. If the temperature response corresponding to a certain thermal diffusion edge appears earlier than predicted, the thermal diffusion delay of that edge can be reduced; if the temperature rise of an adjacent functional unit is higher than predicted, the thermal coupling strength of the corresponding thermal diffusion edge can be increased; conversely, if the temperature response is delayed or the temperature rise is less than predicted, the thermal diffusion delay can be increased or the thermal coupling strength can be decreased accordingly. The corrected directed thermal diffusion graph is used for generating the thermal shadow chain in the next control cycle, making the next cold shadow reservation and task migration more consistent with the actual thermal state of the integrated circuit module 10. Thus, the thermal shadow chain serves as a cold shadow reservation area before high thermal load, as a chain-end migration task segment receiving area during high thermal load, and as a thermal diffusion model calibration area after high thermal load ends. The same link produces different technical functions at different stages, forming a linked heat dissipation optimization process.

[0042] In another implementation, such as Figure 6 As shown, when there are two or more heat pulse source units within the same control cycle, such as the first heat pulse source unit 21a and the second heat pulse source unit 21b, and the heat shadow chains corresponding to different heat pulse source units have overlapping functional units, the overlapping functional units are marked as shared cold shadow units 51. Shared cold shadow units 51 differ from ordinary heat shadow chain functional units in that they do not undertake chain-end migration task segments during the corresponding predicted high heat load period, but are only used to maintain the thermal isolation margin between the first heat pulse source unit 21a and the second heat pulse source unit 21b. By setting shared cold shadow units 51, it is possible to avoid the heat shadow chains of two heat pulse source units simultaneously undertaking tasks at the overlapping position, causing the overlapping area to heat up rapidly and form new hot spots. Shared cold shadow units 51 are equivalent to thermal isolation buffer points between multiple heat diffusion paths, enabling the heat dissipation optimization processes corresponding to multiple heat pulse source units to be carried out collaboratively.

[0043] In a further embodiment, after correcting the directed thermal diffusion graph in step S8, the change in thermal coupling strength before and after correction is used to determine whether the heat dissipation performance of the integrated circuit module 10 has degraded. If the thermal coupling strength of the same thermal diffusion edge decreases or the thermal diffusion delay increases within multiple consecutive control cycles, it indicates that the heat dissipation efficiency of the thermal diffusion path may decrease. This could be due to factors such as local temperature accumulation, aging of the packaging material, obstruction of the heat conduction path, or changes in environmental heat dissipation conditions, leading to slower thermal diffusion. In this case, the priority of selecting the thermal shadow chain associated with the thermal diffusion edge is increased, so that the region with degraded heat dissipation performance forms an earlier pre-cooling shadow reservation in subsequent control cycles. Thus, even if the heat dissipation conditions of the integrated circuit module 10 change over time, this embodiment can still continuously adjust the thermal shadow chain generation strategy through thermal diffusion response feedback, thereby achieving continuous optimization of heat dissipation performance without changing the hardware structure of the integrated circuit module 10.

[0044] This embodiment does not simply select the functional unit with the lower temperature for task migration based on the current temperature. Instead, it first establishes a directed thermal diffusion graph based on the actual thermal diffusion relationship between functional units, then identifies the heat pulse source unit 21 in advance based on the future temperature rise curve, and generates a thermal shadow chain along the thermal diffusion direction toward the module cooling boundary. Before the arrival of high heat load, the thermal shadow chain forms a cold shadow reservation area 41 through cold shadow reservation processing. During the high heat load stage, it receives the task segments migrating from the end functional unit 33 to the beginning functional unit 31 in a step-by-step backfilling order. After the high heat load ends, the directed thermal diffusion graph is corrected in reverse through the temperature drop curve. This method makes the thermal diffusion path, task migration path, and model calibration path interconnected, which can reduce the temperature rise peak of the heat pulse source unit 21, reduce the formation of local hot spots, and improve the heat dissipation control stability of the integrated circuit module 10 during continuous operation.

[0045] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions conceived without inventive effort should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope defined in the claims.

Claims

1. A method for optimizing the heat dissipation performance of an integrated circuit module, characterized in that, Includes the following steps: S1. Obtain the functional unit distribution information, physical adjacency relationship between functional units, temperature sampling information and task sequence to be executed of the integrated circuit module (10). The integrated circuit module (10) includes multiple functional units that can execute tasks independently or collaboratively. The functional unit distribution information includes at least the planar position of each functional unit in the integrated circuit module (10), its relative position with respect to the module cooling boundary, and the corresponding allowable operating temperature. S2. Based on the physical adjacency relationship between the functional units, temperature sampling information, and task execution power consumption, a directed heat diffusion graph is established. The directed heat diffusion graph includes multiple unit nodes and heat diffusion edges connecting adjacent unit nodes. Each unit node corresponds to a functional unit. Each heat diffusion edge records the heat diffusion direction, heat diffusion delay, and heat coupling strength. The heat diffusion direction is from the heat-generating functional unit to the functional unit adjacent to it and close to the module cooling boundary. S3. During the current control cycle, based on the sequence of tasks to be executed and the current temperature of each functional unit, predict the temperature rise curve of each functional unit in the future time window, and determine the functional unit whose predicted temperature rise curve exceeds the preset heat pulse condition as the heat pulse source unit (21). S4. Starting from the heat pulse source unit (21), along the heat diffusion direction towards the module cooling boundary in the heat diffusion directional diagram, select at least three sequentially adjacent functional units to form a heat shadow chain. The heat shadow chain includes a first functional unit (31) close to the heat pulse source unit (21), an intermediate functional unit (32) located on the side of the first functional unit (31) away from the heat pulse source unit (21), and a last functional unit (33) close to the module cooling boundary. Each functional unit in the heat shadow chain has a current heat margin higher than a preset heat margin threshold, and the thermal coupling strength between adjacent functional units in the heat shadow chain is higher than a preset coupling threshold. S5. Before the heat pulse source unit (21) enters the predicted high heat load period, cold shadow reservation processing is performed on the functional units in the heat shadow chain, so that the areas where the first functional unit (31), the middle functional unit (32) and the last functional unit (33) are located form a cold shadow reservation area (41). The cold shadow reservation processing includes delaying, transferring or reducing the non-critical task load on the functional units in the heat shadow chain, so that the temperature of the first functional unit (31) is lower than the temperature of the heat pulse source unit (21), and forming a temperature gradient from the heat pulse source unit (21) to the module cooling boundary. S6. When the heat pulse source unit (21) enters the predicted high heat load period, the tasks to be executed assigned to the heat pulse source unit (21) are divided into source end retention task segments and chain end migration task segments. The source end retention task segments continue to be executed by the heat pulse source unit (21), and the chain end migration task segments are assigned to the heat shadow chain for execution in the order of being filled back from the end functional unit (33) of the heat shadow chain to the middle functional unit (32) and the beginning functional unit (31) level by level. S7. During the process of the heat pulse source unit (21) executing the source end retention task segment and the heat shadow chain executing the chain end migration task segment, the temperature gradient between the heat pulse source unit (21) and the first end functional unit (31), the middle functional unit (32), and the last end functional unit (33) is continuously monitored. If the temperature gradient is lower than the preset gradient threshold, the allocation of the chain end migration task segment to the first end functional unit (31) is suspended, and the chain end migration task segment is allocated to the last end functional unit (33) or other heat shadow chains in priority. S8. After the predicted high heat load period ends, the temperature drop curves of each functional unit in the thermal shadow chain are collected. Based on the actual power consumption of the chain-end migration task segment, the thermal diffusion response formed by the thermal pulse source unit (21) is separated from the temperature drop curve. The thermal diffusion response is used to correct the thermal diffusion delay and thermal coupling strength in the thermal diffusion directed graph for the generation of the thermal shadow chain in the next control cycle.

2. The method for optimizing the heat dissipation performance of an integrated circuit module according to claim 1, characterized in that, In step S2, when establishing the directed heat diffusion graph, adjacent unit pairs are first generated based on the planar position of each functional unit. Then, the heat diffusion direction is determined based on the temperature response sequence of the adjacent unit pairs during the historical task execution process. Finally, the heat diffusion delay is determined based on the time required for adjacent functional units to reach the corresponding temperature rise ratio after a functional unit experiences a temperature rise.

3. The method for optimizing the heat dissipation performance of an integrated circuit module according to claim 1, characterized in that, The preset thermal pulse conditions in step S3 include: the predicted temperature rise value of the functional unit in the future time window exceeds the temperature rise threshold, and the predicted temperature peak arrival time of the functional unit is earlier than the predicted temperature peak arrival time of its adjacent functional units; the functional unit that meets the preset thermal pulse conditions is determined as a thermal pulse source unit (21).

4. The method for optimizing the heat dissipation performance of an integrated circuit module according to claim 1, characterized in that, When forming the thermal shadow chain in step S4, a functional unit that simultaneously meets the following conditions is selected from multiple adjacent functional units of the thermal pulse source unit (21) as the first functional unit (31): First, the functional unit is located between the thermal pulse source unit (21) and the module cooling boundary; second, the thermal coupling strength between the functional unit and the thermal pulse source unit (21) is the largest or the thermal diffusion delay is the shortest; third, the current thermal margin of the functional unit is higher than the current thermal margin of the thermal pulse source unit (21).

5. The method for optimizing the heat dissipation performance of an integrated circuit module according to claim 1, characterized in that, The start time of the cold shadow reservation process in step S5 is determined based on the predicted temperature peak arrival time of the thermal pulse source unit (21), the thermal diffusion delay of the first-end functional unit (31) relative to the thermal pulse source unit (21), and the preset safety advance amount, so that the cold shadow reservation area (41) completes the establishment of a low load state before the temperature of the thermal pulse source unit (21) rises rapidly.

6. The method for optimizing the heat dissipation performance of an integrated circuit module according to claim 1, characterized in that, In step S6, when dividing the task to be executed into source-end retained task segments and chain-end migration task segments, first identify the task segments in the task to be executed whose local data interaction frequency with the thermal pulse source unit (21) is higher than the preset interaction threshold, and use them as source-end retained task segments; then identify the task segments in the task to be executed whose calculation results can be delayed or can be executed in parallel, and use them as chain-end migration task segments.

7. The method for optimizing the heat dissipation performance of an integrated circuit module according to claim 1, characterized in that, The step-by-step backfilling from the end functional unit (33) of the thermal shadow chain to the beginning functional unit (31) in step S6 means that the chain-end migration task segment is first assigned to the end functional unit (33) that is farthest from the thermal pulse source unit (21) and close to the module cooling boundary. When the thermal margin of the end functional unit (33) is lower than the task acceptance threshold, the chain-end migration task segment is then assigned to the intermediate functional unit (32) or the beginning functional unit (31) that is adjacent to and closer to the thermal pulse source unit (21) in the thermal shadow chain.

8. The method for optimizing the heat dissipation performance of an integrated circuit module according to claim 1, characterized in that, If the temperature difference between the thermal pulse source unit (21) and the head end functional unit (31) is less than the preset temperature difference threshold in step S7, the execution frequency of the source end retention task segment executed by the thermal pulse source unit (21) is reduced, or the unexecuted chain end migration task segment is transferred to another thermal shadow chain that does not share the head end functional unit (31) with the thermal pulse source unit (21).

9. The method for optimizing the heat dissipation performance of an integrated circuit module according to claim 1, characterized in that, When there are two or more thermal pulse source units in the same control cycle, including the first thermal pulse source unit (21a) and the second thermal pulse source unit (21b), and the thermal shadow chains corresponding to different thermal pulse source units have overlapping functional units, the overlapping functional units are marked as shared cold shadow units (51). The shared cold shadow units (51) do not undertake the chain end migration task segments during the corresponding predicted high heat load period, but are only used to maintain the thermal isolation margin between different thermal pulse source units.

10. The method for optimizing the heat dissipation performance of an integrated circuit module according to claim 1, characterized in that, After correcting the directed heat diffusion graph in step S8, the heat dissipation performance of the integrated circuit module (10) is determined based on the change in thermal coupling strength before and after the correction. If the thermal coupling strength of the same heat diffusion edge decreases or the heat diffusion delay increases within multiple consecutive control cycles, the priority of selecting the thermal shadow chain associated with the heat diffusion edge is increased so as to form an early cold shadow reservation in the heat dissipation performance decay area.

Images