Integrated circuit heat distribution optimization method integrating situation awareness driving

Abstract

The invention discloses an integrated circuit heat distribution optimization method driven by fusion situation awareness, and particularly relates to the technical field of integrated circuit heat distribution optimization, which comprises the following steps: acquiring a behavior fragment set in an integrated circuit, performing regulation and control trigger sequence adjustment and regulation and control chain atlas construction operation, constructing a regulation and control chain atlas, and establishing a regulation and control chain atlas; performing identification and extraction operation of path traversal on the regulation and control chain atlas, and performing mapping to generate a disturbance input variable set; extracting data in the disturbance input variable set, performing normalization processing and time sequence reconfiguration operation, and constructing a disturbance response track; a regulation and control path structure caused by a temperature control behavior in an integrated circuit is extracted, and a stability identification model is constructed based on path disturbance propagation characteristics and behavior feedback offset, so that dynamic judgment and regulation and control structure optimization of a potential thermal feedback chain in a multi-core environment are realized, and task scheduling imbalance caused by path evolution accumulation is avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of integrated circuit thermal distribution optimization, and more specifically, to an integrated circuit thermal distribution optimization method integrated with situational awareness and driven by. Background Art

[0002] In existing multi-core integrated circuits, thermal management strategies are generally implemented through the coordinated implementation of temperature sensing, dynamic voltage and frequency adjustment, and task migration. This strategy is based on the premise that local temperature conditions drive control responses. It is believed that when a temperature rise occurs in a single processing core or a local area, thermal risks can be mitigated by reducing the frequency or migrating tasks.

[0003] However, in scenarios with multiple concurrent tasks and complex scheduling dependencies, local thermal control often causes task path reconfiguration, which in turn affects the load distribution of other processing units and causes multiple cores to behave in a short period of time. Dynamic voltage and frequency adjustments have physical response lags, and task migration involves cross-core synchronization and context switching, making scheduling behavior temporally incompressible. This results in delayed propagation of local temperature changes within the system through the control mechanism.

[0004] Existing technologies fail to model the association between thermally driven control responses. This results in a system-level temperature control strategy that uses thermal state as input, control response as intermediary, and task behavior stimulation as feedback. Without dynamic assessment of the evolutionary trend of this feedback path, system control behavior can accumulate over multiple responses, causing the thermal control strategy to no longer suppress temperature rise and instead become the primary cause of task structure imbalance, further triggering register configuration errors, data consistency offsets, or scheduling order disruptions.

[0005] This shows that the existing thermal management mechanism fails to identify the disruptive effect of the thermal feedback path on the stability of the scheduling structure under multi-core and multi-path execution conditions, forming a potential error inducement with the appearance of thermal control and the essence of behavioral oscillation. Summary of the Invention

[0006] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides an integrated circuit thermal distribution optimization method driven by integrated situational awareness. By extracting the control path structure caused by temperature control behavior in the integrated circuit and constructing a stability identification model based on the path disturbance propagation characteristics and behavior feedback offset, it realizes dynamic judgment and control structure optimization of potential thermal feedback chains in a multi-core environment, avoiding task scheduling imbalance caused by path evolution accumulation.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a method for optimizing integrated circuit thermal distribution driven by integrated situational awareness, comprising:

[0008] S1: Obtain a set of behavioral fragments in the integrated circuit and perform regulation trigger sequence adjustment and regulation chain map construction operations, build a regulation chain map, perform path traversal identification and extraction operations on the regulation chain map and perform mapping to generate a set of perturbation input variables;

[0009] S2: Extract the data from the disturbance input variable set and perform normalization and time rematching operations to construct the disturbance response trajectory. The disturbance response trajectory is input into the joint disturbance model and sorted by path disturbance index, and the sorted path disturbance index set is output.

[0010] S3: Based on the node extraction results of the sorted path perturbation index set, feature comparison is performed in combination with the sorted path perturbation index set, a path evaluation vector set is constructed, and alternative pathways that meet the conditions are screened for replacement. The updated regulatory chain map is generated in combination with the regulatory chain map;

[0011] S4: Analyze the node state sequence and path structure changes of the updated regulatory chain map, and calculate the multi-parameter change in the control offset analysis to obtain the behavioral control offset vector. Based on the behavioral control offset vector, confirm that the alternative path execution replacement is established, and combine the disturbance features of the confirmed alternative path into training samples to generate a trend judgment model.

[0012] In a preferred embodiment, in S1, a set of behavior segments in the integrated circuit is obtained, the behavior segment set including a task activation sequence and a frequency adjustment record, and a control trigger sequence adjustment is performed through the behavior segment set to output a node sequence;

[0013] Execute the regulatory chain map construction operation on the node sequence to construct a regulatory chain map, which includes active regulatory nodes, reactivated behavior nodes and regulatory trigger edges;

[0014] The control chain map construction operation extracts the task activation sequence and frequency adjustment record of each node in the node sequence to identify the time and action position of each node. Based on the time sequence in the task activation sequence and the control response relationship in the frequency adjustment record, the control trigger relationship between the nodes in the node sequence is identified, and the node pairs with the time sequence and control response dependency trigger relationship are formed into control trigger pairs;

[0015] All control trigger pairs are distinguished, and the nodes corresponding to the control behaviors in the frequency regulation records are marked as active control nodes. The nodes activated or reactivated due to active control triggers are marked as reactivation behavior nodes, and the trigger direction between the two is retained. Control trigger edges are established for all control trigger pairs in the original order of the node sequence to form an initial subgraph structure with path directionality. Based on the scheduling period corresponding to each node in the control trigger pair in the task activation sequence, the initial subgraph structure is grouped and classified to generate a local control subgraph based on the corresponding scheduling period in the task activation sequence. Based on the order of the corresponding scheduling period in the task activation sequence, all local control subgraphs are structurally merged to output a control chain graph containing active control nodes, reactivation behavior nodes and control trigger edges, and retaining the scheduling period sequence mapping relationship.

[0016] In a preferred embodiment, S1 further includes performing a path traversal identification and extraction operation on the regulatory chain map to generate a set of candidate thermal feedback pathways;

[0017] Based on each active control node, reactivation behavior node and control trigger edge in the candidate thermal feedback path set, a data position mapping relationship between the task activation sequence and the frequency adjustment record is established to form a disturbance input variable set. The disturbance input variable set includes the frequency change value corresponding to the active control node extracted according to the data position mapping relationship, the temperature rise data corresponding to the reactivation behavior node, and the task switching frequency involved in the control trigger edge.

[0018] In a preferred embodiment, in S2, a data extraction operation is performed on each node located by a path index in the disturbance input variable set to generate a path disturbance input sequence, wherein the path disturbance input sequence includes a temperature rise sequence, a frequency change sequence, and a behavior switching frequency sequence;

[0019] Normalization and timing rematching operations are performed on the path perturbation input sequence to construct the perturbation response trajectory.

[0020] In a preferred embodiment, S2 further includes inputting the disturbance response trajectory into joint disturbance modeling, performing multivariate variation analysis in the joint disturbance modeling, and outputting a set of path disturbance indices;

[0021] Sort the path perturbation index set by the path perturbation index, and output the sorted path perturbation index set;

[0022] In the joint disturbance modeling, multivariate variation analysis is performed. By horizontally aggregating the temperature rise value, frequency change value, and behavior switching frequency corresponding to each node position in the disturbance response trajectory, a multivariate collaborative input sequence is constructed. The multivariate collaborative input sequence is subjected to inter-node cross-variable difference expansion, adjacent node disturbance magnitude difference statistics, and inter-variable synchronous fluctuation analysis over the entire length of the path in the path disturbance input sequence to identify whether there are disturbance aggregation characteristics such as coordinated growth, drastic deviation, or structural overlap among the multiple variables.

[0023] Based on the disturbance aggregation characteristics, the co-fluctuation amplitude of the variables is statistically analyzed and the aggregation offset strength value is calculated. The aggregation offset strength value of each path in the candidate thermal feedback path set on the overall disturbance dimension is quantified as the path disturbance index, and the path disturbance index is used as the measurement output of the aggregation offset strength.

[0024] In a preferred embodiment, in S3, node extraction is performed on the path that is ranked first in the sorted path disturbance index set, and a set of candidate replacement paths is output;

[0025] The alternative path candidate set is compared with the execution features of the path that ranks first in the sorted path perturbation index set to construct a path evaluation vector set. The path evaluation vector set includes path length, resource conflict coefficient and recorded behavior stability label.

[0026] In a preferred embodiment, S3 also includes conditional judgment on any path in the path evaluation vector set, judging whether the length of any path in the path evaluation vector set is less than the length of the path with the first sorting position in the sorted path disturbance index set, whether the resource conflict coefficient of any path in the path evaluation vector set is lower than the resource conflict coefficient of the path with the first sorting position in the sorted path disturbance index set, and whether the behavior stability score value of any path in the path evaluation vector set is greater than a preset behavior stability score threshold, as three conditions for judgment;

[0027] If all three conditions meet their corresponding judgment conditions, the path replacement operation is performed, otherwise the path with the first ranking position in the sorted path perturbation index set is retained;

[0028] The pathways that have undergone path replacement are subjected to node replacement and edge reconstruction operations on the regulatory chain graph, and the output is the updated regulatory chain graph.

[0029] In a preferred embodiment, in S4, node state sequence extraction and path structure change analysis are performed on the updated regulatory chain graph to generate a behavior state change tensor;

[0030] The behavior state change tensor is input into the control offset analysis, multi-parameter change calculation is performed in the control offset analysis, and the behavior control offset vector is output;

[0031] The node state sequence extraction and path structure change analysis records the activation marks of all behavior nodes in the traversal and updated control chain graph within the corresponding scheduling period in each task activation sequence, and arranges them in order according to the corresponding scheduling period in the task activation sequence to form a discrete time series. The control chain graphs before and after the path replacement operation are then encoded into path sets, and the node connection sequence and the number of control edges of each path in the path set are matched and compared. The structural differences are counted to construct a path change matrix. The path change matrix is ​​used to express the topological changes of the control chain structure before and after the path replacement.

[0032] In a preferred embodiment, S4 further includes performing a threshold judgment on the behavior control offset vector to determine whether the behavior control offset vector value meets a preset behavior control offset vector threshold. If so, confirming that the path replacement operation is established based on the behavior offset influence, otherwise, falling back to the path replacement operation of screening.

[0033] The disturbance response trajectory corresponding to the path that confirms the validity of the path replacement operation under the influence of the behavior deviation is combined with the behavior control deviation vector as a training sample and input into the behavior perception prediction training model to generate a trend determination model;

[0034] In the behavior perception prediction training model, the temperature rise sequence, frequency change sequence, and behavior switching frequency sequence in the disturbance response trajectory are expanded into time series according to the path order of each node in the candidate thermal feedback path set. In the time series, the corresponding cumulative change amplitude, average growth rate, and change amplitude in the local increment are extracted from the temperature rise sequence, frequency change sequence, and behavior switching frequency sequence, respectively, to form three sets of statistical vectors representing path disturbance characteristics. At the same time, the distribution ratio of the control deviation degree among different behavior categories, the overall trend of the deviation change direction, and the structural fracture index of the behavior connectivity are extracted from the behavior control deviation vector to form three sets of control response description quantities representing the control consequence characteristics.

[0035] The three groups of statistical vectors, the three groups of control response description quantities, and the paths where the confirmation path substitution operation is valid under the influence of behavioral deviation are aligned and spliced ​​to form a multidimensional joint feature vector. The multidimensional joint feature vector is used to represent the complete characteristics of the paths where the confirmation path substitution operation is valid under the influence of behavioral deviation under the dimensions of disturbance and feedback influence. Based on the paths where the confirmation path substitution operation is valid under the influence of behavioral deviation, the corresponding disturbance response trajectories and behavioral control deviation vectors are combined into training samples for use by the behavior perception prediction training model. A mapping relationship is established from the joint space of disturbance characteristics and deviation responses to the risk trend of thermal control instability, and a trend judgment model for identifying potential feedback risks in the regulation path is constructed.

[0036] Technical effects and advantages of the present invention:

[0037] 1. This solution, by injecting multivariable disturbances and extracting control response deviations, breaks the assumption that “stability of the control path is safety” and verifies the control stability of the thermal feedback path.

[0038] 2. By constructing a path disturbance index and a control offset description vector group, we comprehensively analyze the temperature rise response severity, frequency regulation hysteresis, and task switching frequency to identify potential abnormal paths that are unstable under disturbances.

[0039] 3. Based on the regulatory chain map and pathway structure change tensor, identify behavioral connectivity breaks and topological changes after regulatory substitution, and determine the structural stability of pathway substitution results;

[0040] 4. Using a perturbation path ranking and candidate path multi-factor comparison mechanism, we conduct low-cost alternative simulations on suspicious thermal feedback paths and construct an alternative path selection strategy with regulatory stability criteria;

[0041] 5. By integrating the disturbance response trajectory and control offset characteristics, multi-dimensional supervised training samples are generated, and the behavior perception prediction model is trained to improve the trend prediction ability of the feedback risk of the control path. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is a flowchart of the framework of the method steps of the present invention.

[0043] Figure 2 A flow chart is constructed for the regulatory chain map of the present invention.

[0044] Figure 3 A flow chart for disturbance modeling and path disturbance index generation of the present invention.

[0045] Figure 4 Updated flowchart for pathway alternative assessment and regulatory chain of the present invention.

[0046] Figure 5This is a flow chart of the behavior deviation analysis and trend model training of the present invention. DETAILED DESCRIPTION

[0047] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0048] Refer to the instruction manual Figure 1-5 According to an embodiment of the present invention, a method for optimizing thermal distribution of integrated circuits driven by integrated situational awareness includes:

[0049] S1: Obtain a set of behavioral fragments in the integrated circuit and perform regulation trigger sequence adjustment and regulation chain map construction operations, build a regulation chain map, perform path traversal identification and extraction operations on the regulation chain map and perform mapping to generate a set of perturbation input variables;

[0050] S2: Extract the data from the disturbance input variable set and perform normalization and time rematching operations to construct the disturbance response trajectory. The disturbance response trajectory is input into the joint disturbance model and sorted by path disturbance index, and the sorted path disturbance index set is output.

[0051] S3: Based on the node extraction results of the sorted path perturbation index set, feature comparison is performed in combination with the sorted path perturbation index set, a path evaluation vector set is constructed, and alternative pathways that meet the conditions are screened for replacement. The updated regulatory chain map is generated in combination with the regulatory chain map;

[0052] S4: Analyze the node state sequence and path structure changes of the updated regulatory chain map, and calculate the multi-parameter change in the control offset analysis to obtain the behavioral control offset vector. Based on the behavioral control offset vector, confirm that the alternative path execution replacement is established, and combine the disturbance features of the confirmed alternative path into training samples to generate a trend judgment model.

[0053] In S1, a set of behavior fragments in the integrated circuit is obtained. The behavior fragment set includes a task activation sequence and a frequency adjustment record. The control trigger sequence is adjusted based on the behavior fragment set, and a node sequence is output. The control trigger sequence adjustment refers to rearranging the triggering order of the control behavior in time based on the task activation sequence and the frequency adjustment record, restoring the causal execution relationship of the behavior nodes in the control chain, and forming a node sequence with a logical order.

[0054] Execute the regulatory chain map construction operation on the node sequence to construct a regulatory chain map, which includes active regulatory nodes, reactivated behavior nodes and regulatory trigger edges;

[0055] The control chain map construction operation extracts the task activation sequence and frequency adjustment record of each node in the node sequence to identify the time and action position of each node. Based on the time sequence in the task activation sequence and the control response relationship in the frequency adjustment record, the control trigger relationship between the nodes in the node sequence is identified, and the node pairs with the time sequence and control response dependency trigger relationship are formed into control trigger pairs;

[0056] All control trigger pairs are distinguished, and the nodes corresponding to the control behaviors in the frequency regulation records are marked as active control nodes. The nodes activated or reactivated due to active control triggers are marked as reactivation behavior nodes, and the trigger direction between the two is retained. Control trigger edges are established for all control trigger pairs in the original order of the node sequence to form an initial subgraph structure with path directionality. Based on the scheduling period corresponding to each node in the control trigger pair in the task activation sequence, the initial subgraph structure is grouped and classified to generate a local control subgraph based on the corresponding scheduling period in the task activation sequence. Based on the order of the corresponding scheduling period in the task activation sequence, all local control subgraphs are structurally merged to output a control chain graph containing active control nodes, reactivation behavior nodes and control trigger edges, and retaining the scheduling period sequence mapping relationship.

[0057] S1 also includes performing a path traversal identification and extraction operation on the regulatory chain graph to generate a set of candidate thermal feedback paths, wherein the path traversal identification and extraction operation refers to performing a complete path search in the regulatory chain graph to identify the path structure containing the regulatory behavior reflux relationship, and forming a set of paths with feedback features;

[0058] Based on each active control node, reactivation behavior node and control trigger edge in the candidate thermal feedback path set, a data position mapping relationship between the task activation sequence and the frequency adjustment record is established to form a disturbance input variable set. The disturbance input variable set includes the frequency change value corresponding to the active control node extracted according to the data position mapping relationship, the temperature rise data corresponding to the reactivation behavior node, and the task switching frequency involved in the control trigger edge.

[0059] In S2, a data extraction operation is performed on the node located by each path index in the disturbance input variable set to generate a path disturbance input sequence. The path disturbance input sequence includes a temperature rise sequence, a frequency change sequence, and a behavior switching frequency sequence. The data extraction operation refers to extracting the recorded temperature rise, frequency change, and behavior switching frequency data in the node located by each path in the disturbance input variable set to construct the path disturbance input sequence.

[0060] Normalization and timing rematching are performed on the path perturbation input sequence to construct a disturbance response trajectory. Normalization and timing rematching refer to the standardization of the numerical domain and alignment of the time axis of various variables in the path perturbation input sequence, so as to maintain a variable mapping structure consistent with the order of path nodes when constructing the disturbance response trajectory.

[0061] It should be noted that in the formula structure involved in this solution, dimensionless terms can serve as proportionality or structural adjustment factors. When combined with quantities with units, they only play a numerical scaling role and do not introduce new physical dimensions. Therefore, they will not change or confuse the overall unit system of expression. This combination of "dimensionless terms and units" can be understood as a composite structural expression commonly used in mathematical and physical modeling, conforming to the principle of dimensional consistency and having a clear physical interpretation basis.

[0062] Secondly, in the formula structure of this scheme, if multiple variables with different physical units are involved, including but not limited to time, mass or energy variables, their joint appearance is to express the collaborative modeling relationship of multiple physical mechanisms. Each variable can be formed into a unified structure through function mapping, ratio combination or normalization adjustment. The units and meanings are clear, and the overall expression conforms to the principle of dimensional consistency and the common formula of engineering modeling.

[0063] In this solution, any design constants, weights, adjustment factors, threshold parameters, and proportional coefficients are adjustable control parameters for different application environments. Their values ​​depend on the target device configuration, data input characteristics, and performance optimization goals. During the implementation phase, they are set within a reasonable range through model verification, performance constraints, or engineering calibration. Although these parameters do not have unique preset values, they have clear adjustment logic and calculation paths, and are part of the deterministic setting process in engineering implementation. The purpose of such setting is to ensure that the solution is both universally adaptable, reproducible, and operable, without affecting its technical clarity and feasibility.

[0064] S2 also includes inputting the disturbance response trajectory into joint disturbance modeling, performing multivariate variation analysis in the joint disturbance modeling, and outputting a set of path disturbance indexes. Joint disturbance modeling refers to using the temperature rise sequence, frequency change sequence, and behavior switching frequency sequence contained in the disturbance response trajectory as input variables to construct a joint modeling mechanism. The joint modeling mechanism is used to capture the dynamic coupling characteristics between multiple variables and the path disturbance propagation mechanism.

[0065] Define the path perturbation index set Θ (p) :

[0066]

[0067] Δζ (p)(t)=|ζ (p) (t)-ζ (p) (t-1)|,ζ (p) (t) = T (p) (t)+Γ (p) (t)+φ (p) (t)

[0068]

[0069] Where p is the path number in the candidate thermal feedback path set; L (p) is the total length of nodes in the candidate thermal feedback path set; t is the node time index in the path disturbance response trajectory; is the disturbance response intensity function; represents the disturbance second-order variation function, which reflects the disturbance acceleration; ζ (p) (t) is the instantaneous value of the disturbance joint variable; Δζ (p) (t) is the instantaneous change amplitude of the disturbance joint variable; κ (p) (t) is the disturbance direction consistency response function; dt is the discrete time index step; T (p) (t) is the temperature rise value of the node with time index t in the path disturbance response trajectory of the path numbered p in the candidate thermal feedback path set; Γ (p) (t) is the frequency change value of the path numbered p in the candidate thermal feedback path set with time index t in the path disturbance response trajectory; φ (p) (t) is the behavior switching frequency of the node with time index t in the path disturbance response trajectory of the path numbered p in the candidate thermal feedback path set; log(·) is a logarithmic function based on the natural logarithm; It represents the temperature rise value on the path numbered p in the candidate thermal feedback path set, and the second-order derivative of the node with time index t in the path disturbance response trajectory; Represents the frequency change value on the path numbered p in the candidate thermal feedback path set, and the second-order derivative of the node with time index t in the path disturbance response trajectory; represents the behavior switching frequency on the path numbered p in the candidate thermal feedback path set, and the second-order derivative of the node with time index t in the path disturbance response trajectory; ζ (p) (t-1) represents the instantaneous value of the disturbance joint variable at the previous moment of the node with time index t in the path disturbance response trajectory; sign(·) is the sign function, which is used to extract directional consistency; exp(·) is the exponential function, which is used to convert the disturbance consistency into a normalized trend consistency index;

[0070] Sort the path perturbation index set by the path perturbation index, and output the sorted path perturbation index set;

[0071] In the joint disturbance modeling, multivariate variation analysis is performed. By horizontally aggregating the temperature rise value, frequency change value, and behavior switching frequency corresponding to each node position in the disturbance response trajectory, a multivariate collaborative input sequence is constructed. The multivariate collaborative input sequence is subjected to inter-node cross-variable difference expansion, adjacent node disturbance magnitude difference statistics, and inter-variable synchronous fluctuation analysis over the entire length of the path in the path disturbance input sequence to identify whether there are disturbance aggregation characteristics such as coordinated growth, drastic deviation, or structural overlap among the multiple variables.

[0072] Based on the disturbance aggregation characteristics, the co-fluctuation amplitude of the variables is statistically analyzed and the aggregation offset strength value is calculated. The aggregation offset strength value of each path in the candidate thermal feedback path set on the overall disturbance dimension is quantified as the path disturbance index, and the path disturbance index is used as the measurement output of the aggregation offset strength.

[0073] In S3, node extraction is performed on the path ranked first in the sorted path disturbance index set, and a candidate set of replacement paths is output. Node extraction refers to obtaining the start node and end node of the path ranked first in the sorted path disturbance index set, and locating the inactivated paths that can be used for replacement in the scheduling graph based on the nodes.

[0074] Compare the execution features of the path with the first ranking position in the sorted path disturbance index set with the candidate set of alternative paths, and construct a set of path evaluation vectors. The path evaluation vector set includes path length, resource conflict coefficient and recorded behavior stability label;

[0075] Define the path evaluation vector set V (h) :

[0076] V (h) =[Λ (h) ,Δ (h) ,Σ (h) ]

[0077]

[0078] Where h is the number of the candidate alternative path; (h) represents the cumulative physical length of the candidate alternative path numbered h; Δ (h) represents the resource conflict metric of the candidate alternative path numbered h; Σ (h) represents the behavioral stability score of the candidate alternative path numbered h; L h represents the number of nodes in the candidate alternative path numbered h; j is the node number; represents the distance between the node numbered j and the successor node in the candidate alternative path numbered h; Indicates the number of resource control units occupied by node j in the candidate replacement path numbered h; TCU j Indicates the total resource units in the scheduling period corresponding to the node numbered j; represents the dependency depth of the node numbered j in the candidate alternative path numbered h; ∈ is the perturbation constant, which prevents the denominator from being zero; Represents the historical stability label value of the node numbered j in the candidate replacement path numbered h; Represents the consistency factor of node numbered j in the candidate replacement path numbered h.

[0079] S3 also includes conditional judgment on any path in the path evaluation vector set, judging whether the length of any path in the path evaluation vector set is less than the length of the path with the first sorting position in the sorted path disturbance index set, whether the resource conflict coefficient of any path in the path evaluation vector set is lower than the resource conflict coefficient of the path with the first sorting position in the sorted path disturbance index set, and whether the behavior stability score value of any path in the path evaluation vector set is greater than a preset behavior stability score threshold, as three conditions for judgment;

[0080] If all three conditions meet their corresponding judgment conditions, the path replacement operation is performed. Otherwise, the path ranked first in the sorted path perturbation index set is retained, and the path ranked first in the sorted path perturbation index set is used to enter the next cycle for re-evaluation. The path replacement operation refers to replacing the path ranked first in the sorted path perturbation index set with any path in the path evaluation vector set that meets the corresponding judgment conditions, thereby "breaking" the potential thermal feedback chain.

[0081] The path that has performed path replacement and the regulatory chain graph are subjected to node replacement and edge reconstruction operations, and the output is an updated regulatory chain graph, where the node replacement and edge reconstruction operations refer to removing the regulatory trigger edge connected backward to the end node of the path that has performed path replacement.

[0082] In S4, node state sequence extraction and path structure change analysis are performed on the updated regulatory chain graph to generate a behavior state change tensor;

[0083] The behavior state change tensor is input into the control offset analysis, and multi-parameter change calculation is performed in the control offset analysis to output a behavior control offset vector, which is used to quantify the impact of the broken chain behavior;

[0084] Define the control offset vector C (p) :

[0085] C (p) =[M (p) ,Υ(p) ,Ξ (p) ]

[0086]

[0087] Among them, M (p) Activate jump change value for behavior; (p) is the change value of the frequency modulation response variance ratio; (p) is the behavior connectivity change index; Z is the total number of scheduling cycles; u is the scheduling cycle index; η (p) (u) represents the number of behavioral activation nodes of the path numbered p in the candidate thermal feedback path set in the scheduling period indexed u; represents the number of behavioral activation nodes of the path numbered p in the candidate thermal feedback path set in the previous scheduling cycle with scheduling cycle index u; Var(·) is the sequence variance operator; represents the frequency adjustment response sequence before the replacement operation of the path numbered p in the candidate thermal feedback path set; represents the frequency adjustment response sequence after the path numbered p in the candidate thermal feedback path set is replaced; It represents the number of valid edges in the regulatory chain graph before the replacement operation of the path numbered p in the candidate thermal feedback path set; It represents the number of valid edges in the regulatory chain graph after the replacement operation of the path numbered p in the candidate thermal feedback path set;

[0088] The node state sequence extraction and path structure change analysis records the activation marks of all behavior nodes in the traversal and updated control chain graph within the corresponding scheduling period in each task activation sequence, and arranges them in order according to the corresponding scheduling period in the task activation sequence to form a discrete time series. The control chain graphs before and after the path replacement operation are then encoded into path sets, and the node connection sequence and the number of control edges of each path in the path set are matched and compared. The structural differences are counted to construct a path change matrix. The path change matrix is ​​used to express the topological changes of the control chain structure before and after the path replacement.

[0089] S4 also includes performing a threshold judgment on the behavior control offset vector to determine whether the behavior control offset vector value meets a preset behavior control offset vector threshold. If so, confirming that the path replacement operation is established based on the behavior offset influence, otherwise, falling back to the path replacement operation of filtering;

[0090] The disturbance response trajectory corresponding to the path that confirms the validity of the path replacement operation under the influence of the behavior deviation is combined with the behavior control deviation vector as a training sample and input into the behavior perception prediction training model to generate a trend determination model. The trend determination model is used to identify the thermal control instability trend caused by the evolution of the control path in feedback situational awareness.

[0091] In the behavior perception prediction training model, the temperature rise sequence, frequency change sequence, and behavior switching frequency sequence in the disturbance response trajectory are expanded into time series according to the path order of each node in the candidate thermal feedback path set. In the time series, the corresponding cumulative change amplitude, average growth rate, and change amplitude in the local increment are extracted from the temperature rise sequence, frequency change sequence, and behavior switching frequency sequence, respectively, to form three sets of statistical vectors representing path disturbance characteristics. At the same time, the distribution ratio of the control deviation degree among different behavior categories, the overall trend of the deviation change direction, and the structural fracture index of the behavior connectivity are extracted from the behavior control deviation vector to form three sets of control response description quantities representing the control consequence characteristics.

[0092] The three sets of statistical vectors and the three sets of control response descriptors are aligned and spliced ​​with the paths where the confirmation path replacement operation is valid under the influence of behavioral deviation to form a multidimensional joint feature vector. The multidimensional joint feature vector is used to represent the complete characteristics of the paths where the confirmation path replacement operation is valid under the influence of behavioral deviation under the dimensions of disturbance and feedback influence. Based on the paths where the confirmation path replacement operation is valid under the influence of behavioral deviation, the corresponding disturbance response trajectories and behavioral control deviation vectors are combined into training samples for use in the behavior perception prediction training model. A mapping relationship is established from the joint space of disturbance characteristics and deviation responses to the risk trend of thermal control instability, and a trend judgment model for identifying potential feedback risks in the control path is constructed.

[0093] The behavior perception prediction training model includes a long short-term memory network model or a random forest model;

[0094] In this scheme, when the long short-term memory network model is applied to the behavior perception prediction training model, the long short-term memory network model receives as input a multi-dimensional joint feature vector formed by concatenating the disturbance response trajectory and the behavior control offset vector. Based on the path node sequence that confirms the validity of the path replacement operation on the behavior offset, it learns the dynamic correlation between the path disturbance characteristics and the control consequence characteristics in the time evolution dimension, and identifies the potential thermal control instability trend of the control path in the feedback chain. The long short-term memory network model retains the long-range dependency characteristics of the path disturbance characteristics between different control nodes through the time step transmission mechanism, so that the trend determination model has the ability to distinguish between the aggregation phenomenon of severe disturbance and the continuity of disturbance slow release, ensuring the adaptability and recognition accuracy of the prediction model to complex control path structures.

[0095] In this scheme, when the random forest model is applied to the behavior perception prediction training model, the random forest model uses the path that confirms that the path replacement operation is valid under the influence of behavior deviation as the index, collects the training samples of the corresponding disturbance response trajectory and the behavior control deviation vector as input, and constructs a nonlinear mapping relationship between the high-dimensional disturbance-control feature space and the feedback risk label through multiple child decision trees; during the training stage, the random forest model uses the disturbance aggregation pattern and the control deviation trend as the key features as the basis for division, and realizes the classification judgment of whether there is a thermal control instability trend in the control path, thereby supporting the situational awareness identification and predictive intervention of potential feedback risks under multi-path execution conditions.

[0096] The following points require general explanation, including but not limited to: Existing thermal management mechanisms for multi-core integrated circuits generally rely on local temperature sensing and static control strategies, primarily addressing single-point temperature rise risks through frequency regulation and task migration, while ignoring the coupling chain between task behavior and control response in the multi-core structure;

[0097] In scenarios with multiple concurrent tasks and complex scheduling paths, local control operations may trigger cross-core behavioral linkages, resulting in delayed propagation of disturbances and feedback accumulation in the control structure. This can cause thermal control strategies to become the cause of task structure imbalance at the system level.

[0098] Traditional control methods do not establish a model for the behavioral transmission relationship in the control path, lack a dynamic judgment and early warning mechanism for the evolution trend of the disturbance chain, and are difficult to identify potential risks of thermal control instability;

[0099] Therefore, a thermal distribution optimization solution that can integrate the control path structure, task behavior changes, and disturbance response mechanism and has dynamic feedback recognition capabilities is urgently needed to achieve the transformation of integrated circuit thermal control strategy from static response to trend perception.

[0100] This plan includes the stages of perturbation variable extraction and regulatory chain map construction:

[0101] Obtaining a set of behavioral fragments in the integrated circuit, performing a regulatory chain graph construction operation after adjusting the regulatory trigger sequence to generate a regulatory chain graph, identifying and extracting the regulatory chain graph, and generating a set of perturbation input variables based on the position mapping relationship between nodes and regulatory trigger edges;

[0102] This stage lays a unified structural foundation for building a path disturbance response model and ensures that the disturbance variables have a traceable structure;

[0103] This solution includes the multivariate modeling and path perturbation index ranking stages:

[0104] Perform data extraction, normalization, and time rematching on the disturbance input variable set to construct a disturbance response trajectory. The disturbance response trajectory is input into the joint disturbance modeling. After performing multivariate variation analysis, the path disturbance index is sorted and the sorted path disturbance index set is output.

[0105] This stage quantifies the impact of each feedback path on the propagation of system disturbances, providing a quantitative evaluation basis for screening key paths and structural replacements;

[0106] This solution includes the stages of path evaluation vector construction and alternative path identification:

[0107] Perform node extraction on the path that ranks first in the sorted path perturbation index set, construct a set of candidate alternative paths, compare the features of the candidate alternative paths with the original path, construct a set of path evaluation vectors, screen the set of path evaluation vectors, and perform path replacement operations on alternative paths that meet the three corresponding judgment conditions. Combined with the regulatory chain map, perform node replacement and edge reconstruction to generate an updated regulatory chain map.

[0108] In this stage, while maintaining the functionality of the scheduling topology, the thermal feedback risk chain is reduced through path structure replacement to interrupt the potential evolution trend of thermal instability;

[0109] This solution includes the behavioral control deviation analysis and trend prediction modeling phase:

[0110] Analyze the node state sequence and path structure changes of the updated regulatory chain map, generate a behavior state change tensor, and calculate multi-parameter changes based on the behavior state change tensor. Output the behavior control offset vector, judge the behavior control offset vector, confirm that the path replacement operation is valid under the influence of the behavior offset, combine the corresponding disturbance response trajectory and the control offset vector into a training sample, input it into the behavior perception prediction training model, and generate a trend determination model.

[0111] This stage is used to identify whether the regulatory pathway has a trend that leads to thermal control instability in its evolved state, and to achieve path risk prediction and feedback intervention through the trend determination model.

[0112] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for optimizing integrated circuit thermal distribution driven by integrated situational awareness, characterized in that: include: S1: Obtain a set of behavioral fragments in the integrated circuit and perform regulation trigger sequence adjustment and regulation chain map construction operations, build a regulation chain map, perform path traversal identification and extraction operations on the regulation chain map and perform mapping to generate a set of perturbation input variables; S2: Extract the data from the disturbance input variable set and perform normalization and time rematching operations to construct the disturbance response trajectory. The disturbance response trajectory is input into the joint disturbance model and sorted by path disturbance index, and the sorted path disturbance index set is output. S3: Based on the node extraction results of the sorted path perturbation index set, feature comparison is performed in combination with the sorted path perturbation index set, a path evaluation vector set is constructed, and alternative pathways that meet the conditions are screened for replacement. The updated regulatory chain map is generated in combination with the regulatory chain map; S4: Analyze the node state sequence and path structure changes of the updated regulatory chain map, and calculate the multi-parameter change in the control offset analysis to obtain the behavioral control offset vector. Based on the behavioral control offset vector, confirm that the alternative path execution replacement is established, and combine the disturbance features of the confirmed alternative path into training samples to generate a trend judgment model.

2. The integrated circuit thermal distribution optimization method integrated with situational awareness drive according to claim 1, characterized in that: In S1, a set of behavior fragments in the integrated circuit is obtained. The behavior fragment set includes a task activation sequence and a frequency adjustment record. Through the behavior fragment set, the control trigger sequence is adjusted and a node sequence is output; Execute the regulatory chain map construction operation on the node sequence to construct a regulatory chain map, which includes active regulatory nodes, reactivated behavior nodes and regulatory trigger edges; The control chain map construction operation extracts the task activation sequence and frequency adjustment record of each node in the node sequence to identify the time and action position of each node. Based on the time sequence in the task activation sequence and the control response relationship in the frequency adjustment record, the control trigger relationship between the nodes in the node sequence is identified, and the node pairs with the time sequence and control response dependency trigger relationship are formed into control trigger pairs; All control trigger pairs are distinguished, and the nodes corresponding to the control behaviors in the frequency regulation records are marked as active control nodes. The nodes activated or reactivated due to active control triggers are marked as reactivation behavior nodes, and the trigger direction between the two is retained. Control trigger edges are established for all control trigger pairs in the original order of the node sequence to form an initial subgraph structure with path directionality. Based on the scheduling period corresponding to each node in the control trigger pair in the task activation sequence, the initial subgraph structure is grouped and classified to generate a local control subgraph based on the corresponding scheduling period in the task activation sequence. Based on the order of the corresponding scheduling period in the task activation sequence, all local control subgraphs are structurally merged to output a control chain graph containing active control nodes, reactivation behavior nodes and control trigger edges, and retaining the scheduling period sequence mapping relationship.

3. The integrated circuit thermal distribution optimization method integrated with situational awareness drive according to claim 2, characterized in that: S1 also includes the identification and extraction operation of performing path traversal on the regulatory chain map to generate a set of candidate thermal feedback pathways; Based on each active control node, reactivation behavior node and control trigger edge in the candidate thermal feedback path set, a data position mapping relationship between the task activation sequence and the frequency adjustment record is established to form a disturbance input variable set. The disturbance input variable set includes the frequency change value corresponding to the active control node extracted according to the data position mapping relationship, the temperature rise data corresponding to the reactivation behavior node, and the task switching frequency involved in the control trigger edge.

4. The integrated circuit thermal distribution optimization method integrated with situational awareness drive according to claim 3, characterized in that: In S2, a data extraction operation is performed on the node located by each path index in the disturbance input variable set to generate a path disturbance input sequence. The path disturbance input sequence includes a temperature rise sequence, a frequency change sequence, and a behavior switching frequency sequence. Normalization and timing rematching operations are performed on the path perturbation input sequence to construct the perturbation response trajectory.

5. The integrated circuit thermal distribution optimization method integrated with situational awareness drive according to claim 4, characterized in that: S2 also includes inputting the disturbance response trajectory into the joint disturbance modeling, performing multivariate variation analysis in the joint disturbance modeling, and outputting a set of path disturbance indices; Sort the path perturbation index set by the path perturbation index, and output the sorted path perturbation index set; In the joint disturbance modeling, multivariate variation analysis is performed. By horizontally aggregating the temperature rise value, frequency change value, and behavior switching frequency corresponding to each node position in the disturbance response trajectory, a multivariate collaborative input sequence is constructed. The multivariate collaborative input sequence is subjected to inter-node cross-variable difference expansion, adjacent node disturbance magnitude difference statistics, and inter-variable synchronous fluctuation analysis over the entire length of the path in the path disturbance input sequence to identify whether there are disturbance aggregation characteristics such as coordinated growth, drastic deviation, or structural overlap among the multiple variables. Based on the disturbance aggregation characteristics, the co-fluctuation amplitude of the variables is statistically analyzed and the aggregation offset strength value is calculated. The aggregation offset strength value of each path in the candidate thermal feedback path set on the overall disturbance dimension is quantified as the path disturbance index, and the path disturbance index is used as the measurement output of the aggregation offset strength.

6. The integrated circuit thermal distribution optimization method integrated with situational awareness drive according to claim 5, characterized in that: In S3, node extraction is performed on the path with the first ranking position in the sorted path disturbance index set, and a set of alternative path candidates is output; The alternative path candidate set is compared with the execution features of the path that ranks first in the sorted path perturbation index set to construct a path evaluation vector set. The path evaluation vector set includes path length, resource conflict coefficient and recorded behavior stability label.

7. The integrated circuit thermal distribution optimization method integrated with situational awareness drive according to claim 6, characterized in that: S3 also includes conditional judgment on any path in the path evaluation vector set, judging whether the length of any path in the path evaluation vector set is less than the length of the path with the first sorting position in the sorted path disturbance index set, whether the resource conflict coefficient of any path in the path evaluation vector set is lower than the resource conflict coefficient of the path with the first sorting position in the sorted path disturbance index set, and whether the behavior stability score value of any path in the path evaluation vector set is greater than a preset behavior stability score threshold, as three conditions for judgment; If all three conditions meet their corresponding judgment conditions, the path replacement operation is performed, otherwise the path with the first ranking position in the sorted path perturbation index set is retained; The pathways that have undergone path replacement are subjected to node replacement and edge reconstruction operations on the regulatory chain graph, and the output is the updated regulatory chain graph.

8. The integrated circuit thermal distribution optimization method integrated with situational awareness drive according to claim 7, characterized in that: In S4, node state sequence extraction and path structure change analysis are performed on the updated regulatory chain graph to generate a behavior state change tensor; The behavior state change tensor is input into the control offset analysis, multi-parameter change calculation is performed in the control offset analysis, and the behavior control offset vector is output; The node state sequence extraction and path structure change analysis records the activation marks of all behavior nodes in the traversal and updated control chain graph within the corresponding scheduling period in each task activation sequence, and arranges them in order according to the corresponding scheduling period in the task activation sequence to form a discrete time series. The control chain graphs before and after the path replacement operation are then encoded into path sets, and the node connection sequence and the number of control edges of each path in the path set are matched and compared. The structural differences are counted to construct a path change matrix. The path change matrix is ​​used to express the topological changes of the control chain structure before and after the path replacement.

9. The integrated circuit thermal distribution optimization method integrated with situational awareness drive according to claim 8, characterized in that: S4 also includes performing a threshold judgment on the behavior control offset vector to determine whether the behavior control offset vector value meets a preset behavior control offset vector threshold. If so, confirming that the path replacement operation is established based on the behavior offset influence, otherwise, falling back to the path replacement operation of filtering; The disturbance response trajectory corresponding to the path that confirms the validity of the path replacement operation under the influence of the behavior deviation is combined with the behavior control deviation vector as a training sample and input into the behavior perception prediction training model to generate a trend determination model; In the behavior perception prediction training model, the temperature rise sequence, frequency change sequence, and behavior switching frequency sequence in the disturbance response trajectory are expanded into time series according to the path order of each node in the candidate thermal feedback path set. In the time series, the corresponding cumulative change amplitude, average growth rate, and change amplitude in the local increment are extracted from the temperature rise sequence, frequency change sequence, and behavior switching frequency sequence, respectively, to form three sets of statistical vectors representing path disturbance characteristics. At the same time, the distribution ratio of the control deviation degree among different behavior categories, the overall trend of the deviation change direction, and the structural fracture index of the behavior connectivity are extracted from the behavior control deviation vector to form three sets of control response description quantities representing the control consequence characteristics. The three sets of statistical vectors, the three sets of control response descriptors, and the paths that confirm the validity of the path substitution operation under the influence of behavioral deviation are aligned and spliced ​​to form a multidimensional joint feature vector. The multidimensional joint feature vector is used to represent the complete characteristics of the paths that confirm the validity of the path substitution operation under the influence of behavioral deviation under the dimensions of disturbance and feedback influence. Based on the path of confirming that the path replacement operation is valid under the influence of behavioral deviation, the corresponding disturbance response trajectory and behavioral control deviation vector are composed into training samples for use by the behavior perception prediction training model. A mapping relationship is established from the joint space of disturbance characteristics and deviation response to the thermal control instability risk trend, and a trend judgment model is constructed to identify the potential feedback risk of the regulation path.

Images