Method and system for screening critical path monitoring points in combination with path activation rate and similarity
By combining a multi-dimensional screening method based on path activation rate and logical topology similarity, the problem of redundant path deployment in integrated circuit design is solved, achieving efficient allocation of monitoring resources and improved accuracy of timing anomaly detection.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SHANGHAI JIAOTONG UNIV
- Filing Date
- 2026-03-27
- Publication Date
- 2026-06-09
AI Technical Summary
In the current technology for integrated circuit design, the critical path monitoring point selection process fails to effectively combine path activation rate and logic topology similarity, resulting in excessive deployment of redundant paths, increasing chip area and power consumption, and insufficient monitoring coverage.
By combining multi-dimensional screening methods based on path activation rate and similarity, including static time series analysis, path activation rate calculation, and logical topology similarity clustering, key path monitoring points are screened out to reduce redundant path deployment.
It reduces chip area and power consumption overhead, and improves the accuracy of timing anomaly detection and the effectiveness of monitoring coverage.
Smart Images
Figure CN121936384B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of electronic design automation algorithms and integrated circuit reliability design, and more specifically, to a method and system for screening critical path monitoring points that combines path activation rate and similarity. Background Technology
[0002] In integrated circuit design, Static Timing Analysis (STA) is a core technology for verifying chip timing performance. STA tools can acquire the timing margins of all paths within a chip, thereby identifying critical paths that affect chip performance. To ensure the timing reliability of chips in actual operation, the industry commonly adopts a technique of inserting in-situ monitors at the end of critical paths. By acquiring and analyzing the timing margins of these paths in real time, rapid detection and response to timing anomalies can be achieved. However, in the monitor selection process, blindly expanding the coverage of monitor points, while improving monitoring comprehensiveness, leads to a surge in monitor hardware size, significantly increasing chip footprint and power consumption. Therefore, achieving a precise balance between controlling hardware overhead and ensuring monitoring effectiveness has become a core problem that current critical path monitor selection technology urgently needs to solve.
[0003] Existing critical path screening technologies are mainly based on STA analysis results. The core logic is to extract paths with a timing margin less than a set threshold as candidate monitoring points. Some optimization schemes combine path activation rate, i.e., the probability of a monitoring point being triggered in the actual working scenario, for secondary screening to reduce the hardware overhead caused by inserting monitors.
[0004] Existing research fails to consider the similarity of logical topologies between paths, leading to a high degree of overlap among numerous traversed logical units. Paths with similar timing characteristics are selected simultaneously, constituting redundant paths. Repeatedly deploying monitors results in overly dense monitor deployment, increasing chip area and power consumption without significantly improving the coverage effectiveness of timing monitoring. To achieve efficient and low-overhead critical path monitoring, a multi-dimensional screening scheme combining timing margin, path activation rate, and path similarity is needed. This scheme should eliminate redundant paths through cluster analysis, optimize monitoring resource allocation, and reduce hardware overhead while ensuring effective monitoring coverage.
[0005] Patent application CN117540675A discloses a critical path ranking method based on graph neural networks, belonging to the technical field of computation, estimation, or counting. The method involves: first, obtaining a set of potential critical paths in a netlist using static timing analysis tools; then, converting the netlist file containing circuit function information into a graph data structure and obtaining an initial training set based on relevant process configuration files; next, inputting the initial training set into a graph neural network for training to obtain a critical unit prediction model, which consists of three parts: a deep structured autoencoder, a nonlinear attribute autoencoder, and a joint error reconstruction module; finally, quantifying the criticality of each path in the potential critical path set based on the critical unit information in the netlist and the path criticality calculation algorithm, and obtaining a process-aware critical path ranking. However, this patent application cannot completely solve the existing technical problems, nor can it meet the needs of this invention. Summary of the Invention
[0006] To address the shortcomings of existing technologies, the purpose of this invention is to provide a method and system for screening key path monitoring points that combines path activation rate and similarity.
[0007] The critical path monitoring point screening method combining path activation rate and similarity provided by the present invention includes:
[0008] Step S1: Obtain timing information of all paths of the target circuit based on static timing analysis (STA), and filter out paths with timing margin less than the preset timing margin threshold according to the preset timing margin threshold to form a critical path set.
[0009] Step S2: Based on a preset benchmark test set, perform signal reversal statistics on the monitoring points of each path in the critical path set, calculate the activation rate of each path, and filter out paths with an activation rate of zero to obtain an effective path set.
[0010] Step S3: Group the paths in the set of valid paths according to their starting points to obtain multiple path groups;
[0011] Step S4: For each path group, calculate the similarity between paths based on the logical topological structure sequence of paths within the group, and cluster the paths according to the similarity, selecting at least one path from each cluster.
[0012] Step S5: Summarize all the paths obtained after filtering to form the final set of monitoring points.
[0013] Preferably, before obtaining the timing information of all paths of the target circuit based on static timing analysis (STA), the method includes: performing full-path timing analysis on the design netlist of the target circuit and removing duplicate endpoints to obtain an initial path set without duplicate endpoints.
[0014] Preferably, the calculation of the activation rate of each path specifically includes: running a benchmark test set to simulate the circuit workload and generating a waveform file that records the endpoint flipping situation; based on the waveform file, statistically analyzing the signal flipping situation of the monitoring points of each path in the critical path set during the simulation process; if the monitoring point signal does not flip, the activation rate of the path is determined to be zero.
[0015] Preferably, the calculation of path similarity based on the logical topology sequence of intra-group paths specifically involves: modeling the path as a valid sequence of logical unit instances traversed, and calculating the sequence similarity ratio (SSR) between two paths based on the longest common prefix, expressed as:
[0016]
[0017] in, This indicates the length of the prefix sequence matched starting from the path's origin. and These are two timing paths being compared; and Representing paths and The total logical depth.
[0018] Preferably, the step of clustering paths based on the similarity specifically includes: sorting the paths within a path group according to their temporal margin, taking the path with the smallest temporal margin in the current group as the clustering benchmark path, calculating the sequence similarity ratio between the remaining paths in the group and the clustering benchmark path, and classifying paths with a sequence similarity ratio exceeding a preset threshold into the same cluster as the clustering benchmark path.
[0019] The step of selecting at least one path from each cluster specifically includes: dynamically determining the number of paths to be selected for each cluster k based on the number of paths contained within the cluster, and selecting the top k paths with the smallest time margin within the cluster to add to the set of monitoring points.
[0020] After filtering the paths within a cluster, all paths of that cluster are removed from the path group, and the path with the smallest time margin among the remaining paths is used as the new clustering baseline path. The clustering and filtering operations are repeated until all paths within the path group have been processed.
[0021] The critical path monitoring point screening system combining path activation rate and similarity provided by the present invention includes:
[0022] Module M1: Based on static timing analysis (STA), it obtains timing information of all paths in the target circuit, and filters out paths with timing margins less than the preset timing margin threshold to form a critical path set.
[0023] Module M2: Based on a preset benchmark test set, perform signal reversal statistics on the monitoring points of each path in the critical path set, calculate the activation rate of each path, and filter out paths with an activation rate of zero to obtain a set of effective paths.
[0024] Module M3: Groups the paths in the set of valid paths according to their starting points to obtain multiple path groups;
[0025] Module M4: For each path group, calculate the similarity between paths based on the logical topological sequence of paths within the group, and cluster the paths according to the similarity, selecting at least one path from each cluster.
[0026] Module M5: Summarizes all paths obtained after filtering from all path groups to form the final set of monitoring points.
[0027] Preferably, before obtaining the timing information of all paths of the target circuit based on static timing analysis (STA), the method includes: performing full-path timing analysis on the design netlist of the target circuit and removing duplicate endpoints to obtain an initial path set without duplicate endpoints.
[0028] Preferably, the calculation of the activation rate of each path specifically includes: running a benchmark test set to simulate the circuit workload and generating a waveform file that records the endpoint flipping situation; based on the waveform file, statistically analyzing the signal flipping situation of the monitoring points of each path in the critical path set during the simulation process; if the monitoring point signal does not flip, the activation rate of the path is determined to be zero.
[0029] Preferably, the calculation of path similarity based on the logical topology sequence of intra-group paths specifically involves: modeling the path as a valid sequence of logical unit instances traversed, and calculating the sequence similarity ratio (SSR) between two paths based on the longest common prefix, expressed as:
[0030]
[0031] in, This indicates the length of the prefix sequence matched starting from the path's origin. and These are two timing paths being compared; and Representing paths and The total logical depth.
[0032] Preferably, the step of clustering paths based on the similarity specifically includes: sorting the paths within a path group according to their temporal margin, taking the path with the smallest temporal margin in the current group as the clustering benchmark path, calculating the sequence similarity ratio between the remaining paths in the group and the clustering benchmark path, and classifying paths with a sequence similarity ratio exceeding a preset threshold into the same cluster as the clustering benchmark path.
[0033] The step of selecting at least one path from each cluster specifically includes: dynamically determining the number of paths to be selected for each cluster k based on the number of paths contained within the cluster, and selecting the top k paths with the smallest time margin within the cluster to add to the set of monitoring points.
[0034] After filtering the paths within a cluster, all paths of that cluster are removed from the path group, and the path with the smallest time margin among the remaining paths is used as the new clustering baseline path. The clustering and filtering operations are repeated until all paths within the path group have been processed.
[0035] Compared with the prior art, the present invention has the following beneficial effects:
[0036] (1) This invention uses path topology similarity clustering to remove highly similar redundant paths, which reduces the number of monitoring points and lowers chip area and power consumption compared to existing technologies.
[0037] (2) This invention combines multi-dimensional screening of “time margin priority + activation rate effectiveness + similarity clustering” to ensure that the selected monitoring points cover key time paths and avoid invalid monitoring, thereby improving the accuracy of time anomaly detection. Attached Figure Description
[0038] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:
[0039] Figure 1 This is a diagram of the algorithm framework. Detailed Implementation
[0040] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.
[0041] Example
[0042] This invention aims to solve the problems of path redundancy, low monitoring efficiency, and unreasonable resource allocation in existing critical path monitoring point screening technologies, and provides a multi-dimensional screening method that combines static time series analysis, path activation rate, and path structure similarity.
[0043] Building upon initial critical path screening based on STA time-series analysis, this paper innovatively integrates a dual screening strategy combining path activation rate and topological similarity. This constructs a multi-dimensional optimization system of "time-series priority - effective activation - topological clustering" to achieve precise screening of critical path monitoring points. The algorithm framework is as follows: Figure 1 As shown.
[0044] The specific technical solution is as follows:
[0045] 1) First, mature STA (Signal-Ahead Analyzer) tools are used to perform full-path timing analysis on the target circuit's design netlist, obtaining timing information for all paths within the chip. During this process, duplicate endpoint insertion points have been removed using existing technology, resulting in an initial path set P_all without duplicate endpoints. A minimum timing margin threshold is set based on system requirements; if the margin is less than this threshold, timing errors are likely to occur. First, paths that have a decisive impact on chip timing performance—that is, paths with a margin less than the set margin threshold—are precisely selected, forming the critical path set P_timing. This step completes the first round of selection based on timing priority, using the timing implementation reported by STA.
[0046] 2) To remove paths with no practical monitoring value, a path activation rate metric is introduced. A benchmark test set is run to simulate the chip's real workload. The activation rate (Act(p)) of each monitoring point in the critical path set P_timing is calculated. The benchmark test set is run again, and waveform files (.vcd files) are obtained through VCS simulation. The .vcd file records the flipping status and timing information of each endpoint, indicating when the flipping occurs. Based on the .vcd file, it can be determined whether the signal at each path monitoring point flipped during the simulation. If it remains stationary (constantly 0 or 1), the activation rate is 0. Paths with an activation rate of 0 are filtered out, resulting in the optimized effective path set P_active.
[0047] 3) Group the set of valid paths P_active according to their startpoints, grouping paths with the same startpoint into a set {G_1, G_2, ..., G_m}. Paths within the same group have similar temporal origins and logical transmission starting points, which ensures the accuracy of subsequent similarity calculations, reduces invalid calculations across groups, and improves the algorithm's execution efficiency.
[0048] 4) For each path group G_i (i=1,2,…,m), each path is modeled as a valid sequence of logical unit instances, and then the Sequence Similarity Ratio (SSR) is defined based on the Longest Common Prefix (LCP). The formula for calculating SSR is:
[0049]
[0050] Here, LCP represents the length of the prefix sequence matched starting from the path's origin, and L represents the total logical depth. When the SSR exceeds a set conservative threshold, these paths are classified as topologically redundant paths. and These are two time-series paths for comparison. The paths within a group are sorted by time margin. The path with the smallest Slack within the current group is used as the clustering benchmark p_seed. The overlap between the remaining paths in the group and the logical units traversed by p_seed is calculated. Paths with high similarity (higher SSR indicates higher similarity, i.e., higher overlap) are grouped into the same cluster C. The number of paths to be filtered, k, is dynamically determined based on the size of cluster C. The top k paths with the smallest Slack within each cluster are added to P_monitor. After removing the cluster, the above process is repeated until all paths within the group have completed clustering and filtering.
[0051] 5) After traversing all path groups and completing clustering, output the final monitoring point set P_monitor. This set is the optimized critical path monitoring point list, which can be directly used to guide the insertion and deployment of in-situ monitors at the end of the chip's critical path.
[0052] This invention also provides a key path monitoring point screening system that combines path activation rate and similarity, comprising:
[0053] Module M1: Based on static timing analysis (STA), it obtains timing information of all paths in the target circuit, and filters out paths with timing margins less than the preset timing margin threshold to form a critical path set.
[0054] Module M2: Based on a preset benchmark test set, perform signal reversal statistics on the monitoring points of each path in the critical path set, calculate the activation rate of each path, and filter out paths with an activation rate of zero to obtain a set of effective paths.
[0055] Module M3: Groups the paths in the set of valid paths according to their starting points to obtain multiple path groups;
[0056] Module M4: For each path group, calculate the similarity between paths based on the logical topological sequence of paths within the group, and cluster the paths according to the similarity, selecting at least one path from each cluster.
[0057] Module M5: Summarizes all paths obtained after filtering from all path groups to form the final set of monitoring points.
[0058] Before obtaining the timing information of all paths of the target circuit based on static timing analysis (STA), the process includes: performing full-path timing analysis on the design netlist of the target circuit and removing duplicate endpoints to obtain an initial path set without duplicate endpoints.
[0059] The calculation of the activation rate of each path specifically includes: running a benchmark test set to simulate the circuit workload and generating a waveform file that records the endpoint flipping situation; based on the waveform file, statistically analyzing the signal flipping situation of the monitoring points of each path in the critical path set during the simulation process; if the monitoring point signal does not flip, the activation rate of the path is determined to be zero.
[0060] The calculation of path similarity based on the logical topology sequence of intra-group paths specifically involves: modeling the path as a valid sequence of logical unit instances traversed, and calculating the sequence similarity ratio (SSR) between two paths based on the longest common prefix, expressed as:
[0061]
[0062] in, This indicates the length of the prefix sequence matched starting from the path's origin. and These are two timing paths being compared; and Representing paths and The total logical depth.
[0063] The step of clustering paths based on the similarity specifically includes: sorting the paths within a path group according to their temporal margin, taking the path with the smallest temporal margin in the current group as the clustering benchmark path, calculating the sequence similarity ratio between the remaining paths in the group and the clustering benchmark path, and classifying paths with a sequence similarity ratio exceeding a preset threshold into the same cluster as the clustering benchmark path.
[0064] The step of selecting at least one path from each cluster specifically includes: dynamically determining the number of paths to be selected for each cluster k based on the number of paths contained within the cluster, and selecting the top k paths with the smallest time margin within the cluster to add to the set of monitoring points.
[0065] After filtering the paths within a cluster, all paths of that cluster are removed from the path group, and the path with the smallest time margin among the remaining paths is used as the new clustering baseline path. The clustering and filtering operations are repeated until all paths within the path group have been processed.
[0066] Those skilled in the art will understand that, in addition to implementing the system, apparatus, and their modules provided by this invention in purely computer-readable program code, the same program can be implemented in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, the system, apparatus, and their modules provided by this invention can be considered a hardware component, and the modules included therein for implementing various programs can also be considered structures within the hardware component; alternatively, modules for implementing various functions can be considered both software programs implementing the method and structures within the hardware component.
[0067] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.
Claims
1. A method for screening key path monitoring points by combining path activation rate and similarity, characterized in that, include: Step S1: Obtain timing information of all paths of the target circuit based on static timing analysis (STA), and filter out paths with timing margin less than the preset timing margin threshold according to the preset timing margin threshold to form a critical path set. Step S2: Based on a preset benchmark test set, perform signal reversal statistics on the monitoring points of each path in the critical path set, calculate the activation rate of each path, and filter out paths with an activation rate of zero to obtain an effective path set. Step S3: Group the paths in the set of valid paths according to their starting points to obtain multiple path groups; Step S4: For each path group, calculate the similarity between paths based on the logical topological structure sequence of paths within the group, and cluster the paths according to the similarity, selecting at least one path from each cluster. Step S5: Summarize all the paths obtained after filtering from all path groups to form the final set of monitoring points; The calculation of path similarity based on the logical topology sequence of intra-group paths specifically involves: modeling the path as a valid sequence of logical unit instances traversed, and calculating the sequence similarity ratio (SSR) between two paths based on the longest common prefix, expressed as: in, This indicates the length of the prefix sequence matched starting from the path's origin. and These are two timing paths being compared; and Representing paths and The total logical depth.
2. The key path monitoring point screening method combining path activation rate and similarity according to claim 1, characterized in that, Before obtaining the timing information of all paths of the target circuit based on static timing analysis (STA), the process includes: performing full-path timing analysis on the design netlist of the target circuit and removing duplicate endpoints to obtain an initial path set without duplicate endpoints.
3. The key path monitoring point screening method combining path activation rate and similarity according to claim 1, characterized in that, The calculation of the activation rate of each path specifically includes: running a benchmark test set to simulate the circuit workload and generating a waveform file that records the endpoint flipping situation; based on the waveform file, statistically analyzing the signal flipping situation of the monitoring points of each path in the critical path set during the simulation process; if the monitoring point signal does not flip, the activation rate of the path is determined to be zero.
4. The critical path monitoring point screening method combining path activation rate and similarity according to claim 1, characterized in that, The step of clustering paths based on the similarity specifically includes: sorting the paths within a path group according to their temporal margin, taking the path with the smallest temporal margin in the current group as the clustering benchmark path, calculating the sequence similarity ratio between the remaining paths in the group and the clustering benchmark path, and classifying paths with a sequence similarity ratio exceeding a preset threshold into the same cluster as the clustering benchmark path. The step of selecting at least one path from each cluster specifically includes: dynamically determining the number of paths to be selected for each cluster k based on the number of paths contained within the cluster, and selecting the top k paths with the smallest time margin within the cluster to add to the set of monitoring points. After filtering the paths within a cluster, all paths of that cluster are removed from the path group, and the path with the smallest time margin among the remaining paths is used as the new clustering baseline path. The clustering and filtering operations are repeated until all paths within the path group have been processed.
5. A critical path monitoring point screening system combining path activation rate and similarity, characterized in that, include: Module M1: Based on static timing analysis (STA), it obtains timing information of all paths in the target circuit, and filters out paths with timing margins less than the preset timing margin threshold to form a critical path set. Module M2: Based on a preset benchmark test set, perform signal reversal statistics on the monitoring points of each path in the critical path set, calculate the activation rate of each path, and filter out paths with an activation rate of zero to obtain a set of effective paths. Module M3: Groups the paths in the set of valid paths according to their starting points to obtain multiple path groups; Module M4: For each path group, calculate the similarity between paths based on the logical topological sequence of paths within the group, and cluster the paths according to the similarity, selecting at least one path from each cluster. Module M5: Summarizes all paths obtained after filtering from all path groups to form the final set of monitoring points; The calculation of path similarity based on the logical topology sequence of intra-group paths specifically involves: modeling the path as a valid sequence of logical unit instances traversed, and calculating the sequence similarity ratio (SSR) between two paths based on the longest common prefix, expressed as: in, This indicates the length of the prefix sequence matched starting from the path's origin. and These are two timing paths being compared; and Representing paths and The total logical depth.
6. The critical path monitoring point screening system combining path activation rate and similarity according to claim 5, characterized in that, Before obtaining the timing information of all paths of the target circuit based on static timing analysis (STA), the process includes: performing full-path timing analysis on the design netlist of the target circuit and removing duplicate endpoints to obtain an initial path set without duplicate endpoints.
7. The critical path monitoring point screening system combining path activation rate and similarity according to claim 5, characterized in that, The calculation of the activation rate of each path specifically includes: running a benchmark test set to simulate the circuit workload and generating a waveform file that records the endpoint flipping situation; based on the waveform file, statistically analyzing the signal flipping situation of the monitoring points of each path in the critical path set during the simulation process; if the monitoring point signal does not flip, the activation rate of the path is determined to be zero.
8. The critical path monitoring point screening system combining path activation rate and similarity according to claim 5, characterized in that, The step of clustering paths based on the similarity specifically includes: sorting the paths within a path group according to their temporal margin, taking the path with the smallest temporal margin in the current group as the clustering benchmark path, calculating the sequence similarity ratio between the remaining paths in the group and the clustering benchmark path, and classifying paths with a sequence similarity ratio exceeding a preset threshold into the same cluster as the clustering benchmark path. The step of selecting at least one path from each cluster specifically includes: dynamically determining the number of paths to be selected for each cluster k based on the number of paths contained within the cluster, and selecting the top k paths with the smallest time margin within the cluster to add to the set of monitoring points. After filtering the paths within a cluster, all paths of that cluster are removed from the path group, and the path with the smallest time margin among the remaining paths is used as the new clustering baseline path. The clustering and filtering operations are repeated until all paths within the path group have been processed.