A control system, method, and computing device for running semiconductor EDA tool jobs in coordination with an LSF platform

By constructing a semiconductor EDA tool job execution control system on the LSF platform, the problem of low resource utilization of existing schedulers has been solved, achieving efficient and automated resource management and task execution, and improving computing resource utilization and job efficiency.

CN123114731APending Publication Date: 2026-11-17HANGZHOU HFC SEMICONDUCTOR CO
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611223043.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-13
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

The existing IBM Spectrum LSF job scheduler suffers from manual bottlenecks in semiconductor EDA tool jobs, failing to achieve unified semantic abstraction of jobs and joint optimization of multi-dimensional resources, resulting in low resource utilization and low job scheduling efficiency.

Method used

A control system for semiconductor EDA tools operating in conjunction with the LSF platform is provided, including a submission and receiving module, a data analysis and processing module, a multi-level task scheduling and buffer processing module, a task execution and distribution module, and an execution monitoring module. Through resource requirement profiling, task decomposition, and multi-level priority queue management, intelligent resource allocation and automated task execution are achieved.

Benefits of technology

It improved the utilization rate of computing resources from 20%~40% to 60%~80%, and realized end-to-end unattended fully automated execution, reducing the need for manual intervention and optimizing the fairness of job turnaround time and resource allocation.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123114731A_ABST
    Figure CN123114731A_ABST
Patent Text Reader

Abstract

This invention provides a control system, method, and computing device for running semiconductor EDA tool jobs in conjunction with the LSF platform. It includes a submission and receiving module for receiving all user-submitted EDA tool jobs; a data analysis and processing module for extracting relationship graphs, breaking down EDA tool jobs, and generating a task list; a multi-level task scheduling and buffering module for real-time task priority determination and resource allocation; a task execution and transmission module for receiving tasks and sending them to the LSF platform; and an execution monitoring module for receiving real-time task completion status and task log data, and feeding back the data to the data analysis and processing module. This invention achieves precise matching and layout of computing resources, significantly improving the average CPU utilization of the cluster and effectively solving the resource fragmentation problem. Simultaneously, through multi-level buffer queues and dynamic priority calculation, it effectively isolates tasks of different priorities and optimizes the average job turnaround time.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application belongs to the field of semiconductor workload technology, specifically relating to a control system, method, and computing device for operating semiconductor EDA tools in conjunction with the LSF platform. Background Technology

[0002] In the semiconductor industry, every stage from R&D to mass production relies heavily on massive central processing unit (CPU) computing power, and the industry generally uses IBM Spectrum LSF job scheduler as the cluster scheduler. However, the manual bottleneck of this cluster scheduler is very obvious. Engineers need to submit jobs one by one and monitor each group. Once a task is abnormally suspended due to CPU resource or license preemption, the entire operation will come to a standstill.

[0003] In the current development of advanced nodes, the operation logic of multiple electronic design automation platforms is inconsistent. The existing IBM Spectrum LSF job scheduler framework lacks the ability to unify the abstraction of job semantics and jointly optimize multi-dimensional resources, and cannot map heterogeneous loads to the same scheduling plane, resulting in unavoidable waiting overhead in the project cycle. Summary of the Invention

[0004] This application is made to solve the above problems, and aims to provide a control system, method and computing device for running semiconductor EDA tools in conjunction with the LSF platform, so as to maximize the utilization of computing resources and improve the efficiency of semiconductor EDA tool operation.

[0005] This application provides a control system for running semiconductor EDA tool jobs in conjunction with the LSF platform, characterized by: a user EDA tool job submission and receiving module for receiving all user-submitted EDA tool jobs; a data analysis and processing module for identifying and classifying EDA tool jobs, determining resource requirements, extracting relationship graphs between EDA tool jobs, and decomposing EDA tool jobs into a task list based on the relationship graphs; a multi-level task scheduling and buffering module for real-time determination of task priorities in the task list and resource allocation for tasks about to be executed; and task execution... The task execution and distribution transmission module receives tasks from the task multi-level scheduling and buffer processing module and sends them to the LSF platform; the execution monitoring module receives data, including real-time task completion status and task logs from the LSF platform, through the task execution and distribution transmission module, and feeds the data back to the data analysis and processing module; when the submission receiving module in the control system receives a new EDA tool job submitted by the user, the data analysis and processing module forms a new task list and adds it to the task multi-level scheduling and buffer processing module, prioritizes the tasks, and sends them to the LSF platform through the task execution and distribution transmission module.

[0006] In one embodiment of this application, the submission and receiving module includes: a front-end interface unit, used to provide information input for EDA tool jobs, upload associated files, and query and reuse historical job information; a back-end access service unit, used to perform unified authentication of user identity, perform security verification on input content, generate a globally unique identifier for EDA tool jobs that pass verification, and record the description information and data storage location of EDA tool jobs; and a storage interface unit, used to store associated files in a temporary area of ​​the shared storage system.

[0007] In one embodiment of this application, the data analysis and processing module includes: a modular parsing engine, used to parse the description information of EDA tool jobs and determine their corresponding processing logic; a resource requirement profiling unit, used to extract the feature information of EDA tool jobs and predict the amount of resources required by EDA tool jobs based on historical data; and a dependency extraction unit, used to obtain the execution order constraint information within EDA tool jobs and generate a relationship graph.

[0008] In one embodiment of this application, the EDA tool job is decomposed into a task list based on a relationship diagram, including: decomposing the EDA tool job into multiple sub-tasks according to the relationship diagram and preset splitting rules, and assigning an independent identifier, execution instructions and runtime environment parameters to each sub-task; generating a task list according to the sub-tasks, the relationship diagram between the sub-tasks and the priority of the tasks, and sending the task list to the task multi-level scheduling and buffer processing module.

[0009] In one embodiment of this application, the task multi-level scheduling and buffer processing module includes a primary waiting queue and a secondary priority sub-queue group, wherein the secondary priority sub-queue group includes multiple secondary priority sub-queues with different priorities.

[0010] In one embodiment of this application, the task multi-level scheduling and buffer processing module performs real-time determination of task priorities in the task list, including: obtaining tasks from the first-level waiting queue through a priority calculation service, and determining the scheduling priority order of tasks based on the task's attribute parameters; the attribute parameters include: the priority of the project associated with the task, the reciprocal of the estimated execution time of the task, and the waiting time of the task in the queue; after the calculation is completed, the tasks are delivered to the corresponding second-level priority sub-queues according to the scheduling priority order.

[0011] In one embodiment of this application, the task multi-level scheduling and buffer processing module further includes a dependency checker. The dependency checker is used to poll the tasks in the secondary priority sub-queue to determine whether the preconditions of the task are met. If the preconditions are met, the task status is set to the dispatchable status. If the preconditions are not met, the task is kept in the original queue for retry later.

[0012] In one embodiment of this application, the execution monitoring module specifically includes: a log collection unit, used to collect the running records generated during the execution of the task; an event processing unit, used to acquire and distribute status change events of the task and EDA tool job; and a visualization unit, used to provide a dynamic display of the EDA tool job status based on the status change events, and to feed back the status change events to the data analysis and processing module.

[0013] This application also provides a control method for running semiconductor EDA tool jobs in conjunction with the LSF platform, characterized by the following steps: a user EDA tool job submission and receiving module receives all user-submitted EDA tool jobs; identifies and categorizes the EDA tool jobs, determines resource requirements, and extracts a relationship graph between the EDA tool jobs, and decomposes the EDA tool jobs according to the relationship graph to generate a task list; determines the priority of tasks in the task list in real time and allocates resources for tasks about to be executed; receives tasks issued by the task multi-level scheduling and buffer processing module and sends them to the LSF platform; receives data containing at least the real-time task completion status and task logs fed back by the LSF platform through the task execution and transmission module, and feeds the data back to the data analysis and processing module; when the submission and receiving module in the control system receives a new user-submitted EDA tool job, the data analysis and processing module forms a new task list and adds it to the task multi-level scheduling and buffer processing module, arranges the task priorities and waits, and sends it to the LSF platform through the task execution and transmission module.

[0014] This application also provides a computing device having the features of: a processor, a computer-readable storage medium, and a communication bus, wherein the computer-readable storage medium stores program instructions, and when the program instructions are executed by the processor, they are capable of implementing the control method for operating a semiconductor EDA tool in conjunction with the LSF platform as described in any one of claims 9.

[0015] The role and effect of the invention: According to the control system, method, and computing device for running semiconductor EDA tools in conjunction with the LSF platform involved in this application, the computing, storage, and licensing requirements of EDA tool operations are quantitatively modeled by the resource requirement profiling unit. Combined with the multi-dimensional intelligent allocation algorithm in the task execution and distribution transmission module, the matching and layout of computing resources can be realized, thereby increasing the average CPU utilization of the cluster from the traditional 20%~40% to 60%~80%, effectively solving the problem of resource fragmentation.

[0016] This application achieves end-to-end unattended fully automated execution by combining automated dependency resolution and task decomposition with closed-loop feedback from dependency checkers and execution monitoring modules, significantly reducing the need for manual intervention. At the same time, through multi-level buffer queues and dynamic priority calculation based on project priority, estimated time countdown, and waiting time, it effectively isolates tasks of different priorities, balances high-priority low-latency response with scheduling fairness, and optimizes the average turnaround time of jobs. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the control system for the operation of semiconductor EDA tools in an embodiment of this application.

[0019] Figure 2 This is a flowchart of a control method for the operation of a semiconductor EDA tool in an embodiment of this application.

[0020] Explanation of reference numerals in the attached figures: 100-Control system, 101-Submission and receiving module, 102-Data analysis and processing module, 103-Task multi-level scheduling and buffer processing module, 104-Task execution and distribution module, 105-Execution monitoring module. Detailed Implementation

[0021] The technical solutions disclosed in this application will be described in detail below with reference to specific embodiments.

[0022] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0023] It should be noted that the illustrations provided in this embodiment are only schematic representations of the basic concept of this application. Therefore, the drawings only show the components related to this application and are not drawn according to the number, shape and size of the components in actual implementation. In actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0024] In this application, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application. Furthermore, the terms "first" and "second" are used only for descriptive and distinguishing purposes and should not be construed as indicating or implying relative importance.

[0025] To make the technical means, creative features, objectives and effects of this application easy to understand, the following embodiments are described in detail with reference to the accompanying drawings.

[0026] This application provides a control system and method for running semiconductor EDA tool jobs in conjunction with the LSF platform. It aims to address the shortcomings of existing cluster schedulers, such as the IBM Spectrum LSF job scheduler, which suffer from low scheduling efficiency, rigid resource allocation, and a lack of a unified scheduling plane in handling EDA (Electronic Design Automation) tools, especially when dealing with similar or overlapping EDA tools from different vendors. The proposed control system and method for running semiconductor EDA tool jobs in conjunction with the LSF platform achieves automated, scalable, and efficient processing of heterogeneous electronic design automation toolchain jobs by constructing a closed-loop intelligent collaborative system encompassing job submission, parsing, decomposition, scheduling, and execution monitoring.

[0027] like Figure 1 As shown, the control system 100 for semiconductor EDA tool operations in conjunction with the LSF platform in this application includes a user EDA tool job submission and receiving module 101, a data analysis and processing module 102, a task multi-level scheduling and buffer processing module 103, a task execution and distribution module 104, and an execution monitoring module 105. When the submission and receiving module 101 receives a new EDA tool job submitted by the user, the data analysis and processing module 102 forms a new task list and adds it to the task multi-level scheduling and buffer processing module 103, prioritizing the tasks and waiting for them to be sent to the LSF platform (Load Sharing Facility Platform) through the task execution and distribution module 104. It is worth noting that this application does not limit the specific deployment method and number of each module in the above-mentioned control system 100. As an example, the task multi-level scheduling and buffer processing module 103 can be deployed as an independent microservice, or it can be integrated with the task execution and distribution module 104 in the same service. The number of each module can be one or more, and their functions can be the same or different, depending on whether the control functions of this application can be realized. Figure 2As shown, based on the aforementioned control system 100, the control method for operating semiconductor EDA tools in conjunction with the LSF platform in this application specifically includes the following steps: S1: The user EDA tool job submission receiving module 101 receives all user-submitted EDA tool jobs. Specifically, S1 further includes the following sub-steps S1-1 to S1-3: S1-1: The front-end interface unit provides the input of job information, the uploading of associated files, and the querying and reuse of historical job information. Users can fill in job-related information and upload necessary design files through a highly real-time and responsive front-end interface.

[0028] S1-2: The backend access service unit performs unified authentication of user identities, executes security checks on input content, generates globally unique identifiers for EDA tool jobs that pass verification, and records the description information and data storage location of EDA tool jobs. Specifically, the backend access service unit integrates with an enterprise-level single sign-on system to perform input verification on user identities and submitted job data to prevent malicious code injection. After successful verification, a globally unique job identifier and chip identifier are generated for each successfully submitted job, and the job's metadata and file storage path are recorded in the database. The job's metadata includes at least the submitting user, submission time, and project name. As one implementation method, the database can be, for example, but not limited to, a relational database or a non-relational database, specifically designed for efficient storage and retrieval of job metadata.

[0029] S1-3: The storage interface unit stores the associated files in a temporary area of ​​the shared storage system. It is worth noting that this application does not limit the specific type and size of the distributed file system; it can be one or more storage nodes with the same or different functions, specifically designed to provide high-throughput, high-reliability temporary file storage and sharing capabilities. The job files include at least scripts, configuration files, and input data.

[0030] S2: The data analysis and processing module 102 identifies and categorizes EDA tool jobs, determines resource requirements, and extracts relationship diagrams between EDA tool jobs. Based on the relationship diagrams, it breaks down the EDA tool jobs to generate a task list. Specifically, S2 further includes sub-steps S2-1 to S2-4: S2-1: The modular parsing engine parses the description information of EDA tool jobs to determine their corresponding processing logic. The modular parsing engine uses a syntax analyzer to parse user-submitted scripts and configuration files, identifies the electronic design automation (EDA) tools called within them, and executes the specific logic corresponding to that EDA tool. For example, an EDA tool could be a simulation tool, a verification tool, an optical proximity correction tool, etc. This achieves a unified abstraction of the job logic for different EDA tool platforms.

[0031] S2-2: The resource requirement profiling unit extracts feature information of EDA tool jobs and predicts the resource requirements of EDA tool jobs based on historical data. Specifically, the job features are vectorized, a historical database is constructed, and a regression prediction model is used to predict the CPU resource value required for the next job with similar features. The job features can include project name, product name, number of layers, chip identifier, data density, file size, etc. For example, the resource requirement profile of an optical proximity correction job can include: the estimated number of CPU cores required, memory size, and estimated runtime. This application does not limit the specific type of regression prediction model; as an example, it can be linear regression, decision tree regression, or a more complex gradient boosting regression tree model.

[0032] S2-3: The dependency extraction unit obtains the execution order constraint information within the EDA tool job and generates a relationship graph. Specifically, it extracts explicit declarations from the job's configuration parameters, analyzes the dependencies between various workflows of each chip within the job, and constructs a sequence of inter-task dependencies. For example, in the process of developing optical proximity correction, an optical proximity correction job may contain multiple steps, such as model calibration, rule checking, and optical proximity correction execution, with a clear sequential order among these steps. The dependency extraction unit is responsible for parsing these relationships and forming a directed acyclic graph structure to describe the dependencies between tasks.

[0033] S2-4: Based on the relationship diagram and the preset splitting rules, the EDA tool job is decomposed into multiple sub-tasks, and each sub-task is assigned an independent identifier, execution instructions and runtime environment parameters; according to the sub-tasks, the relationship diagram between the sub-tasks and the priority of the tasks, a task list is generated and the task list is sent to the task multi-level scheduling and buffer processing module 103.

[0034] S3: The multi-level task scheduling and buffer processing module 103 determines the priority of tasks in the task list in real time and allocates resources for tasks about to be executed. Specifically, S3 further includes sub-steps S3-1 to S3-4: S3-1: The multi-level task scheduling and buffering module 103 includes a primary waiting queue and a secondary priority sub-queue group. The secondary priority sub-queue group includes multiple secondary priority sub-queues with different priorities. All subtasks in the task list generated from S2 are first sent to the primary waiting queue.

[0035] S3-2: Tasks are retrieved from the primary waiting queue via a priority calculation service. The scheduling priority of tasks is determined based on their attribute parameters, including: the project priority associated with the task, the reciprocal of the estimated execution time of the task, and the waiting time already elapsed in the queue. After calculation, tasks are delivered to the corresponding secondary priority sub-queues according to their scheduling priority. The project priority is typically preset by the user or administrator based on the importance of different business processes. The estimated task duration prioritizes shorter jobs to improve system response speed. Furthermore, the elapsed waiting time is primarily used to prevent low-priority tasks from being unable to obtain execution opportunities for extended periods.

[0036] S3-3: The multi-level task scheduling and buffer processing module 103 also includes a dependency checker. The dependency checker is used to poll the tasks in the secondary priority sub-queue to determine whether the preconditions of the task are met. If the preconditions are met, the task status is set to the dispatchable status. If the preconditions are not met, the task is kept in the original queue for retry later.

[0037] S3-4: The task execution and distribution module 104 receives tasks in the distribution status from the task multi-level scheduling and buffer processing module 103. Based on the resource requirement profile generated by the data analysis and processing module 102, it allocates computing resources to the tasks and sends the tasks carrying resource allocation information to the LSF platform. During resource allocation, the task execution and distribution module 104 first filters out all available computing nodes in the cluster that meet the hard requirements based on the resource requirement profile. These hard requirements can include CPU architecture, memory size, specific electronic design automation tool licenses, etc. Secondly, among the nodes that meet the hard requirements, it prioritizes computing nodes with lower current loads to ensure that tasks can quickly obtain resources and begin execution, avoiding hotspot nodes. Furthermore, based on historical data, using indicators such as project, product, layer, chip, density, file size, historical runtime, and average CPU utilization, it uses a regression prediction model to predict the job and allocate suitable computing resources as the CPU value. The prediction model can be a multiple linear regression, support vector regression, or neural network regression model, aiming to predict the optimal number of CPU cores based on historical data characteristics. This avoids excessive redundancy or insufficient resources caused by users manually submitting resources, thus achieving refined resource supply.

[0038] S4: The execution monitoring module 105 receives data, including at least the real-time task completion status and task logs fed back by the LSF platform, through the task execution and distribution transmission module 104, and feeds the data back to the data analysis and processing module 102. Specifically, S4 further includes sub-steps S4-1 to S4-3: S4-1: The log collection unit collects the running records generated during the execution of the task, as well as the standard output and exception output during the task execution process. These records are collected uniformly by the log collection unit and sent to the centralized log system module.

[0039] S4-2: The event processing unit acquires and distributes status change events for tasks and EDA tool jobs. Lifecycle status change events for jobs and tasks flow into a message queue for consumption by downstream monitoring and status update services. For example, lifecycle status change events for jobs and tasks can be pending submission, submitted, in queue, executing, rolled back, successful execution, or failed.

[0040] S4-3: The visualization unit provides a dynamic display of the EDA tool job status based on status change events and feeds these events back to the data analysis and processing module 102. The visualization interface generates an independent lifecycle view for each job, displaying various job statuses in real time. Specifically, the visualization interface displays at least the following statuses: pending submission, submitted, in queue, executing, rolled back, successful execution, and failed execution. This provides users with full-stack monitoring, from global cluster resource heatmaps to the lifecycle of individual tasks, greatly improving system transparency and maintainability.

[0041] S5: When a task is completed, the real-time task completion status reported by the LSF platform is received by the execution monitoring module 105 through the task execution and distribution module 104. The event processing unit obtains the task completion status change event and reports it to the data analysis and processing module 102. The data analysis and processing module 102 updates the task status accordingly. Simultaneously, the dependency checker in the task multi-level scheduling and buffer processing module 103 continuously polls, quickly retrieving all tasks in the waiting state and checking whether the prerequisites of each task are satisfied due to this status update. For tasks with newly satisfied dependencies, their status is set to the deployable state, and then the task execution and distribution module 104 distributes them to the LSF platform, realizing automated flow between subtasks and end-to-end unattended full-process execution.

[0042] When the submission receiving module 101 in the control system 100 receives a new EDA tool job submitted by the user, it forms a new task list through the data analysis and processing module 102 and adds it to the task multi-level scheduling and buffer processing module 103 to prioritize the tasks. The tasks are then sent to the LSF platform through the task execution and distribution transmission module 104, thus forming a complete closed loop.

[0043] The technical solution of this application constructs a complete closed-loop system capable of achieving a high degree of automation and intelligence. The following will describe in detail the complete operation process from job submission to result acceptance, using a specific example, to further illustrate the application of the technical solution in a real-world environment.

[0044] During the job submission phase, user actions include: Login Platform: Access the Optical Proximity Correction Computing Resource Management System through the unified portal and complete the authentication based on the enterprise-level single sign-on system; Create a new job: Users click the "Add" button and fill in basic job information, including project name, design version, description, etc. Upload design files: Upload the runtime scripts such as runme and related design files to the temporary area specified in the distributed file system through the front-end interface unit; Configure operating parameters: Set various operating parameters required for optical proximity correction operation; Confirm Submission: The submission receiving module 101 generates a unique job identifier and chip identifier, and the user receives a confirmation of successful submission.

[0045] At this stage, the system's feedback includes: real-time display of file upload progress and verification results; immediate return of job identifiers for subsequent tracking; and display of the job's queue position based on the current queue status.

[0046] During the status monitoring phase, user actions include: View all submitted jobs in your personal workbench; Quickly locate and view the current processing stage using job identifiers; Observe the changes in the position of the job in the first-level waiting queue and the second-level priority sub-queue of the task multi-level scheduling and buffer processing module 103; By using the visualization interface provided by the monitoring module 105, you can view the dependency graph between subtasks and understand the blocking relationships.

[0047] During this stage, the system feedback will clearly display the various statuses of the job and update the queue position and estimated waiting time in real time.

[0048] During the resource viewing phase, user actions include: View the computing resources allocated to the job, including the number of CPU cores and the amount of memory; The usage of allocated computing resources can be viewed in real time by executing the monitoring module 105; Confirm the status of the required electronic design automation tool license.

[0049] At this stage, system feedback includes: detailed display of estimated CPU, memory, and storage requirements for each subtask; display of actual resource consumption during task execution; and license allocation status and waiting queue status.

[0050] During the execution process tracking phase, user actions include: The execution progress of each subtask is monitored by the monitoring module 105; Click on a specific subtask to view the real-time runtime log output; Observe the execution efficiency of subtasks to identify potential performance bottlenecks.

[0051] During this stage, the system provides real-time streaming log output and supports keyword filtering search.

[0052] During the results acceptance phase, user actions include: Browse the list of output files after the job is completed; Check the work results documents and perform quality verification.

[0053] At this stage, the system will display all output files and reports in a structured manner and provide a comparative analysis of key metrics and expected targets.

[0054] During the problem-solving phase, user actions include: For failed subtasks, review their error logs and diagnostic information; Choose a retry strategy based on the error type in the error log, such as retrying in place or retrying after adjusting parameters; For tasks that fail due to insufficient resources, request additional computing resources; When necessary, the administrator can manually suspend the job or adjust the job execution order.

[0055] During this stage, the system records the retry history, calculates the success rate, and provides real-time impact analysis of resource adjustments.

[0056] During the historical data utilization phase, user operations include: Filter historical jobs by time, project, status, and other criteria; Compare the execution efficiency and quality results of different versions of the job; View resource usage trends and job execution patterns; Save the successful job configuration as a template for later use.

[0057] At this stage, the system will provide multi-dimensional historical data statistical analysis, as well as template library management and quick application functions.

[0058] This application also provides a computing device, including: a processor, a computer-readable storage medium and a communication bus, wherein the computer-readable storage medium stores program instructions, and when the program instructions are executed by the processor, they can realize the aforementioned control method for operating semiconductor EDA tools in conjunction with the LSF platform.

[0059] This application generates a resource requirement profile through the resource requirement profiling unit in the data analysis and processing module 102, quantitatively modeling the computation, storage, and licensing requirements of the job to achieve precise matching of computing resource allocation. Simultaneously, the task execution and distribution module 104, through resource filtering, load balancing, and a multi-dimensional intelligent CPU allocation algorithm, comprehensively considers node load, hardware topology, and data locality to achieve optimal layout of computing resources. Practical verification shows that after adopting this application, the average CPU utilization of the cluster can be increased from 20%~40% in the traditional mode to 60%~80%, effectively solving the resource fragmentation problem.

[0060] Existing complex electronic design automation (EDA) workflows suffer from task dependencies, leading to frequent workflow interruptions, prolonged iteration cycles, and the introduction of human error risks. This application addresses this by automating dependency resolution and task decomposition via a data analysis and processing module 102. A dependency checker in the multi-level task scheduling and buffering module 103 ensures that tasks are automatically set to a deployable state once their prerequisites are met, and are then sent to the LSF platform for execution by the task execution and deployment module 104. Simultaneously, the event processing unit of the execution monitoring module 105 captures task status change events and feeds them back to the data analysis and processing module 102, instantly triggering the dependency checker to inspect and prepare subsequent dependent tasks. This achieves end-to-end unattended, fully automated workflow execution, significantly reducing the need for manual intervention.

[0061] Traditional first-in-first-out (FIFO) queues cannot handle scenarios involving urgent tasks jumping the queue or heterogeneous job deployments, resulting in poor scheduling fairness and high latency for critical tasks. This application sets up a primary waiting queue and a secondary priority sub-queue group in the multi-level task scheduling and buffering module 103. It also uses an independent priority calculation service to calculate priorities based on attributes such as project priority, the reciprocal of the estimated task duration, and the already waited duration, achieving multi-dimensional intelligent decision-making. This architecture effectively isolates tasks of different priorities, ensuring system throughput while providing low-latency responses for high-priority tasks. It optimizes average job turnaround time and queuing time, ensuring that computing resources are prioritized for critical business objectives.

[0062] Existing cluster schedulers rely on command-line tools for native monitoring, lacking real-time graphical displays and resulting in low efficiency in operation, maintenance, and debugging. The execution monitoring module 105 in this application provides full-stack monitoring, from global cluster resource heatmaps to the lifecycle of individual tasks, offering data support for system optimization, capacity planning, and root cause analysis.

[0063] In existing technologies, resource allocation strategies are scattered among individuals, making it difficult to achieve unified optimization and scalable reuse of best practices. This application provides a unified job submission interface through a submission and receiving module 101, and enforces standardization of job submission and management in conjunction with a data analysis and processing module 102. The platform's continuously accumulated historical job data and resource prediction models constitute the enterprise's core digital assets. Users can save successful job configurations as templates for subsequent use, effectively saving labor costs.

[0064] In summary, this application can form a relatively mature system product. When this system product is used to manage the operation of several different semiconductor EDA tools used in optical proximity correction, it can be an optical proximity correction computing resource management system combined with the LSF platform. This can solve a series of problems such as the need for manual intervention in the scheduling of computing resources during the operation of various semiconductor EDA tools, resulting in low operating efficiency and long operation cycles for each job.

[0065] It will be apparent to those skilled in the art that this application is not limited to the details of the exemplary embodiments described above, and that this application can be implemented in other specific forms without departing from the spirit or essential characteristics of this application. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of this application is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within this application. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0066] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A control system for the operation of semiconductor EDA tools in conjunction with the LSF platform, characterized in that, include: The user EDA tool job submission and receiving module is used to receive all the EDA tool jobs submitted by users; The data analysis and processing module is used to identify and classify the EDA tool jobs, determine resource requirements, extract the relationship diagram between the EDA tool jobs, and decompose the EDA tool jobs according to the relationship diagram to generate a task list. The multi-level task scheduling and buffering module is used to determine the priority of tasks in the task list in real time and allocate resources for tasks to be executed. The task execution and distribution module is used to receive the task distributed by the task multi-level scheduling and buffer processing module and send it to the LSF platform; as well as The execution monitoring module is used to receive data containing at least the real-time task completion status and task logs fed back by the LSF platform through the task execution and transmission module, and to feed the data back to the data analysis and processing module; When the submission receiving module in the control system receives a new EDA tool job submitted by the user, the data analysis and processing module forms a new task list and adds it to the task multi-level scheduling and buffering processing module. The priority of the tasks is arranged and they are then sent to the LSF platform through the task execution and distribution transmission module.

2. The control system according to claim 1, characterized in that, The submission receiving module includes: The front-end interface unit is used to provide information input for the EDA tool's job, upload of associated files, and query and reuse of historical job information; The backend access service unit is used to perform unified authentication of user identity, perform security verification on input content, generate a globally unique identifier for the EDA tool job that passes the verification, and record the description information and data storage location of the EDA tool job. The storage interface unit is used to store the associated files in a temporary area of ​​the shared storage system.

3. The control system according to claim 1, characterized in that, The data analysis and processing module includes: A modular parsing engine is used to parse the description information of the EDA tool job and determine its corresponding processing logic; The resource requirement profiling unit is used to extract the feature information of the EDA tool job and predict the amount of resources required by the EDA tool job based on historical data. The dependency extraction unit is used to obtain the execution order constraint information within the EDA tool job and generate the relationship graph.

4. The control system according to claim 1, characterized in that, The step of breaking down the EDA tool job into a task list based on the relationship diagram includes: Based on the relationship diagram and the preset splitting rules, the EDA tool job is decomposed into multiple sub-tasks, and each sub-task is assigned an independent identifier, execution instructions, and runtime environment parameters. Based on the subtasks, the relationship diagram between the subtasks, and the priority of the tasks, the task list is generated and sent to the task multi-level scheduling and buffer processing module.

5. The control system according to claim 1, characterized in that: The multi-level task scheduling and buffering module includes a primary waiting queue and a secondary priority sub-queue group, wherein the secondary priority sub-queue group includes multiple secondary priority sub-queues with different priorities.

6. The control system according to claim 5, characterized in that, The multi-level task scheduling and buffering module performs real-time priority determination of tasks in the task list, including: The task is obtained from the first-level waiting queue through the priority calculation service, and the scheduling priority of the task is determined based on the attribute parameters of the task. The attribute parameters include: the project priority associated with the task, the reciprocal of the estimated execution time of the task, and the waiting time of the task in the queue; After the calculation is completed, the task is delivered to the corresponding secondary priority sub-queue according to the scheduling priority order.

7. The control system according to claim 1, characterized in that: The task multi-level scheduling and buffer processing module also includes a dependency checker, which is used to poll the tasks in the secondary priority sub-queue to determine whether the preconditions of the tasks are met. If the aforementioned preconditions have been met, then the status of the task is set to be ready for distribution. If the aforementioned preconditions are not met, the task will be retained in the original queue for retry later.

8. The control system according to claim 1, characterized in that, The execution monitoring module specifically includes: The log collection unit is used to collect the running records generated during the execution of the task; The event processing unit is used to acquire and distribute status change events of the task and the EDA tool job; The visualization unit is used to provide a dynamic display of the EDA tool's job status based on the status change event, and to feed back the status change event to the data analysis and processing module.

9. A control method for the operation of semiconductor EDA tools in conjunction with the LSF platform, characterized in that, Includes the following steps: The user EDA tool job submission and receiving module receives all the EDA tool jobs submitted by users; The EDA tool jobs are categorized and classified, resource requirements are determined, and a relationship diagram between the EDA tool jobs is extracted. Based on the relationship diagram, the EDA tool jobs are decomposed to generate a task list. The priority of tasks in the task list is determined in real time, and resources are allocated to tasks that are about to be executed. Receive the task issued by the task multi-level scheduling and buffer processing module and send it to the LSF platform; The task execution and transmission module receives data containing at least the real-time task completion status and task logs fed back by the LSF platform, and feeds the data back to the data analysis and processing module. When the submission receiving module in the control system receives a new EDA tool job submitted by the user, the data analysis and processing module forms a new task list and adds it to the task multi-level scheduling and buffering processing module. The priority of the tasks is arranged and they are then sent to the LSF platform through the task execution and distribution transmission module.

10. A computing device, characterized in that, include: Processor, computer-readable storage medium, and communication bus, The computer-readable storage medium stores program instructions, which, when executed by the processor, enable the control method for operating a semiconductor EDA tool in conjunction with the LSF platform as described in any one of claims 9.