EDA platform operation and maintenance method, system, device and readable storage medium

By constructing a three-level digital twin model and license-aware reinforcement learning, the problems of resource conflicts and unreproducible signing results in EDA computing clusters are solved, realizing a unified platform for intelligent operation and maintenance and traceable and reproducible results.

CN122389773APending Publication Date: 2026-07-14SUZHOU HUAXIN MICROELECTRONICS
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
SUZHOU HUAXIN MICROELECTRONICS
Filing Date
2026-05-26
Publication Date
2026-07-14

AI Technical Summary

Technical Problem

Existing EDA computing cluster operation and maintenance technologies cannot detect micro-architectural resource conflicts such as memory and bandwidth contention, and lack fine-grained forward-looking optimization for expensive EDA licenses, resulting in idleness and deadlock; the signing results are unreproducible due to dependence on drift and environmental differences, and lack automated consistency determination and difference attribution.

Method used

A three-tiered digital twin model is constructed, which combines a memory bandwidth contention model to quantify resource conflicts. A license-aware reinforcement learning and rule masking mechanism is introduced to generate optimal scheduling instructions. Environment reconstruction is achieved through snapshot records and reproduction packages to ensure that the results are traceable and reproducible.

Benefits of technology

It achieves accurate prediction and avoidance of microarchitectural resource conflicts, avoids ineffective scheduling and deadlocks, ensures consistent reproduction of approval results and automated attribution of differences, and forms a unified intelligent operation and maintenance platform covering design, operation and approval verification.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122389773A_ABST
    Figure CN122389773A_ABST
Patent Text Reader

Abstract

The application provides an EDA platform operation and maintenance method, system, device and readable storage medium. The method comprises the following steps: collecting infrastructure indexes, job scheduling states, EDA tool running logs and license occupation data of an EDA computing cluster, and constructing a three-level digital twin containing a tool chain behavior model, a resource state model and a design process dependency model; defining an action space containing task distribution, node selection and parameter configuration, imposing a rule mask on the action space based on EDA engineering hard constraints, and filtering actions that violate license quota, process library compatibility requirements or design process dependency logic; inputting the effective actions after the mask into a simulation engine, deducing the performance of the candidate scheduling strategy in the virtual time domain, and generating optimal scheduling instructions based on the deduction result and executing the instructions. Compared with the prior art, the application can accurately predict and avoid the noisy neighbor effect. At the same time, invalid scheduling and deadlock are avoided.
Need to check novelty before this filing date? Find Prior Art