Method and apparatus for repairing semiconductor device, model training method and apparatus

The maintenance model trained by deep neural networks solves the problem of semiconductor equipment maintenance relying on human experience, enabling rapid and efficient fault location and repair, and improving the reliability and safety of equipment operation.

CN122173935APending Publication Date: 2026-06-09SUZHOU WINMAX TECH CORP
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
SUZHOU WINMAX TECH CORP
Filing Date
2026-05-08
Publication Date
2026-06-09

AI Technical Summary

Technical Problem

The maintenance of existing semiconductor manufacturing equipment relies on human experience, resulting in long fault location and repair times, significant losses from unplanned downtime, and difficulty in quickly and efficiently resolving complex faults.

Method used

The maintenance model is trained using a deep neural network model. By classifying maintenance actions into multi-level action sets such as component level, system level, and termination judgment, and by using reinforcement learning and experience playback mechanisms, maintenance strategies are automatically evaluated and recommended.

Benefits of technology

It improves the efficiency of fault location and repair, reduces unplanned downtime, enhances the interpretability and operability of decisions, and supports adaptive dynamic maintenance decisions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122173935A_ABST
    Figure CN122173935A_ABST
Patent Text Reader

Abstract

This application provides a method and apparatus for repairing semiconductor equipment, as well as a model training method and apparatus. The semiconductor equipment repair model training method includes the following steps: acquiring a training data set and an original model; classifying the repair action data according to the target of the repair action to obtain a multi-level repair action set; obtaining reinforcement learning rules based on the execution result of the repair action and the reward function, wherein the reward function calculates corresponding positive or negative reward values ​​based on whether the fault is eliminated, whether a new fault is introduced, or whether the situation worsens after the repair action; training the original model based on the training data set and the reinforcement learning rules, outputting the evaluation value corresponding to each repair action in the multi-level repair action set, and storing the training data using an experience playback mechanism to obtain a repair model. The technical solution of this application can quickly locate the cause of the fault, facilitating efficient repair of equipment faults.
Need to check novelty before this filing date? Find Prior Art