CN-122028709-A - Semiconductor diagnosis method, apparatus, device and storage medium based on multiple variables

Abstract

The application provides a semiconductor diagnosis method, a device, equipment and a storage medium based on multiple variables, wherein the semiconductor diagnosis method comprises the steps of responding to an abnormality detection request aiming at a target manufacturing system and acquiring grouping process variables of each target sensor group in the target manufacturing system; the method comprises the steps of obtaining a grouping process variable, carrying out feature extraction on the grouping process variable to obtain a grouping variable feature corresponding to the grouping process variable, carrying out anomaly detection on the grouping variable feature by utilizing an anomaly detection model corresponding to the target sensor grouping to obtain a variable detection label corresponding to the grouping variable feature, and determining a system detection result of the target manufacturing system based on the variable detection label corresponding to each target sensor grouping. The technical scheme of the application can effectively reduce influence of irrelevant variables in the semiconductor abnormality detection process and improve the interpretability and accuracy of the abnormality detection process.

Inventors

• XU MINGGUANG

Assignees

• 格创东智(武汉)科技有限公司

Dates

Publication Date

20260512

Application Date

20260128

Claims (14)

1. 1. A semiconductor diagnostic method based on multiple variables, the semiconductor diagnostic method comprising: acquiring a grouping process variable of each target sensor grouping in a target manufacturing system in response to an anomaly detection request for the target manufacturing system; Extracting characteristics of the grouping process variables to obtain grouping variable characteristics corresponding to the grouping process variables; Performing anomaly detection on the group variable characteristics by using an anomaly detection model corresponding to the target sensor group to obtain variable detection labels corresponding to the group variable characteristics; and determining a system detection result of the target manufacturing system based on the variable detection label corresponding to each target sensor group.
2. 2. The semiconductor diagnostic method according to claim 1, wherein the performing anomaly detection on the group variable feature using the anomaly detection model corresponding to the target sensor group to obtain a variable detection tag corresponding to the group variable feature comprises: obtaining an abnormality detection model corresponding to the target sensor group, wherein the abnormality detection model comprises a first abnormality detection module, a second abnormality detection module and a label fusion module; performing first anomaly detection on the grouping variable characteristics by using the first anomaly detection module to obtain a first variable detection tag; performing second anomaly detection on the grouping variable characteristics by using a second anomaly detection module to obtain a second variable detection tag; And evaluating the first variable detection tag and the second variable detection tag by using the tag fusion module, and determining the variable detection tag corresponding to the grouping variable characteristic.
3. 3. The semiconductor diagnostic method as claimed in claim 2, wherein the second abnormality detection module includes at least two different abnormality detection units and a detection fusion unit, wherein an input end of the abnormality detection unit is used for receiving the group variable characteristics, an output end of the abnormality detection unit is connected with an input end of the detection fusion unit, and an output end of the detection fusion unit is connected with the tag fusion module as an output end of the second abnormality detection module; The step of performing a second anomaly detection on the group variable feature by using a second anomaly detection module to obtain a second variable detection tag includes: Calculating an initial feature distribution parameter of the group variable feature by using the anomaly detection unit; determining variable characteristic distribution parameters corresponding to the grouping variable characteristics based on the detection fusion unit and the initial characteristic distribution parameters; determining characteristic deviation parameters of the grouping variable characteristics according to the variable characteristic distribution parameters and preset target characteristic distribution parameters; and determining and generating a second variable detection tag according to the characteristic deviation parameter and a preset deviation threshold value.
4. 4. The semiconductor diagnostic method according to claim 2, wherein the evaluating the first variable detection tag and the second variable detection tag by the tag fusion module to determine the variable detection tag corresponding to the group variable feature comprises: when the first variable detection tag and the second variable detection tag are both normal detection tags, determining the normal detection tags as variable detection tags corresponding to the grouping variable characteristics; And when the first variable detection tag or the second variable detection tag is an abnormality detection tag, determining the abnormality detection tag as a variable detection tag corresponding to the grouping variable characteristic.
5. 5. The method according to any one of claims 1 to 4, characterized by further comprising, before the abnormality detection of the group variable feature using the abnormality detection model corresponding to the target sensor group: Grouping the sensor devices in the target manufacturing system to obtain target sensor groups corresponding to the target manufacturing system; And acquiring a grouping variable sample corresponding to the target sensor grouping, and training an initial model by using the grouping variable sample to obtain an anomaly detection model corresponding to the target sensor grouping.
6. 6. The method of claim 5, wherein the obtaining the group variable samples corresponding to the target sensor group, training an initial model using the group variable samples, and obtaining the anomaly detection model corresponding to the target sensor group, comprises: Acquiring an original variable sample corresponding to the target manufacturing system, and performing data separation on the original variable sample based on the machine module and/or the signal type of the target sensor group to obtain a group variable sample of the target sensor group; Performing quantity enhancement on the grouping variable samples to obtain grouping enhancement samples; Performing fault injection processing on the grouping enhanced samples by using a fault injection model and a target fault mode to obtain target grouping samples; Training the initial model by using the target grouping sample to obtain an anomaly detection model corresponding to the target sensor grouping.
7. 7. The semiconductor diagnostic method of claim 6, wherein the performing the number enhancement on the group variable samples results in group enhanced samples, comprising: Performing similarity retrieval on a sample database associated with the target manufacturing system based on the group variable samples to obtain first enhancement samples associated with the group variable samples in the sample database; generating samples by using a data enhancement model and the grouping variable samples to obtain a second enhancement sample; generating grouping enhanced samples according to the grouping variable samples, the first enhanced samples and the second enhanced samples.
8. 8. The method of claim 6, wherein performing fault injection processing on the packet enhanced samples using a fault injection model and a target fault mode to obtain target packet samples, comprises: Accessing a fault mode library, and acquiring a target fault mode corresponding to the target sensor group in the fault mode library; Configuring the grouping enhanced samples by utilizing the fault injection model and the target fault mode to obtain grouping fault samples corresponding to the grouping enhanced samples; and generating a target grouping sample according to the grouping enhanced sample and the grouping fault sample.
9. 9. The semiconductor diagnostic method according to claim 6, wherein training the initial model by using the target group sample to obtain an anomaly detection model corresponding to the target sensor group comprises: Extracting the characteristics of the target grouping samples to obtain grouping sample characteristics, wherein the grouping sample characteristics comprise grouping enhancement characteristics corresponding to grouping enhancement samples and grouping fault characteristics corresponding to grouping fault samples; performing supervision training on a first initial detection module in the initial model by utilizing the grouping enhancement features and the grouping fault features to obtain a first abnormal detection module; Semi-supervised training is carried out on a first initial detection module in the initial model by utilizing the grouping enhancement features to obtain a second abnormal detection module; And generating an anomaly detection model corresponding to the target sensor group according to the first anomaly detection module and the second anomaly detection module.
10. 10. The method according to claim 1, wherein after determining the system detection result of the target manufacturing system based on the variable detection label corresponding to each of the target sensor groups, further comprising: responding to a detection marking event aiming at the system detection result, and counting marking frequency data of the detection marking event; If the marking times data is larger than or equal to a marking times threshold value, sample updating is carried out based on the detection marking event and the target grouping sample, and an updating grouping sample is obtained; Updating the abnormal detection model by using the update packet sample to obtain an update detection model; And updating and detecting the grouping process variable by using the updating and detecting model to obtain an updating and detecting result.
11. 11. The method according to claim 1, wherein after determining the system detection result of the target manufacturing system based on the variable detection label corresponding to each of the target sensor groups, further comprising: Acquiring variable distribution data of the grouping process variable and training distribution data corresponding to the grouping process variable; calculating a variable offset parameter of the group process variable according to the variable distribution data and the training distribution data; And if the variable offset parameter is larger than a preset offset threshold, updating the abnormal detection model to obtain an updated detection model.
12. 12. A semiconductor diagnostic device, characterized in that the semiconductor diagnostic device comprises: a variable acquisition module configured to acquire a grouping process variable for each target sensor grouping in a target manufacturing system in response to an anomaly detection request for the target manufacturing system; The characteristic extraction module is configured to perform characteristic extraction on the grouping process variable to obtain grouping variable characteristics corresponding to the grouping process variable; the anomaly identification module is configured to perform anomaly detection on the grouping variable characteristics by utilizing an anomaly detection model corresponding to the grouping of the target sensors to obtain variable detection labels corresponding to the grouping variable characteristics; and an anomaly detection module configured to determine a system detection result of the target manufacturing system based on the variable detection tags corresponding to each of the target sensor groups.
13. 13. A semiconductor diagnostic device, characterized in that the semiconductor diagnostic device comprises: one or more processors; Memory, and One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the processor to implement the steps of the semiconductor diagnostic method of any one of claims 1 to 11.
14. 14. A computer-readable storage medium, characterized in that a computer program is stored thereon, which computer program is loaded by a processor to perform the steps of the semiconductor diagnostic method according to any one of claims 1 to 11.

Description

Semiconductor diagnosis method, apparatus, device and storage medium based on multiple variables Technical Field The present application relates to the field of computer technology, and in particular, to a semiconductor diagnosis method, apparatus, device and storage medium based on multiple variables. Background Currently, there are a large number of complex manufacturing equipment and process variables such as temperature, pressure, flow, radio frequency power, etc., that are highly coupled to each other in the manufacturing process links of the semiconductor manufacturing process. In order to ensure the accuracy of semiconductor manufacturing, abnormality detection is required in a manufacturing process link, however, the existing abnormality detection mode is based on a univariate threshold value, and the abnormality detection mode has low detection accuracy, so that the false alarm rate and the false alarm rate are high, and the existing semiconductor manufacturing detection scene cannot be satisfied. Disclosure of Invention The embodiment of the application provides a semiconductor diagnosis method, device, equipment and storage medium based on multiple variables, which aim to solve the technical problem that production abnormality cannot be accurately detected in the semiconductor manufacturing process in the prior art. In one aspect, embodiments of the present application provide a semiconductor diagnostic method based on multiple variables, the semiconductor diagnostic method comprising the steps of: acquiring a grouping process variable of each target sensor grouping in a target manufacturing system in response to an anomaly detection request for the target manufacturing system; Extracting characteristics of the grouping process variables to obtain grouping variable characteristics corresponding to the grouping process variables; Performing anomaly detection on the group variable characteristics by using an anomaly detection model corresponding to the target sensor group to obtain variable detection labels corresponding to the group variable characteristics; and determining a system detection result of the target manufacturing system based on the variable detection label corresponding to each target sensor group. In one possible implementation manner of the present application, the performing anomaly detection on the group variable feature by using the anomaly detection model corresponding to the target sensor group to obtain a variable detection tag corresponding to the group variable feature includes: obtaining an abnormality detection model corresponding to the target sensor group, wherein the abnormality detection model comprises a first abnormality detection module, a second abnormality detection module and a label fusion module; performing first anomaly detection on the grouping variable characteristics by using the first anomaly detection module to obtain a first variable detection tag; performing second anomaly detection on the grouping variable characteristics by using a second anomaly detection module to obtain a second variable detection tag; And evaluating the first variable detection tag and the second variable detection tag by using the tag fusion module, and determining the variable detection tag corresponding to the grouping variable characteristic. In one possible implementation manner of the application, the second abnormality detection module comprises at least two different abnormality detection units and a detection fusion unit, wherein the input end of the abnormality detection unit is used for receiving the group variable characteristics, the output end of the abnormality detection unit is connected with the input end of the detection fusion unit, and the output end of the detection fusion unit is used as the output end of the second abnormality detection module to be connected with the tag fusion module; The step of performing a second anomaly detection on the group variable feature by using a second anomaly detection module to obtain a second variable detection tag includes: Calculating an initial feature distribution parameter of the group variable feature by using the anomaly detection unit; determining variable characteristic distribution parameters corresponding to the grouping variable characteristics based on the detection fusion unit and the initial characteristic distribution parameters; determining characteristic deviation parameters of the grouping variable characteristics according to the variable characteristic distribution parameters and preset target characteristic distribution parameters; and determining and generating a second variable detection tag according to the characteristic deviation parameter and a preset deviation threshold value. In one possible implementation manner of the present application, the evaluating, by the tag fusion module, the first variable detection tag and the second variable detection tag to determine a variable detection tag corr