CN-122022578-A - Manufacturing detection method, device, equipment and storage medium based on track morphology

Abstract

The application provides a manufacturing detection method, a device, equipment and a storage medium based on track form, wherein the manufacturing detection method comprises the steps of responding to an abnormality detection request aiming at a target manufacturing system and acquiring manufacturing process data of the target manufacturing system; and determining a system detection result of the manufacturing process data according to the data evaluation parameters and an abnormality evaluation threshold value of the abnormality detection model. The technical scheme of the application can effectively detect the abnormality of each manufacturing process and node in the target manufacturing system, and improves the accuracy, efficiency and interpretability of the abnormality detection.

Inventors

• XU MINGGUANG

Assignees

• 格创东智(武汉)科技有限公司

Dates

Publication Date

20260512

Application Date

20260128

Claims (15)

1. 1. A manufacturing detection method based on a track morphology, the manufacturing detection method comprising: Acquiring manufacturing process data of a target manufacturing system in response to an abnormality detection request for the target manufacturing system; reasoning is carried out by utilizing an anomaly detection model and the manufacturing process data to obtain data evaluation parameters corresponding to the manufacturing process data; and determining a system detection result of the manufacturing process data according to the data evaluation parameter and an abnormality evaluation threshold of the abnormality detection model.
2. 2. The manufacturing inspection method according to claim 1, wherein the reasoning using the anomaly detection model and the manufacturing process data to obtain data evaluation parameters corresponding to the manufacturing process data comprises: performing data segmentation on the manufacturing process data according to the number of target windows of the target data windows and window width data to obtain window manufacturing data corresponding to each target data window; Carrying out distributed modeling on manufacturing data in a window in a target data window to obtain manufacturing distribution data corresponding to the manufacturing data in the window; And calculating data evaluation parameters of the manufacturing data in the window according to the abnormality detection model and the manufacturing distribution data.
3. 3. The manufacturing inspection method of claim 2, wherein the calculating data evaluation parameters of the intra-window manufacturing data from the anomaly detection model and the manufacturing distribution data comprises: calculating likelihood probability data corresponding to the manufacturing distribution data and the target distribution data by using the anomaly detection model; calculating the abnormal likelihood probability of the manufacturing data in the window according to the likelihood probability data and the target evaluation coefficient; and determining data evaluation parameters of the manufacturing data in the window according to the likelihood probability data and the target quantiles.
4. 4. The manufacturing inspection method of claim 1, wherein the determining the system inspection result of the manufacturing process data based on the data evaluation parameter and an anomaly evaluation threshold value of the anomaly detection model comprises: Acquiring an abnormality evaluation threshold of the abnormality detection model, wherein the abnormality evaluation threshold is an abnormality critical threshold calculated by a training evaluation parameter mean value of training evaluation parameters and a training evaluation parameter standard of the abnormality detection model in a training process; If the data evaluation parameter is smaller than the abnormal evaluation threshold, determining that the system detection result of the manufacturing process data is a normal detection result; and if the data evaluation parameter is larger than the abnormality evaluation threshold, determining that the system detection result of the manufacturing process data is an abnormality detection result.
5. 5. The method of claim 1, wherein the reasoning using the anomaly detection model and the manufacturing process data further comprises, before obtaining the data evaluation parameters corresponding to the manufacturing process data: Acquiring training sample data corresponding to the target manufacturing system; And training the initial model by using the training sample data to obtain an anomaly detection model.
6. 6. The method according to claim 5, wherein training the initial model using the training sample data to obtain the abnormality detection model comprises: preprocessing the training sample data based on the model type of the initial model to obtain preprocessed sample data; Performing data division on the preprocessed samples based on a preset sample division ratio to obtain target training data sets corresponding to the preprocessed samples, wherein the target training data sets comprise training sample sets and test sample sets; and training the initial model by using the target training data set to obtain an anomaly detection model.
7. 7. The manufacturing inspection method according to claim 6, wherein preprocessing the training sample data based on the model type of the initial model to obtain preprocessed sample data comprises: screening the training sample data based on the model type of the initial model to obtain screened sample data; performing data filtering on the screening sample data based on the data fluctuation level of the screening sample data to obtain filtering sample data; and carrying out data alignment on the filtered sample data based on a target time scale to obtain preprocessed sample data.
8. 8. The method of claim 6, wherein training the initial model using the target training data set to obtain an anomaly detection model comprises: Calculating the number of target windows and window width data according to the data sampling information and the data length information of the target training data set; Window division is carried out on the target training data set based on the number of the target windows and the window width data, so that intra-window training data corresponding to each training data window are obtained; carrying out distributed modeling on the intra-window training data corresponding to the training data window to obtain training distribution data corresponding to the intra-window training data; and training the initial model based on the training distribution data and the intra-window training data to obtain an abnormality detection model.
9. 9. The method of claim 8, wherein training the initial model based on the training distribution data and intra-window training data to obtain an anomaly detection model comprises: Calculating an interval threshold parameter of a target confidence interval of the initial model based on the training distribution data; Determining training evaluation parameters of the intra-window training data based on the training distribution data and the intra-window training data using the initial model; Calculating an abnormal evaluation threshold value of the initial model based on a training evaluation parameter mean value and a training evaluation parameter standard deviation corresponding to the training evaluation parameters; Training the initial model based on the interval threshold parameter and the abnormality evaluation threshold to obtain an abnormality detection model.
10. 10. The manufacturing inspection method according to claim 9, wherein training the initial model based on the interval threshold parameter and the abnormality evaluation threshold value to obtain an abnormality inspection model includes: Training the initial model based on the interval threshold parameter and the abnormality evaluation threshold to obtain a model to be verified; performing performance verification on the model to be verified to obtain performance verification indexes corresponding to the model to be verified; and if the performance verification index is larger than a preset verification index threshold, determining the model to be verified as an abnormal training model.
11. 11. The method according to claim 6, wherein training the initial model using the training sample data to obtain the anomaly detection model, further comprises: Responding to a model update request aiming at the abnormal detection model, and acquiring update sample data corresponding to the model update request; generating an updated training data set corresponding to the abnormality detection model according to the updated sample data and the training sample data; Performing update training on the abnormal detection model by using the update training data set to obtain an update detection model; and determining a target detection model of the model update request according to the update detection model and the abnormality detection model.
12. 12. The manufacturing inspection method of claim 11, wherein the determining the target inspection model of the model update request from the update inspection model and the anomaly inspection model comprises: Acquiring an update performance index of the update detection model and update distribution difference data of the update detection model; if the updated performance index is smaller than the performance verification index of the abnormal detection model, determining the abnormal detection model as a target detection model of the model update request; And if the updated performance index is greater than the performance verification index and the updated distribution difference data is greater than a preset difference threshold, determining the updated detection model as a target detection model of the model update request.
13. 13. A manufacturing inspection device is characterized in that, the manufacturing detection device includes: a data acquisition module configured to acquire manufacturing process data of a target manufacturing system in response to an abnormality detection request for the target manufacturing system; the abnormality evaluation module is configured to utilize the abnormality detection model and the manufacturing process data to infer so as to obtain data evaluation parameters corresponding to the manufacturing process data; an anomaly detection module configured to determine a system detection result of the manufacturing process data based on the data evaluation parameters and an anomaly evaluation threshold value of the anomaly detection model.
14. 14. A manufacturing inspection apparatus, characterized in that the manufacturing inspection apparatus 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 manufacturing inspection method of any one of claims 1 to 12.
15. 15. A computer-readable storage medium, having stored thereon a computer program, the computer program being loaded by a processor to perform the steps of the manufacturing inspection method of any of claims 1 to 12.

Description

Manufacturing detection method, device, equipment and storage medium based on track morphology Technical Field The present application relates to the field of anomaly detection technologies, and in particular, to a method, an apparatus, a device, and a storage medium for manufacturing detection based on a track shape. Background At present, a plurality of processing procedures of different processes are generally involved in the semiconductor manufacturing process, and the processing quality of each processing procedure directly influences the final yield, so that abnormal detection is required to be carried out on data in the processing process in the semiconductor manufacturing process, thereby improving the semiconductor manufacturing yield, but the existing abnormal detection method has the defects of dependence on manual experience, high detection cost and high false alarm rate due to different processing procedure equipment and parameters. Disclosure of Invention The embodiment of the application provides a manufacturing detection method, device, equipment and storage medium based on track morphology, which aim to solve the technical problem of higher error detection and error report rate in the semiconductor manufacturing process in the prior art. In one aspect, an embodiment of the present application provides a manufacturing inspection method, including the steps of: Acquiring manufacturing process data of a target manufacturing system in response to an abnormality detection request for the target manufacturing system; reasoning is carried out by utilizing an anomaly detection model and the manufacturing process data to obtain data evaluation parameters corresponding to the manufacturing process data; and determining a system detection result of the manufacturing process data according to the data evaluation parameter and an abnormality evaluation threshold of the abnormality detection model. In one possible implementation manner of the present application, the reasoning using the anomaly detection model and the manufacturing process data to obtain the data evaluation parameters corresponding to the manufacturing process data includes: performing data segmentation on the manufacturing process data according to the number of target windows of the target data windows and window width data to obtain window manufacturing data corresponding to each target data window; Carrying out distributed modeling on manufacturing data in a window in a target data window to obtain manufacturing distribution data corresponding to the manufacturing data in the window; And calculating data evaluation parameters of the manufacturing data in the window according to the abnormality detection model and the manufacturing distribution data. In one possible implementation of the present application, the calculating the data evaluation parameter of the intra-window manufacturing data according to the anomaly detection model and the manufacturing distribution data includes: calculating likelihood probability data corresponding to the manufacturing distribution data and the target distribution data by using the anomaly detection model; calculating the abnormal likelihood probability of the manufacturing data in the window according to the likelihood probability data and the target evaluation coefficient; and determining data evaluation parameters of the manufacturing data in the window according to the likelihood probability data and the target quantiles. In one possible implementation of the present application, the determining the system detection result of the manufacturing process data according to the data evaluation parameter and the abnormality evaluation threshold of the abnormality detection model includes: Acquiring an abnormality evaluation threshold of the abnormality detection model, wherein the abnormality evaluation threshold is an abnormality critical threshold calculated by a training evaluation parameter mean value of training evaluation parameters and a training evaluation parameter standard of the abnormality detection model in a training process; If the data evaluation parameter is smaller than the abnormal evaluation threshold, determining that the system detection result of the manufacturing process data is a normal detection result; and if the data evaluation parameter is larger than the abnormality evaluation threshold, determining that the system detection result of the manufacturing process data is an abnormality detection result. In one possible implementation manner of the present application, before the reasoning is performed by using the anomaly detection model and the manufacturing process data to obtain the data evaluation parameter corresponding to the manufacturing process data, the method further includes: Acquiring training sample data corresponding to the target manufacturing system; And training the initial model by using the training sample data to obtain an anomaly detection model. In one