CN-116057546-B - Error cause estimating device and estimating method

CN116057546BCN 116057546 BCN116057546 BCN 116057546BCN-116057546-B

Abstract

The present invention uses an error cause estimation device, which includes a data preprocessing unit that generates training data having a format suitable for input to a machine learning model using processing target data, and a model tree generation unit that generates a learning model for detecting errors, that is, an error detection model, using the training data as input, and generates a model tree that expresses a relationship between the error detection models using a tree structure in which the error detection model is a node. Thus, even if the error cause is not labeled in advance, a learning model for detecting errors for a plurality of types of errors that occur can be generated.

Inventors

Jitian Thailand is great
ISHIKAWA MASANORI
Sasajima Second University
Takanoh Masanori
HAYAKAWA KOICHI

Assignees

株式会社日立高新技术

Dates

Publication Date: 20260512
Application Date: 20200917

Claims (10)

1. An estimating device for estimating a cause of an error in an inspection operation performed by a semiconductor inspection device, comprising: A data preprocessing section for generating training data having a format suitable for input to the machine learning model using the processing object data, and A model tree generating unit that generates an error detection model, which is a learning model for detecting errors, using the training data as an input, and generates a model tree that expresses a relationship between the error detection models using a tree structure in which the error detection model is a node, The processing object data includes setting parameters and measurement results of an inspection object of the semiconductor inspection apparatus, The model tree generation unit includes: A clustering error detection model generation unit for generating a clustering error detection model by taking data of incomplete clusters in the training data as input and learning an error detection rule according to a difference in trend between normal data and error data; a model analysis unit that calculates a value indicating a sensitivity of a degree to which the feature amount input as the training data contributes to the output of the error detection model; A data clustering unit that clusters data based on the feature amount and the sensitivity value; a clustering completion judging unit that judges whether or not the clustering is completed with respect to the clustered data; A causal error detection model generation part which takes the clustered data as input, learns error detection rules according to the difference of the trend of normal data and error data, generates a causal error detection model, and And a model connection unit that connects the error detection model for clustering and the error detection model for causal analysis based on a flow of clustering of data, and generates the model tree composed of a phylogenetic tree having the error detection model for clustering and the error detection model for causal analysis as nodes.
2. The error cause inference apparatus according to claim 1, wherein, The error detection model for clustering has a simple model structure compared with the error detection model for the reasons.
3. The error cause inference apparatus according to claim 1, wherein, Also comprises an error cause deducing part, The error cause estimation unit includes: a model evaluation unit that evaluates performance of the generated divided cause error detection model, using, as input, error cause estimation target data generated by the data preprocessing unit for data of the processing target data for which an error cause is to be estimated, the data having a format suitable for input to the machine learning model; A model selecting unit for selecting 1 or more model of error detection due to the difference in the evaluation value of the performance, and And a related parameter extraction unit that extracts a branch in which the selected partial cause error detection model is located in the model tree, and extracts a related parameter of the error detection model included in the branch.
4. The error cause inference apparatus according to claim 3, wherein, The error cause estimation unit further includes: and an error cause probability calculation unit that calculates a probability of a candidate error cause, using the correlation parameter extracted by the correlation parameter extraction unit as an input.
5. The error cause inference apparatus according to claim 4, wherein, When the model selecting unit selects a plurality of the partial cause error detection models, the probability obtained based on each of the partial cause error detection models is corrected using the model evaluation value obtained by the model evaluating unit.
6. The error cause inference apparatus according to claim 5, wherein, Also included is a model database that stores the error detection model for clustering and the causal error detection model in association with information about these models.
7. The error cause estimation device according to claim 6, wherein the error cause estimation device is configured to, In the case where there are a plurality of versions of the error detection model for clustering and the error detection model for dividing cause stored in the model database, a user can use a terminal to replace a model within the model tree with another model stored in the model database.
8. The error cause inference apparatus according to claim 5, wherein, The probabilities before and after correction are displayed on the terminal.
9. The error cause estimation device according to claim 6, wherein the error cause estimation device is configured to, Information associated with the selected error detection model is displayed by a user selecting an error detection model included in the model tree via a terminal.
10. A method for estimating the cause of an error in an inspection operation performed by a semiconductor inspection apparatus, comprising: A data preprocessing step in which training data having a format suitable for input to a machine learning model is generated using the processing object data, and A model tree generating step of generating an error detection model, which is a learning model for detecting an error, with the training data as an input, and generating a model tree expressing a relationship between the error detection models by using a tree structure in which the error detection model is a node, The processing object data includes setting parameters and measurement results of an inspection object of the semiconductor inspection apparatus, The model tree generating step comprises the following steps: A step of generating an error detection model for clustering, wherein data of incomplete clusters in the training data is used as input, error detection rules are learned according to different trends of normal data and error data, and the error detection model for clustering is generated; a model analysis step of calculating a value of sensitivity indicating to what degree a feature quantity input as the training data contributes to an output of the error detection model; A data clustering step in which data is clustered based on the values of the feature quantity and the sensitivity; A clustering completion judging step, wherein whether the clustering is completed or not is judged for the clustered data; A step of generating a causal error detection model, in which the clustered data is used as input, error detection rules are learned according to the difference of the trend of normal data and error data, and a causal error detection model is generated, and And a model connection step in which a data-based clustering process connects the error detection model for clustering and the error detection model for causal clustering, thereby generating the model tree composed of a phylogenetic tree having the error detection model for clustering and the error detection model for causal clustering as nodes.

Description

Error cause estimating device and estimating method Technical Field The present invention relates to an estimation device and an estimation method for error causes. Background Semiconductor metrology (inspection) devices and semiconductor inspection (inspection) devices perform inspection and measurement operations for each inspection point on the surface of a semiconductor wafer according to set parameters called recipe. In general, in adjusting recipe parameters, each item is manually optimized by an engineer according to the properties of a measurement/inspection object, the characteristics of a device, and the like. Thus, for example, when the characteristics of a recipe or an apparatus, which is not sufficiently adjusted, are changed with time, there is a possibility that an error occurs in the inspection operation and the measurement operation. Such an error is called a recipe error as an error due to the content of the recipe. When a recipe error occurs, the cause location is typically determined by a maintenance engineer analyzing the device internal data from the semiconductor metrology device and the semiconductor inspection device. However, as semiconductors become finer and more versatile, the number of recipes and the number of recipe setting items increase, and the formation of recipes becomes complicated. Therefore, it takes time to determine the cause of the formulation error, which becomes one of the causes of the decrease in the operation rate of the apparatus. Patent document 1 discloses a technique for shortening training time by reducing training data by saving the work amount of a true value generation operation of training data, and provides a pattern inspection system for inspecting an image of an inspection target pattern by using a recognizer configured by machine learning based on an image of the inspection target pattern of an electronic device and data used for manufacturing the inspection target pattern, wherein an image selecting section (training data selecting section) selects a training pattern image for machine learning from a plurality of pattern images based on pattern data and pattern images stored in a storage section, and clusters data having a plurality of position coordinates of the homomorphic pattern stored for each pattern into 1 or more clusters. Prior art literature Patent literature Patent document 1 Japanese patent application laid-open No. 2020-35282 Disclosure of Invention Technical problem to be solved by the invention The training data selecting unit of the pattern inspection system described in patent document 1 clusters data of a plurality of position coordinates based on the isomorphic pattern. However, the training data selecting unit is not based on the degree of contribution of each parameter to the error, and it is considered that there is room for improvement. The purpose of the present invention is to generate a learning model that can detect errors in a variety of types of errors that can occur without adding a label to the cause of the error in advance. Further, the present invention aims to reduce the number of label adding processes (labeling). Technical means for solving the problems The device for estimating the cause of an error includes a data preprocessing unit that generates training data having a format suitable for input to a machine learning model using processing target data, and a model tree generation unit that generates an error detection model, which is a learning model for detecting an error, using the training data as input, and generates a model tree that expresses a relationship between the error detection models using a tree structure having the error detection model as a node. Effects of the invention According to the present invention, a learning model can be generated, and errors can be detected for various types of errors that can occur without adding a label to the cause of the error in advance. Further, according to the present invention, the number of tag adding processes can be reduced. Drawings Fig. 1 is a block diagram showing an example of an information processing system including an error cause estimation device according to an embodiment. Fig. 2 is a block diagram showing the model tree generating unit of fig. 1. Fig. 3 is a flowchart showing steps in the model tree generating section of fig. 2. Fig. 4 is a schematic diagram showing a part of the steps of fig. 3, namely, a step of error data clustering and error detection model generation in the model tree generation section. Fig. 5 is a block diagram showing the error cause estimating unit of fig. 1. Fig. 6 is a flowchart showing a procedure for calculating the error cause probability in the error cause estimating unit. Fig. 7 is a schematic diagram showing a structure in which error cause probabilities obtained by analyzing a plurality of error detection models are synthesized and visualized. Fig. 8 is a terminal screen showing an example of the error cause