JPH11344450A - Instruction data forming method and defect classifying method and apparatus thereof - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a defect classifying method and apparatus equipped with a learning function constituted so as to accurately classify the kind of a defect from the image signal of a detect, in order to estimate the generation cause of the defect by supporting the formation of accurate instruction data. SOLUTION: In an instruction data forming method for classifying the kind of a defect based on a defect image signal, a plurality of feature quantities are calculated with respect to a plurality of defect images for instruction and the classifying parameters shown by the discrimination function between categories in a feature space based on the calculated feature quantities, to diagnose the statistical significance showing the corresponding relation between a plurality of instruction defects and the category corresponding to the kind of the defects, and an instruction defect image having capacity lowering possibility is displayed on a screen to correct instruction data.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method of classifying a defect type based on an image signal obtained by imaging a defect found in a defect inspection in a manufacturing process of a semiconductor wafer or the like and estimating a cause of the defect occurrence. The present invention relates to a defect classifying method and an apparatus therefor.

[0002]

2. Description of the Related Art Japanese Patent Laying-Open No. 8-21803 discloses a defect type judging device. In this configuration, defect information (area, shape, position, brightness, etc.) extracted from a defect image is given as an input of a neuro processing unit, and a defect type is obtained as an output. The neuro processing unit is trained so that each defect type appears. As a result, the defect information extracted from the classification target image is input to the neuro processing unit, and the defect type is determined from the output.

[0003]

The learning type defect classifying apparatus has a problem in that learning data presented by a user tends to include inconsistent data. In particular, in the classification of images, the above-mentioned tendency is remarkable because the user can easily make an ambiguous determination. However, in the above-described conventional technology, even when a user gives inconsistent data, the system tries to cope with it in a general-purpose manner, and there is a possibility that the performance as a whole may be reduced.

SUMMARY OF THE INVENTION In view of the above problems, it is an object of the present invention to support creation of accurate teaching data for classifying a defect type from a defect image signal in order to estimate the cause of the defect. An object of the present invention is to provide a teaching data creating method. Another object of the present invention is to provide a learning function capable of accurately classifying a defect type from a defect image signal in order to estimate a cause of the defect by being able to support creation of accurate teaching data. It is an object of the present invention to provide a defect classification method and an apparatus therefor.

[0005]

In order to achieve the above object, the present invention relates to a teaching data generating method for classifying the types of defects based on a defect image signal, comprising a plurality of teaching defect images. The teaching data is created by assigning a category corresponding to the type of defect to the teaching defect data, a plurality of feature amounts are calculated for each of the teaching defect images, and the assigned category and the calculated feature amount are calculated. The discriminant function between the categories in the feature space of the teaching defect image is calculated based on the above, and the teaching data including the correspondence between the teaching defect image and the category is diagnosed using the discrimination function, and at least the performance can be reduced. And displaying the defective teaching image on the screen and correcting the teaching data. Further, the present invention, a teaching defect image detecting step of detecting a plurality of teaching defect images by imaging a plurality of teaching defects in advance,
A plurality of teaching defect images detected in the teaching defect image detection process are displayed on a display means, and a category is assigned to the displayed plurality of teaching defect images, so that the teaching defect image and the category are displayed. A plurality of desired feature amounts for each of a plurality of teaching defects are calculated from a plurality of teaching defect images detected in the category teaching process for teaching the correspondence of the plurality of teaching defect images detected in the teaching defect image detecting process; A category providing step of calculating a discriminant function between categories in the feature amount space of the teaching defect image based on the category and the calculated feature amount, classifying each teaching defect using the discriminant function, and assigning a category; Comparing the category taught in the category teaching process with the category assigned in the category assigning process for each teaching defect, and if the two categories are different, Also displayed on the display means that information and,
A correction or exclusion teaching process for correcting the category of the teaching defect or excluding the teaching defect and teaching the correspondence between the teaching defect and the category based on the displayed information. And a learning step of learning and acquiring teaching data indicating a correspondence relationship between a defect and a category corresponding to a defect type, and a defect image of the inspection target by capturing an image of a defect found by inspection with respect to the inspection object. And a classifying step of classifying the type of the defect from the teaching data acquired in the learning step based on the detected defect image of the inspection object. .

Further, the present invention provides a teaching defect image detecting step of detecting a plurality of teaching defect images by imaging a plurality of teaching defects in advance, and providing a plurality of teaching defects detected in the teaching defect image detecting step. Displaying a defect image on the display means, assigning a category to the plurality of displayed defect images, and teaching the correspondence between the defect image and the category. A plurality of desired feature quantities are calculated for each of the plurality of teaching defects from the plurality of teaching defect images detected in the image detection process, and a relationship between the plurality of desired feature quantities for each of the calculated teaching defects is calculated. A category assigning step of classifying each teaching defect from and assigning a category, and a category taught in the category teaching step and a category assigned in the category assigning step for each teaching defect. In comparison, if the two categories are different, at least the information is displayed on the display means, and based on the displayed information, the teaching defect is corrected or the teaching defect is excluded and the teaching defect and the category are excluded. A learning step of learning and acquiring teaching data indicating a correspondence relationship between a feature amount of a plurality of teaching defects and a category corresponding to the type of the defect. Imaging a defect found by inspection of the inspection object, detecting a defect image of the inspection object, calculating a desired feature amount of the defect from the detected defect image of the inspection object, A classifying step of classifying the type of the defect from the teaching data acquired in the learning step based on a desired feature amount of the defect thus obtained.

Further, the present invention provides a teaching defect image detecting step of detecting a plurality of teaching defect images by imaging a plurality of teaching defects in advance, and a plurality of teaching defects detected in the teaching defect image detecting step. Displaying a defect image on the display means, assigning a category to the plurality of displayed defect images, and teaching the correspondence between the defect image and the category. A plurality of desired feature quantities are calculated for each of the plurality of teaching defects from the plurality of teaching defect images detected in the image detection process, and a relationship between the plurality of desired feature quantities for each of the calculated teaching defects is calculated. A category assigning step of calculating parameters for classifying the category from the above, classifying each teaching defect based on the calculated classification parameter and assigning a category; and Comparing the category taught in the category teaching process and the category given in the category giving process, when both categories are different, at least the information is displayed on the display means, based on the displayed information, A correction or exclusion teaching process for correcting the category of the teaching defect or excluding the teaching defect to teach the correspondence between the teaching defect and the category, and corresponding to a plurality of teaching defect feature amounts and defect types. A learning step of learning and acquiring teaching data indicating a correspondence relationship with a category to be inspected, and detecting a defect image of the inspection target by imaging a defect found by inspection with respect to the inspection object. A desired characteristic amount of the defect is calculated from the defect image of the inspection target, and the type of the defect is determined from the teaching data acquired in the learning step based on the calculated desired characteristic amount of the defect. A defect classification method characterized by having a classification step similar.

Further, according to the present invention, when classifying a category for each teaching defect in the category assignment step in the learning step of the defect classification method, a plurality of desired feature quantities are selected from the calculated feature quantities. It is characterized by having a feature amount selecting step. According to the present invention, there is provided a teaching defect image detecting step of imaging a plurality of teaching defects in advance and detecting a plurality of teaching defect images, and a plurality of teaching defect images detected in the teaching defect image detecting step. Is displayed on the display means,
A category teaching process for teaching the correspondence between the teaching defect image and the category by assigning a category to the displayed teaching defect images; and a plurality of teachings detected in the teaching defect image detecting process. Calculating a plurality of desired feature amounts for each of the plurality of teaching defects from the defect image for use;
A category classification evaluation for calculating a discriminant function between categories in the feature amount space of the teaching defect image based on the assigned category and the calculated feature amount, and classifying each teaching defect using the discrimination function. A category classification evaluation value calculating step of calculating a value, and displaying the category classification evaluation value calculated in the category classification evaluation value calculation step for each teaching defect on the display means, based on the displayed category classification evaluation value. A correction or exclusion teaching process for correcting the category of the teaching defect or excluding the teaching defect and teaching the correspondence between the teaching defect and the category, and having a plurality of teaching defect feature amounts and defect types. A learning step of learning and acquiring teaching data indicating a correspondence relationship with a corresponding category;
Detecting a defect image of the inspection target by imaging a defect found by inspection with respect to the inspection target, calculating a desired feature amount of the defect from the detected defect image of the inspection target,
A classification step of classifying the type of the defect from the teaching data acquired in the learning step based on the calculated desired feature amount of the defect. Further, in the present invention, in the category classification evaluation value calculating step in the learning step of the defect classification method, when calculating the category classification evaluation value for classifying each teaching defect, a plurality of desired feature values are calculated from the calculated feature amounts. It is characterized by having a feature amount selecting step of selecting a feature amount.

Further, the present invention provides a teaching defect image detecting step in which a plurality of teaching defects are imaged in advance and a plurality of teaching defect images are detected, and a plurality of teaching defects detected in the teaching defect image detecting step. Calculating a plurality of feature amounts for each of the plurality of teaching defects from the defect image for use, and providing the teaching defect in a feature space based on a relationship between the plurality of feature amounts calculated in the feature calculating process. Is displayed on the display means, and the correspondence between the teaching defect and the category is taught by assigning a category to the teaching defect according to the position of the teaching defect in the displayed feature amount space. A learning process of learning and acquiring teaching data indicating a correspondence relationship between a feature amount of a plurality of teaching defects and a category corresponding to the type of the defect, and Inspected and found The defect is imaged for the defect to be inspected, a defect image of the inspection target is detected, a desired characteristic amount of the defect is calculated from the detected defect image of the inspection target, and the desired characteristic amount of the defect is calculated based on the calculated desired characteristic amount. A classifying step of classifying the type of the defect from the teaching data acquired in the learning step. According to the present invention, there is provided a teaching defect image detecting step of imaging a plurality of teaching defects in advance and detecting a plurality of teaching defect images, and a plurality of teaching defect images detected in the teaching defect image detecting step. A feature amount calculating step of calculating a plurality of feature amounts for each of the plurality of teaching defects, and a feature amount selecting step of selecting a desired plurality of feature amounts from the feature amounts calculated in the feature amount calculating step. The teaching defect calculated in the feature amount calculating step is located on a feature amount space based on a relationship between a plurality of desired feature amounts selected in the feature amount selecting step, and is displayed on display means. And a category teaching process for teaching the correspondence between the teaching defect and the category by assigning a category to the teaching defect according to the position of the teaching defect in the feature amount space. Defect features and defect types A learning step of learning and acquiring teaching data indicating a correspondence relationship with the category corresponding to the category, and detecting a defect image of the inspection target by imaging a defect found by inspection with respect to the inspection object. A desired characteristic amount of the defect is calculated from the detected defect image of the inspection target, and based on the calculated desired characteristic amount of the defect, the type of the defect is determined from the teaching data acquired in the learning step. And a classifying step of classifying the defect.

The present invention is also characterized in that the defect classification method further comprises a defect occurrence cause estimation step of estimating a defect occurrence cause based on the type of the defect classified in the classification step. I do. Further, the present invention is a defect classification method for classifying a defect type based on an image signal obtained by imaging a defect found in a defect inspection of an inspection object, wherein a user previously classifies a defect image for teaching into a defect image. The teaching process of teaching the correspondence between the teaching defect image and the category by giving the category A, and the giving by which the calculation means classifies and gives the category B based on the feature amount of the teaching defect image and the given category A. A display step of displaying a teaching defect image in which the category A and the category B do not match, and a correction or exclusion step of correcting the category A or excluding the displayed teaching defect from the teaching image. And a learning step of acquiring teaching data indicating a correspondence between the teaching defect and the category corresponding to the type of the defect. The present invention also relates to a defect classification method for classifying a defect type based on an image obtained by imaging a defect found in a defect inspection of an inspection object. , A teaching process of teaching the correspondence between the teaching defect image and the category to the system, a calculation process of calculating a classification evaluation value (continuous amount) for each category of the teaching defect image, and a calculation process of the calculation process. Display the classification evaluation value of the teaching defect image on the display means,
A correction or exclusion process of correcting or excluding the category A from the image for teaching based on the display result; and a process of teaching a system to correspond to the image based on the corrected category A, wherein a plurality of teaching defects and defects are included. And a learning step of acquiring teaching data indicating a correspondence relationship with a category corresponding to the type.

The present invention also relates to a defect classification method for classifying a defect type based on an image obtained by imaging a defect found in a defect inspection of an inspection object. And displaying each teaching defect image in a feature amount space based on the extracted feature amount, selecting a defect based on the display result, and displaying the selected teaching defect image. And acquiring a teaching data indicating a correspondence relationship between the plurality of teaching defects and the category corresponding to the type of the defect, including a category assigning step in which the user assigns a category A to the teaching defect image based on the display result. It is characterized by having a learning step to perform. Further, the present invention provides a defect image detecting means for capturing a plurality of teaching defects in advance and detecting a plurality of teaching defect images, and a plurality of defect images based on the plurality of teaching defect images detected by the defect image detecting means. Calculating means for classifying each of the teaching defects and assigning a category, a plurality of teaching defect images detected by the defect image detecting means, and a category classified by the calculating means for the plurality of teaching defects. A learning unit that obtains teaching data indicating a correspondence relationship between a plurality of teaching defects and a category corresponding to the type of the defect, and stores the teaching data in a storage unit. Image detecting means for picking up an image of a defect found by inspection and detecting the defect image of the inspection target, and storing the learning unit based on the defect image of the inspection target detected by the defect image detecting means. Stored in the means And a defect classification apparatus, characterized in that it consists of a classification unit with the teaching data and a calculation means for classifying a type of the defect.

Further, the present invention provides a defect image detecting means for picking up a plurality of teaching defects and detecting a plurality of teaching defect images in advance, and a plurality of teaching defect images detected by the defect image detecting means. Calculating means for calculating a plurality of desired feature quantities for each of the plurality of teaching defects; and positioning the teaching defect on a feature quantity space based on a relationship between the plurality of desired feature quantities calculated by the calculating means. Display means for displaying, and further displaying a category assigned to the teaching defect, acquiring teaching data indicating a correspondence relationship between a feature amount of a plurality of teaching defects and a category corresponding to the defect type. A learning unit for storing the defect in the storage unit and storing the defect in the storage unit; a defect image detection unit for capturing an image of a defect found by inspection of the inspection object to detect a defect image of the inspection target; Inspected object A classifying unit that calculates a desired feature amount of the defect from the defect image, and classifies the type of the defect from the teaching data stored in the storage unit of the learning unit based on the calculated desired feature amount of the defect; And a defect classification apparatus characterized in that: Further, the present invention provides a defect image detecting means for detecting a plurality of teaching defect images by imaging a plurality of teaching defects in advance, and a plurality of teaching defects from the plurality of teaching defect images detected by the defect image detecting means. Calculating means for calculating a plurality of desired feature amounts for each of the teaching defects, and calculating a category classification evaluation value for classifying each teaching defect from the relationship between the calculated desired plurality of feature amounts for each of the teaching defects. And display means for displaying the category classification evaluation value calculated by the calculation means for each teaching defect, and further displaying a category assigned to the teaching defect, wherein a plurality of feature amounts of the teaching defects are provided. A learning unit that acquires teaching data indicating a correspondence relationship between the data and a category corresponding to the type of defect and stores the data in a storage unit; defect An image is detected, a desired feature amount of the defect is calculated from the detected defect image of the inspection target, and stored in the storage unit of the learning unit based on the calculated desired feature amount of the defect. A defect classification device configured to classify the type of the defect from teaching data.

The present invention also relates to a defect classifying apparatus for classifying a defect type based on an image obtained by imaging a defect found in a defect inspection of an object to be inspected. A category assigning means for calculating a parameter and assigning a category to the teaching defect image based on the calculated classification parameter; a display means for displaying the teaching defect image assigned a category by the category assigning means; Storage means for acquiring and storing teaching data indicating a correspondence relationship between the teaching defect and a category corresponding to the type of the defect; Further, the present invention is a defect classification apparatus for classifying a defect type based on an image obtained by imaging a defect found in a defect inspection of an inspection object, wherein a classification parameter is calculated based on a teaching defect image. Calculating means for calculating a classification evaluation value (continuous value) for each category of the teaching defect image based on the calculated classification parameter; and a classification evaluation value for each category of the teaching defect image calculated by the calculation means. Is displayed, and storage means for acquiring and storing teaching data indicating a correspondence relationship between a plurality of teaching defects and categories corresponding to the types of the defects. Further, the present invention is a defect classification device that classifies a defect type based on an image obtained by imaging a defect found in the defect inspection of the inspection object,
A feature amount extracting unit that extracts a feature amount from the teaching defect image; a display unit that displays a teaching defect image located in a feature space based on the feature amount extracted by the feature amount extracting unit;
Storage means for acquiring and storing teaching data indicating a correspondence relationship between a plurality of teaching defects and a category corresponding to the defect type;

As described above, according to the above configuration,
Statistically significant teaching data can be efficiently created using the learning-type defect classifier, and the performance of the defect classifier can be utilized at a high level. As a result, the defect classification accuracy is improved, and The cause of the abnormality in the manufacturing process can be estimated with high accuracy, and the effect of improving the abnormality in the manufacturing process can be exerted.

[0015]

Embodiments of the present invention will be described with reference to the drawings. FIG. 1 is a diagram for explaining the role of a defect classification method and apparatus according to the present invention in a semiconductor manufacturing process. A semiconductor device product is manufactured on a substrate 1 such as a wafer through hundreds of manufacturing process steps 1 to n, and it may take about 100 days to complete. However, the quality of the semiconductor device product is determined in the probe inspection step 1t in which almost all the manufacturing processes are completed. For this reason, in order to improve the yield, means for estimating the quality of the process in the middle of the manufacturing process is indispensable. For this,
For example, the substrate 1 processed in the manufacturing process 1 is subjected to an external appearance inspection by an external inspection apparatus (an optical inspection apparatus or an inspection apparatus using an electron beam) 2, and the substrate 1 is processed. The quality of process processing is determined based on the appearance abnormality of the wiring pattern formed on the substrate. When an abnormality in the process processing is confirmed, it is necessary to take a countermeasure, but the defect classification device 3 as an information collecting means plays an important role. That is, although the location and number of defects can be grasped from the inspection result 4 of the visual inspection device 2, information about the contents cannot be obtained. Therefore, the defect classification device 3 classifies the type of the defect (whether it is a foreign substance, a pattern defect, a flaw, or other) based on the image 5 at the location of the defect obtained from the appearance inspection device 2. Then, in the quality management system 7,
By displaying the occurrence frequency 6 for each failure mode using the display means based on the classified information, the operator can narrow down the countermeasure candidates in the countermeasure candidate narrowing step 8 (a). By the way, as the manufacturing process step 1, a film forming step, a resist coating step, an exposure / developing step,
It comprises an etching step, a resist stripping step, and the like, and is constituted by a production line on which various processing apparatuses are installed. Therefore, there are various processing apparatuses that generate various types (categories) of defects.

In the embodiment shown in FIG. 1, since the frequency of occurrence of foreign matters is higher than that of pattern defects, it can be seen that measures to prevent the occurrence of foreign matters should be taken. That is, the operator performs the cause estimation step 8
In (b), the cause of the defect having a high frequency of occurrence is preferentially estimated, and in the countermeasure step 9, a countermeasure is taken against a process processing apparatus that executes a process process based on the estimated cause, thereby quickly. The yield can be improved. The defect classification apparatus 3 according to the present invention is an apparatus that automates the defect classification work conventionally performed visually, and requires high classification performance. Next, a defect classification apparatus and a classification procedure according to the present invention will be described with reference to FIGS. FIG. 2 is a configuration diagram showing one embodiment of the defect classification device according to the present invention.
FIG. 3 is a diagram for explaining a classification procedure performed by the defect classification device shown in FIG. FIG. 4 is a diagram showing an example of a teaching data creation procedure, a screen, and teaching data. FIG. 5 is a diagram for explaining the principle of learning.
The description will be made with reference to FIG. 2 according to the classification procedure of FIG. Reference is made to FIGS. 4 and 5 where appropriate.

The classification procedure is divided into a learning step shown in FIG. 3A and a classification step shown in FIG.

First, the learning step shown in FIG. 3A executed by the defect classification device according to the present invention will be described.
In the learning step, in step S31, defect images for creating teaching data are collected and stored in the image recording device (image storage device) 20, as will be specifically described below. That is, first, as shown in FIG. 2, the substrate 12 ′ on which a teaching defect for creating teaching data is formed by the substrate transport device 11 controlled by the substrate transport control unit 24.
Is mounted on the stage 13. On the other hand, the host computer 15 receives defect coordinate information corresponding to the teaching data substrate 12 ′ inspected and selected by the visual inspection device 2 or the like from the host system via the network 14. The host computer 15 sends an instruction to the stage controller 16 with reference to the received defect coordinate information,
3 to move the teaching defect on the substrate 12 'to the observation position. Then, the imaging device 18 such as a TV camera captures an image of the defect obtained via the optical system 17 and records an image signal of the captured defect on the image recording device 20 via the image input device 19. By repeating the above process for the designated defect, the teaching defect image signal for creating the teaching data is accumulated in the image recording device 20, and the teaching defect image signal is collected.
As described above, the substrate 12 'is for creating teaching data in advance.

Next, in step S32, teaching data is created (category A is assigned) by an operation of the host computer 15 by an operator. 4A shows the flow of the teaching data creation, FIG. 4B shows the screen of the monitor 23 in the teaching data creation, and FIG. 4C shows the teaching data stored in the storage device 25. It is an example of data. As shown in FIG. 4A, step S5
1, the operator operates the input means 22 such as a keyboard.
, The host computer 15 calls the defect image signal i = 1 for the teaching data recorded in the image recording device (image storage device) 20 in step S52. To be displayed on the monitor 23. At the same time, in the process of sequentially observing the defect images for teaching, in order to register similar defect images as the same category, the category and defect name (type of defect: foreign matter, pattern defect, scratch, etc.) Is input to the host computer 15 using the input means 22 and registered in the storage device 25. Thereby, the host computer 15 can display the category list 32 indicating the correspondence between the category and the defect name on the screen of the monitor 23. In step S53, as shown in FIG. 4B, the operator displays the teaching defect image i = 31 displayed on the monitor 23 screen.
1 is observed, and the category A (33 on the screen of the monitor 23) is inputted by the input means 22 for the teaching defect image i = 1 with reference to the category list (correspondence relationship between the category and the defect name) 32. The category A is stored in the storage device 25 for the defect number i = 1 indicating the teaching defect image. Until the processing is completed for all the teaching defects collected and recorded in the image recording device 20 in step S54, the teaching defect image i = i + 1 using the input unit 22 by the operator in step S55.
In step S52, the host computer 15 repeatedly calls the teaching defect image signal i = i + 1 recorded in the image recording device 20 and displays it on the screen of the monitor 23 in step S52. . Further, in step S53, the operator sequentially displays the teaching defect images i = i displayed on the monitor 23.
+1 is observed, and referring to the category list (correspondence relationship between categories and defect names) 32 displayed on the screen of the monitor 23, the input means 22 is sequentially input to the teaching defect image i = i + 1.
By inputting and assigning a category, the storage device 25 sequentially stores defect numbers i = i + 1 indicating defect images for teaching.

That is, as shown in FIG.
3 shows a teaching defect image 31 and a category list 3
2 is displayed, the operator selects a corresponding category number from the list 32 and inputs the category number using the category number input unit 33 while observing the teaching defect image 31. Here, the category is a number given to the type of defect, and is determined according to the type of defect. For example, the number of foreign substances is 1, the number of pattern defects is 2, the number of scratches is 3, and the others are 4. The category input by the operator using the input unit 22 is stored in the storage device 25 as teaching data in the host computer 15. here,
It should be noted that the category is not determined in advance, but is determined in the process of sequentially observing the teaching defect image i. That is, since the overall tendency of the teaching defect images collected in the image recording device 20 is unknown, it is necessary to register similar defects as the same category while observing the teaching defect images i in order. For this reason, the teaching data tends to include inconsistent data. Next, in step S33, the image processing apparatus 21 extracts various feature amounts 34 (including feature amount selection) for each teaching defect image i and provides them to the host computer 15 to make them correspond to the category 35. 4C is stored in the storage device 25 as teaching data shown in FIG. That is, the image processing device 21 reads each of the teaching defect image signals i stored in the image recording device 20, performs image processing on each of the read teaching defect image signals i, and executes the teaching defect image signal. Image i
Are extracted and provided to the host computer 15 and stored in the storage device 25. Here, the various feature amounts of the defect image include color information, shape, size, and the like of the defect image. For example, category 1 foreign matter is dark and nearly circular,
Category 2 pattern defects have the same color as the peripheral patterns and are complex in shape, and category 3 flaws have a small area but a long length. Therefore, the image processing device 2
1 is a predetermined type of feature quantity 1 to 5 (FIG. 4C)
In the example, the host computer 1 calculates 5 ways 34).
5, and stored in the storage device 25 as teaching data corresponding to the defect number. Here, it should be noted that a feature amount effective for classification cannot be uniquely determined. Defects vary depending on the product or process targeted by the defect classification device 3. For this reason, a large number of feature amounts (color information, shape, size, etc.) are prepared in a conceivable range, and only the feature amounts effective for classification of collected defects are selected and used. The method of selecting the feature amount will be described later. FIG.
The feature extraction in step S33 shown in FIG.

Next, at step S34, the host computer 15 determines at step S32 that the category A
Learning (calculation of classification parameters) is performed based on the teaching data to which is added, and the calculated classification parameters 26 are stored in the storage device 25. FIGS. 5A and 5B are diagrams for explaining the principle of learning. FIG. 5A shows step S3.
3 is a diagram in which teaching data to which a category A is assigned in FIG. 2 is plotted in a feature amount space. Here, × 42 is category 1, □ 43 is category 2, X1 is a feature amount 1 and X2 is a feature amount 2 axis. Learning performed in the host computer 15 means calculating a discriminant function between the categories given in step S32, and the function is referred to as a classification parameter 26. When the discriminant function is illustrated, 46 shown in FIG. 5B is a category boundary. Many methods are known for calculating the discriminant function. In this specification, a discriminant analysis, which is a typical method, will be described as an example. Hereinafter, description will be made by taking discriminant analysis as an example, but the present invention can be similarly realized in other methods.

Next, a discriminant analysis which is a method of calculating a discriminant function between categories, which is the classification parameter 26, based on a plurality of feature amounts will be described with reference to FIG.
In the figure, a point 47 represents a center of gravity of category 1, a point 48 represents a center of gravity of category 2, and an ellipse 49 represents a group of points having the same distance from the center of gravity. Here, the distance is a value obtained by standardizing the Euclidean distance in the feature amount space by the variance of the teaching data, and is called a Mahalanobis distance. The boundary line 46 is a set of points equidistant from the center of gravity of category 1 and category 2. The discriminant function between categories for the unknown point 50 (corresponding to a defect) is given by the following (Equation 1).

[0023]

Z1 · 2 = D1 ² −D2 ² where Z1 · 2: Discriminant function between category 1 and category 2 D1 ² : Mahalanobis square distance from the center of gravity of category 1 D2 ² : Mahalanobis with the center of gravity of category 2 Square distance For this reason, classification can be performed by the following (Equation 2) based on Z1 · 2.

[0024]

[Mathematical formula-see original document] Z1 * 2 <0 → belongs to category 1 Z1 / 2> 0 → belongs to category 2 Here, in FIG. 5 (a) and FIG. Note that a dimensional vector space is constructed.

Next, the substrate to be inspected (wafer) processed by the processing apparatus installed in the actually desired processing step according to the present invention is extracted in a substrate unit or a lot unit, and the appearance inspection is performed. A description will be given of the classification process shown in FIG. 3B executed by the defect classification device for the substrate to be inspected in units of substrates or lots which are inspected by the device 2 and determined to be defective by the appearance inspection device 2. In the classification step, in step S41, as will be described in detail below, a substrate to be inspected (wafer) actually processed by a processing apparatus in a desired processing step is processed in units of a substrate or lot. A defect image of the substrate to be inspected is collected and inspected by the visual inspection device 2 and determined as a defect by the visual inspection device 2 for each substrate or lot, and is collected in the image recording device (image storage device) 20. Is stored. That is, first, as shown in FIG.
The substrate 12 to be inspected, which is determined as a defect by the appearance inspection device 2 by the substrate transfer device 11 controlled by the above, is mounted on the stage 13 in units of substrates or lots. On the other hand, the host computer 15 receives, via the network 14, defect coordinate information corresponding to the inspected substrate 12 in which defects have been inspected by the visual inspection device 2 or the like on a substrate or lot basis from the host system. The host computer 15 sends an instruction to the stage control unit 16 with reference to the received defect coordinate information, and moves the stage 13 to move the defect on the inspection target substrate 12 to the observation position. The imaging device 18 such as a TV camera includes the optical system 1
An image of the defect obtained through the image pickup device 7 is captured, and an image signal of the captured defect is recorded in the image recording device 20 through the image input device 19. By repeating the above processing for the defect inspected by the visual inspection device 2, the defect image signal for classifying the category is stored in the image recording device 20 in correspondence with the defect number to be classified, and the defect to be classified is An image signal will be collected.

Next, in step S 42, the image processing device 21 extracts various feature amounts (including feature amount selection) from the defect image to be classified and provides the extracted images to the host computer 15. Is stored. That is, the image processing device 21 reads out the defect image signals to be classified stored in the image recording device 20 in association with the defect numbers to be classified, and performs image processing on each of the read defect image signals. Then, various characteristic amounts (referred to as classification data) of the defect image are extracted, provided to the host computer 15, and stored in the storage device 25. Here, the various feature amounts of the defect image include color information, shape, size, and the like of the defect image. The classification data stored in the storage device 25 is shown in FIG.
It has the same form as the teaching data of (c), but the category column 35 is blank. Therefore, in step S43, the host computer 15
Classification data (various feature amounts) for each defect stored in the
And the teaching data calculated by the learning process and indicated by the classification parameter (discrimination function between categories in the feature amount space) 26 stored in the storage device 25. The category (corresponding to the type of defect). Is performed to estimate the classification category B27, and the estimated classification category B (corresponding to the type of defect) 27 is stored in the storage device 25 in association with the defect number to be classified.

As described above, the stored classification category B2
7 is transferred to the quality management system 7 which is a higher-order system via the network 14, and is displayed in the quality management system 7 with respect to a failure mode (classification category B corresponding to the type of defect) as shown in FIG. Analysis such as narrowing down candidates and estimating the cause of occurrence of defects is performed. Learning type classification system (defect classification device) as described above
In 3, the teaching data creation is the most important. If the operator can only create inconsistent teaching data, the functions of the system cannot be fully utilized.
For example, on the screen shown in FIG. 4B, it is conceivable that the operator makes an input error in the teaching category 33 with respect to the teaching defect image 31. When an input error is made in this way, as shown in FIG. 5C, the data becomes teaching data in which the noise 53 is mixed in the cluster 52 separated into the category 1 and the category 2 in the feature amount space. Further, when the operator observes the teaching defect image 31 on the screen shown in FIG. 4B, for example, it may be confused whether the image belongs to category 1 or category 2. When the teaching category 33 is input by mistake as described above, the input category becomes inconsistent and becomes teaching data in which the boundary in the feature space intersects as shown in FIG. 5D. Further, when the operator is inexperienced and inputs the teaching category 33 for the teaching defect image 31, the categories become inconsistent and different categories are uniformly distributed in the feature amount space as shown in FIG. There is a possibility that inconsistent teaching data may be created. In such a case, the system does not operate normally and the defect classification performance is significantly reduced. In such a case, the conventional system does not provide a guide for what and how to improve the operator.

Next, in the process of creating teaching data by the operator according to the present invention, diagnosis / evaluation of teaching data and a function of supporting creation are provided, so that even an inexperienced operator can provide a high level of teaching data. A configuration in which the defect classification apparatus can be operated with the classification performance will be described with reference to FIGS. That is, the teaching data diagnosis function according to the first embodiment of the present invention will be described with reference to FIGS. FIG. 6 is a flowchart of the teaching data diagnosis function. FIG. 7 is an explanatory diagram of a screen for realizing the teaching data diagnosis function. The operation of the operator described below is performed through input means 22 such as a keyboard, and the result is displayed on the monitor 2.
3 is displayed. The operation result of the operator is also displayed on the monitor 23. Further, the processing for the operator's operation is performed by the host computer 15, and the result is stored in the storage device 25 or the like of the host computer 15.

The operation up to the final defect determination step S54 shown in FIG. 6 is the same as the operation flow for creating teaching data shown in FIG. In particular, the present invention is characterized in that diagnosis of the teaching data created once as described above is performed. Hereinafter, this diagnostic procedure will be described. First, in step S56, the host computer 15 selects a feature amount using the teaching data created up to step S54 and stored in the storage device 25. That is, the host computer 15 selects only the feature amount that effectively functions in classifying the teaching data from the feature amounts prepared in advance. The procedure for selecting a feature amount is known as, for example, a variable selection method in discriminant analysis. In the variable selection method, the statistical significance is determined by comparing the discrimination efficiencies before and after adding a variable (the Mahalanobis square distance between the centroids represented by the above-mentioned (Equation 1)).

Next, in step S57, the host computer 15 uses the selected feature amount to classify the classification parameter (discrimination function between categories in the feature amount space) 2
6 is calculated, and in step S58, the teaching data is classified based on the calculated classification parameters.

By the way, when good teaching data is created, the classification category B27, which is the classification result, should exactly match the category A given in advance by the operator. Here, there is a high possibility that the user has made a mistake in the teaching data in which the categories A and B do not match (is different). For this reason, the host computer 15
At 59, the teaching data in which the categories A and B do not match (is different) is displayed on the screen of the monitor 23 to prompt the operator to confirm. FIG. 7A shows an example in which the teaching data in which the categories A and B do not match is displayed. The feature value space 59 on the right side of the screen is illustrated for explanation. The image at the left end of the screen is displayed to refer to teaching defects (for example, a dark round defect and a bright star defect) in which the categories A and B match 1 or 2. The image in the middle of the screen and the image at the left end are teaching defects (for example, bright round shape defects, two kinds of bright polygonal defects, and dark star shape defects) whose categories do not match. In the numbers below the image, the left side 62 of → indicates category A, and the right side 63 of → indicates category B. In the example shown in the figure, the teaching defects 60 (a) and 61 (a) have a round shape, and the operator pays attention to the roundness and sets the category to 1, and the teaching defects 60 (b) and 61 (c) are complicated. The operator pays attention to the shape and the category 2
It is estimated that However, the host computer 1
In No. 5, the teaching defects 61 (a) and 60 (b) were determined to be category 2, and the teaching defects 60 (a) and 61 (c) were determined to be category 1. This is because in the process of creating the teaching data, the operator originally assigned the category noting the shape but the brightness of the teaching defect. In the feature amount space, a dark defect 60 (a) and a bright defect 61 (b) form a separated cluster, and a dark defect cluster 60 (a) has a bright defect 61 (a) and a bright defect cluster 60.
A dark defect 61 (c) is mixed in (b). This is noise generated by the user changing or confusing the classification criteria in the process of categorizing. Therefore, the host computer 15 recognizes these noises 61 (a) and 61 (a).
A value that deviates from the original classification parameter under the influence of (c) is output. Taking the discriminant analysis as an example, the variance of the teaching data increases due to the influence of noise, the degree of separation between the two groups decreases, and the intersection area 64 increases.

Therefore, in step S60, the operator corrects or corrects the category A using the input means 22 to the host computer 15 while looking at the screen of the monitor 23, as shown in FIGS. Delete from teaching data. That is, in order to correct the classification parameter (the discriminant function between categories in the feature space), as shown in FIG. 7B, in the defects 61 (a) and 61 (c) where the categories A and B do not match. The operator corrects the category A using the input means 22. Defect 61 (a)
Of the defect 61 (c) from 2 to 1, and the category of the defect 61 (c) from 1 to 2. FIG. 13C shows the result of correcting the category 65 of the defect which is a noise and repeating the learning sequence by the host computer 15 again. As a result, the variance of the teaching data was reduced, the degree of separation between the two groups was improved, and the classification performance was improved.

When the host computer 15 classifies the teaching data according to the re-determined classification parameters, the defects 61 (b) and 61 (d) are displayed on the screen of the monitor 23 as teaching data in which the categories A and B do not match. You. These are difficult to discriminate in either category from the defect appearance, and are not suitable as teaching data.
For this reason, the operator uses the input unit 22 to
By excluding these defects 66 from the teaching data as shown in (d), it is possible to reduce ambiguous determination and to select only defects having clear features as teaching data. As a result, as a result of learning and classifying again,
As shown in FIG. 7E, the variance of the teaching data is further reduced, and the second group 67 is clearly separated, so that a good classification parameter can be calculated. The above operation is repeated until a defective image in which the categories A and B do not match is not displayed on the screen. As described above, the above embodiment is characterized in that only a defect image corresponding to a specific condition is displayed on the screen. In other words, the teaching data must be presented to the operator without fail in order to make the result of human observation true. At this time, by selecting and displaying only the defects that may adversely affect the classification, the user can efficiently correct the teaching data.

Next, a teaching data evaluation function according to a second embodiment of the present invention will be described with reference to FIGS. FIG. 8 is a flowchart of the teaching data evaluation function. FIG. 9 is an explanatory diagram of a screen for realizing the teaching data evaluation function. The flowchart of FIG. 8 is similar to that of FIG. 6, and the difference from FIG. 6 is that a classification evaluation value calculation procedure (step S61) is included. Generally, in the classification process, the reliability of the classification is often quantitatively required. For example, in the discriminant analysis method, as described above, the probability of belonging to a category is quantitatively output by the discriminant function represented by Expression (1). As a result, the membership probability can be evaluated for each category. In general, in many cases, the category of a defect image cannot be clearly determined. In this case, the operator can create teaching data based on the membership probability calculated by the host computer (system) 15. For example, a certain defect image is one of categories 1 to 5,
It is assumed that the distance between categories 1 and 3 is smaller than the others and is substantially equal. As a result, the user can select the category from the above two types as a guideline instead of selecting the category from the five types.

FIG. 9A is an example of a screen displaying the result of classifying the teaching data. The upper part 69 (a), 6 of the figure
9 (b) and 69 (c) show images of the defect whose category A is category 1 given in advance by the operator.
(D), 69 (e), and 69 (f) display images of defects in which category A is category 2, which is given in advance by the operator. Here, in step S62, in the upper part, the classification evaluation value 70 of category 1 calculated quantitatively as the belonging probability of the category by the discriminant function by the host computer 15 is from left to right in ascending order (in descending order of belonging probability). Note that they are displayed side by side. Similarly,
In step S62, the lower row displays the category evaluation values 71 of category 2 in ascending order (in descending order of membership probability) from left to right. As shown in FIG.
In (c), the classification evaluation values are greatly reversed (Category 1
It can be quantitatively grasped that the category 2 evaluation value 0.2) is incorrect for the evaluation value 0.8. Although the classification evaluation value of the defect 69 (b) is reversed as compared with that of the category A, the difference between the categories is so small that it cannot be clearly separated. FIGS. 9B to 9E show a process in which the categories are corrected in the same order as in FIG. 7 and re-learned in step S60. Compared with FIG. 7, there is an advantage that the correction of the category A can be quantitatively determined by referring to the classification evaluation value. FIG.
9 (e), the belonging probability 71 (a) of category 1 is improved, and the effect of the teaching data correction can be quantitatively grasped.

In the second embodiment described above, in step S62, as shown in FIG.
Characteristically, (a) to 69 (f) are displayed on the screen of the monitor 23 with reference to the classification evaluation values 70 and 71. That is, a defect having a high classification evaluation value is a defect image representing the category of the teaching data created by the operator, and a defect having a low classification evaluation value is a guideline that is a vague defect as the teaching data. For this reason, even if a defect happens to coincide with the category A assigned by the operator in advance and the category B classified by the system, if the classification evaluation value is low, it can be determined that the defect is not appropriate as the teaching data. As described above, even in the case of the ambiguous defect image which had to be determined from the observation result of the defect image in the first embodiment, accurate determination can be performed by referring to the classification evaluation value. Next, a teaching data creation support function according to a third embodiment of the present invention will be described with reference to FIGS. In the first and second embodiments, the operator needs to assign the category A to the teaching data in advance. This is a method of correcting a once assigned category using a diagnosis function or an evaluation function. The third embodiment aims at suggesting the tendency of the teaching data before the operator assigns a category to assist the creation of the teaching data. This allows
It is possible to prevent assignment of an ambiguous classification category from the beginning, and it is possible to efficiently execute the condition setting of the defect classification device 3. FIG. 10 is a flowchart of the teaching data creation support function. The feature of the third embodiment is that the teaching-less clustering process is performed in step S63 before the creation of the teaching data shown in FIG. By performing the clustering without proper teaching, there is an effect that a feature amount that effectively contributes to the classification can be selected.

In the clustering without teaching, it is difficult to automatically determine the optimal clustering because there is no guideline for selecting the feature amount and the number of clusters is unknown. Therefore, interactive processing with the user is practical. FIG.
FIG. 1 shows an embodiment of a processing procedure of interactive clustering without teaching. In step S111, the host computer 15 calculates all feature amounts (m) in advance from the teaching data stored in the storage device 25. Next, step S
At 112, the operator inputs an arbitrary number of clusters to the host computer 1 using the input means 22 such as a keyboard.
Enter 5 Next, in step S114, the host computer 15 selects i features from among all the stored features (m) by inputting i features using the input means 22 to obtain mCi features. Create a quantity group. The host computer 15 determines in step S1
At 15, the principal component analysis is performed for each group, and i
Reduce the dimensional vector to j dimensions. Principal component analysis is to perform linear transformation of the original data. If the ratio of the eigenvalues of the principal components is high, the dimension can be reduced while reducing the loss of information. As a result, since clustering is performed on a j-dimensional vector, there is an advantage that the calculation time can be reduced.

Next, in step S116, the host computer 15 performs a clustering process in the j-dimensional feature space. This clustering process is a general method and various methods are known (Okuno et al., Multivariate Analysis Method, 1971, Nikkagirenren Shuppansha, pp391-
412). When mCi types of clustering results are obtained by the clustering process, the host computer 15
Evaluates the validity of these clustering results and ranks them in step S117. It is known that the evaluation of the clustering validity is better as the evaluation value of the following equation (3) is smaller.

[0039]

Where tr: trace of a matrix W: sum of variances in each cluster In step S119, the host computer 15 displays the clustering results on the monitor 23 in descending or ascending order of the evaluation values. Here, step S1
At 18, the host computer 15 reduces the dimensions of the j-dimensional space to two dimensions by principal component analysis for display. Next, in step S120, the operator
The displayed clustering result is visually checked to determine the validity of the cluster. As a criterion for judgment, as shown in FIG. 12, defects are displayed as separated clusters in the feature space, and significant defects are achieved by displaying and observing representative defects of each cluster on a monitor. That is being done. If it is determined in step S120 that the cluster is an invalid defect, in step S121, it is determined that the selected number of feature amounts i is inappropriate (NO). Is changed for the feature number selection number i. In addition, even if it is determined in step S121 that the feature amount selection number i is appropriate (YES), step S12 is performed.
In the case where the cluster is a bad defect in S2, S1
At 24, the number of clusters (for example, the number of clusters is changed from 2 to 3) is changed for the host computer 15 using the input unit 22, and the above process is repeated again.

As described above, when an appropriate cluster is generated in step S122, it is estimated that the selected i feature amounts contribute to the classification. Therefore, the host computer 15 learns based on the selected i feature amounts.

Next, a screen for realizing the teaching data creation support function will be described with reference to FIG. The host computer 15 configures a feature amount space as shown in FIG. 12 based on the feature amounts selected in the flow shown in FIG. 11, and plots teaching data. Here, displaying the teaching data on the screen serves as a guideline when the category A is assigned according to the distance in the feature amount space. The purpose of referring to a plurality of feature spaces is that the statistically significant classification does not always match the category desired by the operator. When three or more feature values are selected, a two-dimensional feature value space including the first principal component and the second principal component is displayed by principal component analysis. FIG. 12 shows two feature quantity spaces 85.
Is displayed. The number of displayed feature amount spaces can be specified by the user. One point 86 (□,
When (corresponding to a defect) is selected, □ 87 is displayed for the corresponding defect in the feature amount space 2 and a defect image 88 is displayed. This is referred to as category 1 (89). Next, when another point 90 (○) in the feature amount space 1 is selected, the corresponding defect in the feature amount space 2 is displayed as ○ 91 and the defect image 92 is displayed. □
Since (86, 87) and （(90, 91) belong to different clusters in the two feature amount spaces, it can be seen that ○ (90, 91) should be a different category from category 1. This is referred to as category 2 (93). In this case, $ 94 is located at the boundary between the two categories in the feature space 2, but $ 95 in the feature space 1 is close to ９０90. After confirming the displayed defect image 96, it can be set to category 2 (97).

In the third embodiment, the host computer (system) before the operator assigns a category
15 is characterized by displaying the feature amount space by verifying the statistical significance. Here, it should be noted that the image of the defect displayed on the screen is displayed on the screen corresponding to a point in the feature amount space. Even in the case of performing ambiguous categorization using only defect images, the degree of belonging to the category can be qualitatively grasped by referring to the distance in the feature space, and the teaching data can be obtained with high reproducibility. It can be created. By learning based on the teaching data created as described above, defect classification can be performed with high classification performance.

[0043]

According to the present invention, statistically significant teaching data can be efficiently created using a learning-type defect classifier, and the performance of the defect classifier can be utilized at a high level. As a result, the accuracy of defect classification is improved, and the cause of the abnormality in the manufacturing process can be estimated with high accuracy, which is effective in improving the abnormality of the manufacturing process. Further, according to the present invention, when creating accurate teaching data for classifying a type of a defect from an image signal of the defect in order to estimate a cause of occurrence of the defect, inconsistent teaching data given by a user is pointed out. By doing so, it is possible to obtain accurate teaching data, and as a result, it is possible to improve the accuracy of classifying defects, and to more accurately estimate the cause of abnormality in the manufacturing process.

[Brief description of the drawings]

FIG. 1 is a diagram for explaining the role of a defect classification method and apparatus according to the present invention in a semiconductor manufacturing process.

FIG. 2 is a configuration diagram showing one embodiment of a defect classification device according to the present invention.

FIG. 3 is a diagram for explaining a processing procedure performed by the defect classification device shown in FIG. 2;

FIG. 4 is a diagram illustrating an example of a teaching data creation procedure, a monitor screen, and teaching data.

FIG. 5 is a diagram for explaining the principle of learning in the learning process.

FIG. 6 is a flowchart illustrating a teaching data diagnosis function according to the first embodiment of the present invention.

FIG. 7 is an explanatory diagram of a monitor screen for realizing a teaching data diagnosis function according to the first embodiment of the present invention.

FIG. 8 is a flowchart illustrating a teaching data evaluation function according to a second embodiment of the present invention.

FIG. 9 is an explanatory diagram of a monitor screen for realizing a teaching data evaluation function according to a second embodiment of the present invention.

FIG. 10 is a flowchart illustrating a teaching data creation support function according to a third embodiment of the present invention.

FIG. 11 is an explanatory diagram of an example of an interactive teachingless clustering processing procedure;

FIG. 12 is an explanatory diagram of a monitor screen that realizes a teaching data creation support function according to a third embodiment of the present invention.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Substrate, 1t ... Probe inspection process, 2 ... Visual inspection device, 3 ... Defect classification device, 4 ... Visual inspection result, 5 ... Defect image, 6 ... Frequency of occurrence by failure mode, 7 ... Quality control system, 8 ( a) ... narrowing down of countermeasure candidates, 8 (b) ... estimation of cause, 9 ... countermeasure (countermeasure process), 11 ... board transfer device, 1
2, 12 ': substrate, 13: stage, 14: network, 15: host computer, 16: stage controller, 17: optical system, 18: TV camera, 19: image input device, 20: image recording device, 21 ... Image processing device, 22
... keyboard, 23 ... monitor, 25 ... storage device, 26 ...
Classification parameters, 27: Classification category B, 31: Defect image, 32: Category list, 33: Category number input unit, 34: Five kinds of feature values, 35: Category column.

Claims (13)

[Claims] 1. A teaching data creating method for classifying a defect type based on a defect image signal, wherein the teaching is performed by assigning a category corresponding to the defect type to a plurality of teaching defect images. Data for each of the teaching defect images, calculating a plurality of feature amounts for each of the teaching defect images, and discriminating between the categories in the feature amount space of the teaching defect image based on the assigned categories and the calculated feature amounts. Calculate the function, diagnose the teaching data consisting of the correspondence between the teaching defect image and the category using the discriminant function,
A teaching data creation method, characterized by displaying at least a teaching defect image having a possibility of deteriorating performance on a screen and correcting the teaching data. 2. A teaching defect image detecting step of detecting a plurality of teaching defect images by imaging a plurality of teaching defects in advance, and a plurality of teaching defect images detected in the teaching defect image detecting step. A category teaching step of displaying on the display means and assigning a category to the displayed plurality of teaching defect images to teach the correspondence between the teaching defect images and the categories; and a teaching defect image detecting step. A plurality of desired feature amounts are calculated for each of the plurality of teaching defects from the plurality of detected teaching defect images, and a feature amount space of the teaching defect image is calculated based on the assigned category and the calculated feature amounts. A discriminant function between categories is calculated, and each teaching defect is classified using the discriminant function to assign a category; and each teaching defect is taught in the category teaching process. Comparing the category with the category assigned in the category assigning process, and when the two categories are different, displaying at least the information on the display means, and, based on the displayed information, the category for the teaching defect is determined. Having a correction or exclusion teaching process of teaching the correspondence between the teaching defect and the category by excluding the correction or teaching defect,
A learning step of learning and acquiring teaching data indicating a correspondence relationship between a plurality of teaching defects and a category corresponding to the type of the defect; and capturing and inspecting a defect found by inspecting the inspection object. And a classifying step of classifying the type of the defect from the teaching data acquired in the learning step based on the detected defect image of the inspection target based on the detected defect image of the inspection target. Defect classification method. 3. A teaching defect image detecting step of imaging a plurality of teaching defects in advance and detecting a plurality of teaching defect images, and a plurality of teaching defect images detected in the teaching defect image detecting step. A category teaching step of displaying on the display means and assigning a category to the displayed plurality of teaching defect images to teach the correspondence between the teaching defect images and the categories; and a teaching defect image detecting step. A plurality of desired feature amounts are calculated for each of the plurality of teaching defects from the detected plurality of teaching defect images, and each of the plurality of desired defect amounts is calculated based on a relationship between the calculated plurality of desired feature amounts for each of the calculated teaching defects. A category assigning step of classifying defects and assigning a category; and comparing a category taught in the category teaching step with a category assigned in the category assigning step for each teaching defect. If Gori are different, and displayed on the display means at least that information, based on the information the display,
A correction or exclusion teaching process for correcting the category of the teaching defect or excluding the teaching defect to teach the correspondence between the teaching defect and the category, and corresponding to a plurality of teaching defect feature amounts and defect types. A learning step of learning and acquiring teaching data indicating a correspondence relationship with a category to be inspected, and detecting a defect image of the inspection target by capturing an image of a defect found by inspection with respect to the inspection object. Calculating a desired feature amount of the defect from the defect image of the inspection target,
A classifying step of classifying the type of the defect from the teaching data acquired in the learning step based on the calculated desired feature amount of the defect. 4. A teaching defect image detecting step of detecting a plurality of teaching defect images by capturing a plurality of teaching defects in advance, and a plurality of teaching defect images detected in the teaching defect image detecting step. A category teaching step of displaying on the display means and assigning a category to the displayed plurality of teaching defect images to teach the correspondence between the teaching defect images and the categories; and a teaching defect image detecting step. A plurality of desired feature amounts are calculated for each of the plurality of teaching defects from the detected plurality of teaching defect images, and a category is classified based on a relationship between the calculated plurality of desired feature amounts for each of the calculated teaching defects. A category assignment step of calculating parameters for performing the teaching, classifying each teaching defect based on the calculated classification parameter and assigning a category, and the category teaching step for each teaching defect. The instructed category is compared with the category assigned in the category assigning process. If the two categories are different, at least the information is displayed on the display means, and based on the displayed information, a defect for teaching is determined. A correction or exclusion teaching process for correcting the category or excluding the teaching defect and teaching the correspondence between the teaching defect and the category; and providing a feature amount of the plurality of teaching defects and a category corresponding to the defect type. A learning step of learning and acquiring teaching data indicating the correspondence, and detecting a defect image of the inspection target by imaging a defect found by inspection with respect to the inspection target, and detecting the defect image of the inspection target. Calculate the desired feature amount of the defect from the defect image of
A classifying step of classifying the type of the defect from the teaching data acquired in the learning step based on the calculated desired feature amount of the defect. 5. A feature amount for selecting a plurality of desired feature amounts from the calculated feature amounts when classifying a category for each teaching defect in the category assigning step in the learning step according to claim 3 or 4. A defect classification method comprising a selection step. 6. A teaching defect image detecting step of detecting a plurality of teaching defect images by imaging a plurality of teaching defects in advance, and a plurality of teaching defect images detected in the teaching defect image detecting step. A category teaching step of displaying on the display means and assigning a category to the displayed plurality of teaching defect images to teach the correspondence between the teaching defect images and the categories; and a teaching defect image detecting step. A plurality of desired feature amounts are calculated for each of the plurality of teaching defects from the plurality of detected teaching defect images, and a feature amount space of the teaching defect image is calculated based on the assigned category and the calculated feature amounts. A category classification evaluation value calculating step of calculating a category classification evaluation value for classifying each teaching defect by using the classification function, and The category classification evaluation value calculated in the category classification evaluation value calculation process is displayed on the display means, and based on the displayed category classification evaluation value, the category is corrected for the teaching defect or the teaching defect is excluded by excluding the teaching defect. Has a correction or exclusion teaching process for teaching the correspondence between a defect and a category, and learns and acquires teaching data indicating a correspondence between a feature amount of a plurality of teaching defects and a category corresponding to the type of the defect; A learning step, detecting a defect image of the inspection object by imaging the defect found by inspection of the inspection object, and calculating a desired feature amount of the defect from the detected defect image of the inspection object. ,
A classifying step of classifying the type of the defect from the teaching data acquired in the learning step based on the calculated desired feature amount of the defect. 7. A method according to claim 6, wherein when calculating a category classification evaluation value for classifying each teaching defect, a plurality of desired features are calculated from the calculated feature amounts. A defect classification method characterized by having a feature amount selecting step of selecting an amount. 8. A teaching defect image detecting step in which a plurality of teaching defects are imaged in advance to detect a plurality of teaching defect images, and a plurality of teaching defect images detected in the teaching defect image detecting step. A feature amount calculating step of calculating a plurality of feature amounts for each of the plurality of teaching defects, and positioning the teaching defect on a feature amount space based on a relationship between the plurality of feature amounts calculated in the feature amount calculating process. A category teaching process of displaying on the display means and assigning a category to the teaching defect in accordance with the position of the teaching defect in the displayed feature amount space, thereby teaching the correspondence between the teaching defect and the category; A learning step of learning and acquiring teaching data indicating a correspondence relationship between a feature amount of a plurality of teaching defects and a category corresponding to the type of the defect; About defects Imaging and detecting a defect image of the inspection target, calculating a desired feature amount of the defect from the detected defect image of the inspection target,
A classifying step of classifying the type of the defect from the teaching data acquired in the learning step based on the calculated desired feature amount of the defect. 9. A teaching defect image detecting step in which a plurality of teaching defects are imaged in advance to detect a plurality of teaching defect images, and a plurality of teaching defect images detected in the teaching defect image detecting step. A feature value calculating step of calculating a plurality of feature quantities for each of the plurality of teaching defects; a feature quantity selecting step of selecting a desired plurality of feature quantities from the feature quantities calculated in the feature quantity calculating step; The teaching defect calculated in the feature amount calculating step is located on a feature amount space based on a relationship between a plurality of desired feature amounts selected in the feature amount selecting step and displayed on a display means, and the displayed defect is displayed. A category teaching process of teaching a correspondence between the teaching defect and the category by assigning a category to the teaching defect in accordance with the position of the teaching defect in the feature amount space; Features and defect types A learning step of learning and acquiring teaching data indicating a correspondence relationship with a corresponding category; and detecting a defect image of the inspection target by imaging a defect found by inspection with respect to the inspection object. Calculate the desired feature amount of the defect from the defect image of the inspected object,
A classifying step of classifying the type of the defect from the teaching data acquired in the learning step based on the calculated desired feature amount of the defect. 10. A defect occurrence cause estimating step of estimating a defect occurrence cause based on a type of the defect classified in the classification step. Or 6 or 7 or 8 or 9
The described defect classification method. 11. A defect image detecting means for imaging a plurality of teaching defects in advance and detecting a plurality of teaching defect images, and a plurality of defect images for teaching based on the plurality of teaching defect images detected by said defect image detecting means. Calculating means for classifying each of the teaching defects and assigning a category thereto; a plurality of teaching defect images detected by the defect image detecting means; and a category classified for the plurality of teaching defects by the calculating means. A learning unit that obtains teaching data indicating a correspondence relationship between a plurality of teaching defects and a category corresponding to the type of the defect, and stores the teaching data in a storage unit. Defect image detecting means for picking up an image of a defect found by inspection and detecting a defect image of the inspection object; and storage means of the learning unit based on the defect image of the inspection object detected by the defect image detecting means. Stored in And a calculating unit for classifying the type of the defect from the teaching data. 12. A defect image detecting means for picking up a plurality of teaching defects and detecting a plurality of teaching defect images in advance, and a plurality of teaching defects from the plurality of teaching defect images detected by the defect image detecting means. Calculating means for calculating a plurality of desired feature quantities for each of the defects; and displaying and displaying the teaching defect on a feature quantity space based on a relationship between the plurality of desired feature quantities calculated by the calculating means. Display means for displaying a category assigned to the teaching defect, and acquiring and storing teaching data indicating a correspondence relationship between a plurality of feature amounts of the teaching defect and a category corresponding to the type of the defect. A defect image detecting means for imaging a defect found by inspection of the inspection object to detect a defect image of the inspection object; and a detection object detected by the defect image detection means. From the defect image A classifying unit configured to calculate a desired characteristic amount of the defect and classify the type of the defect from the teaching data stored in the storage unit of the learning unit based on the calculated desired characteristic amount of the defect; A defect classifying device. 13. A defect image detecting means for picking up a plurality of teaching defects in advance and detecting a plurality of teaching defect images, and a plurality of teaching defects from the plurality of teaching defect images detected by the defect image detecting means. Calculating means for calculating a plurality of desired feature amounts for each of the defects and calculating a category classification evaluation value for classifying each teaching defect from the relationship between the calculated plurality of desired feature amounts for each of the calculated teaching defects; Display means for displaying the category classification evaluation value calculated by the calculation means for each teaching defect, and further displaying a category assigned to the teaching defect, A learning unit that acquires teaching data indicating a correspondence relationship with a category corresponding to a type of a defect and stores the data in a storage unit; and a defect of the inspection target by capturing an image of a defect found by inspection of the inspection object. Picture Detecting an image, calculating a desired feature amount of the defect from the detected defect image of the inspection target,
A defect classifying unit configured to classify the type of the defect from the teaching data stored in the storage unit of the learning unit based on the calculated desired feature amount of the defect.