Method and its apparatus for classifying defects

• US 7,583,832 B2
• Filed: 02/28/2003
• Issued: 09/01/2009
• Est. Priority Date: 08/12/2002
• Status: Active Grant

• Current Assignee: Hitachi, Ltd.
• Original Assignee: Hitachi, Ltd.
• Inventors: Takagi, Yuji, Okuda, Hirohito, Honda, Toshifumi, Miyamoto, Atsushi
• Primary Examiner(s): Mariam; Daniel G
• Assistant Examiner(s): Seth; Manav
• Application Number:
  US10/378,016
• Publication Number:
  US 20040028276A1
• Time in Patent Office: 2,377 Days
• Field of Search: 382141-152, 382155-161, 382224-228, 348/61, 348/86, 348/91, 348/92, 348125-134, 348/87
• US Class Current: 382/141
• CPC Class Codes: G06T 7/001 using an image reference approach

Abstract

A highly reliable defect automatic classifying method and apparatus capable of flexibly coping with a request for classifying a defect given by each user without having to collect lots of teach data items. A classifying class arrangement is defined by a user by combining classes supplied by the system itself or classes defined by the user and, further, a priori knowledge on the defect class is given by the user as a restriction so as to carry out restricted learning.

Claims

1. A method for classifying defects with a classification system including a computer and imaging apparatus, the method comprising steps of:

obtaining with the imaging apparatus a defect image of a defect present in a sample;

calculating with the computer feature amounts of said defect from said defect image;

classifying said defect image with the computer into one of a plurality of classes based on said feature amounts and specified classification criteria to obtain a classification result of said defect image; and

displaying said classification result of said defect image,

wherein said specified classification criteria are specified by the following steps;

selecting a set of classification criteria from a plurality of classification criteria displayed at the classification system, the set of classification criteria being associated with a predetermined classifying rule stored at the computer,

modifying a class arrangement of the classification system based on input from a user by performing at least one of deleting, dividing or integrating the selected set of classification criteria with respect to the class arrangement, and

adapting the classifying rule automatically in response to the modified class arrangement so as to avoid an empty area in a feature space of the classification system corresponding to the classifying rule.

2. A method for classifying defects as claimed in claim 1 wherein said step of defining classification criteria includes arbitrarily combining predetermined classes.

3. A method for classifying defects as claimed in claim 1, wherein said step of assigning includes:

displaying said sample defect image on a display screen;

displaying said classes for classification on said display screen; and

specifying, on said display screen on which said sample defect images and said classes for classification are displayed, classes for classification that are to be associated with corresponding respective sample defect images.

4. A method for classifying defects as claimed in claim 1, further comprising a step of receiving, at the computer, a restriction condition to be given to a value of said feature amount, said restriction condition being specified by the user and corresponding to the information displayed at the classification system.

5. A method for classifying defects as claimed in claim 1, wherein in said step of classifying said defect image, a classifier is used comprising a rule-based classifier or a learning classifier or a combination thereof.

7. A method for classifying defects as claimed in claim 1, further comprising allocating the empty area to another class in the class arrangement of the computer.

8. A method for classifying defects as claimed in claim 7, wherein the empty area is allocated to a closest class in the feature space.

9. A method for classifying defects as claimed in claim 7, wherein the empty area is allocated based on learning with teach data.

10. A method for classifying defects as claimed in claim 1, further comprising:

detecting a redundant feature in the feature space when the class arrangement is modified; and

adjusting the feature space absent user input so as to delete the redundant feature.

6. A method of classifying defects with a classification system including a computer and imaging apparatus, comprising:

establishing a plurality of user-defined defect categories by altering a class arrangement of the classification system in response to a user input, wherein the user input directs at least one predetermined category of the classification system to be deleted, divided or integrated with respect to other predetermined categories of the classification system;

detecting a void in a feature space of the classification system based on the altered class arrangement;

updating classification rules associated with the plurality of user-defined categories to eliminate the void;

obtaining from a storage unit of the computer learning samples representative of defects occurring on a substrate;

for each defect category in the plurality of user-defined defect categories;

associating one or more of the learning samples with the defect category,

restricting at least one feature amount of the defect category in accordance with a corresponding user input, and

training a classifier with the learning samples such that feature amounts of the learning samples not within a feature space determined by restricting the at least one feature amount of the defect category are ignored by the classifier;

obtaining with the imaging apparatus a defect image of a defect present on the substrate;

calculating with the computer feature amounts of the defect image corresponding to a feature space defined by the plurality of defect categories;

classifying with the computer the defect image into a target category based on the calculated feature amounts to obtain a classification result; and

displaying at the classification system the classification result on a display screen,

wherein the classification result indicates that the defect image corresponds to a region of the feature space defined by the learning samples and restrictions of the target category.

11. A method for classifying defects as claimed in claim 6, further comprising allocating a region of the feature space corresponding to the void to another class in the class arrangement of the computer.

12. A method for classifying defects as claimed in claim 11, wherein the region is allocated to a closest class in the feature space.

13. A method for classifying defects as claimed in claim 11, wherein the region is allocated based on learning with teach data.

14. A method for classifying defects as claimed in claim 6, further comprising:

detecting a redundant feature in the feature space based on the altered class arrangement; and

adjusting the feature space absent user input so as to delete the redundant feature.