Defect classification method and apparatus, and defect inspection apparatus

A defect classification method to classify defects by using a classifier of instruction type on the basis of features of the defects extracted on the basis of detected signals acquired from a defect inspection apparatus, the defect classification method comprising:

• a priority order specification process for previously specifying target classification performance of purity and accuracy for each of defect classes, whole and in worst case, with priority order;
• and
• a classifier construction process for constructing the classifier of instruction type by evaluating whether the specified target sort performance is satisfied every item while comprehensively changing a learning algorithm and a learning parameter of the classifier of instruction type on the basis of instruction of defect classes and feature data respectively associated therewith beforehand, searching for a learning algorithm and a learning parameter that are favorable in a result of evaluation conducted every item, and setting a sort condition on the basis of learning using the learning algorithm and learning parameter obtained by the searching.

A defect classification apparatus to classify defects by using a classifier having a binary tree structure on the basis of features of the defects extracted on the basis of detected signals acquired from a defect inspection apparatus, the defect classification apparatus comprising:

• classifier construction means for constructing the classifier having the binary tree structure by setting a branch condition for each branch point in the binary tree structure on the basis of instruction of defect classes and feature data respectively associated therewith beforehand, the branch condition including defect classes respectively belonging to groups located on both sides of the branch point, a feature to be used for branch, and a discriminant reference.

A defect classification apparatus to classify defects by using a classifier of instruction type on the basis of features of the defects extracted on the basis of detected signals acquired from a defect inspection apparatus, the defect classification apparatus comprising:

• priority order specification means for previously specifying target classification performance of purity and accuracy for each of defect classes, whole and in worst case, with priority order;
• and
• classifier construction means for constructing the classifier of instruction type by evaluating whether the specified target classification performance is satisfied every item while comprehensively changing a learning algorithm and a learning parameter of the classifier of instruction type on the basis of instruction of defect classes and feature data respectively associated therewith beforehand, searching for a learning algorithm and a learning parameter that are favorable in a result of evaluation conducted every item, and setting a classify condition on the basis of learning using the learning algorithm and learning parameter obtained by the searching.

A defect inspection apparatus including a stage for retaining a substrate to be inspected, an image detection unit for detecting an image of the substrate to be inspected, a defect decision unit for detecting detects by comparing two corresponding areas formed to originally become same in shape, a feature extraction unit for extracting features of the defect on the basis of an image concerning each of the defects detected by the defect decision unit, and a defect classify unit to classify defects by using a classifier having a binary tree structure on the basis of the features of the defects extracted by the feature extraction unit, wherein the defect inspection apparatus comprises:

• classifier construction means for constructing the classifier having the binary tree structure used in the defect classifier by setting a branch condition for each branch point in the binary tree structure on the basis of instruction of defect classes and feature data respectively associated therewith beforehand, the branch condition comprising defect classes respectively belonging to groups located on both sides of the branch point, a feature to be used for branch, and a discriminant reference.

A defect inspection apparatus including a rotational stage for retaining a substrate to be inspected, a signal detection unit for applying laser light onto the substrate to be inspected, detecting scattered light and outputting a detected signal, a defect decision unit for detecting detects by comparing the detected signal detected by the signal detection unit with a predetermined threshold, a feature extraction unit for extracting features of each of the defects detected by the defect decision unit, and a defect classify unit to classify defects by using a classifier having a binary tree structure on the basis of the features of the defects extracted by the feature extraction unit, wherein the defect inspection apparatus comprises:

• classifier construction means for constructing the classifier having the binary tree structure used in the defect classifier by setting a branch condition for each branch point in the binary tree structure on the basis of instruction of defect classes and feature data respectively associated therewith beforehand, the branch condition comprising defect classes respectively belonging to groups located on both sides of the branch point, a feature to be used for branch, and a discriminant reference.