US 6,148,099 AGranted

Method and apparatus for incremental concurrent learning in automatic semiconductor wafer and liquid crystal display defect classification

AssigneeNeoPath, Inc.

InventorsShih-Jong J. Lee, Chih-Chau L. Kuan

Life of this patent

Estimating the term…

Loading patent…

Abstract

An incremental concurrent learning method starts with providing potential defects and fabrication information and a primary classification rule and secondary classification rule selection from a knowledge defect database from multiple products with different process cycles. The method then performs a truth inquiry to update a classification rule database for use by the primary classification rule and secondary classification rule selection. The method performs a primary defect classification and checks the confidence of the classification, and performs a secondary defect classification if the confidence is not high. If the confidence of the secondary defect classification is not high, a new defect may have been discovered and a novelty defect detection step is performed to define artifacts or potential new defect types to provide information for the truth inquiry.

Claims35

Dependent claims

A method for incremental concurrent learning in automatic defect classification comprising the steps of:

(a) providing defect images with truth labels;
(b) performing a context sensitive segmentation to produce a segmented image output;
(c) extracting a set of comprehensive features from a segmented image output to provide a feature output;
(d) performing automatic feature selection and transformation of the feature output having an application feature set output;
(e) automatic classification rule generation to provide classification rules for automatic feature selection and transformation and to provide initial classification rules;
(f) providing new potential defects;
(g) performing a context sensitive segmentation of the new potential defects;
(h) performing feature extraction of potential defects to provide extracted potential defects;
(i) performing feature classification of the extracted potential defects using the initial classification rules to provide defect classification;
(j) providing potential defects and fabrication information;
(k) providing a primary classification rule and secondary classification rule selection from a defect knowledge database from multiple products with different process cycles;
(l) performing primary defect classification having a classification output;
(m) checking a confidence of the classification output to decide whether to perform a secondary defect classification;
(n) if the confidence is not high, performing a secondary defect classification;
(o) checking the confidence of a secondary defect classification and if the confidence of the secondary defect classification is high, defining a classification outcome and providing the classification outcome to the primary classification rule and secondary classification rule selection step;
(p) if the confidence of the secondary defect classification is not high, questioning the classification outcome and sending potential defects to a novelty defect detection step;
(q) performing the novelty defect detection step to define artifacts or potential new defect types to provide information for a truth inquiry;
and
(r) performing a truth inquiry to update a classification rule database for use by the primary classification rule and secondary classification rule selection.

16 more elements

A method for incremental concurrent learning in automatic defect classification comprising the step of:

(a) providing defect images with truth labels;
(b) performing a context sensitive segmentation to produce a segmented image output;
(c) extracting a set of comprehensive features from a segmented image output to provide a feature output;
(d) performing automatic feature selection and transformation of the feature output having an application feature set output;
(e) automatic classification rule generation to provide classification rules for automatic feature selection and transformation and to provide initial classification rules;
(f) providing new potential defects;
(g) performing a context sensitive segmentation of the new potential defects;
(h) performing feature extraction of potential defects;
(i) performing feature classification of the extracted potential defects using the initial classification rules to provide defect classification;
(j) providing potential defects and fabrication information;
(k) selecting a multiple stage classification rule from a defect knowledge database from multiple products with different process cycles;
(l) performing multiple stage classification having at least one classification result and at least one confidence;
(m) checking the at least one confidence and if the confidence is high, defining a classification outcome and providing the classification outcome to the multiple stage classification rule selection step;
(n) if the at least one confidence is not high, questioning the classification outcome and sending the potential defects to the novelty defect detection module;
(o) performing a novelty defect detection step to define artifacts or potential new defect types to provide information for a truth inquiry;
and
(p) performing the truth inquiry to update a classification rule database for use by the multiple stage classification rule selection.

14 more elements

A method for incremental concurrent learning comprising the steps of:

(a) providing potential defects and fabrication information;
(b) providing a primary classification rule and secondary classification rule selection from a defect knowledge database from multiple products with different process cycles;
(c) performing a truth inquiry to update a classification rule database for use by the primary classification rule and secondary classification rule selection step;
(d) performing primary defect classification having a classification output;
(e) checking a confidence of the classification to decide whether to perform a secondary defect classification;
(f) checking the confidence of a secondary defect classification and if the confidence of the second defect classification is high, defining the classification outcome and providing the classification outcome to the primary classification rule and secondary classification rule selection step;
(g) if the confidence of the secondary defect classification is not high, questioning the classification outcome and sending the potential defects to a novelty defect detection step;
and
(h) performing the novelty defect detection step to define artifacts or a potential new defect type to provide information for the truth inquiry.

6 more elements

A method of incremental concurrent learning comprising the steps of:

(a) providing potential defects and fabrication information;
(b) providing a multiple classification rule selection from a defect knowledge database from multiple products with different process cycles;
(c) performing a truth inquiry to update a classification rule database for use by the multiple classification rule selection;
(d) performing multiple stage classification having at least one classification output and at least one confidence output;
(e) checking the at least one confidence and if the confidence of the at least one classification is high, defining the classification outcome and providing the classification outcome to the multiple classification rule selection step;
(f) if the at least one confidence is not high., questioning the classification outcome and sending the potential defects to the novelty defect detection module;
and
(g) performing a novelty defect detection step to define artifacts or potential new defect type to provide information for the truth inquiry.

5 more elements

A method of incremental concurrent learning comprising the steps of:

(a) providing potential defects and fabrication information;
(b) providing a classification rule selection from a defect knowledge database from multiple products with different process cycles;
(c) performing a truth inquiry to update a classification rule database for use by the classification rule selection;
(d) performing a single stage classification having at least one classification output and at least one confidence output;
(e) checking the at least one confidence and if the confidence of the at least one classification is high, defining the classification outcome and providing the classification outcome to the classification rule selection step;
(f) if the at least one confidence is not high, questioning the classification outcome and sending the potential defects to the novelty defect detection module;
and
(g) performing a novelty defect detection step to define artifacts or potential new defect type to provide information for the truth inquiry.

5 more elements

Dependencies are inferred from the claim text.