Optimization of unknown defect rejection for automatic defect classification

• US 9,715,723 B2
• Filed: 04/19/2012
• Issued: 07/25/2017
• Est. Priority Date: 04/19/2012
• Status: Active Grant

• Current Assignee: Applied Materials Israel Limited (Applied Materials Incorporated)
• Original Assignee: Applied Materials Israel Limited (Applied Materials Incorporated)
• Inventors: Shlain, Vladimir, Kaizerman, Idan, Rozenman, Efrat, Greenberg, Gadi
• Primary Examiner(s): Lefkowitz, Sumati
• Assistant Examiner(s): Wang, Carol
• Application Number:
  US13/451,500
• Publication Number:
  US 20130279795A1
• Time in Patent Office: 1,923 Days
• CPC Class Codes: G06F 18/2411 based on the proximity to a decision surface, e.g. support vector machines
• CPC Class Codes: G06F 18/2433 Single-class perspective, e.g. one-against-all classification; Novelty detection; Outlier detection
• CPC Class Codes: G06T 2207/10061 from scanning electron microscope
• CPC Class Codes: G06T 2207/20076 Probabilistic image processing
• CPC Class Codes: G06T 2207/20081 Training; Learning
• CPC Class Codes: G06T 2207/30148 Semiconductor; IC; Wafer
• CPC Class Codes: G06T 7/0004 Industrial image inspection
• CPC Class Codes: G06V 10/764 using classification, e.g. of video objects
• CPC Class Codes: G06V 10/987 with the intervention of an operator

Abstract

A method for defect classification includes storing, in a computer system, a definition of a region in a feature space. The definition is associated with a class of defects and comprises a kernel function comprising a parameter. The parameter determines a shape of the region. A confidence threshold for automatic classification of at least one defect associated with the class is received. A value of the parameter associated with the confidence threshold is selected. Inspection data for a plurality of defects detected in one or more samples under inspection is received. The plurality of defects for the class are automatically classified using the kernel function and the selected value of the parameter.

Claims

1. A method for defect classification comprising:

storing a definition of a region in a feature space, wherein the definition is associated with a defect class and comprises a kernel function, wherein the kernel function comprises a kernel parameter that is associated with a plurality of parameter values, wherein each of the parameter values for the kernel parameter is used to determine a different shape of the region based on the kernel function and corresponds to a different desired rejection rate for automatic classification of a plurality of unclassified defects for the defect class;

receiving, by a processing device, a confidence threshold that corresponds to a desired rejection rate for automatic classification of at least one defect associated with the defect class;

selecting, by the processing device, one of the parameter values of the plurality of parameter values for the kernel parameter that determines the shape of the region and corresponds to the desired rejection rate associated with the confidence threshold for automatic classification of the plurality of unclassified defects;

receiving, by the processing device, inspection data for the plurality of unclassified defects detected in one or more wafer samples under inspection;

automatically classifying, by the processing device, the plurality of unclassified defects for the defect class using the kernel function and the selected one of the plurality of parameter values of the kernel parameter that determines the shape of the region and corresponds to the desired rejection rate for the plurality of unclassified defects to generate a first plurality of classification results, wherein the plurality of unclassified defects are classified with a class label associated with the region based on a location of the unclassified defects in the region with the shape determined from the kernel parameter with the one of the plurality of parameter values that is selected based on the confidence threshold that corresponds to the desired rejection rate;

obtaining a second plurality of classification results for one or more of the plurality of unclassified defects for the defect class; and

combining the first plurality of classification results and the second plurality of classification results.

2. The method of claim 1, wherein selecting the one of the values of the kernel parameter comprises:

associating each of a plurality of confidence levels with a unique parameter value of the plurality of parameter values.

3. The method of claim 2, wherein associating each of the plurality of confidence levels comprises:

applying the kernel function to training data to obtain a set of training results using the plurality of parameter values; and

for each of the plurality of confidence levels, assigning an optimal parameter value to the confidence level based on the training results.

4. The method of claim 3, wherein assigning the optimal parameter value to the confidence level comprises:

selecting a parameter value for the confidence threshold to result in rejection of a subset of the plurality of defects corresponding to the confidence threshold during an automatic classification of the training data using the kernel function.

5. The method of claim 1, wherein the definition further is of a plurality of regions in the feature space, each of the regions in the feature space is defined by a one-class support vector machine comprising the kernel function, and wherein automatically classifying the plurality of unclassified defects comprises distinguishing between a plurality of first defects belonging to the class and a plurality of second defects that are classified as unknown defects with respect to the class using the one-class support vector machine.

6. The method of claim 5, wherein automatically classifying the plurality of unclassified defects comprises applying a multi-class support vector machine to assign each unclassified defect to a respective defect class, and identifying at least one non-decidable defect from the plurality of unclassified defects, wherein the non-decidable defect is in an overlap area between the plurality of regions of at least two of the defect classes in the feature space, and wherein the automatic classification is performed by a first inspection modality, and wherein the at least one non-decidable defect is provided to a second inspection modality to obtain the second plurality of classification results.

7. The method of claim 5, wherein the one-class support vector machine defines a mapping of the inspection data to a hypersphere in the feature space, and wherein the confidence level and the value of the kernel parameter defines a hyperplane that divides the hypersphere to create a hyperspherical cap, wherein the hyperspherical cap is mapped to a region in feature space containing the plurality of first defects classified as belonging to the class.

8. An apparatus comprising:

a memory to store a definition of a region in a feature space, wherein the definition is associated with a defect class and comprises a kernel function, wherein the kernel function comprises a kernel parameter that is associated with a plurality of parameter values, wherein each of the parameter values for the kernel parameter is used to determine a different shape of the region based on the kernel function and corresponds to a different desired rejection rate for automatic classification of a plurality of unclassified defects for the defect class; and

a processor, operatively coupled with the memory, to;

receive a confidence threshold that corresponds to a desired rejection rate for automatic classification of at least one defect associated with the defect class;

select one of the parameter values of the plurality of parameter values for the kernel parameter that determines the shape of the region and corresponds to the desired rejection rate associated with the confidence threshold for automatic classification of the plurality of unclassified defects;

receive inspection data for the plurality of unclassified defects detected in one or more wafer samples under inspection;

automatically classify the plurality of unclassified defects for the defect class using the kernel function and the selected one of the plurality of parameter values of the kernel parameter that corresponds to the desired rejection rate for the plurality of unclassified defects to generate a first plurality of classification results;

obtain a second plurality of classification results for one or more of the plurality of unclassified defects for the class; and

combine the first plurality of classification results and the second plurality of classification results, wherein the plurality of unclassified defects are classified with a class label associated with the region based on a location of the unclassified defects in the region with the shape determined from the kernel parameter with the one of the plurality of parameter values that is selected based on the confidence threshold that corresponds to the desired rejection rate.

9. The apparatus of claim 8, wherein the processor is to associate each of a plurality of confidence levels with a unique parameter value of the plurality of parameter values.

10. The apparatus of claim 9, wherein to associate each of the plurality of confidence levels, the processor is to apply the kernel function to training data to obtain a set of training results using the plurality of parameter values, and for each of the plurality of confidence levels, to assign an optimal parameter value to the confidence level based on the training results.

11. The apparatus of claim 10, wherein to assign the optimal parameter value to the confidence level, the processor is to select a parameter value for the confidence threshold to result in rejection of a subset of the plurality of defects corresponding to the confidence threshold during an automatic classification of the training data using the kernel function.

12. The apparatus of claim 8, wherein the region in the feature space is defined by a one-class support vector machine comprising the kernel function, and wherein to automatically classify the plurality of unclassified defects, the processing device is to distinguish between a plurality of first defects belonging to the class and a plurality of second defects that are classified as unknown defects with respect to the class using the one-class support vector machine.

13. The apparatus of claim 12, wherein to automatically classify the plurality of unclassified defects, the processor is to apply a multi-class support vector machine to assign each unclassified defect to a respective defect class, and to identify at least one non-decidable defect from the plurality of unclassified defects, wherein the non-decidable defect is in an overlap area between a plurality of regions of at least two of the defect classes in the feature space, and wherein the automatic classification is performed by a first inspection modality, and wherein the at least one non-decidable defect is provided to a second inspection modality to obtain the second plurality of classification results.

14. The apparatus of claim 12, wherein the one-class support vector machine defines a mapping of the inspection data to a hypersphere in the feature space, and wherein the confidence level and the value of the kernel parameter define a hyperplane that divides the hypersphere to create a hyperspherical cap, wherein the hyperspherical cap is mapped to a region in feature space containing the plurality of first defects classified as belonging to the class.

15. A non-transitory computer readable storage medium having instructions that, when executed by a processing device, cause the processing device to perform operations comprising:

storing a definition of a region in a feature space, wherein the definition is associated with a defect class and comprises a kernel function, wherein the kernel function comprises a kernel parameter that is associated with a plurality of parameter values, wherein each of the parameter values for the kernel parameter is used to determine a different shape of the region based on the kernel function and corresponds to a different desired rejection rate for automatic classification of a plurality of unclassified defects for the defect class;

receiving a confidence threshold that corresponds to a desired rejection rate for automatic classification of at least one defect associated with the defect class;

selecting one of the parameter values of the plurality of parameter values for the kernel parameter that determines the shape of the region and corresponds to the desired rejection rate associated with the confidence threshold for automatic classification of the plurality of unclassified defects;

receiving inspection data for the plurality of unclassified defects detected in one or more wafer samples under inspection;

automatically classifying the plurality of unclassified defects for the class using the kernel function and the selected one of the plurality of parameter values of the kernel parameter that determines the shape of the region and corresponds to the desired rejection rate for the plurality of unclassified defects to generate a first plurality of classification results, wherein the plurality of unclassified defects are classified with a class label associated with the region based on a location of the unclassified defects in the region with the shape determined from the kernel parameter with the one of the plurality of parameter values that is selected based on the confidence threshold that corresponds to the desired rejection rate;

obtaining a second plurality of classification results for one or more of the plurality of unclassified defects for the class; and

combining the first plurality of classification results and the second plurality of classification results.

16. The non-transitory computer readable storage medium of claim 15, wherein to select the one of the values of the kernel parameter, the processing device performs operations comprising:

associating each of a plurality of confidence levels with a unique parameter value of the plurality of parameter values.

17. The non-transitory computer readable storage medium of claim 16, wherein to associate each of the plurality of confidence levels, the processing device performs operations comprising:

applying the kernel function to training data to obtain a set of training results using the plurality of parameter values; and

for each of the plurality of confidence levels, assigning an optimal parameter value to the confidence level based on the training results.

18. The non-transitory computer readable storage medium of claim 17, wherein to assign the optimal parameter value to the confidence level, the processing device performs operations comprising:

selecting a parameter value for the confidence threshold to result in rejection of a subset of the plurality of defects corresponding to the confidence threshold during an automatic classification of the training data using the kernel function.

19. The non-transitory computer readable storage medium of claim 15, wherein the region in the feature space is defined by a one-class support vector machine comprising the kernel function, and wherein automatically classifying the plurality of unclassified defects comprises distinguishing between a plurality of first defects belonging to the class and a plurality of second defects that are classified as unknown defects with respect to the class using the one-class support vector machine.

20. The non-transitory computer readable storage medium of claim 19, wherein to automatically classify the plurality of unclassified defects, the processing device performs operations comprising:

applying a multi-class support vector machine to assign each unclassified defect to a respective defect class, and identifying at least one non-decidable defect from the plurality of unclassified defects, wherein the non-decidable defect is in an overlap area between a plurality of regions of at least two of the defect classes in the feature space, and wherein the automatic classification is performed by a first inspection modality, and wherein the at least one non-decidable defect is provided to a second inspection modality to obtain the second plurality of classification results.

21. The non-transitory computer readable storage medium of claim 19, wherein the one-class support vector machine defines a mapping of the inspection data to a hypersphere in the feature space, and wherein the confidence level and the value of the parameter define a hyperplane that divides the hypersphere to create a hyperspherical cap, wherein the hyperspherical cap is mapped to a region in feature space containing the plurality of first defects classified as belonging to the class.