Training a learning based defect classifier

• US 10,713,534 B2
• Filed: 08/22/2018
• Issued: 07/14/2020
• Est. Priority Date: 09/01/2017
• Status: Active Grant

• Current Assignee: KLA-Tencor Corporation
• Original Assignee: KLA-Tencor Corporation
• Inventors: Brauer, Bjorn
• Primary Examiner(s): Azarian, Seyed H
• Application Number:
  US16/109,631
• Publication Number:
  US 20190073566A1
• Time in Patent Office: 692 Days
• Field of Search: 382100, 382103, 382108, 382141, 382144, 382149, 382156, 382168, 382173, 382181, 382219-220, 382254, 382275-276, 382289, 382305, 382321, 438 7, 438 18, 438 69, 438 77
• CPC Class Codes: G06F 18/217 Validation; Performance evaluation; Active pattern learning techniques
• CPC Class Codes: G06F 18/24143 Distances to neighbourhood prototypes, e.g. restricted Coulomb energy networks [RCEN]
• CPC Class Codes: G06T 2207/10061 from scanning electron microscope
• CPC Class Codes: G06T 2207/20081 Training; Learning
• CPC Class Codes: G06T 2207/20084 Artificial neural networks [ANN]
• CPC Class Codes: G06T 2207/30148 Semiconductor; IC; Wafer
• CPC Class Codes: G06T 7/0002 Inspection of images, e.g. flaw detection
• CPC Class Codes: G06V 10/764 using classification, e.g. of video objects
• CPC Class Codes: G06V 10/776 Validation; Performance evaluation
• CPC Class Codes: G06V 10/82 using neural networks
• CPC Class Codes: G06V 2201/06 Recognition of objects for industrial automation

Abstract

Methods and systems for training a learning based defect classifier are provided. One method includes training a learning based defect classifier with a training set of defects that includes identified defects of interest (DOIs) and identified nuisances. The DOIs and nuisances in the training set include DOIs and nuisances identified on at least one training wafer and at least one inspection wafer. The at least one training wafer is known to have an abnormally high defectivity and the at least one inspection wafer is expected to have normal defectivity.

Claims

1. A system configured to train a learning based defect classifier, comprising:

an inspection subsystem comprising at least an energy source and a detector, wherein the energy source is configured to generate energy that is directed to wafers, wherein the detector is configured to detect energy from the wafers and to generate output responsive to the detected energy, and wherein the wafers comprise at least one training wafer known to have an abnormally high defectivity and at least one inspection wafer expected to have normal defectivity; and

one or more computer subsystems configured for;

detecting defects on the at least one training wafer and the at least one inspection wafer by applying a defect detection method to the output generated for the at least one training wafer and the at least one inspection wafer, respectively;

identifying defects of interest on the at least one training wafer and the at least one inspection wafer by determining which of the defects detected on the at least one training wafer and the at least one inspection wafer, respectively, are the defects of interest;

identifying nuisances on the at least one training wafer and the at least one inspection wafer by determining which of the defects detected on the at least one training wafer and the at least one inspection wafer, respectively, are the nuisances;

generating a training set of defects comprising the identified defects of interest and the identified nuisances by combining information for the identified defects of interest and the identified nuisances;

identifying one or more locations on the at least one training wafer and the at least one inspection wafer at which none of the defects are detected;

adding images generated for the identified one or more locations to the training set of defects as one or more non-defective locations on the at least one training wafer and the at least one inspection wafer; and

training a learning based defect classifier with the training set of defects.

2. The system of claim 1, wherein the at least one training wafer and the at least one inspection wafer are formed using the same one or more fabrication processes except that the same one or more fabrication processes have produced the abnormally high defectivity on the at least one training wafer.

3. The system of claim 1, wherein the at least one training wafer comprises at least one wafer on which at least one fabrication process known to cause the abnormally high defectivity on wafers is performed.

4. The system of claim 1, wherein the at least one training wafer comprises at least one wafer having one or more materials formed thereon, and wherein the one or more materials are known to cause the abnormally high defectivity to be detected on the at least one training wafer.

5. The system of claim 1, wherein the at least one training wafer comprises at least one wafer on which at least one fabrication process is performed with one or more parameters of the at least one fabrication process known to be suboptimal.

6. The system of claim 1, wherein the one or more computer subsystems are further configured for artificially creating one or more examples of the defects of interest.

7. The system of claim 1, wherein said generating comprises selecting which of the identified defects of interest and the identified nuisances are included in the training set of defects such that the training set of defects comprises two or more of the defects of interest identified on the at least one training wafer, two or more of the defects of interest identified on the at least one inspection wafer, two or more of the nuisances identified on the at least one training wafer, and two or more of the nuisances identified on the at least one inspection wafer.

8. The system of claim 1, wherein said generating comprises selecting which of the identified defects of interest and the identified nuisances are included in the training set of defects such that the training set of defects comprises an equal number of the identified defects of interest and the identified nuisances.

9. The system of claim 1, wherein the one or more computer subsystems are further configured for applying the trained learning based defect classifier to the defects detected on the at least one training wafer or the at least one inspection wafer and not included in the training set of defects.

10. The system of claim 9, wherein the one or more computer subsystems are further configured for modifying the trained learning based defect classifier using results of applying the trained learning based defect classifier to the defects detected on the at least one training wafer or the at least one inspection wafer and not included in the training set of defects.

11. The system of claim 1, wherein the one or more computer subsystems are further configured for detecting defects on another inspection wafer by applying the defect detection method to the output generated for the other inspection wafer and applying the trained learning based defect classifier to the defects detected on the other inspection wafer.

12. The system of claim 11, wherein the one or more computer subsystems are further configured for modifying the trained learning based defect classifier using results of applying the trained learning based defect classifier to the defects detected on the other inspection wafer.

13. The system of claim 1, wherein the one or more computer subsystems are further configured for sampling the defects detected on the at least one training wafer or the at least one inspection wafer based on results generated by the trained learning based defect classifier.

14. The system of claim 1, wherein the one or more computer subsystems are further configured for detecting defects on another inspection wafer by applying the defect detection method to the output generated for the other inspection wafer, applying the trained learning based defect classifier to the defects detected on the other inspection wafer, and sampling the defects detected on the other inspection wafer based on results generated by the trained learning based defect classifier for the other inspection wafer.

15. The system of claim 14, wherein the one or more computer subsystems are further configured for creating another defect classifier based on the sampled defects.

16. The system of claim 15, wherein the other defect classifier is not a learning based type defect classifier.

17. The system of claim 1, wherein the one or more computer subsystems are further configured for detecting defects on other inspection wafers by applying the defect detection method to the output generated for the other inspection wafers, setting up a nuisance filter based on results generated by the trained learning based defect classifier, and applying the nuisance filter to information for the defects detected on the other inspection wafers.

18. The system of claim 1, wherein the trained learning based defect classifier is configured for separating the nuisances from the defects of interest, and wherein the one or more computer subsystems are further configured for detecting defects on another inspection wafer by applying the defect detection method to the output generated for the other inspection wafer and performing nuisance filtering for the other inspection wafer by inputting information for the defects detected on the other inspection wafer into the trained learning based defect classifier.

19. The system of claim 1, wherein the one or more computer subsystems are further configured for inputting information for defects detected on other inspection wafers into the trained learning based defect classifier, and wherein training the learning based defect classifier with the training set of defects reduces wafer-to-wafer variation in performance of the trained learning based defect classifier for the other inspection wafers.

20. The system of claim 1, wherein the trained learning based defect classifier is configured for separating the defects of interest from the nuisances.

21. The system of claim 1, wherein the trained learning based defect classifier is configured for separating the defects of interest from the nuisances and for separating different types of the defects of interest from each other.

22. The system of claim 1, wherein the one or more computer subsystems are further configured for detecting defects on other inspection wafers by applying the defect detection method to the output generated for the other inspection wafers and altering one or more parameters of an inspection recipe for the other inspection wafers based on results of applying the trained learning based defect classifier to one or more of the defects detected on the at least one training wafer, the defects detected on the at least one inspection wafer, or the defects detected on the other inspection wafers.

23. The system of claim 1, wherein the learning based defect classifier comprises a convolutional neural network.

24. The system of claim 1, wherein the inspection subsystem is configured as an optical inspection subsystem.

25. The system of claim 1, wherein the inspection subsystem is configured as an electron beam inspection system.

26. A non-transitory computer-readable medium, storing program instructions executable on a computer system for performing a computer-implemented method for training a learning based defect classifier, wherein the computer-implemented method comprises:

detecting defects on at least one training wafer and at least one inspection wafer by applying a defect detection method to output generated for the at least one training wafer and the at least one inspection wafer, respectively, by a detector of an inspection subsystem, wherein the at least one training wafer is known to have an abnormally high defectivity, wherein the at least one inspection wafer is expected to have normal defectivity, wherein the inspection subsystem comprises at least an energy source and the detector, wherein the energy source is configured to generate energy that is directed to the wafer, and wherein the detector is configured to detect energy from the wafer and to generate the output responsive to the detected energy;

identifying defects of interest on the at least one training wafer and the at least one inspection wafer by determining which of the defects detected on the at least one training wafer and the at least one inspection wafer, respectively, are the defects of interest;

identifying nuisances on the at least one training wafer and the at least one inspection wafer by determining which of the defects detected on the at least one training wafer and the at least one inspection wafer, respectively, are the nuisances;

generating a training set of defects comprising the identified defects of interest and the identified nuisances by combining information for the identified defects of interest and the identified nuisances; and

training a learning based defect classifier with the training set of defects.

27. A computer-implemented method for training a learning based defect classifier, comprising:

detecting defects on at least one training wafer and at least one inspection wafer by applying a defect detection method to output generated for the at least one training wafer and the at least one inspection wafer, respectively, by a detector of an inspection subsystem, wherein the at least one training wafer is known to have an abnormally high defectivity, wherein the at least one inspection wafer is expected to have normal defectivity, wherein the inspection subsystem comprises at least an energy source and the detector, wherein the energy source is configured to generate energy that is directed to the wafer, and wherein the detector is configured to detect energy from the wafer and to generate the output responsive to the detected energy;

identifying defects of interest on the at least one training wafer and the at least one inspection wafer by determining which of the defects detected on the at least one training wafer and the at least one inspection wafer, respectively, are the defects of interest;

identifying nuisances on the at least one training wafer and the at least one inspection wafer by determining which of the defects detected on the at least one training wafer and the at least one inspection wafer, respectively, are the nuisances;

generating a training set of defects comprising the identified defects of interest and the identified nuisances by combining information for the identified defects of interest and the identified nuisances; and

training a learning based defect classifier with the training set of defects, wherein said detecting, said identifying the defects of interest, said identifying the nuisances, said generating, and said training are performed by one or more computer subsystems coupled to the inspection subsystem.

28. The method of claim 27, wherein the at least one training wafer and the at least one inspection wafer are formed using the same one or more fabrication processes except that the same one or more fabrication processes have produced the abnormally high defectivity on the at least one training wafer.

29. The method of claim 27, wherein the at least one training wafer comprises at least one wafer on which at least one fabrication process known to cause the abnormally high defectivity on wafers is performed.

30. The method of claim 27, wherein the at least one training wafer comprises at least one wafer having one or more materials formed thereon, and wherein the one or more materials are known to cause the abnormally high defectivity to be detected on the at least one training wafer.

31. The method of claim 27, wherein the at least one training wafer comprises at least one wafer on which at least one fabrication process is performed with one or more parameters of the at least one fabrication process known to be suboptimal.

32. The method of claim 27, wherein the one or more computer subsystems are further configured for artificially creating one or more examples of the defects of interest.

33. The method of claim 27, further comprising identifying one or more locations on the at least one training wafer and the at least one inspection wafer at which none of the defects are detected and adding images generated for the identified one or more locations to the training set of defects as one or more non-defective locations on the at least one training wafer and the at least one inspection wafer.

34. The method of claim 27, wherein said generating comprises selecting which of the identified defects of interest and the identified nuisances are included in the training set of defects such that the training set of defects comprises two or more of the defects of interest identified on the at least one training wafer, two or more of the defects of interest identified on the at least one inspection wafer, two or more of the nuisances identified on the at least one training wafer, and two or more of the nuisances identified on the at least one inspection wafer.

35. The method of claim 27, wherein said generating comprises selecting which of the identified defects of interest and the identified nuisances are included in the training set of defects such that the training set of defects comprises an equal number of the identified defects of interest and the identified nuisances.

36. The method of claim 27, further comprising applying the trained learning based defect classifier to the defects detected on the at least one training wafer or the at least one inspection wafer and not included in the training set of defects.

37. The method of claim 36, further comprising modifying the trained learning based defect classifier using results of applying the trained learning based defect classifier to the defects detected on the at least one training wafer or the at least one inspection wafer and not included in the training set of defects.

38. The method of claim 27, further comprising detecting defects on another inspection wafer by applying the defect detection method to the output generated for the other inspection wafer and applying the trained learning based defect classifier to the defects detected on the other inspection wafer.

39. The method of claim 38, further comprising modifying the trained learning based defect classifier using results of applying the trained learning based defect classifier to the defects detected on the other inspection wafer.

40. The method of claim 27, further comprising sampling the defects detected on the at least one training wafer or the at least one inspection wafer based on results generated by the trained learning based defect classifier.

41. The method of claim 27, further comprising detecting defects on another inspection wafer by applying the defect detection method to the output generated for the other inspection wafer, applying the trained learning based defect classifier to the defects detected on the other inspection wafer, and sampling the defects detected on the other inspection wafer based on results generated by the trained learning based defect classifier for the other inspection wafer.

42. The method of claim 41, further comprising creating another defect classifier based on the sampled defects.

43. The method of claim 42, wherein the other defect classifier is not a learning based type defect classifier.

44. The method of claim 27, further comprising detecting defects on other inspection wafers by applying the defect detection method to the output generated for the other inspection wafers, setting up a nuisance filter based on results generated by the trained learning based defect classifier, and applying the nuisance filter to information for the defects detected on the other inspection wafers.

45. The method of claim 27, wherein the trained learning based defect classifier is configured for separating the nuisances from the defects of interest, and wherein the method further comprises detecting defects on another inspection wafer by applying the defect detection method to the output generated for the other inspection wafer and performing nuisance filtering for the other inspection wafer by inputting information for the defects detected on the other inspection wafer into the trained learning based defect classifier.

46. The method of claim 27, further comprising inputting information for defects detected on other inspection wafers into the trained learning based defect classifier, and wherein training the learning based defect classifier with the training set of defects reduces wafer-to-wafer variation in performance of the trained learning based defect classifier for the other inspection wafers.

47. The method of claim 27, wherein the trained learning based defect classifier is configured for separating the defects of interest from the nuisances.

48. The method of claim 27, wherein the trained learning based defect classifier is configured for separating the defects of interest from the nuisances and for separating different types of the defects of interest from each other.

49. The method of claim 27, further comprising detecting defects on other inspection wafers by applying the defect detection method to the output generated for the other inspection wafers and altering one or more parameters of an inspection recipe for the other inspection wafers based on results of applying the trained learning based defect classifier to one or more of the defects detected on the at least one training wafer, the defects detected on the at least one inspection wafer, or the defects detected on the other inspection wafers.

50. The method of claim 27, wherein the learning based defect classifier comprises a convolutional neural network.

51. The method of claim 27, wherein the inspection subsystem is configured as an optical inspection subsystem.

52. The method of claim 27, wherein the inspection subsystem is configured as an electron beam inspection system.