Automated defect spatial signature analysis for semiconductor manufacturing process

• US 5,982,920 A
• Filed: 01/08/1997
• Issued: 11/09/1999
• Est. Priority Date: 01/08/1997
• Status: Expired due to Term

• Current Assignee: Lockheed Martin Energy Research Corporation Oak Ridge National Laboratory
• Original Assignee: Lockheed Martin Energy Research Corporation Oak Ridge National Laboratory
• Inventors: Karnowski, Thomas P., Gleason, Shaun S., Sari-Sarraf, Hamed, Tobin, Kenneth W. Jr.
• Primary Examiner(s): Boudreau, Leo H.
• Assistant Examiner(s): Tadayon, Bijan
• Application Number:
  US08/780,569
• Time in Patent Office: 1,035 Days
• Field of Search: 382/144, 382/145, 382/146, 382/147, 382/149, 382/150, 382/308, 382/218, 382/259, 382/286, 382/293, 382/141, 382/142, 382/143, 382/148, 382/151, 382/152, 382/181, 382/190 , 382/195, 382/203, 382/209, 382/217, 382/224, 382/225, 348/87, 371/22.1, 364/490, 364/468.28, 702/35, 250/492.2, 250/559.42, 250/398, 438/17, 430/296, 324/158.1, 356/237
• US Class Current: 382/145
• CPC Class Codes: G06T 2207/30148 Semiconductor; IC; Wafer
• CPC Class Codes: G06T 7/0004 Industrial image inspection

Abstract

An apparatus and method for performing automated defect spatial signature alysis on a data set representing defect coordinates and wafer processing information includes categorizing data from the data set into a plurality of high level categories, classifying the categorized data contained in each high level category into user-labeled signature events, and correlating the categorized, classified signature events to a present or incipient anomalous process condition.

Claims

1. A method of performing automated defect spatial signature analysis on a data set representing defect coordinates and wafer processing information, comprising the steps of:

categorizing data from the data set into a plurality of high level event categories including global events, curvilinear events, amorphous events, and micro-structure events;

classifying the categorized data contained in each high level category into user-labeled signature events; and

correlating the categorized, classified signature events to a present or incipient anomalous process condition.

2. A method according to claim 1, wherein the categorizing step includes generating a grey-scale density image ρ

(x,y) from the data set and inputting the grey-scale density image to means for performing the classifying step.

3. A method according to claim 2, wherein the density image is a composite of several overlaying events within the data set.

4. A method according to claim 3, further comprising reducing the data set to simpler, non-overlapping or nearly non-overlapping sets that can be individually analyzed to assign the user-labeled signature events.

5. A method according to claim 1, wherein the global events category is defined as including any unclustered single pixel events, the curvilinear category is defined as including any curvilinear and radial clustered events associated with wafer spinning processes and mechanical surface damage, polishing anomalies, the amorphous category is defined as including nonlinear, tightly grouped cluster events arising from various processes, including lithography, etch, watermarks, and particle stains, and the microstructure category is defined as a collection of distributed single-pixel events which contain laser distributions of defects which reside on the image sub-pixel level and are related to small fine micro-scratching from chemical and mechanical polishing for wafer planarization.

6. A method of performing automated defect spatial signature analysis on a data set representing defect coordinates and wafer processing information, comprising the steps of:

forming a density image, ρ

(x,y), from the data set;

parsing the density image into two categories based on defect density values;

low-density, potentially random events and higher-density, potentially clustered events, thereby segmenting random defect distributions from other signature type events;

classifying the data contained in each category into user-labeled signature events; and

correlating the categorized, classified signature events to a present or incipient anomalous process condition.

7. A method according to claim 6, further comprising re-evaluating the low-density, random event image using a connectivity analysis to determine if some of the events should be removed and merged with the clustered-objects image.

8. A method according to claim 7, further comprising moving the initially random events back into the clustered image if connected groups of low-density defect pixels are found.

9. A method according to claim 6, further comprising binarizing the images for higher-density clustered events and processing the binarized imaged to account for nearest-neighbor relationships and relationships between large connected groups.

10. A method according to claim 9, wherein the processing includes applying a dilation technique to the clustered images so that neighboring objects are grouped into single objects, thereby pulling in some scattered defects around a single event.

11. A method according to claim 10, further comprising determining the features of the resulting image, wherein the features include area, elongation, compactness, and proximity, and then combining the measured features using fuzzy logic-type constraints.

12. A method according to claim 11, further comprising breaking the combined features into initial approximations of a curvilinear mask, M_C (x,y), and an amorphous mask, M_A (x,y).

13. A method according to claim 12, wherein the mask is a binary image used to filter out areas of the target image, whereby if the mask value is 0, areas are removed and a remainder of the target image where the value is 1 passes through, and wherein the masks are eventually used to generate three final event images on a display device coupled to a computer.

14. A method according to claim 13, further comprising directionally dilating objects based on surrounding structure prior to final feature measurement.

15. A method according to claim 14, wherein each object in the initial curvilinear mask, M_C (x,y), is measured to determine a direction of desired elongation for a selected object, based on the distance and size of neighboring objects.

16. A method according to claim 6, wherein the data set is wafer map data.

17. A method according to claim 6, wherein the data set is flat panel display data.

18. An apparatus for performing automated defect spatial signature analysis on a data set representing defect coordinates and wafer processing information, comprising:

means for categorizing data from the data set into a plurality of high level categories including global events, curvilinear events, amorphous events, and micro-structure events;

means for classifying the categorized data contained in each high level category into user-labeled signature events; and

means for correlating the categorized, classified signature events to a present or incipient anomalous process condition.

19. An apparatus according to claim 18, wherein the categorizing means includes means for generating a grey-scale density image ρ

(x,y) from the data set and means for inputtng the grey-scale density image to the classifying means.

20. An apparatus according to claim 19, wherein the density image is a composite of several overlaying events within the data set.

21. An apparatus according to claim 20, further comprising means for reducing the data set to simpler, non-overlapping or nearly non-overlapping sets that can be individually analyzed to assign the user-labeled signature events.

22. An apparatus according to claim 19, wherein the global events category is defined as including any unclustered single pixel events, the curvilinear category is defined as including any curvilinear and radial clustered events associated with wafer spinning processes and mechanical surface damage, polishing anomalies, the amorphous category is defined as including nonlinear, tightly grouped cluster events arising from various processes, including lithography, etch, watermarks, and particle stains, and the micro-structure category is defined as a collection of distributed single-pixel events which contain linear distributions of defects which reside on the image sub-pixel level and are related to small fine micro-scratching from chemical and mechanical polishing for wafer planarization.

23. An apparatus according to claim 19, further comprising means for forming a density image, ρ

(x,y), and parsing the density image into two categories based on defect density values;

low-density, potentially random events and higher-density, potentially, clustered events, thereby segmenting random defect distributions from other signature type events.

24. An apparatus according to claim 23, further comprising means for re-evaluating the low-density, random event image using a connectivity analysis to determine if some of the events should be removed and merged with the clustered-objects image.

25. An apparatus according to claim 24, further comprising means for moving the initial random events back into the clustered image if connected groups of low-density defect pixels are found.