Method and system for detection of tool performance degradation and mismatch

• US 8,725,667 B2
• Filed: 03/31/2009
• Issued: 05/13/2014
• Est. Priority Date: 03/08/2008
• Status: Active Grant

• Current Assignee: Tokyo Electron Limited
• Original Assignee: Tokyo Electron Limited
• Inventors: Kaushal, Sanjeev, Patel, Sukesh Janubhai, Sugishima, Kenji
• Primary Examiner(s): Chaki, Kakali
• Assistant Examiner(s): PELLETT, DANIEL T
• Application Number:
  US12/416,018
• Publication Number:
  US 20090240366A1
• Time in Patent Office: 1,869 Days
• Field of Search: None
• US Class Current: 706/14
• CPC Class Codes: G05B 13/0265 the criterion being a learning criterion
• CPC Class Codes: G05B 19/4065 Monitoring tool breakage, life or condition
• CPC Class Codes: G05B 2219/33041 Structure optimization and learning of artificial neural network by genetic algorithm
• CPC Class Codes: G05B 2219/37252 Life of tool, service life, decay, wear estimation
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 5/04 Inference or reasoning models
• CPC Class Codes: Y02P 90/80 Management or planning

Abstract

Autonomous biologically based learning tool system(s) and method(s) that the tool system(s) employs for learning and analysis of performance degradation and mismatch are provided. The autonomous biologically based learning tool system includes (a) one or more tool systems that perform a set of specific tasks or processes and generate assets and data related to the assets that characterize the various processes and associated tool performance; (b) an interaction manager that receives and formats the data, and (c) an autonomous learning system based on biological principles of learning. Objectively generated knowledge gleaned from synthetic or production data can be utilized to determine a mathematical relationship among a specific output variable and a set of associated influencing variables. The generated relationship facilitates assessment of performance degradation of a set of tools, and performance mismatch among tools therein.

Claims

1. A learning system, comprising:

a processor; and

a computer-readable storage medium storing computer executable components and communicatively coupled to the processor to facilitate operation of the computer executable components, the computer executable components comprising;

a component configured to select a set of variables associated with data generated based on a first unit of equipment in a tool system and a second unit of equipment in the tool system, wherein the tool system is configured to produce an asset, and wherein the set of variables includes a reference variable and a set of influencing variables that influences the reference variable;

an objective learning engine configured to learn autonomously a relationship among the reference variable and the set of influencing variables associated with the first unit of equipment and the second unit of equipment, wherein the objective learning engine is configured to remove an influencing variable from the set of influencing variables in response to a determination that a weight associated with the influencing variable is below a threshold value and a previous weight associated with the influencing variable is above the threshold value, and wherein the relationship among the reference variable and the set of influencing variables is autonomously adjusted in response to the removal of the influencing variable from the set of influencing variables in order to account for the removal of the influencing variable from the set of influencing variables; and

a performance assessment component configured to compare a measured value of the reference variable and a predicted value of the reference variable based on the relationship among the reference variable and the set of influencing variables to determine a performance degradation or a performance mismatch of the first unit of equipment and the second unit of equipment.

2. The system of claim 1, wherein the asset includes at least one semiconductor device, the at least one semiconductor device comprising at least one of a plasma-discharge based flat panel display (FPD), an organic light emitting diode (OLED)-based FPD, a liquid crystal display (LCD)-based FPD, or elements thereof which comprise at least one of a thin-film-transistor active matrix array, a color filter, or a polarizer.

3. The system of claim 1, wherein the performance assessment component is configured to generate a ranking of the set of influencing variables associated with the performance degradation or the performance mismatch.

4. The system of claim 3, wherein the data includes at least one of manufacturing run data, synthetic manufacturing data, or test manufacturing data.

5. The system of claim 1, wherein the data selector is configured to select the reference variable via at least one of an automated analysis or an actor input.

6. The system of claim 5, wherein the automated analysis includes an arrangement of candidate reference variables in accordance with a variation magnitude of each of the candidate reference variables in a manufacturing process of the asset.

7. The system of claim 6, wherein the data selector is further configured to select at least one of a low variation variable, a medium variation variable, or a high variation variable.

8. The system of claim 1, wherein the data selector is further configured to select a set of influencing variables based at least in part on one of a universal selection, a semi-universal selection, a knowledge based selection, an empirical selection, or an actor-driven selection.

9. The system of claim 4, wherein the data selector includes a formatter component configured to convey selected data in a matrix representation.

10. The system of claim 9, wherein the matrix representation includes a reference variable and at least one variable in the set of influencing variables associated with the reference variable.

11. The system of claim 10, wherein the data selector includes a component configured to select reference data from the data associated with the set of units of equipment in the tool system that produces the asset, the reference data is at least a portion of the data.

12. The system of claim 11, wherein the data selector is further configured to select reference data via an actor input.

13. The system of claim 12, wherein the formatter component is further configured to generate a training matrix based at least in part on the selected reference data.

14. The system of claim 13, wherein the data is collected at a predetermined sampling rate.

15. The system of claim 14, wherein the reference data includes data from a group of non-contiguous steps in an ordered set of steps for the asset.

16. The system of claim 15, wherein the reference data includes average data that originates from at least one of a step level, a wafer level, a lot level, an active matrix array manufacturing, or a wet-clean cycle.

17. The system of claim 16, wherein the reference data includes average data that originates in a predetermined cycle within a process to manufacture the asset, the asset comprises a flat panel display device.

18. The system of claim 14, wherein the training matrix is generated for one or more instances of a process to manufacture the asset, the one or more instances include at least one of a step in the ordered set of steps for the asset, a wafer run in a set of tools, a lot run in a set of tools, or a set of cycles within the process.

19. The system of claim 3, wherein the relationship among the reference variable and the set of influencing variables for the selected data is a mathematical functional relationship, wherein the mathematical functional relationship is autonomously learned without reliance on an external bias.

20. The system of claim 19, wherein the mathematical functional relationship conveys a difference among a measured value of the reference variable and a predicted value thereof.

21. The system of claim 20, wherein the mathematical function relationship conveys a variation of the difference among a measured value of the reference variable and a predicted value thereof.

22. The system of claim 19, wherein the selected data is production data for at least one instance in a process to manufacture the asset.

23. The system of claim 19, wherein the performance assessment component includes an analysis component configured to summarize a distribution of differences between a measured value and a predicted value through a set of statistical-mechanics functions.

24. The system of claim 23, wherein a set of actor-supplied key performance indicator thresholds are utilized to determine at least one of an abnormal performance degradation or an abnormal performance mismatch.

25. The system of claim 23, wherein at least one of the abnormal performance degradation or the abnormal performance mismatch is determined for at least one of a reference tool or an instrument thereof.

26. The system of claim 24, wherein the analysis component includes a formal analysis component configured to determine the ranking of the set of influencing variables through unsupervised theorem proving algorithms and expression analysis of the mathematical functional relationship.

27. The system of claim 26, wherein the analysis component includes a numerical analysis component configured to establish the ranking of the set of influencing variables through computation of a gradient vector of the mathematical functional relationship for the set of influencing variables.

28. The system of claim 27, wherein at least one of the performance degradation or a performance mismatch of the first unit of equipment and the second unit of equipment is profiled according to the ranking of the set of influencing variables and a timestamp associated with the performance degradation or the performance mismatch.

29. A method, comprising:

measuring a value of a reference variable associated with a first tool that manufactures at least one asset and a second tool that manufactures the at least one asset;

measuring a value for each variable in a set of influencing variables that influences the reference variable;

predicting a value for the reference variable based on a learned formal relationship among the reference variable and the set of influencing variables associated with the first tool that manufactures the at least one asset and the second tool that manufactures the at least one asset;

removing an influencing variable from the set of influencing variables in response to a determination that a weight associated with the influencing variable satisfies a function of a threshold value and a previous weight associated with the influencing variable does not satisfy the function of the threshold value;

adjusting the learned formal relationship among the reference variable and the set of influencing variables in response to the removing of the influencing variable from the set of influencing variables in order to account for the removing of the influencing variable from the set of influencing variables; and

reporting at least one of performance degradation of at least one tool or performance mismatch amongst the first tool that manufactures the at least one asset and the second tool that manufactures the at least one asset in response to a difference among the predicted value for the reference variable and the measured value for the reference variable being above a performance metric threshold.

30. The method of claim 29, wherein the reporting includes generating a profile that is associated with the performance degradation or the performance mismatch and includes the set of influencing variables ordered according to a ranking criterion;

and retaining the profile for the set of influencing variables and learned functions in response to a difference among the predicted value for the reference variable and the measured value for the reference variable being above a performance metric threshold.

31. The method of claim 30, wherein reporting at least one of the performance degradation of the at least one tool or the performance mismatch includes conveying performance information to an actor, the performance information comprises at least one of the profile or the set of influencing variables ordered according to the ranking criterion with a time stamp associated with the performance degradation or the mismatch event.

32. The method of claim 31, wherein generating the profile associated with the performance degradation includes performing a sensitivity analysis based at least in part on at least one of a percentage impact of the set of influencing variables on a predicted value for the reference variable, or a numeric derivative assessment of the learned function of the set of influencing variables.

33. The method of claim 29, further comprising

receiving a set of data associated at least in part with the manufacture of the at least one asset;

selecting a set of reference data for a set of variables; and

determining the reference variable from the set of variables and selecting the set of influencing variables.

34. The method of claim 33, wherein the at least one asset include at least one of a semiconductor device, the semiconductor device comprising at least one of a plasma-discharge based flat panel display (FPD), an organic light emitting diode (OLED)-based FPD, or a liquid crystal display (LCD)-based FPD.

35. The method of claim 34, wherein the set of data includes at least one of production run data, synthetic manufacturing data, or test manufacturing data.

36. The method of claim 35, further comprising formatting the received set of data, wherein formatting the received data includes generating a matrix representation for at least one of the test manufacturing data or the production run data.

37. The method of claim 36, wherein the matrix representation of the test manufacturing data is generated at one or more instants during operation of the first tool that manufactures the at least one asset and the second tool that manufactures the at least one asset.

38. The method of claim 36, wherein the matrix representation of the production run data is generated at one or more instants during operation of the first tool that manufactures the at least one asset and the second tool that manufactures the at least one asset.

39. The method of claim 36, wherein generating the matrix representation for at least one of test manufacturing data or the production run data is generated at one or more production instances, the one or more production instances include at least one of a step level in one or more recipes, or a plurality of steps either contiguous or disjointed in the one or more recipes.

40. The method of claim 36, wherein generating the matrix representation for at least one of the test manufacturing data or the production run data is generated via received data at the asset level.

41. The method of claim 33, wherein selecting the set of reference data for the set of variables includes at least one of selecting a portion of the received data or receiving input from an actor.

42. The method of claim 41, wherein the set of reference data for the set of variables includes data from the first tool that manufactures the at least one asset and the second tool that manufactures the at least one asset, and wherein the data is collected at a predetermined sampling rate.

43. The method of claim 42, wherein the set of reference data for the set of variables includes data from a group of non-contiguous steps in one or more recipes to produce the at least one asset.

44. The method of claim 43, wherein the set of reference data for the set of variables includes average data that originates from at least one of a step level, a wafer level, a lot level, an active matrix array manufacturing, or a wet-clean cycle.

45. The method of claim 33, wherein selecting the set of influencing variables is based at least on one of a universal selection, a semi-universal selection, a knowledge based selection, an empirical selection, or an actor-driven selection.

46. The method of claim 29, further comprising updating a learned mathematical function in response to the removing the influencing variable from the set of influencing variables.

47. A non-transitory computer-readable storage medium comprising a set of instructions that, when executed by a computer, cause the computer to perform the following acts:

measuring a set of values of a reference variable associated with a first tool for manufacturing an asset and a second tool for manufacturing the asset;

measuring a set of values for each variable in a set of influencing variables that influences the reference variable;

learning autonomously a formal relationship among the reference variable and the set of influencing variables associated with the first tool for manufacturing the asset and the second tool for manufacturing the asset;

predicting a set of values for the reference variable based on the learned formal relationship;

removing an influencing variable from the set of influencing variables in response to a determination that a first weight associated with the influencing variable is above a threshold value and a second weight associated with the influencing variable is below the threshold value;

modifying the formal relationship among the reference variable and the set of influencing variables in response to the removing of the influencing variable from the set of influencing variables in order to account for the removing of the influencing variable from the set of influencing variables; and

reporting performance degradation of at least a tool or performance mismatch of the first tool for manufacturing the asset and the second tool for manufacturing the asset when a difference among a predicted value of the reference variable and a measured value of the reference variable is above a key performance indicator threshold.

48. The non-transitory computer-readable storage medium of claim 47, wherein the reporting includes ordering the set of influencing variables according to a ranking criterion relevant to the performance degradation or the performance mismatch.

49. An apparatus, comprising:

means for identifying a set of variables and data associated with a first unit of equipment in a tool system that produces an asset and a second unit of equipment in the tool system that produces the asset, wherein the set of variables includes a reference variable and a set of influencing variables that influences the reference variable;

means for autonomously learning a functional relationship among the reference variable and the set of influencing variables associated with the first unit of equipment and the second unit of equipment;

means for removing a selected influencing variable from the set of influencing variables in response to a determination that a weight associated with the selected influencing variable is below a set value and a previous weight associated with the selected influencing variable is above the set value;

means for autonomously updating the learned functional relationship among the reference variable and the set of influencing variables in response to the removing of the selected influencing variable from the set of influencing variables in order to account for the removing of the selected influencing variable from the set of influencing variables; and

means for assessing at least one of performance degradation or performance mismatch of the first unit of equipment and the second unit of equipment in the tool system based at least in part on values of the reference variable predicted through the learned functional relationship.

50. The apparatus of claim 49, wherein the means for assessing includes means for generating a relevancy profile associated with the performance degradation or the performance mismatch which includes the set of influencing variables ordered according to a ranking criterion.