Automated diagnosis using wafer tracking databases

• US 5,483,636 A
• Filed: 03/29/1995
• Issued: 01/09/1996
• Est. Priority Date: 02/03/1993
• Status: Expired due to Term

• Current Assignee: Texas Instruments, Inc.
• Original Assignee: Texas Instruments, Inc.
• Inventors: Saxena, Sharad
• Primary Examiner(s): Beausoliel, Jr., Robert W.
• Assistant Examiner(s): LE, DIEU MINH T
• Application Number:
  US08/415,081
• Time in Patent Office: 286 Days
• Field of Search: 371/15.1, 371/13, 371/29.1, 371/16.5, 371/21.1, 371/20.4, 371/22.1, 371/22.5, 371/25.1, 371/22.6, 371/40.1, 324/73.1, 324/158.1, 364/249, 364/267.4, 364/267.7, 364/282.1, 365/103, 365/114, 365/174, 365/212, 365/208, 395/575, 437/8
• US Class Current: 714/25
• CPC Class Codes: H01L 22/20 Sequence of activities consisting of a plurality of measurements, corrections, marking or sorting steps
• CPC Class Codes: H01L 2223/54473 for use after dicing
• CPC Class Codes: H01L 23/544 Marks applied to semiconductor devices or parts, e.g. registration marks, alignment structures, wafer maps test patterns for characterising or monitoring manufacturing processes H01L22/00
• CPC Class Codes: H01L 2924/00 Indexing scheme for arrangements or methods for connecting or disconnecting semiconductor or solid-state bodies as covered by H01L24/00
• CPC Class Codes: H01L 2924/0002 Not covered by any one of groups H01L24/00, H01L24/00 and H01L2224/00

Abstract

A system and method for isolating one or more causes of wafer misprocessing. A list of interesting queries (10) is generated. During wafer processing (15), processing parameters are measured (20) and a wafer tracking database (25) is created. The list of queries (10) may be filtered (30) before the queries are tested for interestingness. Interestingness is determined by outlier calculation (35) and trend analysis (40) on data stored in the wafer tracking database (25). Queries found to be interesting are displayed (50).

Claims

1. An automated method for isolating one or more causes of misprocessing in a semiconductor process comprising the steps of:

a. processing a plurality of wafers;

b. measuring a plurality of process parameters during said processing for each of said plurality of wafers;

c. creating a wafer tracking database which contains said plurality of process parameters and a plurality of identifying information associated with each wafer;

d. generating a first plurality queries wherein each query in said first plurality of queries corresponds to at least one of said plurality of process parameters;

e. applying each query of said first plurality of queries to said wafer tracking database to obtain a set of observations for each query of said first plurality of queries;

f. automatically determining from a pattern of each of said set of observations whether each query of said first plurality of queries is interesting for fault isolation;

g. dividing said first plurality of queries into a second plurality of queries wherein said second plurality contains queries determined to be interesting for fault isolation; and

h. displaying said second group of queries so that at least one cause of misprocessing can be determined.

2. The method of claim 1 wherein the step for determining whether each query of said first plurality of queries is interesting comprises determining whether each query contains outliers, wherein queries found to contain outliers are interesting.

3. The method of claim 2 wherein, for each query in said first plurality of queries, said step of determining whether each query contains outliers comprises the steps of:

a. converting each of said set of observations associated with the query to an integer so as to create a set of integers (y₁, y₂, . . . y_n);

b. calculating a first central tendency (c) according to a formula as follows;

##EQU4## wherein L defined to be Elias'"'"'s codelength; and

choosing the smallest D(x) as said first central tendency (c);

c. creating a set of deviations by encoding a difference between each integer in the query and said first central tendency according to a formula as follows;

space="preserve" listing-type="equation">α

.sub.i =L(|y.sub.i -c|),

wherein each encoded difference is a deviation in the set of deviations;

d. calculating a second central tendency from the set of deviations according to a formula as follows;

##EQU5## and, e. designating the deviations higher than said second central tendency as said outliers.

4. The method of claim 1 wherein the step for determining whether each query of said first plurality of queries is interesting comprises determining whether each query contains trends.

5. The method of claim 4, wherein the step of determining whether each query contains trends comprises the step of:

determining, for each query in said first plurality of queries, whether trends occur in the set of observations associated with said query by dividing the set of observations associated with the query into a plurality of subsets and checking each subset for a trend.

6. A method for isolating one or more causes of semiconductor misprocessing comprising the steps of:

a. creating a wafer tracking database for storing a plurality of process parameters and a plurality of identifying information associated with each wafer;

b. generating a first plurality of queries wherein each query is associated with at least one of said plurality of process parameters;

c. applying said first plurality of queries to said wafer tracking database to obtain a set of observations for each query of said first plurality of queries;

d. filtering said first plurality of queries to create a second plurality of queries having a first predetermined criterion;

e. detecting outliers in the set of observations for each of said second plurality of queries;

f. creating a third plurality of queries containing queries from said second plurality of queries having outliers;

g. detecting trends in the set of observations for each of said second plurality of queries;

h. adding queries from said second plurality of queries having a trend to said third plurality of queries;

i. filtering said third plurality of queries to create a fourth plurality of queries having a second predetermined criterion; and

j. listing said fourth plurality of queries.

7. The method of claim 6 wherein, for each query in said second plurality of queries, said step of detecting said outliers comprises the steps of:

a. converting each of said set of observations associated with the query to an integer so as to create a set of integers (y₁, y₂, . . . y_n);

b. calculating a first central tendency (c) according to a formula as follows;

##EQU6## and choosing a smallest D(x) as said first central tendency (c);

c. creating a set of deviations by encoding a difference between each integer in the query and said first central tendency according to;

space="preserve" listing-type="equation">α

.sub.i =L(|y.sub.i -c|),

wherein each encoded difference is a deviation in the set of deviations;

d. calculating a second central tendency from the set of deviations according to the formula;

##EQU7## and, e. designating the deviations higher than said second central tendency as said outliers.

8. The method of claim 6, wherein the step of detecting trends comprises:

determining, for each query in said second plurality of queries, whether trends occur in the set of observations associated with said query by dividing the set of observations associated with the query into a plurality of subsets and checking each subset for a trend.

9. The system of claim 6, wherein said first predetermined criterion is yield loss.

10. The system of claim 6, wherein said second predetermined criterion is interestingness score.

11. A system for extracting information from a database comprising:

a. a wafer tracking database for storing a plurality of parameters and a plurality of identifying information associated with a plurality of wafers;

b. a query generator for generating a plurality of queries wherein each of said plurality of queries is associated with at least one of said plurality of parameters; and

c. a query evaluator connected to said query generator and said wafer tracking database for applying said plurality of queries to said wafer tracking database to determine a set of observations and automatically determining from a pattern in said set of observations whether each of said plurality of queries is interesting for fault isolation.

12. A system for isolating one or more causes of semiconductor misprocessing comprising:

a. a wafer tracking database for storing a plurality of process parameters and a plurality of identifying information associated with a plurality of processed wafers;

b. a query generator connected to said wafer tracking database and operable to generate a first plurality of queries and apply said first plurality of queries to said wafer tracking database to generate a set of observations for each query; and

c. an outlier detector connected to the query generator operable to determine whether the set of observations for each of said first plurality of queries contains outliers as determined by the set of observations.

13. The system of claim 12, further comprising a trend detector operable to determine whether said set of observations for each of said first plurality of queries contains trends.

14. The system of claim 12, wherein said outlier detector is operable to create a second plurality of queries, said second plurality of queries containing queries from said first plurality of queries containing outliers.

15. The system of claim 14, further comprising a display for listing said second plurality of queries.

16. The system of claim 13, wherein said trend detector is operable to add queries from said first plurality of queries having a trend at a predetermined significance level to a second plurality of queries.

17. The system of claim 14, further comprising:

a. a first domain filter connected between said query generator and said outlier detector for eliminating queries from said first plurality of queries;

b. a trend detector operable to add queries from said first plurality of queries having a trend at a predetermined significance level to said second plurality of queries;

c. a second domain filter for creating a third plurality of queries from said second plurality of queries; and

d. a display for listing said third plurality of queries.

18. The system of claim 12, wherein, for each query in said first plurality of queries, said outlier detector comprises:

a. means for converting each of said set of observations associated with the query to an integer so as to create a set of integers (y₁, y₂, . . . y_n);

b. means for calculating a first central tendency (c) according to;

##EQU8## and choosing the smallest D(x) as said first central tendency (c);

c. means for creating a set of deviations by encoding the difference between each integer in the query and said first central tendency according to;

space="preserve" listing-type="equation">α

.sub.i =L(|y.sub.i -c|),

wherein each encoded difference is a deviation in the set of deviations;

d. means for calculating a second central tendency from the set of deviations according to the formula;

##EQU9## and, e. means for designating the deviations higher than said second central tendency as said outliers.

19. The system of claim 17, wherein, for each query in said first plurality of queries, said trend detector is operable to perform Kendall'"'"'s τ

-coefficient test on a plurality of subsets of said set of observations associated with the query.

20. A system for isolating one or more causes of semiconductor misprocessing comprising:

a. a wafer tracking database for storing a plurality of process parameters and a plurality of identifying information associated with each of said plurality of process parameters;

b. a query generator connected to said wafer tracking database for generating a first plurality of queries wherein each query is associated with at least one of said plurality of process parameters;

c. a first domain filter connected to said query generator for creating a second plurality of queries from said first plurality of queries based on a first predetermined criterion;

d. an outlier detector connected to said first domain filter operable to create a third plurality of queries containing queries from said second plurality of queries containing outliers;

e. a trend detector connected to said first domain filter operable to add queries from said second plurality of queries having a trend to said third plurality of queries;

f. a second domain filter connected to said trend detector for creating a fourth plurality of queries from said third plurality of queries based on a second predetermined criterion; and

,

g. a display connected to said second domain filter for listing said fourth plurality of queries.

21. The system of claim 20, wherein said first domain filter is a yield loss filter.

22. The system of claim 20, wherein said second domain filter is an interestingness score filter.

23. The system of claim 20, wherein, for each query in said first plurality of queries, said outlier detector comprises:

a. means for converting each of said set of observations associated with the query to an integer so as to create a set of integers (y_i, y₂, . . . y_n);

b. means for calculating a first central tendency (c) according to;

##EQU10## and choosing the smallest D(x) as said first central tendency (c);

c. means for creating a set of deviations by encoding the difference between each integer in the query and said first central tendency according to;

space="preserve" listing-type="equation">α

.sub.i =L(|y.sub.i -c|);

wherein each encoded difference is a deviation in the set of deviations;

d. means for calculating a second central tendency from the set of deviations according to;

##EQU11## and, e. means for designating the deviations higher than said second central tendency as said outliers.

24. The system of claim 20, wherein, for each query in said second plurality of queries, said trend detector is operable to perform Kendall'"'"'s τ

-coefficient test on a plurality of subsets of said set of observations associated with the query.