METHOD AND SYSTEM FOR SEMICONDUCTOR PROCESS CONTROL AND MONITORING BY USING PCA MODELS OF REDUCED SIZE

• US 20090276077A1
• Filed: 02/18/2009
• Published: 11/05/2009
• Est. Priority Date: 04/30/2008
• Status: Active Grant

• Current Assignee: GlobalFoundries, Inc.
• Original Assignee: GlobalFoundries, Inc.
• Inventors: Good, Richard, KOST, Daniel
• Granted Patent: US 8,103,478 B2
• Time in Patent Office: Days
• US Class Current: 700/110
• CPC Class Codes: G06F 18/2135 based on approximation criteria, e.g. principal component analysis
• CPC Class Codes: H01L 22/20 Sequence of activities consisting of a plurality of measurements, corrections, marking or sorting steps

Abstract

By dividing a complex set of parameters of a production process in forming semiconductor devices into individual blocks, respective PCA models may be established for each block and may thereafter be combined by operating on summary statistics of each model block in order to evaluate the complete initial parameter set. Thus, compared to conventional strategies, a significant reduction of the size of the combined PCA model compared to a single PCA model may be obtained, while also achieving an enhanced degree of flexibility in evaluating various subsets of parameters.

Claims

1. A method, comprising:

obtaining a plurality of historical measurement data sets, each of said plurality of historical measurement data sets related to a respective parameter set and measured during processing of semiconductor devices in a manufacturing environment;

establishing a model for each respective parameter set by using a principal component analysis technique and a respective one of said plurality of measurement data sets related to said respective parameter set;

obtaining a first measurement data set corresponding to a first parameter set of said parameter sets;

obtaining a second measurement data set corresponding to a second parameter set of said parameter sets;

applying a first model corresponding to said first parameter set to said first measurement data set;

applying a second model corresponding to said second parameter set to said second measurement data set; and

evaluating said first and second measurement data sets by combining a first statistical value set obtained from said first model and a second statistical value set obtained from said second model.

2. The method of claim 1, wherein said first and second sets of statistical values comprise at least a first metric for a model prediction error, a second metric for a model internal error, a third metric for the number of principal components and one or more metrics for characterizing eigenvector matrices and a correlation matrix.

3. The method of claim 1, wherein said first model is established in a first data processing unit and a second model is established in a second data processing unit.

4. The method of claim 3, further comprising storing said first and second sets of statistical values in a common database that is accessible by said first and second data processing units.

5. The method of claim 1, further comprising obtaining a combined measurement data set and selecting at least two or more of said models so as to adapt a combined model comprised of said at least two or more models to said combined set of measurement data.

6. The method of claim 2, further comprising determining a first limit and a second limit for said first and second metrics, respectively, by using the chi-square inverse function, the matrix of the non-principal components, the correlation matrix and the square of the correlation matrix of the first and second models.

7. The method of claim 6, wherein said one or more metrics for characterizing eigenvector matrices and a correlation matrix of each said model comprises a fourth metric obtained by said correlation matrix and a matrix represented by non-eigenvectors of each said model and a fifth metric obtained by the square of said correlation matrix and said matrix represented by non-eigenvectors of each said model.

8. The method of claim 7, wherein combining said first and second sets of statistical values for evaluating said first and second measurement data sets comprises determining statistical limits for a combined model corresponding to said combined first and second measurement data sets by using said first, second, third, fourth and fifth metrics of the first and second models.

9. The method of claim 8, wherein determining said statistical limits of said combined model comprises determining a first combined limit for a model prediction error of said combined model by using a sum of said fourth metrics and a sum of said fifth metrics of each said model used in said combined model.

10. The method of claim 9, further comprising determining a second limit for a model internal error of said combined model by using a sum of said third metrics of each said model used in said combined model.

11. The method of claim 10, further comprising determining a third limit by using said first and second limits, a sum of the fourth metrics of said first and second models, a sum of said fifth metrics of said first and second models and a sum of said third metrics of said first and second models.

12. A method of fault detection in a semiconductor manufacturing process sequence, the method comprising:

applying a first PCA model to a first set of measurement data related to a process result of said semiconductor manufacturing process sequence and corresponding to a first parameter set;

determining a first set of summary statistical values for evaluating said first parameter set;

applying a second PCA model to a second set of measurement data related to said process result of said semiconductor manufacturing process sequence and corresponding to a second parameter set;

determining a second set of summary statistical values for evaluating said second parameter set; and

combining said first and second sets of summary statistical values so as to commonly evaluate said first and second measurement data.

13. The method of claim 12, further comprising building said first model on the basis of first historical measurement data related to said first set of measurement data and building said second model on the basis of second historical measurement data related to said second set of measurement data and storing a first set of statistical key values of said first model and a second set of statistical key values of said second model in a database.

14. The method of claim 13, wherein said first model is built in a first data processing unit and said second model is built in a second data processing unit.

15. The method of claim 12, wherein said first and second sets of measurement data correspond to measurements performed on the same substrate after performing said manufacturing process, wherein said first and second parameters sets are different.

16. The method of claim 15, further comprising applying a third PCA model to a third set of measurement data corresponding to a third set of parameters, wherein said third set of measurement data relates to said process result and wherein a third set of summary statistical values is generated.

17. The method of claim 16, wherein said third set of parameters differs from said first and second parameters sets and said first, second and third sets of measurement data are obtained from the same substrate.

18. The method of claim 16, wherein said third set of parameters differs from said first and second parameters sets and said first, second and third sets of measurement data define a combined set of measurement data, and wherein said first, second and third models are applied separately to a combined set of measurement data of a plurality of substrates.

19. The method of claim 12, further comprising establishing a plurality of PCA models for a plurality of parameter sets measured after performing said manufacturing process, and evaluating at least a subset of said plurality of parameter sets and said first and second parameter set by selecting a combined model by using stored statistical key values of said first and second PCA models and at least one of said plurality of PCA models, wherein said combined model corresponds to said at least a subset.

20. A fault detection system, comprising:

a database comprising a plurality of PCA models and a corresponding set of statistical key values obtained by applying each PCA model to a respective set of measurement data corresponding to a respective set of process parameters to be monitored during the processing of semiconductor devices; and

a fault detection module connected to said database and configured to retrieve summary statistics of at least some of said PCA models and to combine said at least some summary statistics to provide a combined statistical evaluation of at least some of said parameter sets.