Semiconductor yield management system and method

• US 8,380,472 B2
• Filed: 04/29/2008
• Issued: 02/19/2013
• Est. Priority Date: 10/28/2004
• Status: Active Grant

• Current Assignee: Onto Innovation Inc.
• Original Assignee: Rudolph Technologies Incorporated (Onto Innovation Inc.)
• Inventors: Wang, Weidong, Buckheit, Jonathan B.
• Primary Examiner(s): ALHIJA, SAIF A
• Application Number:
  US12/150,676
• Publication Number:
  US 20080281566A1
• Time in Patent Office: 1,757 Days
• Field of Search: 703/22, 703/7, 700/121
• US Class Current: 703/7
• CPC Class Codes: H01L 22/20 Sequence of activities consisting of a plurality of measurements, corrections, marking or sorting steps

Abstract

A system and method for yield management are disclosed wherein a data set containing one or more prediction variable values and one or more response variable values is input into the system. The system can process the input data set to remove prediction variables with missing values and data sets with missing values based on a tiered splitting method to maximize usage of all valid data points. The processed data can then be used to generate a model that may be a decision tree. The system can accept user input to modify the generated model. Once the model is complete, one or more statistical analysis tools can be used to analyze the data and generate a list of the key yield factors for the particular data set.

Claims

1. A yield management apparatus, comprising:

means for receiving at least one case describing a semiconductor fabrication process, where a case comprises at least one prediction variable and at least one response variable corresponding to the at least one prediction variable;

means for identifying, based at least in part on the at least one prediction variable, a number of times that a tool is used during the semiconductor fabrication process;

means for producing at least one additional variable for the at least one case, the additional variable having a value equal to the identified number of times that the tool is used in the semiconductor fabrication process;

means for generating a processed data set by adding the at least one additional variable to the at least one case;

means for generating a model based on the processed data set,

where the model describes the relationship between the at least one prediction variable, the at least one additional variable and the corresponding at least one response variable for the semiconductor fabrication process, the model is a decision tree, and the decision tree is tiered split using at least a first prediction variable of the at least one prediction variable;

means for analyzing the generated model to identify at least one factor that affected a yield of the semiconductor fabrication process, where the at least one factor is identified out of the at least one prediction variable; and

means for outputting the at least one factor that affected the yield of the semiconductor fabrication process.

2. The yield management apparatus of claim 1, where the receiving means is further for receiving an input data set comprising a plurality of cases describing the semiconductor fabrication process.

3. The yield management apparatus of claim 2, where the indentifying means and the producing means are used for each of the plurality of cases in the input data set and the generating means is configured to generate the model based on processed data sets for each of the plurality of cases in the input data set.

4. The yield management apparatus of claim 2, further comprising means for filtering out cases comprising outliers from the input data set.

5. The yield management apparatus of claim 2, where the plurality of cases describing the semiconductor fabrication process comprise at least one case missing at least one of the at least one prediction variable and at least one case comprising the at least one of the at least one prediction variable, further comprising means for filtering out the at least one case missing at least one of the at least one prediction variable.

6. The yield management apparatus of claim 1, further comprising means for adjusting at least one process step of the semiconductor fabrication process based at least in part on the generated model.

18. The yield management apparatus of claim 1, where the number of times that a tool is used during the semiconductor fabrication process magnifies an impact of the tool on the yield of the semiconductor fabrication process.

19. The yield management apparatus of claim 1, where the yield of the semiconductor fabriction process describes a number of functional devices produced by the process as compared to a total number of devices produced.

20. The yield management apparatus of claim 1, where the at least one case comprises a variable which identifies a time when an associated tool was used.

7. A yield management method, comprising:

receiving at least one case describing a semiconductor fabrication process, where a case comprises at least one prediction variable and at least one response variable corresponding to the at least one prediction variable;

identifying, based at least in part on the at least one prediction variable, a number of times that a tool is used during the semiconductor fabrication process;

producing at least one additional variable for the at least one case, the additional variable having a value equal to the identified number of times that the tool is used in the semiconductor fabrication process;

generating, by a processor, a processed data set by adding the at least one additional variable to the at least one case;

generating a model based on the processed data set, where the model describes the relationship between the at least one prediction variable, the at least one additional variable and the corresponding at least one response variable for the semiconductor fabrication process, the model is a decision tree, and the decision tree is tiered split using at least a first prediction variable of the at least one prediction variable;

analyzing the generated model to identify at least one factor that affected a yield of the semiconductor fabrication process, where the at least one factor is identified out of the at least one prediction variable; and

outputting the at least one factor that affected the yield of the semiconductor fabrication process.

8. The yield management method of claim 7, where receiving further comprises receiving an input data set comprising a plurality of cases describing the semiconductor fabrication process.

9. The yield management method of claim 8, where indentifying and producing are used for each of the plurality of cases in the input data set and generating generates the model based on processed data sets for each of the plurality of cases in the input data set.

10. The yield management method of claim 8, further comprising filtering out cases comprising outliers from the input data set.

11. The yield management method of claim 7, further comprising adjusting at least one process step of the semiconductor fabrication process based at least in part on the generated model.

12. The yield management method of claim 7, where indentifying and producing are repeated for each tool used in the semiconductor fabrication process.

13. A non-transitory computer readable memory tangibly encoded with a software application executable by a processing unit to perform actions comprising:

receiving at least one case describing a semiconductor fabrication process, where a case comprises at least one prediction variable and at least one response variable corresponding to the at least one prediction variable;

identifying, based at least in part on the at least one prediction variable, a number of times that a tool is used during the semiconductor fabrication process;

producing at least one additional variable for the at least one case, the additional variable having a value equal to the identified number of times that the tool is used in the semiconductor fabrication process;

generating, by a processor, a processed data set by adding the at least one additional variable to the at least one case;

generating a model based on the processed data set, where the model describes the relationship between the at least one prediction variable, the at least one additional variable and the corresponding at least one response variable for the semiconductor fabrication process, the model is a decision tree, and the decision tree is tiered split using at least a first prediction variable of the at least one prediction variable;

analyzing the generated model to identify at least one factor that affected a yield of the semiconductor fabrication process, where the at least one factor is identified out of the at least one prediction variable; and

means for outputting the at least one factor that affected the yield of the semiconductor fabrication process.

14. The computer readable memory of claim 13, where receiving further comprises receiving an input data set comprising a plurality of cases describing the semiconductor fabrication process.

15. The computer readable memory of claim 14, where indentifying and producing are used for each of the plurality of cases in the input data set and generating generates the model based on processed data sets for each of the plurality of cases in the input data set.

16. The computer readable memory of claim 14, where the actions further comprise filtering out cases comprising outliers from the input data set.

17. The computer readable memory of claim 13, where the actions further comprise adjusting at least one process step of the semiconductor fabrication process based at least in part on the generated model.