Expert knowledge methods and systems for data analysis

Abstract

A method for adjusting a data set defining a set of process runs, each process run having a set of data corresponding to a set of variables for a wafer processing operation is provided. A model derived from a data set is received. A new data set corresponding to one process run is received. The new data set is projected to the model. An outlier data point produced as a result of the projecting is identified. A variable corresponding to the one outlier data point is identified, the identified variable exhibiting a high contribution. A value for the variable from the new data set is identified. Whether the value for the variable is unimportant is determined. A normalized matrix of data is created, using random data and the variable that was determined to be unimportant from each of the new data set and the data set. The data set is updated with the normalized matrix of data.

1 . A method for updating a data set defining a set of process runs, each process run having a set of data corresponding to a set of variables for a wafer processing operation, comprising:
 receiving a data set;   performing scaling to the data set;   performing principal component analysis to the scaled data set to generate a model;   receiving new data;   projecting the new data to the model;   identifying outlier data points based on the projecting;   examining a contribution plot corresponding to one of the outlier data points;   identifying a variable that corresponds to the one outlier data point which provides a high contribution in the contribution plot;   determining that the variable is unimportant;   creating a desensitizing set of data for the variable based on a standard deviation of the data set and a randomization of the new data; and   augmenting the data set with the desensitizing set of data.   
   
   
       2 . The method of  claim 1 , wherein determining that the variable is unimportant is performed with expert knowledge. 
   
   
       3 . The method of  claim 2 , wherein expert knowledge is knowledge of behavior of the variable. 
   
   
       4 . A method for adjusting a data matrix defining a set of process runs each process run having a set of data corresponding to a set of variables for a wafer processing operation, comprising:
 receiving a data matrix of N rows and M columns where N equals a number of process runs and M equals a number of variables in the data matrix;   receiving a new set of data with M variables wherein at least one variable corresponds to an outlier and is unimportant based on expert input;   generating a normally distributed random vector containing N−1 rows;   generating a one vector containing N−1 rows of ones;   determining a standard deviation of data corresponding to the variable in the data matrix;   multiplying the standard deviation by the normally distributed random vector producing a first vector;   multiplying the data corresponding to the variable from the new data by the one vector producing a second vector;   adding the first vector to the second vector producing a third vector;   creating an expert desensitizing matrix where the Mth column contains the third vector and the remaining columns are made up of data corresponding to the remaining variables; and   creating a new data matrix where the data matrix is augmented by the expert desensitizing matrix.   
   
   
       5 . An expert system for desensitizing variables based on engineering knowledge, comprising:
 a first database that includes data for process runs;   a second database that includes corresponding models of the data;   a processor coupled to the first and second databases; and   logic that identifies outliers and variable contributions that caused the outliers, the logic being further configured to incorporate expert engineering knowledge in an examination of the variable contributions, and the logic adjusts the data in order that future process runs properly identify resulting outliers as faults.   
   
   
       6 . An expert system for desensitizing variables based on engineering knowledge as recited in  claim 5 , wherein the expert system enables proper fault detection due to a desensitizing of variable contributions that can cause false positive faults. 
   
   
       7 . An expert system for desensitizing variables based on engineering knowledge as recited in  claim 5 , wherein the variables represent a range of variables defining changes in a design of equipment used to perform the process runs.