Semiconductor yield management system and method

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.

1 . A yield management system, comprising: 
 means for processing an input data set comprising one or more prediction variables and one or more response variables containing data about a particular semiconductor process, the processing means comprising tiered splitting means wherein: 
 the tiered splitting means enables user selection of at least one prediction variable to generate processed data; and  
   means for generating a model based on the processed data.    
   
   
       2 . The system of  claim 1  wherein the model is a decision tree.  
   
   
       3 . The system of  claim 1 , further comprising means for analyzing the model using a statistical tool to generate one or more key yield factors based on the input data set.  
   
   
       4 . The system of  claim 1  wherein the tiered splitting means further comprises means for enabling user selection of a predetermined value for the selected prediction variable and means for removing data contained in the input data set for which the selected prediction variable has missing values and values different from the predetermined value, to generate the processed data.  
   
   
       5 . A yield management system, comprising: 
 means for processing an input data set comprising one or more prediction variables and one or more response variables containing data about a particular semiconductor process, the processing means comprising auto-categorization means wherein: 
 the auto-categorization means enables user selection for binning at least one response variable contained in the input data set into a class; and  
   means for generating a model based on the processed data.    
   
   
       6 . The system of  claim 5  wherein the model is a decision tree.  
   
   
       7 . The system of  claim 5 , further comprising means for analyzing the model using a statistical tool to generate one or more key yield factors based on the input data set.  
   
   
       8 . The system of  claim 5  wherein the auto-categorization means enables user selection for binning the at least one response variable contained in the input data set into a class using data clustering.  
   
   
       9 . The system of  claim 8  wherein the auto-categorization means further enables the user to enter a number of categories to determine if small clusters are to be excluded.  
   
   
       10 . The system of  claim 8  wherein the data clustering is performed using a nearest neighbor methodology.  
   
   
       11 . The system of  claim 8  wherein the auto-categorization means comprises means to provide a preview to enable the user to view results and make adjustments.  
   
   
       12 . A yield management system, comprising: 
 means for processing an input data set comprising one or more prediction variables and one or more response variables containing data about a particular semiconductor process, the processing means comprising outlier filtering means to enable user selection of a filter for removing response variables contained in the input data set, the outlier filtering means comprising one or more of the following filters:    1) Mean±N*std, wherein:              Mean   =       ∑     i   =   1     n     ⁢           ⁢       x   i     /   n         ,     
     ⁢     std   =           ∑     i   =   1     n     ⁢           ⁢       (       x   i     -   Mean     )     2         /     (     n   -   1     )         ,   and            N is a threshold value selected by the user, whereby the system removes data outside the range of Mean±N*std; and    2) Median±N*MAD, wherein:            MAD   =       ∑     i   =   1     n     ⁢           ⁢              x   i     -   Mean          /     n   .                 
   
   
       13 . A yield management system, comprising: 
 means for processing an input data set comprising one or more prediction variables and one or more response variables containing data about a particular semiconductor process, the processing means comprising tool usage parameter means to identify from prediction variables in the input data set a number of times that each tool is used during the semiconductor fabrication process, the tool usage parameter means determining a number that equals the number of times that each tool is used in each case contained in the input data set for the semiconductor fabrication process under analysis and producing an additional variable for each case having a value equal to the number; and 
 means for generating a model based on the processed data.  
   
   
   
       14 . A yield management system, comprising: 
 means for processing an input data set comprising one or more prediction variables and one or more response variables containing data about a particular semiconductor process, the processing means comprising treat integer as categorical means to designate an integer corresponding to a response variable as a categorical variable.    
   
   
       15 . The system of  claim 14  wherein the treat integer as categorical means enables a user to selectively designate a response variable in a list as a categorical variable.  
   
   
       16 . A yield management system, comprising: 
 means for processing an input data set comprising one or more prediction variables and one or more response variables containing data about a particular semiconductor process; and    means for generating a model based on the processed data, the model being a decision tree, wherein the model generating means comprises: 
 means for generating a linear type split for use in constructing the model comprising means to identify that a response variable and a prediction variable have a linear relationship; and  
 means to construct a decision tree having a predetermined number of sub-nodes using a fitted regression line, the predetermined number of sub-nodes being greater than two  
   
   
   
       17 . The system of  claim 16 , further comprising means for analyzing the model using a statistical tool to generate one or more key yield factors based on the input data set.  
   
   
       18 . A yield management system, comprising: 
 means for processing an input data set comprising one or more prediction variables and one or more response variables containing data about a particular semiconductor process; and    means for generating a model based on the processed data, the model being a decision tree, wherein the model generating means comprises: 
 means for generating a range type split using a split rule of the form a1≦X<a2, where X is a variable and a1 and a2 are real numbers.  
   
   
   
       19 . The system of  claim 18 , further comprising means for analyzing the model using a statistical tool to generate one or more key yield factors based on the input data set.  
   
   
       20 . A yield management system, comprising: 
 means for processing an input data set comprising one or more prediction variables and one or more response variables containing data about a particular semiconductor process; and    means for generating a model based on the processed data, the model being a decision tree, wherein the model generating means comprises means for providing user control in formulating rules for splitting nodes of the decision tree comprising at least one of: 
 means for considering tool and date parameters jointly, whereby a tool parameter and its corresponding date parameter are considered together as a split candidate;  
   means for considering tool and event parameters jointly, whereby a tool parameter and a related event are considered together as a split candidate;    means for considering maximum class distinction, whereby the model generating means builds the model based on a split that provides the greatest distinction of a class of categorical response variable;    means for parameter weighting to weight one or more variables, whereby the model generating means calculates an internal score for each variable based on its statistical significance and multiplies the score by its weight to obtain an overall score in determining a split parameter;    means for preferring simple splits of one or more categorical variables responsive to user selection from a range of preference values;    means for specifying minimum purity responsive to user selection of 1) a class of interest, 2) a threshold value for the selected class, and 3) a response variable, wherein purity is defined as all the cases in a node having the same response;    means for specifying minimum group size responsive to user selection of a threshold value, whereby the model generating means does not consider a further split when a node contains fewer cases than the selected threshold value;    means for specifying a maximum number of descendants in response to selection by the user of a predetermined cut-off level, whereby the model generating means does not generate subsequent splits when the decision tree reaches the predetermined cut-off level; and 
 means for raw data mapping to link a binned variable, which is treated as a categorical variable, to its original form.  
   
   
   
       21 . The system of  claim 20 , further comprising means for analyzing the model using a statistical tool to generate one or more key yield factors based on the input data set.  
   
   
       22 . A yield management system, comprising: 
 means for processing an input data set comprising one or more prediction variables and one or more response variables containing data about a particular semiconductor process; and    means for generating a model based on the processed data, the model being a decision tree, wherein the model generating means comprises means for providing user control for splitting nodes of the decision tree comprising means for applying a new cut rule, whereby: 
 if the variable is categorical, the user may select any combination of classes of the variable and include them in a first sub-node, the remainder of the data being included in a second sub-node; and  
 if the variable is continuous, applying one of the following split formats responsive to user selection: 
 1) a default type of the form a≦X;  
 2) a range type of the form a1≦X<a2; and  
 3) a linear type of the form X<a1, X in [a1, a2], X in [a2, a3], X>a3), wherein X is the continuous variable and a, a1, a2, and a3 are real numbers.  
 
   
   
   
       23 . The system of  claim 22 , further comprising means for analyzing the model using a statistical tool to generate one or more key yield factors based on the input data set.  
   
   
       24 . A yield management system, comprising: 
 means for processing an input data set comprising one or more prediction variables and one or more response variables containing data about a particular semiconductor process; and    means for generating one or more models based on the processed data, wherein the model generating means is responsive to user selection of a group of variables for the model building to simultaneously generate a model for each of the variables selected by the user.    
   
   
       25 . The system of  claim 24  wherein the model is a decision tree.  
   
   
       26 . The system of  claim 24 , further comprising means for analyzing the model using a statistical tool to generate one or more key yield factors based on the input data set.  
   
   
       27 . A yield management system, comprising: 
 means for processing an input data set comprising one or more prediction variables and one or more response variables containing data about a particular semiconductor process;    means for generating a model based on the processed data, the model being a decision tree;    means for modifying the model based on user input;    means for analyzing the model using a statistical tool to generate one or more key yield factors based on the input data set; and    means for redisplaying a setup means to enable a user to modify previous selections    
   
   
       28 . A yield management system, comprising: 
 means for processing an input data set comprising one or more prediction variables and one or more response variables containing data about a particular semiconductor process;    means for generating a model based on the processed data, the model being a decision tree;    means for modifying the model based on user input;    means for analyzing the model using a statistical tool to generate one or more key yield factors based on the input data set; and    means for collapsing a node responsive to user selection when the user decides that the split of the node is unnecessary and for expanding the node responsive to user selection when the user wants to examine aggregate statistics.    
   
   
       29 . The system of  claim 28  wherein the means for expanding the node expands a previously collapsed node, so that the node returns to its original length.  
   
   
       30 . A yield management method, comprising: 
 processing an input data set comprising one or more prediction variables and one or more response variables containing data about a particular semiconductor process, the processing comprising tiered splitting wherein: 
 the tiered splitting enables user selection of a prediction variable to generate processed data; and  
   generating a model based on the processed data.    
   
   
       31 . The method of  claim 30  wherein the model is a decision tree.  
   
   
       32 . The method of  claim 30 , further comprising analyzing the model using a statistical tool to generate one or more key yield factors based on the input data set.  
   
   
       33 . The system of  claim 30  wherein the tiered splitting further comprises enabling user selection of a predetermined value for the selected prediction variable and removing data contained in the input data set for which the selected prediction variable has missing values and values different from the predetermined value, to generate the processed data.  
   
   
       34 . A yield management method, comprising: 
 processing an input data set comprising one or more prediction variables and one or more response variables containing data about a particular semiconductor process, the processing comprising auto-categorizing wherein: 
 auto-categorizing enables user selection for binning at least one response variable contained in the input data set into a class; and  
   generating a model based on the processed data.    
   
   
       35 . The method of  claim 34  wherein the model is a decision tree.  
   
   
       36 . The method of  claim 34 , further comprising analyzing the model using a statistical tool to generate one or more key yield factors based on the input data set.  
   
   
       37 . The method of  claim 34  wherein auto-categorizing enables user selection for binning at least one response variable contained in the input data set into a class using data clustering  
   
   
       38 . The method of  claim 37  wherein the auto-categorizing further enables the user to enter a number of categories to determine if small clusters are to be excluded.  
   
   
       39 . The method of  claim 37  wherein the data clustering is performed using a nearest neighbor methodology.  
   
   
       40 . The method of  claim 37  wherein the auto-categorizing comprises providing a preview to enable the user to view results and make adjustments.  
   
   
       41 . A yield management method, comprising: 
 processing an input data set comprising one or more prediction variables and one or more response variables containing data about a particular semiconductor process, the processing comprising outlier filtering to enable user selection of a filter for removing response variables contained in the input data set comprising one or more of the following filters:    1) Mean±N*std, wherein:              Mean   =       ∑     i   =   1     n     ⁢           ⁢       x   i     /   n         ,     
     ⁢     std   =           ∑     i   =   1     n     ⁢           ⁢       (       x   i     -   Mean     )     2         /     (     n   -   1     )         ,   and            N is a threshold value selected by the user, whereby the method removes data outside the range of Mean±N*std; and    2) Median±N*MAD, wherein:            MAD   =       ∑     i   =   1     n     ⁢           ⁢              x   i     -   Mean          /     n   .                 
   
   
       42 . A yield management method, comprising: 
 processing an input data set comprising one or more prediction variables and one or more response variables containing data about a particular semiconductor process, the processing comprising identifying from prediction variables in the input data set a number of times that each tool is used during the semiconductor fabrication process to determine a number that equals the number of times that each tool is used in each case contained in the input data set for the semiconductor fabrication process under analysis and producing an additional variable for each case having a value equal to the number; and    generating a model based on the processed data.    
   
   
       43 . A yield management method, comprising: 
 processing an input data set comprising one or more prediction variables and one or more response variables containing data about a particular semiconductor process, the processing comprising designating an integer corresponding to a response variable as a categorical variable responsive to user selection.    
   
   
       44 . The method of  claim 43 , further comprising enabling a user to selectively designate a variable in a list as a categorical variable.  
   
   
       45 . A yield management method, comprising: 
 processing an input data set comprising one or more prediction variables and one or more response variables containing data about a particular semiconductor process; and    generating a model based on the processed data, the model being a decision tree, wherein the model generating comprises: 
 generating a linear type split for use in constructing the model to identify that a response variable and a prediction variable have a linear relationship and to construct the decision tree having a predetermined number of sub-nodes using a fitted regression line, the predetermined number of sub-nodes being greater than two.  
   
   
   
       46 . The method of  claim 45 , further comprising analyzing the model using a statistical tool to generate one or more key yield factors based on the input data set.  
   
   
       47 . A yield management method, comprising: 
 processing an input data set comprising one or more prediction variables and one or more response variables containing data about a particular semiconductor process; and    generating a model based on the processed data, the model being a decision tree, wherein the model generating comprises: 
 generating a range type split using a split rule of the form a1≦X<a2, where X is a variable and a1 and a2 are real numbers.  
   
   
   
       48 . The method of  claim 47 , further comprising analyzing the model using a statistical tool to generate one or more key yield factors based on the input data set.  
   
   
       49 . A yield management method, comprising: 
 processing an input data set comprising one or more prediction variables and one or more response variables containing data about a particular semiconductor process;    generating a model based on the processed data, the model being a decision tree, wherein the model generating comprises providing user control in formulating rules for splitting nodes of the decision tree comprising at least one of: 
 considering tool and date parameters jointly, whereby a tool parameter and its corresponding date parameter are considered together as a split candidate;  
 considering tool and event parameters jointly, whereby a tool parameter and a related event are considered together as a split candidate;  
 considering maximum class distinction, whereby generating a model builds the model based on a split that provides the greatest distinction of a class of categorical response variable;  
 weighting one or more variables, whereby the model generating calculates an internal score for each variable based on its statistical significance and multiplies the score by its weight to obtain an overall score in determining a split parameter;  
 preferring simple splits of one or more categorical variables responsive to user selection from a range of preference values;  
 specifying minimum purity responsive to user selection of 1) a class of interest, 2) a threshold value for the selected class, and 3) a response variable, wherein purity is defined as all the cases in a node having the same response;  
 specifying minimum group size responsive to user selection of a threshold value, whereby the model generating does not consider a further split when a node contains fewer cases than the selected threshold value;  
 specifying a maximum number of descendants in response to selection by the user of a predetermined cut-off level, whereby the model generating does not generate subsequent splits when the decision tree reaches the predetermined cut-off level; and  
 mapping to link a binned variable, which is treated as a categorical variable, to its original raw data form.  
   
   
   
       50 . The method of  claim 49 , further comprising analyzing the model using a statistical tool to generate one or more key yield factors based on the input data set.  
   
   
       51 . A yield management method, comprising: 
 processing an input data set comprising one or more prediction variables and one or more response variables containing data about a particular semiconductor process; and    generating a model based on the processed data, the model being a decision tree, wherein the model generating comprises providing user control for splitting nodes of the decision tree comprising applying a new cut rule, whereby: 
 if the variable is categorical, the user may select any combination of classes of the variable and include them in a first sub-node, the remainder of the data being included in a second sub-node; and  
 if the variable is continuous, applying one of the following split formats responsive to user selection: 
 1) a default type of the form a≦X;  
 2) a range type of the form a1≦X<a2; and  
 3) a linear type of the form X<a1, X in [a1, a2], X in [a2, a3], X>a3), wherein X is the continuous variable and a, a1, a2, and a3 are real numbers.  
 
   
   
   
       52 . The method of  claim 51 , further comprising analyzing the model using a statistical tool to generate one or more key yield factors based on the input data set.  
   
   
       53 . A yield management method, comprising: 
 processing an input data set comprising one or more prediction variables and one or more response variables containing data about a particular semiconductor process; and    generating one or more models based on the processed data, wherein the model generating is responsive to user selection of a group of variables for the model building to simultaneously generate a model for each of the variables selected by the user.    
   
   
       54 . The method of  claim 53  wherein the model is a decision tree.  
   
   
       55 . The method of  claim 53 , further comprising analyzing the model using a statistical tool to generate one or more key yield factors based on the input data set.  
   
   
       56 . A yield management method, comprising: 
 processing an input data set comprising one or more prediction variables and one or more response variables containing data about a particular semiconductor process;    generating a model based on the processed data, the model being a decision tree;    modifying the model based on user input;    analyzing the model using a statistical tool to generate one or more key yield factors based on the input data set; and    redisplaying a setup to enable a user to modify previous selections.    
   
   
       57 . A yield management method, comprising: 
 processing an input data set comprising one or more prediction variables and one or more response variables containing data about a particular semiconductor process;    generating a model based on the processed data, the model being a decision tree;    modifying the model based on user input;    analyzing the model using a statistical tool to generate one or more key yield factors based on the input data set; and    collapsing a node responsive to user selection when the user decides that the split of the node is unnecessary and expanding the node responsive to user selection when the user wants to examine aggregate statistics.    
   
   
       58 . The method of  claim 57  wherein expanding the node expands a previously collapsed node, so that the node returns to its original length.