Industrial process fault detection using principal component analysis
First Claim
1. A method for monitoring an industrial process, said method comprising the steps of:
- obtaining sensor data corresponding to a plurality of product units being processed in accordance with the industrial process;
forming a sample matrix of data representing at least two of the product units, wherein the sample matrix is formed from at least a portion of the sensor data;
computing a plurality of singular vectors of the sample matrix;
reducing the plurality of singular vectors to a principal set of singular vectors;
computing principal components of sensor data corresponding to at least one additional product unit processed subsequent to the product units represented in the sample matrix;
computing a predicted data vector for the additional product unit;
calculating a residual data vector for the additional product unit using the predicted data vector for the additional product unit and a measured data vector corresponding to the additional product unit, the measured data vector comprising sensor data obtained for the additional product unit;
calculating a scalar metric from the residual data vector for the additional product unit; and
categorizing the additional product unit based on the value of the scalar metric mass spectrometer data, electrical sensor data, and RF sensor data.
1 Assignment
0 Petitions
Accused Products
Abstract
A method and system for use in monitoring/evaluating industrial process such as, for example, plasma processes useful in the fabrication of semiconductor chips, microelectromechanical devices, and the like on semiconductor wafers and the like are provided. In one embodiment, a plasma process fault detection module (100) includes a data selection sub-module (101), a model building/updating sub-module (102), a principal component analysis (PCA) analysis sub-module (103), a model maintenance sub-module (104), a wafer categorization sub-module (105), and a data output sub-module (106). The data selection sub-module (101) obtains selected optical emissions spectra (OES) data for each wafer that is processed. The model building/updating sub-module (102) constructs multiple models from the selected OES data for a number of wafers. The PCA analysis sub-module (103) utilizes PCA techniques to determine whether the selected OES data for a particular wafer differs significantly from that expected for a normal wafer as represented by the models. The model maintenance sub-module (104) saves and retrieves models for different processes, associating the current wafer with the correct process. The wafer categorization sub-module (105) categorizes each wafer based on a scalar metric characterizing the residual spectrum vector. The data output sub-module (106) outputs the results that are obtained to a user.
-
Citations
43 Claims
-
1. A method for monitoring an industrial process, said method comprising the steps of:
-
obtaining sensor data corresponding to a plurality of product units being processed in accordance with the industrial process;
forming a sample matrix of data representing at least two of the product units, wherein the sample matrix is formed from at least a portion of the sensor data;
computing a plurality of singular vectors of the sample matrix;
reducing the plurality of singular vectors to a principal set of singular vectors;
computing principal components of sensor data corresponding to at least one additional product unit processed subsequent to the product units represented in the sample matrix;
computing a predicted data vector for the additional product unit;
calculating a residual data vector for the additional product unit using the predicted data vector for the additional product unit and a measured data vector corresponding to the additional product unit, the measured data vector comprising sensor data obtained for the additional product unit;
calculating a scalar metric from the residual data vector for the additional product unit; and
categorizing the additional product unit based on the value of the scalar metric mass spectrometer data, electrical sensor data, and RF sensor data. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18)
-
-
19. A method for monitoring a plurality of plasma processes, said method comprising the steps of:
-
obtaining sensor data for a plurality of wafers being processed in accordance with a plurality of plasma processes, wherein the sensor data is obtained at a plurality of times during the processing of each wafer;
forming at least one of a plurality sample matrices from at least a portion of the sensor data, wherein each sample matrix formed is associated with one of the plasma processes;
computing singular vectors for each sample matrix formed in said forming step;
reducing the singular vectors computed for each sample matrix to a principal set of singular vectors for each sample matrix, each sample matrix and its corresponding principal set of singular vectors comprising a model associated with the plasma process with which the sample matrix is associated;
storing each model associated with the plasma processes that the wafers are processed in accordance therewith;
associating an additional wafer processed subsequent to plasma processing of the wafers represented in the sample matrix with one of the plurality of plasma processes;
retrieving the model for the plasma process associated with the additional wafer;
computing principal components, a predicted data vector, and a residual data vector for the additional wafer, using the principal set of singular vectors from the retrieved model;
calculating a scalar metric from the residual data vector for the additional wafer;
categorizing the additional wafer based on the value of the scalar metric. - View Dependent Claims (20, 21, 22, 23, 24, 25, 26, 27, 28, 29)
-
-
30. A system for use in monitoring industrial processes, said system comprising:
-
a data selection module operable to obtain at least one data vector corresponding to each of a plurality of product units being processed;
a sample matrix building module operable to build at least one sample matrix from data vectors corresponding to at least two of the product units;
a principal component analysis module operable to compute principal singular vectors for a sample matrix input thereto, the sample matrix and principal singular vectors thereof comprising model data, said principal component analysis module being further operable to compute, from the principal singular vectors associated with the sample matrix input thereto, principal components, a predicted data vector, and a residual data vector for each additional product unit that is processed subsequent to processing of the product units represented in the sample matrix input thereto;
a model maintenance module operable to identify a process identity for a product unit currently being processed from among a plurality of process identities, store model data associated with the identified process identity, and retrieve stored model data associated with the identified process identity; and
a product unit categorization module operable to categorize each additional product unit that is processed subsequent to processing of the product units represented in the sample matrix based a scalar metric calculated from each residual data vector for each additional product unit. - View Dependent Claims (31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43)
-
Specification