Multivariate control of semiconductor processes

• US 7,151,976 B2
• Filed: 09/16/2005
• Issued: 12/19/2006
• Est. Priority Date: 09/17/2004
• Status: Active Grant

• Current Assignee: MKS Instruments AB (Sartorius AG)
• Original Assignee: MKS Instruments Incorporated
• Inventors: Lin, Kuo-Chin
• Primary Examiner(s): CABRERA, ZOILA E
• Application Number:
  US11/229,153
• Publication Number:
  US 20060111804A1
• Time in Patent Office: 459 Days
• Field of Search: 700108-110, 700/121, 702/183, 702/184, 702/182, 702/185
• US Class Current: 700/108
• CPC Class Codes: G05B 19/4184 characterised by fault tolerance, reliability of production system
• CPC Class Codes: G05B 2219/31357 Observer based fault detection, use model
• CPC Class Codes: G05B 2219/45031 Manufacturing semiconductor wafers
• CPC Class Codes: Y02P 90/02 Total factory control, e.g. smart factories, flexible manufacturing systems [FMS] or integrated manufacturing systems [IMS]
• CPC Class Codes: Y02P 90/80 Management or planning

Abstract

A method for detecting a fault condition of a manufacturing process involving acquiring data and developing a model based on a plurality of manufacturing related variables for a plurality of outputs of a manufacturing process. The method also involves identifying which of the manufacturing related variables have a substantial affect on the model for detecting faulty outputs of the manufacturing process. The method also involves updating the model by using the manufacturing related variables identified as having a substantial affect on the model to improve the ability of the model to detect faulty outputs of the manufacturing process.

Claims

1. A method for detecting a fault condition of a manufacturing process, comprising:

a) acquiring data associated with a plurality of manufacturing related variables for a plurality of outputs of a manufacturing process;

b) developing a model based on the data for detecting faulty outputs of the manufacturing process;

c) automatically identifying which of the manufacturing related variables have a substantial effect on the model for detecting faulty outputs of the manufacturing process using a mathematical algorithm; and

d) updating the model by using the manufacturing related variables identified in step c) to improve the ability of the model to detect faulty outputs of the manufacturing process,

wherein using a mathematical algorithm comprises,

i) identifying a linear equation of a selected variable to the origin of the loadings associated with the mode;

ii) identifying linear equations of the manufacturing related variables that are perpendicular to the linear equation of step i);

iii) solving the simultaneous equations between every manufacturing related variable linear equation and the linear equation of step i) to obtain a projection point of each manufacturing related variable on the linear equation of step i);

iv) determining the distance between each projection point to the origin;

v) calculating a correlation coefficient based on a ratio of distance between the origin and each projected point relative to distance of the selected variable to the origin; and

vi) identifying the substantial manufacturing related variables by selecting the manufacturing related variables that have a correlation coefficient greater than a predefined number.

2. The method of claim 1, wherein steps b), c) and d) comprise utilizing a multivariate analysis technique.

3. The method of claim 1, wherein step b) comprises computing principal components and loadings associated with the model using a non-linear iterative partial least square method and singular value decomposition.

4. The method of claim 1, wherein step c) comprises utilizing a partial least square discrimination analysis method.

5. The method of claim 1, wherein the selected variable is a fault condition or a metrology metric value.

6. The method of claim 1, wherein the predefined number is selected by a user.

7. A method for detecting a fault condition of a manufacturing process, comprising:

a) identifying which of a plurality of manufacturing related variables for a plurality of outputs of a manufacturing process have a substantial effect on a model for detecting faulty outputs of a manufacturing process by;

i) identifying a linear equation of a selected variable to the origin of the loadings associated with the model;

ii) identifying linear equations of the manufacturing related variables that are perpendicular to the linear equation of step i);

iii) solving the simultaneous equations between every manufacturing related variable linear equation and the linear equation of step i) to obtain a projection point of each manufacturing related variable on the linear equation of step i);

iv) determining the distance between each projection point to the origin;

v) calculating a correlation coefficient based on a ratio of distance between the origin and each projected point relative to distance of the selected variable to the origin; and

vi) identifying the substantial manufacturing related variables by selecting the manufacturing related variables that have a correlation coefficient greater than a predefined number; and

b) automatically updating the model by using the manufacturing related variables identified in step a) to improve the ability of the model to detect faulty outputs of the manufacturing process.

8. The method of claim 7, wherein steps a) and b) comprise utilizing a multivariate analysis technique.

9. The method of claim 7, wherein step a) comprises utilizing a partial least square discrimination analysis method.

10. The method of claim 7, wherein the selected variable is a fault condition or a metrology metric value.

11. The method of claim 7, wherein the predefined number is selected by a user.