Data processing: generic control systems or specific application – Specific application – apparatus or process – Product assembly or manufacturing
Reexamination Certificate
2001-11-06
2004-01-13
Picard, Leo (Department: 2125)
Data processing: generic control systems or specific application
Specific application, apparatus or process
Product assembly or manufacturing
C700S051000
Reexamination Certificate
active
06678569
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention generally relates to tool control system and solves the problem of mis-processing wafer lots/wafers by monitoring specific physical parameters of the tool set in real time. These parameters include any measurable quantifications relating to the process or tool, such as gas flow, temperature, pressure, power, etc.
2. Description of Related Art
A manufacturing line is generally controlled using three logistic functions: lots must be routed to the proper tool set; recipes for various process steps must be routed to an individual process tool; and wafer lots must be matched to tool and then to a recipe. Further, two control functions exist. The first control function is to track the logistics functions. The second is to control the process tool parameters (e.g., to decide when a tool malfunction or tool fault has occurred.)
The tool process parameters form a multivariate data set. The monitoring of such datasets has been the subject of many works, some of which are mentioned below. The set of tool data to be analyzed is very large and can be considered a three-dimensional dataset.
The first dimension of the three-dimensional dataset is the process parameter. Examples of process parameters include chamber pressure, temperature, gas flows, and RF power. The second dimension is time, with data for each process parameter being taken at regular times during wafer processing. The third dimension is the batching of data as defined by wafer.
Thus, the dataset consists of a time series for each process parameter, for each wafer processed. To reduce this dataset to a few indicators of tool performance (or health), i.e., a few surrogate variables, a data reduction scheme is necessary.
Nomikos (P. Nomikos and J. F. MacGregor, “Multivariate SPC Charts for Monitoring Batch Processes,”
Technometrics,
vol. 37, No. 1, pp. 41-59, February 1995), incorporated herein by reference, discuss one such scheme by rearranging the three-dimensional dataset into a two-dimensional matrix and then performing a conventional principal components analysis (PCA). Spanos C. J. Spanos, H. F. Guo, A. Miller and J. Leville-Parrill, “Real-Time Statistical Process Control Using Tool Data,”
IEEE Trans. on Semicond. Manuf,
vol. 5, No. 4, pp. 308-318, November 1992, incorporated herein by reference, describes a method to process time series (without regard to batching by wafer) using a time series filter. The Spanos article then describes using a Hotelling T
2
function to reduce groupings of n time series points to a single surrogate variable.
Lee (S. F. Lee, E. D. Boskin, H. C. Liu, E. H. Wen and C. J. Spanos, “RTSPC: A Software Utility for Real-Time SPC and Tool Data Analysis,”
IEEE Trans. on Semicond. Manuf,
vol. 8, No. 1, pp. 17-25, February 1995), incorporated herein by reference, describes a second method to reduce the dataset by collapsing the time series for each wafer using the average value of the time series, or the length of the time series (i.e., the process step). U.S. Pat. No. 5,442,562, incorporated herein by reference, describes a general method of reducing a plurality of process intermediate process variables (such as principal components) to a single surrogate variable, determining which intermediate variable is outside a predetermined limit, determining which process variable is the primary contributor and then correcting that process variable automatically via a computer.
When a two-dimensional dataset is to be monitored, for example, tool parameters such as pressure, gas flow, or temperature vs. wafer identity are to be monitored, a statistic called the Hotelling Function (T
2
) can be used (see for example Doganaksoy, (N. Doganaksoy, F. W. Faltin and W. T. Tucker (1991)), “Identification of Out of Control Quality Characteristics in a Multivariate Manufacturing Environment,”
Comm. Statist.—Theory Meth.,
20, 9, pp. 2775-2790), incorporated herein by reference, where:
T
2
=
(
n
ref
⁢
n
new
)
⁢
(
n
ref
-
p
)
(
n
ref
⁢
n
new
)
+
(
n
ref
-
1
)
⁢
p
⁢
(
x
-
x
_
)
T
⁢
S
-
1
⁢
(
x
-
x
_
)
where x=vector of measured values, {overscore (x)}=vector of reference means (based on history), and S−1 is the inverse covariance matrix (based on history), nref is the size of the reference sample, nnew is the size of the new sample and p is the number of variables. This function obeys an F statistic, F(p, nref-p,alpha). Where alpha is the probability corresponding to a desired false call rate (presuming the distributions of the individual parameters are normal).
Another related technique for reducing a two-dimensional dataset to a surrogate variable is known as principal component analysis (PCA) which is incorporated by Jackson (J. Edward Jackson,
A User's Guide to Principal Components,
John Wiley & Sons, Inc. (1991) M. J. R. Healy,
Matrices for Statistics,
Oxford Science Publications (1986)), incorporated herein by reference. In this method, the eigenvectors of the covariance matrix (or alternately the eigenvectors of the correlation matrix) form a set of independent intermediate variables, consisting of linear combinations of the original process variables. These principal components may be monitored separately, by taking only the most significant or by monitoring the residuals. Alternatively, the principal components may be monitored in aggregate, in which case the sum of squares distance of a particular sample from its principal components is identical to the Hotelling T
2
surrogate variable.
One difficulty with multivariate methods is that, once an out-of-control situation is detected, determining which process variables caused the problem is difficult, based on a single or few indicator variables. This is often known as the T
2
decomposition problem. Doganaksoy, above, provides a method to solve the decomposition problem, that orders the process variables in order of descending normalized difference from their means, i.e., (x
i
−{overscore (x)}
i
)/&sgr;
i
, the univariate statistic.
Runger (G. C. Runger, F. B. Alt, “Contributors to a multivariate statistical Process control chart signal,” Commun.
Statist.—Theory Meth.
25(10) 2203-2213 (1996)), incorporated herein by reference, provides another such method by considering the change in the T
2
score if the process parameter is not included (the “drop 1” T
2
score, a
D
i).
However, the conventional systems are not adequately able to distinguish between variations that do not affect products and variations that do affect products. The invention solves this and other problems of the conventional systems, as discussed below.
SUMMARY OF THE INVENTION
It is, therefore, an object of the present invention to provide a structure and method for controlling a manufacturing tool, including measuring different manufacturing parameters of the tool, transforming a plurality of time series of the manufacturing parameters into intermediate variables based on restrictions and historical reference statistics, generating a surrogate variable based on the intermediate variables, if the surrogate variable exceeds a predetermined limit, identifying a first intermediate variable, of the intermediate variables, that caused the surrogate variable to exceed the predetermined limit and identifying a first manufacturing parameter associated with the first intermediate variable, and inhibiting further operation of the tool until the first manufacturing parameter has been modified to bring the surrogate value within the predetermined limit.
The surrogate variable comprises T
2
, where:
T
2
=
(
(
x
-
x
_
)
σ
)
T
⁢
R
-
1
⁢
(
(
x
-
x
_
)
σ
)
=
z
T
⁢
R
-
1
⁢
z
,
where x comprises the intermediate variables, {overscore (x)} comprises a historical sensor value, &sgr; comprises a historical standard deviation sensor value, R
−1
comprises an inverse correlation matrix, z comprises mean and standard deviation normalized values. Further, x and &sgr; are user-adjustable which provides a substantial advantage over conventional system
Bunkofske Raymond J.
Colt, Jr. John Z.
Ghias Asif
McGill James J
Pascoe Nancy T.
Garland Steven R.
International Business Machines - Corporation
Kotulak, Esq. Richard M.
McGinn & Gibb PLLC
Picard Leo
LandOfFree
User configurable multivariate time series reduction tool... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with User configurable multivariate time series reduction tool..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and User configurable multivariate time series reduction tool... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3245632