Browse Prior Art Database

Process Diagnosis and Tool Control by Statistical Template Matching

IP.com Disclosure Number: IPCOM000106135D
Original Publication Date: 1993-Sep-01
Included in the Prior Art Database: 2005-Mar-20
Document File: 4 page(s) / 147K

Publishing Venue

IBM

Related People

Gifford, GG: AUTHOR [+4]

Abstract

Disclosed is a method of characterizing a manufacturing tool when it is running well and a means for detecting and identifying changes from this baseline condition. Ideally this is implemented for feedback control of the process tool.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 41% of the total text.

Process Diagnosis and Tool Control by Statistical Template Matching

      Disclosed is a method of characterizing a manufacturing tool
when it is running well and a means for detecting and identifying
changes from this baseline condition.  Ideally this is implemented
for feedback control of the process tool.

      Statistical templates provide a framework for handling multiple
and aggregate sensors like  photodiode arrays.  Aggregate sensors
greatly improve the signal-to-noise ratio of the condition being
monitored.  The proposed system features control of temporally
varying conditions within a process run, visualization of any
variance of this temporal behavior from the mean, and visualization
of the coexisting variance among multiple sensors (aggregate and
nonaggregate).

      Data is collected by multiple sensors that are monitoring a
production tool or apparatus.  These are combined into a vector
representing the state of the tool at time  t sub '0'.  The data from
each sensor is scaled to create a smooth gradation of integer values
from 0 to  2 sup 'N' - 1 where N is the number which produces the
word length in bits that is the most efficient integer representation
for the target processor.  An array of these vectors are collected
over time.  The values in this array, ranging from 0 to 2 sup 'N' - 1
can be interpreted as gray-level pixels of a two-dimensional image.
One axis of this image is the channel number of each sensor and the
other axis is time.  When interpreted as an image, the magnitude of
each number in the array is represented by its brightness or
darkness.  When treating the templates as images, we open the
possibility of using image processing algorithms for improved
visualization and detection.

      Typically, a series of data vectors is collected at regular
time intervals,  T sub 's' , when the tool or apparatus is running
well, i.e., product produced by the tool meets all known
specifications.  This array forms an image which characterizes how
the tool or apparatus was running during the time data was collected.
A series of such images result, one for each production run that is
monitored.  This is used to generate an electronic signature of the
tool and process which consists of two derived images.  The first
image, the "mean image"  I sub mu , is produced by taking the average
of the series of images recorded when the tool was running well.  The
second image, the "deviation image" I sub sigma, is produced by
calculating the statistical deviation from the "mean image" for each
pixel of the image series.

      An important feature is a technique for correcting for small
offsets in time or channel number between the "mean image" and the
subsequent data sets of interest.  Normalized correlation is used to
align one image or template to another before comparing them.  A
diagram of the overall template building and deviation detection
schemes is shown in the Figure.

The process o...