Image analysis – Applications – Manufacturing or product inspection
Reexamination Certificate
1999-02-25
2003-08-12
Ahmed, Samir (Department: 2623)
Image analysis
Applications
Manufacturing or product inspection
C382S145000, C356S237500
Reexamination Certificate
active
06606401
ABSTRACT:
FIELD AND BACKGROUND OF THE INVENTION
The present invention is of an inspection method for detecting irregularities (hereafter referred to as defects) in two-dimensional periodic structures such as wafer dies or photomasks, and so forth. The invention permits sequential inspection of these periodic structures in real-time, which includes the examination of periods at the edge of the two-dimensional structure with no loss of throughput.
Periodic structures such as semiconductor wafer dies, memory cells, and photomasks require inspection during manufacture in order to detect the presence of defects as they appear thus reduce production costs. Such inspection cannot be performed entirely manually, since manual inspection would be too intensive and would require many hours of human labor. Instead, inspection is performed automatically, by moving the object containing the structure relative to an optical system for inspecting at least a portion of the object. For the sake of clarity it is convenient to model the system as a camera of limited width obtaining sequential images of a portion of the object, in a process known as “scanning”, until the entire desired area is scanned.
Each area of the object which is scanned by a single stroke of the camera is a “swath”. For wafer dies, the width of the swath is typically less than the width of the die
For wafers, the swath which includes only a single period (single die) of the wafer is defined as a “die swath”. A swath covering the same portion of the die for all dies in the wafer is defined as a “virtual swath”. A virtual swath features the images of a number of die swaths, preferably substantially all die swaths in the wafer, concatenated into a long string of die swath images taken from substantially the same portion of each die in the wafer.
An example of the three swath types is given in background art
FIG. 1A. A
wafer
10
features a plurality of dies
12
which are organized into rows
14
. Each die
12
is shown with a die swath
16
in substantially the same location for all dies
12
. A set of die swaths
16
from each row
14
is a swath
18
. All swaths
18
together form a virtual swath.
A classical detection process is based on the analysis of matching signals obtained from a number of periods. The detection of defects is based on a statistical approach, meaning the probability that a defect will exist on the same location within adjacent dies is very low. Hence detection is based on locating irregularities through the use of three-die comparison method which is shown in FIG.
1
B.
FIG. 1B
shows a swath
20
which features five die swaths
22
from five dies, labeled as “A”, “B”, “C”, “D” and “E”. The intensity difference for images of each pair of adjacent die swaths
22
is compared to a threshold value, the output of the comparison being a comparison signal
26
. When intensity difference exceeds the threshold value, comparison signal
26
is said to be significant. Therefore, the proper threshold must be set, such that the system is sensitive enough to detect low contrast defects and robust enough to ignore high contrast noise. Hence, threshold values should represent a tight estimator of pixel noise.
In
FIG. 1B
, comparison signals
26
are labeled as AB, BC, CD and DE. Each comparison signal
26
is an image marking the position of irregularities where there is a significant deviation between the signals obtained for each pair of adjacent die swaths
22
. Various algorithms have been proposed for filtering the signals and for determining the proper threshold at which a different signal from one die indicates a potential defect in the die. Examples of such algorithms are disclosed in U.S. Pat. No. 5,537,669.
A defect image
28
is the result of the defect identification procedure performed with a pair of adjacent comparison signals
26
. A defect identification procedure marks a defect in a certain die if irregularities appear at the same location in the comparison with its neighboring dies.
Since defects are expected to be both statistically random and relatively infrequent events, any defect is statistically unlikely to appear in the same location on two or three wafer dies. Thus, by performing a defect identification procedure between comparison signals obtained from adjacent dies, the presence of a defect
25
(if any) can be detected and the detection of random noise is reduced.
The term “random noise” refers to noise that is introduced in the intensity comparison procedure. Such comparison typically has an increased variance, thus there is a finite probability that the intensity comparison will exceed the threshold value to produce random noise
24
when a defect is absent. At typical threshold values the probability of such event is small for one comparison and is almost zero for two such events on the same pixel. Hence the defect identification operation should reduce the occurrence of such events to zero or near-zero levels.
A defect in the die labeled as “B” yields a significant comparison signal
26
between die swaths
22
labeled as “A” and “B” as well as between the dies labeled as “B” and “C”, such that the defect identification operation is true at the position of the defect.
The advantage of this method is that the defect identification operation results in the cancellation of much of the noise since a defect should produce significant comparison signals
26
for two adjacent die swaths
22
. In addition, such a process is particularly suitable for a real-time image processing system, since the steps required for image acquisition and processing are well defined and are performed repetitively. These steps are as follows. First, an image of a die swath for die “A”, or “die swath A”, is grabbed and stored in the system memory. Next, an image of a die swath for die “B”, or “die swath B”, is grabbed and stored. Each image is grabbed as a plurality of frames, which are processing units within a die swath. Each incoming frame of die swath B is aligned to the corresponding frame of die swath A for comparison, such that a reliable comparison signal is obtained.
As all of the images for die swath B are grabbed, a comparison image is produced which is termed image AB. Next, images for die swath C are grabbed and comparison image BC is generated. The defect identification operation which is performed between images AB and BC permits the defects found in die B to be detected. Unfortunately, this method is not effective for detecting defects at edge dies such as dies A and E. For example, defects at die A may be detected as the result of the defect identification operation between AB and BC. Such defects produce a significant comparison signal at AB while the corresponding portion of the BC signal is free of such irregularities. Hence, the defect identification operation for an edge die, such as die A, is performed only once and is sensitive to the presence of high contrast noise, the thereby giving a misleading result.
Thus, the inspection of edge dies has two difficulties. First, such detection requires additional processing steps, for example in order to perform an additional defect identification operation, which are not included in the typical processing path for the remainder of the wafer, reducing system throughput. In addition, this operation produces a significant number of false positive results for the detection of defects because of the presence of random noise.
These problems of the detection of defects at edge dies are known in the art, and have a number of currently known but deficient solutions. The first solution is simply to ignore all edge dies during the inspection process, declaring all such dies to be unfit. This solution is clearly disadvantageous, since eliminating all edge dies is inefficient and costly. The second solution is to increase the comparison threshold such that significant differences need to be even greater for edge dies. This solution eliminates most of the random noise, but also reduces the detection sensitivity. The third solution is to confirm the presence of defects o
Ahmed Samir
Bali Vikkram
Friedman Mark M.
Tokyo Seimitsu Co. Ltd.
LandOfFree
Inspection of edge periods of two-dimensional periodic... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Inspection of edge periods of two-dimensional periodic..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Inspection of edge periods of two-dimensional periodic... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3112277