U.S. patent application number 11/548259 was filed with the patent office on 2007-06-14 for image alignment method, comparative inspection method, and comparative inspection device for comparative inspections.
This patent application is currently assigned to Hitachi, Ltd.. Invention is credited to Shunji Maeda, Takafumi Okabe, Kaoru Sakai.
Application Number | 20070133863 11/548259 |
Document ID | / |
Family ID | 18684987 |
Filed Date | 2007-06-14 |
United States Patent
Application |
20070133863 |
Kind Code |
A1 |
Sakai; Kaoru ; et
al. |
June 14, 2007 |
Image Alignment Method, Comparative Inspection Method, and
Comparative Inspection Device for Comparative Inspections
Abstract
The present invention provides a high-precision alignment
method, device and code for inspections that compare an inspection
image with a reference image and detect defects from their
differences. In one embodiment an inspection image and a reference
image are divided into multiple regions. An offset is calculated
for each pair of sub-images. Out of these multiple offsets, only
the offsets with high reliability are used to determine an offset
for the entire image. This allows high-precision alignment with
little or no dependency on pattern density or shape, differences in
luminance between images, and uneven luminance within individual
images. Also, detection sensitivity is adjusted as necessary by
monitoring alignment precision.
Inventors: |
Sakai; Kaoru; (Yokohama,
JP) ; Maeda; Shunji; (Yokohama, JP) ; Okabe;
Takafumi; (Yokohama, JP) |
Correspondence
Address: |
TOWNSEND AND TOWNSEND AND CREW, LLP
TWO EMBARCADERO CENTER
EIGHTH FLOOR
SAN FRANCISCO
CA
94111-3834
US
|
Assignee: |
Hitachi, Ltd.
Tokyo
JP
|
Family ID: |
18684987 |
Appl. No.: |
11/548259 |
Filed: |
October 10, 2006 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
11140350 |
May 27, 2005 |
7127126 |
|
|
11548259 |
Oct 10, 2006 |
|
|
|
09802687 |
Mar 8, 2001 |
7020350 |
|
|
11140350 |
May 27, 2005 |
|
|
|
Current U.S.
Class: |
382/151 |
Current CPC
Class: |
G06T 7/001 20130101;
G06T 7/32 20170101; G01N 21/95607 20130101; G06T 2207/30148
20130101; G03F 1/84 20130101 |
Class at
Publication: |
382/151 |
International
Class: |
G06K 9/00 20060101
G06K009/00 |
Foreign Application Data
Date |
Code |
Application Number |
Jun 15, 2000 |
JP |
2000-184563 |
Claims
1. An apparatus for inspecting a specimen, comprising. an image
detection means for detecting an image of a specimen on which
plural patterns are formed; an offset determining means for
determining offsets between said detected image and a reference
image stored in a memory by selecting patterns among said plural
patterns which are advantageous to determine the offsets; an
alignment means for aligning said detected image and said reference
image using said offsets determined by the offset determining
means; and a comparing means for comparing said aligned inspection
image and said reference image to detect a defect candidate; and a
feature extracting unit for extracting a feature of said defect
candidate.
2. An apparatus according to the claim 1, wherein said offset
determining means calculates offsets between the patterns of said
detected image and the patterns of said reference image by
evaluating correlations between shift amount in X-direction and
shift amount in Y-direction and a reliability measurement over
portions of a full image of the pattern.
3. An apparatus according to the claim 2, wherein said offset
determining means calculates offsets between plural images divided
from said detected image and plural images divided from said
reference image in parallel.
4. An apparatus for inspecting a specimen, comprising. an image
detection means having an illuminating unit to illuminate a
specimen on. which patterns are formed with light and a detecting
unit to detect an image of the specimen; a processor which
determines an offset between said detected image and a reference
image, correcting the offset between said detected image and a
reference by using a data of said determined offset, compares said
detected image and said reference image whose offset is corrected
to detect a defect candidate, and extracts a feature of said defect
candidate; and an output means for outputting information on said
extracted feature of said defect candidate, wherein said processor
determines said offset between said detected image and a reference
image by selecting patterns among said plural patterns which are
advantageous to determine the offset by evaluating correlations
between shift amount in X-direction and shift amount in Y-direction
and a reliability measurement over portions of a full image of the
pattern.
5. An apparatus according to the claim 5, wherein said processor
calculates offsets between plural images divided from said detected
image and plural images divided from said reference image in
parallel.
6. A method for inspecting a specimen, comprising: detecting an
image of a specimen on which plural patterns are formed;
determining an offset between said detected image and a reference
image stored in a memory by selecting patterns among said plural
patterns which are advantageous to determine the offsets; aligning
said detected image and said reference image using said determined
offset; comparing said aligned inspection image and said reference
image to detect a defect candidate; and extracting a feature of
said defect candidate.
7. A method according to the claim 6, wherein in the step of
determining an offset, said offset between the patterns of said
detected image and the patterns of said reference image is
determined by evaluating correlations of shift amount between the
patterns of said detected image and the patterns of said reference
image in X-direction and Y-direction and a reliability measurement
over portions of a full image of the pattern.
8. A method according to the claim 7, wherein said offsets between
plural images divided from said detected image and plural images
divided from said reference image are calculated in parallel.
9. A method according to the claim 8, wherein in the step of
aligning, said plural images divided from said detected image and
plural images divided from said reference image are aligned in
parallel, and in the step of said comparing, said aligned plural
images divided from said detected image and plural images divided
from said reference image are compared in parallel.
10. A method for inspecting a specimen, comprising. detecting an
image of a specimen on which patterns are formed; processing said
detected image for detecting defect candidate by determining an
offset between said detected image and a reference image, aligning
said detected image and said reference image using a data of said
calculated offset; comparing said aligned inspection image and said
reference image to detect a defect candidate, and extracting a
feature of said defect candidate; and outputting information on
said extracted feature of said defect candidate, wherein in the
step of processing, said offset is determined by selecting patterns
among said plural patterns which are advantageous to determine the
offset by evaluating correlations between shift amount in
X-direction and shift amount in Y-direction and a reliability
measurement over portions of a full image of the pattern.
Description
CROSS-REFERENCES TO RELATED APPLICATIONS
[0001] The present application is a continuation application and
claims priority from U.S. application Ser. No. 09/802,687, filed
Mar. 8, 2001, and Japanese Patent Application No 2000-184563, filed
on Jun. 15, 2000, which is incorporated by reference.
BACKGROUND OF THE INVENTION
[0002] Japanese laid-open patent publication number Hei 05-264467
describes a known pattern inspection method in which defects are
detected by comparing an inspection image and a reference
image.
[0003] In this method, a line sensor sequentially images an
inspected object on which a repeated pattern is arranged in an
orderly manner. A comparison is made with an image that has been
delayed by a time interval corresponding to the pitch of the
repeated pattern, and inconsistencies are detected as pattern
defects. However, in practice the stage may vibrate or the
inspected object may be tilted so that the two images are not
always aligned. Thus, an offset must be determined for the image
that has been delayed by the pitch of the repeated pattern. The two
images are then aligned based on the determined offset. Then,
differences between the images, e.g., luminance differences, are
determined. If a difference is greater than a defined threshold
value, it is determined to be a defect. Otherwise, it is determined
to not be defective.
[0004] A standard alignment method for two images is an alignment
method where the information from each of the full images are used
to calculate an offset all at once. Problems related to this
standard method are illustrated by FIGS. 1 and 2. FIG. 1 shows
front-view drawings of examples of full images 110, 112, and 114
that tend to lead to failed alignments. FIG. 2 shows front-view
drawings of examples of detection results 120, 122, 124 for full
images that have failed alignments.
[0005] In comparative inspections, alignment of two full images is
generally performed by using edges within the images as information
for offset detection. An offset is calculated so that the edge
offsets between the full images are minimized. There can be cases,
as in FIG. 1 image 110 where there is only an elliptical pattern 21
at the right end of the image, in which there is very little edge
information, i.e., the proportion of edges relative to the entire
alignment region (hereinafter referred to as pattern density) is
very low. There are other cases, as in FIG. 1 image 112 where there
are many vertical edges 22 but there is only a rectangular pattern
23 oriented horizontally, in which edges are predominantly in a
specific direction. There are other cases, as in FIG. 1 image 114
where there are many very small circular patterns 24 with only one
black dot 25, in which a fine-pitch pattern dominates. In all of
these cases, there is a high probability that an offset calculation
error will be generated. Thus, methods that calculate offsets based
on the information for entire images, as in the conventional
technology, have a high probability that erroneous detections
(false positives) will be generated by the offset.
[0006] The detection results from full images, such as in FIG. 1,
with failed offsets are shown in FIG. 2 and will generate erroneous
detections. FIG. 2 images 120, 122, 124 show erroneous detection
positions 31, 32, 33, 34, 35. Furthermore, the possibility of
errors is increased even more if there is luminance shading within
an image or if there is uneven brightness between images. One
method of reducing erroneous detections due to alignment is to
lower the sensitivity. However, not all alignment errors cause a
false detection. In addition lowering the sensitivity, lowers the
detection rate for true defects. Thus unnecessary lowering of the
sensitivity due to alignment errors should be avoided.
[0007] Thus there is a need for improved alignment techniques in
which detection errors are minimally caused by alignment errors due
to pattern density or shape, and/or for improved techniques for
determining when the detection sensitivity needs to be lowered in
comparison inspections.
BRIEF SUMMARY OF THE INVENTION
[0008] The present invention provides a high-precision alignment
method, device and code for inspections that compare an inspection
image with a reference image and detect defects from their
differences. In an embodiment an inspection image and a reference
image are divided into multiple regions. An offset is calculated
for each pair of sub-images. Out of these multiple offsets, only
the offsets with high reliability are used to determine an offset
for the entire image. This allows high-precision alignment with
little or no dependency on pattern density or shape, differences in
luminance between images, and uneven luminance within individual
images. Also, detection sensitivity is adjusted as necessary by
monitoring alignment precision. One embodiment of the present
invention is suited for use in an alignment method implemented for
visual inspection of semiconductor wafers. Other embodiments
include use of a Semiconductor Electron Microscope, X-ray system,
Focus Ion beam, Transparent Electron Microscope, and the like.
[0009] A second embodiment of the present invention provides a
comparative inspection device including: a stage on which an object
is mounted and which moves the object; a detector detecting an
image of the object, the image including a plurality of small
inspection image regions, and outputting an image signal; and an
image processing unit for determining offsets for the inspection
image regions when compared with corresponding reference image
regions, and for determining an offset having a high reliability
out of the offsets for the inspection image regions. The offset
with high reliability is used to align an entire inspection image
and an entire reference image.
[0010] In the second embodiment, the reliability for a small region
can be evaluated as being a high reliability, if patterns in the
small region are dense and can be evaluated as being a low
reliability, if the patterns are sparse. Also, the full image
reliability can be evaluated by comparison with an offset predicted
from past variations of offsets.
[0011] A third embodiment of the present invention provides a
method for aligning comparative inspection images including the
following steps: an image detection step detecting an image of an
object, the image having a plurality of inspection images of small
regions, and outputting an image signal; an offset determining step
determining offsets for the plurality of inspection images; an
offset selection step selecting an offset with a high reliability;
and a step for aligning an entire inspection image and an entire
reference image using the offset with high reliability.
[0012] A forth embodiment of the present invention provides a
method for aligning two images by comparing an inspection image and
a reference image. When the inspection image and the reference
image are compared and aligned, each of the two received images is
divided into a plurality of regions, offsets are calculated for
corresponding divided images, and an offset for an entire image is
determined from a plurality of calculated offsets. In an alternate
embodiment, when determining an offset for the entire image using
the plurality of calculated offsets, the offset for the entire
image is determined using only offsets with high reliability. Also,
when images are received consecutively, reliability of the
determined offset for the entire image is evaluated and, if the
reliability is low, the entire image is aligned using a past offset
having a high reliability. Also, when images are received
consecutively, if an evaluation of reliability for the determined
offset determines that reliability is high, the offset for a
current image is stored as reference data for subsequent
alignments. Also, when images are received consecutively and offset
reliability is to be evaluated, past offsets with high reliability
are collected and these are compared with a current image offset to
determined reliability.
[0013] A fifth embodiment of the present invention provides a
method for comparative inspection including the following steps: a
step for dividing an inspection image into a plurality of regions;
a step for dividing a reference image into a plurality of regions;
a step for calculating offsets between corresponding inspection
images and reference images; a step for determining an offset for
an entire image by using only offsets with high reliability
selected from offsets calculated for divided images; a step for
evaluating reliability of the offset that was finally determined;
and a step for adjusting detection sensitivity based on the finally
determined evaluation result when detecting defects through image
comparison.
[0014] Alternatively, the fifth embodiment can also include a step
for aligning entire images using a past offset with high
reliability if the determined offset for the entire image has a low
reliability. Alternatively, there can also be a step for evaluating
whether or not an alignment error is critical or not for an image
to which the finally determined offset is applied and a step for
adjusting detection sensitivity based on the reliability evaluation
result and the criticality evaluation result.
[0015] A sixth embodiment of the present invention provides a
comparative inspection device including: means for dividing and
inputting two images to be processed; means for simultaneously
calculating offsets for individual images input in a divided
manner; means for evaluating reliability of offsets calculated for
each divided image and calculating an offset for an entire image
based on offsets having high reliability; means for evaluating
reliability of the calculated offset for the entire image and using
the evaluation to determine a final offset; means for aligning the
divided images using the final offset; means for monitoring
alignment precision; means for adjusting detection sensitivity
based on monitoring results; and means for outputting inspection
results along with detection sensitivity. Alternatively, the means
for inputting the images includes a stage moving an inspected
object, an illumination source illuminating the inspected object, a
sensor imaging the inspected object and outputting the image as a
plurality of divided regions, an A/D converter circuit digitizing
an output signal from the sensor, and a memory for storing this;
and the offset calculating means includes an arithmetic module
simultaneously calculating offsets from image data stored in the
memory as a plurality of divided images.
[0016] A seventh embodiment of the present invention provides a
computer program product stored on a computer readable medium for
aligning a first image having a circuit pattern in a semiconductor
material with a second image. The computer program product
includes: code for dividing the first image into a plurality of
regions; code for dividing the second image into a corresponding
plurality of regions; code for determining a first region offset of
a first region of said plurality of regions from a first
corresponding region of the corresponding plurality of regions; and
code for using the first region offset in determining an image
offset for the first image.
[0017] These and other embodiments of the present invention are
described in more detail in conjunction with the text below and
attached figures.
BRIEF DESCRIPTION OF THE DRAWINGS
[0018] FIG. 1 shows examples of front-view drawings of images that
tend to lead to failed alignment;
[0019] FIG. 2 shows examples of front-view drawings of detection
results for images that lead to failed alignment;
[0020] FIG. 3 shows an example of several patterns, a correlation
matrix, and three dimensional (3-D) graphs showing distributions of
correlation values for the patterns;
[0021] FIG. 4 shows a schematic drawing of an architecture of an
embodiment of a comparative inspection device according to the
present invention;
[0022] FIG. 5 shows an embodiment of a bright field inspection
device of the present invention;
[0023] FIG. 6 shows an example of a front-view drawing of a region
on a chip to be aligned;
[0024] FIG. 7 shows a front-view drawing showing a sample image
received by the image processing unit of the comparative inspection
device according to an embodiment of the present invention;
[0025] FIG. 8 shows a schematic diagram describing operations
performed by the image processing unit of the comparative
inspection device according to an embodiment of the present
invention;
[0026] FIG. 9 shows a schematic drawing describing a technique for
calculating offsets according to an embodiment of the present
invention;
[0027] FIG. 10 is a schematic diagram showing an example of images
to be processed by one embodiment of the present invention;
[0028] FIGS. 11a to 11d show examples of schematic diagrams
describing offsets in sub-images;
[0029] FIG. 12 show curves which track the variations in offsets in
the comparative inspection device of an embodiment of the present
invention; and
[0030] FIG. 13 shows a schematic drawing of a chip describing an
example of a method for calculating pattern density in a chip of an
embodiment of the present invention.
DETAILED DESCRIPTION OF THE INVENTION
[0031] The following is a description of the embodiments of the
present invention, with references to the figures.
[0032] In the case of detecting misalignment between an image to be
inspected, i.e., inspection image, and a reference image, the
positional displacement or offset of the inspection image from the
reference image is calculated so as to minimize the displacement of
the edges between the images. However, as shown in FIG. 1 above
there are several cases where there is a high probability of error
in the offset calculation. Some of the cases are: 1) the edge
density is low; 2) the number of edges in a specified direction is
high, and 3) the number of repetitive patterns of fine pitch is
high. Thus when the prior art method of calculating an amount of
positional offset based on information about the entire image is
used, there is a high probability of producing a false detection
due to an incorrect offset at the location of the irregular
pattern.
[0033] In order to show the reasons why such incorrect offset
calculations occur, an example of detecting a positional
displacement between an image to be inspected and a reference image
using a correlation matrix is provided for the purposes of
illustration.
[0034] FIG. 3 shows an example of several patterns, a correlation
matrix, and three dimensional (3-D) graphs showing distributions of
correlation values for the patterns. In FIG. 3 an Inspection image
130 overlays a Reference image 132 in order to calculate the
correlation matrix. An example correlation matrix is shown as 134
with the center 135 of the matrix 134 indicating zero offset, i.e.,
the Inspection image 130 is directly over the Reference image 132.
More specifically, the two images 130 and 132, where a positional
displacement is to be calculated are shifted relatively in a
direction of x by -m to +m pixels and in a direction of y by -i to
+i pixels. For each shift a correlation value is calculated and the
value is entered in a correlation matrix with the position in the
matrix given by the x, y shift. The number of matrix elements
is:
[0035] M=((2*m+1)).times.((2*i+1). In FIG. 3, it is assumed that
m=i=3 is applied. For images 130 and 132, a total number of M=49
values are calculated in response to an amount of shift of the
image 140 over itself. Then, an amount of shift when the value
becomes a maximum value from these correlation values becomes an
amount of positional displacement between the images. In Image 140
the correlation matrix is 134 showing the x-direction shift 136 and
the y direction shift 138. The maximum value is found at the center
135 (when an amount of shift is 0 in both a direction X and a
direction Y), so that an amount of positional displacement of the
image 140 becomes 0 pixels in both a direction of X and a direction
of Y.
[0036] A three dimensional graph showing the x-y offset versus
correlation value for 140 (or matrix 134) is shown as graph 142. As
can be seen, where there is present a sufficient amount of patterns
in the image 140, that is, in the case where a high pattern density
is present, the distribution of values as indicated in graph 142
has a single peak that is near 1.0. However, at a location where no
pattern is present as indicated in image 144, no peak is present in
graph 146 and also its maximum correlation values are also low. In
the case where the pattern is present only at a specified direction
as indicated in image 148, the correlation values become increased
in a direction of the pattern, and a ridge is produced in graph
150. In image 152 a plurality of high peak values are produced on
the graph 154 in the case of repetitive dot patterns at a finer
pitch than the image shift range. In the cases of patterns as
indicated in images 144, 148, and 152, it is hard to find one true
peak value.
[0037] Thus the above example illustrates that there are cases in
which correlation alone over a full image may not be sufficient to
determine an offset. However, as shown later by an embodiment of
the present invention, correlations plus the use of reliability
measurements over portions of a full image allow location of an
offset for the image.
[0038] Next, a comparative inspection device and misalignment
detection method will be described, with references to FIG. 4.
[0039] FIG. 4 is a schematic drawing of the architecture of an
embodiment of a comparative inspection device according to the
present invention. A bright field inspection device is used to
inspect a semiconductor wafer. In the figure, a semiconductor wafer
1 is the inspected object. The semiconductor wafer 1 is secured to
a stage 2, which then moves the semiconductor wafer 1. A detector 3
captures the image of the semiconductor wafer. An A/D converter 4
digitizes the output signal from the detector 3. An image
processing module 5 detects defects by comparing two captured
images. The figure also shows a general control module 6.
[0040] FIG. 5 shows an embodiment of a bright field inspection
device of the present invention. Item 11 denotes a specimen to be
inspected such as a semiconductor wafer; 12 denotes a stage for
mounting the item 11 to be inspected and for moving the item 11; 13
denotes a sensing unit. This apparatus includes a light source 201
for irradiating the item to be inspected 11; an illuminating
optical system 202 for condensing light irradiated from the light
source 201; an objective lens 203 for imaging an optical image
obtained by irradiating the item 11 to be inspected with
irradiating light condensed by the illuminating optical system and
reflecting the illuminated light from the item to be inspected; and
an image sensor 204 for receiving the imaged optical image and
converting it into an image signal corresponding to brightness. 15
denotes an image processing unit for calculating a defect candidate
in the item 11 in reference to an image detected by the sensing
unit 13. The image processing unit 15 includes an A/D converter 14
for converting an input signal from the sensing unit 13 into a
digital signal; a pre-processing unit 205 for performing image
corrections such as a shading correction and a dark level
correction or the like with the digital signal; a delay memory 206
for storing a digital signal of the previous chip as a reference
image signal; a positional displacement detecting unit 207 for
detecting a positional displacement between a digital signal (an
inspection image signal) detected by the sensing unit 13 and the
reference image signal of the delay memory 206; an image comparing
section 208 for performing a positional alignment between the
detected image, i.e., inspection image, and the reference image by
application of the calculated positional displacement amount.
Section 208 also compares both image signals and outputs a part
where a non-coincidence portion is larger than a specified
threshold value as a defect candidate; and a feature extracting
section 209 for calculating the coordinates of defect candidate or
an amount of feature or the like. 16 denotes an entire control
section comprised of a user-interface section 210 having a display
means and an input means for accepting a variation of inspection
parameters (threshold values applied in the image comparison) from
a user or displaying detected defect information; a memory device
211 for storing feature s of detected defect candidates or images
and the like; and a CPU for performing various kinds of controls.
212 denotes a mechanical controller for driving the stage 12 in
response to a control instruction from the control section 16.
Further, the image processing unit 15 and the sensing unit 13 and
the like are also driven by instructions from the control section
16.
[0041] In one embodiment the positional displacement detecting unit
207 divides each inputted inspection image and reference image into
a number of (n) regions. The positional displacement or offset
calculations are carried out for every region. 207-1, 207-2 . . .
207-n denote positional displacement calculating units in the n
channels for calculating in parallel, positional displacements for
each of the region sets (a region set includes a region of an
inspection image and a region of a corresponding reference image).
213 denotes a CPU for sensing a positional displacement or offset
for the entire image based on the "n" positional displacement
amounts determined by the positional displacement calculation
units.
[0042] As shown in the front-view drawing in FIG. 6, the inspected
semiconductor wafer 11 is formed with multiple identically
patterned chips 41, 42 arranged in an orderly manner. In one
embodiment of the inspection device in FIG. 5, images of identical
positions on two chips, e.g., areas 41a, 42a on the chips 41, 42 in
FIG. 6, are compared and differences are detected as defects. The
following is a description of this operation. The general control
module 16 moves the semiconductor wafer 11 continuously using the
stage 12. In tandem with this, the detector 13 captures chip
images. The sensor 204 in the detector 13 is divided into multiple
sensor channels along the direction perpendicular to the direction
in which the stage moves. For example, the sensor may include 32
parallel sensor channels, where each sensor channel has 128 sensing
elements. The incoming signal is sent to the A/D converter 14 as
multiple sub-signals. The A/D converter 14 converts the analog
signal divided into multiple signals into a digital signal, which
is sent to the pre-processing unit 205 of image processing unit 15.
The delay memory 206 is used so that an image signal that has been
delayed by an interval corresponding to one chip's worth of
movement in the stage 12, i.e., the image signal of the chip
previous to the one that is currently set up for inspection, is
sent to positional displacement unit 207 together with the image
signal of the current chip.
[0043] FIG. 7 is a front-view drawing showing a sample image 52
received by the image processing unit 15 of the comparative
inspection device according to an embodiment of the present
invention. An image 52 of a specific area of chip 51 is made up of
a plurality of regions 52a, 52b, 52c, . . . , 52n. The plurality of
regions are sent in parallel to the image processing unit 15 from
sensing unit, i.e., detector, 13. For example, region 52a is sent
to positional displacement calculating unit 207-1, region 52b is
sent to positional displacement calculating unit 207-2, and region
52n is sent to positional displacement calculating unit 207-n. The
plurality of regions are also sent to delay memory 206 for use as
reference when the next chip is inspected. In the positional
displacement calculating units, each 52 region (inspection region)
is compared with the stored 52 region (reference region) from a
previous chip. For example the region 52a (inspection region) is
compared with a previous 52a (reference region) stored in delay
memory 206 in positional displacement calculating unit 207-1 and an
offset determined. The same procedure is done for 52b which is
asynchronously used in positional displacement calculating unit
207-2. Next detector 13 inputs image 52' which includes regions,
52a', 52b', 52c', . . . , 52n'. These 52' regions are sent to
positional displacement calculating units in the image processing
unit 15 and stored in delay memory 206. And the above process for
regions 52 is repeated for regions 52'. The image signals of the
two chips, sent in succession as the stage 12 is moved, will not
correspond to precisely identical positions if the stage vibrates
or if the wafer set on the stage is tilted. Thus, the offset
between the two successively received images must be calculated and
the images must be aligned before performing comparative
inspection.
[0044] Next, an example of a method for detecting offsets used by
the image processing unit 15 will be described, with references to
FIG. 8.
[0045] FIG. 8 is a schematic diagram for the purpose of describing
operations performed by the image processing unit of the
comparative inspection device according to an embodiment of the
present invention. In this example, an bright field inspection
device is used. As shown in FIG. 8, the image signal for the
current chip, received from the detector 13, and the image signal
of the chip detected previously, received from the delay memory 62,
are divided into one or multiple sections (in this example, N
sections are assumed). These sections are processed in parallel.
For example, the area 52 to be detected in a chip is divided up
into regions 52a, 52b, . . . , 52n, which are detected by the
detector 13 and sent to the image processing unit 15. Thus, in this
embodiment the incoming image has as the processing unit fixed-size
images divided up along lines parallel to the direction in which
the stage 12 moves. For example in FIG. 7 the stage is illustrated
to move in the direction 50. The images are detected in time order
of 52, then 52', 52'' and so on. Since each image is made up of
regions, region 52a is sent to the first processing system or first
channel 61a, then after region 52a is processed, region 52a' is
next sent to the first channel 61a and so forth. The processing
systems for the images of the regions 52a, 52b, . . . , 52n are
referred to as channels. In FIG. 8, for example, the image of the
region 52a along with the image of the corresponding region of the
previous image delayed by a delay memory 62 are processed together
by a first channel 61a. The image of the region 52b along with the
image of the corresponding region of the previous image delayed by
the delay memory 62 are processed together by a second channel 61b.
The image of the region 52n along with the image of the
corresponding region of the previous image delayed by the delay
memory 62 are processed together by an n-th channel 61n.
[0046] There are various methods that can be used to calculate
offsets. In this case, FIG. 9 will be used to describe a method
that calculates offsets through normalized cross-correlations.
[0047] FIG. 9 is a schematic diagram for the purpose of describing
a method for calculating offsets according to one embodiment of the
present invention. In the figure, inspection images 71a, 71b, . . .
, 71n are the images of the regions 52a, 52b, . . . , 52n,
respectively. Reference images 72a, 72b, . . . , 72n are the
corresponding images from the previous chip. The inspection image
71a and the reference image 72a delayed by the delay circuit 62 are
sent to the first channel 61a of the image processing unit 15. The
inspection image 71b and the reference image 72b delayed by the
delay circuit 62 are sent to the second channel 61b of the image
processing unit 15. The inspection image 71n and the reference
image 72n delayed by the delay circuit 62 are sent to the n-th
channel 61n of the image processing unit 15. Then, for each of the
channels 61a-61n, the two images are given relative offsets of
-m-+m image elements in the x direction and -m-+m image elements in
the y direction. A total of M correlations are calculated, where
M=((2.times.m+1).times.(2.times.m+1))
[0048] And the x, y position where the maximum correlation value is
found for each channel is determined to be the offset for that
channel, i.e., 73a, 73b, . . . 73n.
[0049] For example, let m=3, and let the reference and inspection
images for 71b and 72b of channel 2, each be a square of size 128
elements by 128 elements. The reference image is fixed and the
correlation calculations are done for an inner square that is
(-64+m) to (64-m) (in this case m=3, so the inner square is -61 to
61) in the x-direction and (-64+m) to (64-m) in the y-direction
(-61 to 61). The inspection image is 128 by 128 and has the same
origin as the reference image. The inspection image is shifted from
-3 to 3 elements with respect to the reference image in the x
direction, -3 to 3 elements with respect to the reference image in
the y direction, and a correlation is calculated for each inner
square associated with an offset, including zero offset. The result
is a (2 m+1).times.(2 m+1), or in this example a 7.times.7, matrix
of correlation values (M=49) with the center matrix value
representing x, y zero offset. The maximum correlation value in the
7.times.7 matrix is found and the x, y offsets associated with that
maximum value is the offset for channel 2, i.e., 73b. The maximum
correlation value and corresponding offset are determined for each
of the N channels. Correlation values are in the range of -1.0-1.0,
with completely identical images resulting in a correlation of
1.0.
[0050] In this example, offsets are determined using normalized
cross-correlation and evaluating reliability using correlations.
However, it would also be possible to use a method involving the
total sum of concentration differences between images. In this
case, offsets are calculated for each channel and corresponding
indexes are set up to evaluate reliability.
[0051] Unlike the conventional technology, this embodiment does not
calculate offsets by using entire images in the processing unit. As
described above, the region 52 is split up into regions 52a-52n and
detected to provide inspection images 71a-71n and reference images
72a-72n. These images are used to determine offsets for channels
61a-61n. In other words, offsets are determined for small regions,
and an offset for the entire image is determined using only the
offsets from the channels having the highest reliability. This
allows detection errors caused by bad offsets to be kept below a
specified rate regardless of the densities or shapes of patterns.
Of course, the ideal detection error rate is 0%, but factors such
as imaging conditions can lead to degraded comparison images. In
this embodiment, detection errors due to bad offsets are greatly
reduced, if not eliminated, regardless of pattern density or shape,
and errors can be kept to 10% or less, and in another specific
embodiment, 1% or less, even if there is image degradation.
[0052] FIG. 10 is a schematic diagram showing an example of images
to be processed by one embodiment of the present invention. Using
these images, the specific advantages of this embodiment will be
described. The two images 71, 72 to be inspected have almost no
patterns, with only inspection images 71c, 71n and reference images
72c, 72n containing patterns in predetermined directions. These two
patterns are formed in different directions.
[0053] Offset detection is performed primarily using edge
information in the conventional method. However, pattern density is
extremely low with these types of images, and calculating offsets
from entire images as in the conventional technologies can result
in a high offset calculation error rate simply from slight
brightness shading within images or uneven brightness between
images. If offsets are determined for each of the channels 61a-61n
and the offset with the highest reliability is selected and applied
to all channels, the selected offset will be for the channel 61c or
the channel 61n, which are the channels in which the images contain
patterns. However, both the channel 61c and the channel 61n contain
edges that are predominantly in specific directions. Thus, it is
likely that there will be calculation errors in the direction of
the arrows shown in FIG. 11a or FIG. 11b relative to the correct
offset. This is because there is little edge information in the
direction of the arrows. As a result, applying the offset for
either the channel 61c or the channel 61n to all channels will
generate an offset error in the channel containing the other
pattern, leading to a detection error.
[0054] In contrast, with the offset method of this embodiment, the
correct offset can be calculated, as will be described below with
references to FIG. 11c and FIG. 11d. FIGS. 11a to 11d show
schematic diagrams for the purpose of describing offsets in
sub-images. This embodiment does not use the offsets for channels
with low reliability, such as the channels 61a, 61b that process
the empty images 71a, 71b, 72a, 72b, and the like. A composite
offset for the entire image is calculated only with the offsets of
the channel 61c (not shown) associated with image 71c, and the
channel 61n, which contain patterns. Since the offsets of the
channel 61c and the channel 61n have high reliability in the
directions of the arrows in FIG. 11c and FIG. 11d, the correct
offset can be calculated in both the x direction and the y
direction by using only the offsets for the directions having high
reliability. Note there is a lot of edge information in the
direction of the arrows. As a result, detection errors caused by
offset errors influenced by pattern density or shapes can be
greatly reduced. Also, since statistics needed for determining
offsets, e.g., normalized cross-correlation or sums of
concentration differences, are calculated within the small regions
52a-52n, the effect of luminance shading within an image and uneven
brightness between images is minimized, thus reducing detection
errors caused by offset errors.
[0055] This embodiment uses only channel information that has high
reliability, e.g., out of the images 71a-71n, 72a-72n, only the
images with the highest pattern densities. From this, reliability
is evaluated for calculated offsets for the entire image. If
calculated offsets have low reliability or the calculated offsets
from all the channels do not contain a single offset with high
reliability, the offset is not determined from the current image.
Instead, alignment is performed using the offset from at least one
iteration back. For example, if pattern density is extremely low as
in the images in FIG. 10, or if an offset with high reliability
cannot be found for any of the channels, this embodiment refers to
a past offset from images having adequate pattern densities and
uses that offset. Conversely, if the current alignment image
contains adequate pattern density and the calculated offset has a
high reliability, the offset is stored in memory as reference data
for subsequent images that can only provide offsets with low
reliability.
[0056] An example of a method for evaluating full-image offset
reliability will be described with references to FIG. 12.
[0057] FIG. 12 show curves 80 and 84 which track the variations in
offsets in the comparative inspection device. In the figure, the
horizontal axis represents the number of inspections and the
vertical axis represents offset value. The reliability of a
calculated offset can be evaluated by examining the relationship
between the calculated offset position and the curve formed by the
past offsets in the curve. For example, curve 80 is an example
where past offsets are plotted over time. Due to cyclic vibration
in the device or the like, the offsets show cyclic variations. In
devices such as this one that show cyclic offset variations, the
path of past offsets can be plotted to predict the current offset
through extrapolation. In the figure, an offset 82 indicates the
current predicted offset. If the offset that is actually calculated
is close to predicted offset 82, the offset is inferred to have a
high reliability. In one embodiment, an offset can be considered to
be reliable if there is an error of no more than +/-a predetermined
tolerance value, for example, 5%, between the actually calculated
offset and the predicted offset 82. In this manner, calculated
offsets can be evaluated more accurately by evaluating offset
reliability through comparison with accumulated past offset
values.
[0058] In the above description, the offset varies cyclically and
prediction values are calculated through extrapolation. However,
other methods can be used to calculate prediction values.
[0059] For example, as shown in FIG. 10 curve 84, only very recent
offset values can be considered and the average of these values can
be used as the prediction value. In the figure, an offset 86 is the
average offset value. If the current measured offset is close to
the offset 86, the measured offset can be evaluated as being
correct. Regardless of the method used to calculate prediction
values, the past offsets that are considered must have high
reliability. Thus, if the offset calculated for the current
alignment image has a high reliability, it is saved as history data
to be used for calculating subsequent prediction values.
[0060] Another example of a method for evaluating reliability is to
use the pattern density of the alignment image. For example, since
the offset error rate will be low if there are adequate patterns in
the x direction and the y direction, these cases are determined to
have a high reliability. One example of a method for measuring
pattern density is to determine differentials in the x direction
and the y direction for each image element in the full image. There
are generally various types of operators for calculating
differentials. One example will be described with references to
FIG. 13.
[0061] FIG. 13 is a schematic diagram of an alignment image for the
purpose of describing a method for calculating pattern density in a
chip of an embodiment of the present invention. The figure shows an
alignment image 90. An image element E 92 in the chip image 90 will
be examined. Using adjacent values, for example, the signal
amplitude values for image elements B 94, D 95, F 98, and H 96,
i.e., v(B), v(D), V(F), and v(H), the following calculations are
performed.
[0062] Differential in signal value in the x direction at
E=(v(B)+v(H)-2.times.v(E))
[0063] Differential in signal value in the y direction at
E=(v(D)+v(F)-2.times.v(E))
[0064] Then, an evaluation is made based on the proportion (e.g.,
10% or higher) of image elements having values greater than a
predetermined threshold value relative to the total number of image
elements.
[0065] Also, reliability can be evaluated using a combination of
multiple combinations, e.g., evaluating using both comparisons with
past offset values and current image pattern densities.
[0066] By checking reliability in this manner, the alignment
precision based on calculated offsets can be monitored using a
computer to see if positional alignment is performed correctly. If
alignment precision is determined to be bad, a high possibility of
detection error due to alignment error is assumed and the detection
sensitivity is lowered beforehand.
[0067] If, for example, alignment precision is monitored based on
pattern density, images with low densities will, of course, have
low alignment reliability. However, if a pattern contains no
patterns at all, for example, there is a high probability that an
offset is mistakenly determined to be accurate, and there will be
no detection error even if alignment takes place with a faulty
offset. Similarly, detection errors will not take place when the
image region contains only patterns in a specific direction, such
as in FIG. 3 image 148. Therefore, in this embodiment, if alignment
precision is found to be bad, an evaluation is made to determine if
the failed alignment is critical for that image, i.e., if it will
lead to a detection error.
[0068] In one embodiment, edge information, both direction and
value, is checked for each channel, if it is above a fixed
threshold. If it is, then the feature is critical. In FIG. 1 image
112, for example, if there is a misalignment in the vertical
direction, then there is a high probability that a horizontal
feature 23 will give a false defect reading. Thus, horizontal
feature 23 is critical and detection sensitivity may be lowered.
However, for vertical feature 22 a change in the vertical direction
is unlikely to produce a false defect reading. Thus vertical
feature 22 is not critical, i.e., a failed alignment is acceptable
and the detection sensitivity is not lowered. By determining the
criticality for a particular image in this manner, needless
reductions in detection sensitivity are avoided. By accurately
monitoring alignment precision and determining criticality of
alignment errors in the current image, detection sensitivity can be
reliably reduced when necessary to reduce detection errors (false
positives) while also keeping the reductions of detection
sensitivity to the necessary minimum. This makes high-sensitivity
inspection possible over a wider range.
[0069] The description above presents an example where an
embodiment of an image alignment method, a comparative inspection
device, and an alignment precision monitoring method according to
an embodiment of the present invention are implemented for a bright
field inspection device for semiconductor wafers. However, the
present invention can also be implemented for aligning comparison
images in electron beam pattern inspection systems and DUV
inspection systems.
[0070] Also, the inspected object is not restricted to
semiconductor wafers, and anything for which defect detection is
performed through image comparison can be used. For example, the
present invention can be used for TFT substrates, photomasks,
printed plates, and the like.
[0071] In order to overcome the problems of the conventional
inspection technology, one embodiment of the present invention as
described above provides a comparative inspection that compares an
inspection image and a reference image and detects defects based on
the difference between these images. The two images being compared
are each divided into multiple regions, and offsets are calculated
for each of the smaller regions. Of the calculated offsets, only
the ones with high reliability are used to calculate an offset for
the entire image. Thus, a high-precision alignment technology that
is not dependent on pattern density or shape can be provided.
[0072] This also allows a highly sensitive comparative inspection
method and device that reduces detection errors caused by alignment
errors to be provided. Also, by monitoring alignment precision and
reducing inspection sensitivity only when there is failed
alignment, the reduction of detection sensitivity can be kept to
the necessary minimum, and a highly sensitive comparative
inspection method and device can be provided.
[0073] One specific embodiment of the present invention provides
alignment precision in some cases of about 1 image element, and
more specifically in some cases, 0.1 image element or less, and a
rate of detection error caused by alignment errors of less than
10%, and more specifically in some case less than 1%, regardless of
pattern density or shape and even when the comparison images are
degraded.
[0074] Although the above functionality has generally been
described in terms of specific hardware and software, it would be
recognized that the invention has a much broader range of
applicability. For example, the software functionality can be
further combined or even separated. Similarly, the hardware
functionality can be further combined, or even separated. The
software functionality can be implemented in terms of hardware or a
combination of hardware and software. Similarly, the hardware
functionality can be implemented in software or a combination of
hardware and software. Any number of different combinations can
occur depending upon the application.
[0075] Many modifications and variations of the present invention
are possible in light of the above teachings. Therefore, it is to
be understood that within the scope of the appended claims, the
invention may be practiced otherwise than as specifically
described.
* * * * *