U.S. patent application number 12/030019 was filed with the patent office on 2008-08-07 for defect image classifying method and apparatus and a semiconductor device manufacturing process based on the method and apparatus.
This patent application is currently assigned to Hitachi, Ltd.. Invention is credited to Seiji ISOGAI, Toshiei KUROSAKI, Ryo NAKAGAKI, Kenji OBARA, Yasuhiro OZAWA, Yuji TAKAGI.
Application Number | 20080187212 12/030019 |
Document ID | / |
Family ID | 18306356 |
Filed Date | 2008-08-07 |
United States Patent
Application |
20080187212 |
Kind Code |
A1 |
OBARA; Kenji ; et
al. |
August 7, 2008 |
Defect Image Classifying Method and Apparatus and a Semiconductor
Device Manufacturing Process Based on the Method and Apparatus
Abstract
According to the present invention, techniques including a
method and apparatus for classifying and displaying images are
provided. In an embodiment of the present invention a defect image
classification method using inspected objects is provided. The
method includes defect images obtained from at least one inspected
object. Next a set of defect images is classified into a specified
category, which has a feature. The defect images are arranged for
display according to the feature and then displayed. The arranging
of the defect images may also be based on an evaluation value for
each defect image. Another embodiment provides a defect image
classification method using inspected objects. Defect images are
obtained from at least one inspected object. Next the defect images
are classified into a plurality of categories and at least two
information items for example, a defect distribution diagram
showing locations of defects in the inspected object, information
associated with a category of the plurality of categories, and a
defect size distribution, are displayed.
Inventors: |
OBARA; Kenji; (Ebina,
JP) ; TAKAGI; Yuji; (Kamakura, JP) ; NAKAGAKI;
Ryo; (Kawasaki, JP) ; OZAWA; Yasuhiro; (Abiko,
JP) ; KUROSAKI; Toshiei; (Hitachinaka, JP) ;
ISOGAI; Seiji; (Hitachinaka, 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: |
18306356 |
Appl. No.: |
12/030019 |
Filed: |
February 12, 2008 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
11506310 |
Aug 17, 2006 |
7356177 |
|
|
12030019 |
|
|
|
|
09716077 |
Nov 17, 2000 |
7113628 |
|
|
11506310 |
|
|
|
|
Current U.S.
Class: |
382/149 |
Current CPC
Class: |
G06T 2207/30148
20130101; G06T 7/001 20130101 |
Class at
Publication: |
382/149 |
International
Class: |
G06K 9/00 20060101
G06K009/00 |
Foreign Application Data
Date |
Code |
Application Number |
Nov 29, 1999 |
JP |
11-337196 |
Claims
1. A defect image classification system for classifying defect
images detected by a defect detection device, comprising: a display
means for displaying plural uncategorized defect images on a
screen; defect categorization means for specifying a defect
category of a defect selected from among the plural uncategorized
defect images by displaying an image of the selected defect in an
area on the screen that is associated with the defect category;
feature calculating means for calculating features from an image of
the selected defect having the defect category specified by the
defect categorization means; and rearranging means for rearranging
a display order of the plural uncategorized defect images in an
area on the screen outside of the area that is associated with the
defect category, the display order of the plural uncategorized
defect images being based on their similarity with the calculated
feature.
2. A defect image classification system according to the claim 1,
wherein the display order of the plural uncategorized defect images
is further based on defect size, brightness, color, and shape of
the plural uncategorized defect images.
3. A defect image classification system according to the claim 1,
wherein the display means is further for displaying each of the
plural uncategorized defect images with substantially the same
scale.
4. A method for classifying defect images detected by a defect
detection device, comprising the steps of: displaying plural
uncategorized defect images on a screen; specifying a category of a
defect from among the plural uncategorized defect images displayed
on the screen by displaying an image of the defect in an area on
the screen which is associated with a defect category; calculating
one or more features of the image of the defect specified in the
step of specifying; and arranging a display order of the plural
uncategorized defect images displayed on the screen in an area of
the screen outside of the area associated with the defect category,
wherein the display order is based on similarity with the
calculated feature.
5. A method for classifying defect images according to the claim 4,
wherein the display order is further based on size, brightness,
color, and shape of each of the plural uncategorized defect
images.
6. A method for classifying defect images according to the claim 4,
wherein in the plural uncategorized defect images are displayed on
the screen with substantially equal scale.
Description
CROSS REFERENCE TO RELATED APPLICATION(S)
[0001] This application is a continuation of U.S. patent
application Ser. No. 11/506,310, filed Aug. 17, 2006, which is a
continuation of U.S. patent application Ser. No. 09/716,077, filed
Nov. 17, 2000, (now U.S. Pat. No. 7,113,628) which is related to
and claims priority from Japanese Patent Application No. 11-337196
filed on Nov. 29, 1999, all of which are incorporated by reference
herein in their entirety for all purposes.
[0002] This application is related to the following commonly
assigned, non-provisional patent application, the disclosure of
which is herein incorporated by reference in its entirety: U.S.
patent application Ser. No. 09/661,182 filed Sep. 13, 2000 (now
U.S. Pat. No. 6,553,323).
BACKGROUND OF THE INVENTION
[0003] This invention relates to a method and apparatus for
classifying defect images obtained by imaging of contaminations or
defects detected on semiconductor wafers and a semiconductor device
manufacturing process based on the method and apparatus.
[0004] A semiconductor device is manufactured by making a wafer as
a substrate go through several processing steps including exposure,
development and etching. After a specific step among these
processing steps is over, the locations and sizes of contaminations
or defects on the wafer (hereinafter collectively referred to as
"defects") are checked by a particle inspection system or a defect
inspection system. Enlarged images of all or some of the
contaminations or defects detected by this inspection are produced
using a magnifying imaging device such as an optical microscope or
SEM (Scanning Electron Microscope) to obtain detailed information
on their size, shape and texture (surface pattern). This
information has been used in order to identify the step which is
the particle or defect source.
[0005] In recent years, a magnifying imaging system which
incorporates a function to automatically produce enlarged images of
contaminations or defects based on inspection result data obtained
from the particle or defect inspection system, or ADR (Auto Defect
Review), has been developed, which has made it possible to collect
a huge volume of image data in a shorter time. However, as an
increasing number of images are obtained, operating personnel must
consume a greater amount of time and energy to analyze the images.
In order to improve the working efficiency in such image analyses,
a function to automatically classify the obtained defect images
into categories according to their features, ADC (Auto Defect
Classification), has been developed. Automatic classification of
defect images using this function has contributed to improvement in
analysis working efficiency by allowing operating personnel to
analyze only defect images of a category to be noted or make an
analysis paying attention only to the number of defects in each
category.
[0006] A method to realize ADC has been disclosed in Japanese
Published Unexamined Patent Application No. Hei 8-21803. In this
method, a standard image for each defect category, called "teacher
image data," is prepared and this data is used to execute a neural
network learning session; then, for automatic classification, the
data thus learned is used to decide to which category each input
defect image should belong.
[0007] An example of a method to realize the function to
automatically produce enlarged images of contaminations or defects
based on inspection result data obtained from the particle or
defect inspection system, or ADR, has been disclosed in Japanese
Published Unexamined Patent Application No. Hei 9-139406 in which
secondary electronic images from an electron microscope are used to
observe contaminations or defects at optimum magnification ratios
based on inspection result data obtained from the particle or
defect inspection system.
[0008] However, in order to realize a system for auto defect
classification as mentioned above, the user must teach the system
the feature of each category by some means, whether a neural
network as mentioned above is used or not, or whatever category
judgment algorithm is used. In other words, it is necessary to
teach the system the relationships between images and the
categories they should belong to, for example, in the following
ways: the user observes a defect image and decides to which
category it should belong, or the user prepares data which
describes the features of an image group which the user considers
as belonging to a category. As the user handles more images to
prepare higher quality teaching data, the user has to spend more
time and energy in this work. Thus there is a need for an efficient
way to provide the system with teaching data.
[0009] In addition, after teaching has occurred and automatic
classification has been done, there still is the problem of
effectively presenting the defect data to the user for analysis.
Thus there is a need for an effective display of classified defect
image data, so that the user may detect/evaluate the defects and,
if needed, improve parts of the semiconductor device manufacturing
process.
BRIEF SUMMARY OF THE INVENTION
[0010] According to the present invention, techniques including a
method and apparatus for classifying and displaying images are
provided.
[0011] An object of one embodiment of this invention is to provide
a defect image classifying method and apparatus capable of
improving the teaching efficiency in automatically classifying
defect images according to teaching data given in advance, as well
as a semiconductor device manufacturing process based on the method
and apparatus; that is to provide, in an embodiment, a defect image
classifying method and apparatus in which teaching data required to
classify huge volumes of image data concerning contaminations or
defects obtained from the particle inspection system or defect
inspection system can be created while saving the user time and
labor, and a semiconductor device manufacturing process based on
the method and apparatus.
[0012] An object of another embodiment of this invention is to
provide a defect image classifying method and apparatus in which
the result of automatic classification is presented to the user in
a user-friendly manner, and a semiconductor device manufacturing
process based on the method and apparatus.
[0013] According to one embodiment of this invention, defect images
of objects of inspection, which have been produced by imaging of
the objects, are classified into one or more categories. Next, the
defect images are arranged and displayed on the screen in a manner
to match the feature of a specified category of images.
[0014] In another embodiment of this invention, defect images of
objects of inspection, which have been obtained by imaging of the
objects, are classified into categories. Next, as a result of
classifying defect images into categories, at least two among three
types of information: a distribution diagram showing the locations
of defects in the objects, information on the categories, and
information on defect size distribution, are displayed on a display
screen simultaneously, and if a distribution diagram appears on the
screen, either the shape or color, or both, of points showing the
locations of defects in the distribution (scatter) diagram, is
varied depending on the category or defect size or their
combination.
[0015] In yet another embodiment of this invention a defect image
classification system for classifying defect images, produced from
at least one object of inspection by imaging, into a plurality of
categories, is provided. The system has software which includes: a
feature calculating module to calculate features of the defect
images; categorization module to specify a set of defect images to
match a specified category; and evaluation module to calculate
evaluation values of the defect images using the features of the
defect images calculated by the feature calculating module and
using the features of the set of defect images. In addition, a
display unit with a screen is provided to show the defect images;
and a sorting software module is provided to alter a display order
of the defect images displayed by the display.
[0016] In a further embodiment of this invention, a semiconductor
device manufacturing process is designed to include the following
steps: a processed substrate which has undergone prescribed
processing steps is inspected for contaminations or defects; images
of contaminations or defects in this processed substrate are
produced; these defect images are processed and classified into
categories; a defect distribution for a category related to trouble
in a specific processing step is monitored; and if the defect
distribution is found abnormal by this monitoring, information to
warn of the abnormality is outputted.
[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 is an explanatory view showing the role of defect
image classifying apparatus according to an embodiment of this
invention in the whole semiconductor manufacturing process.
[0019] FIG. 2 is a flowchart showing the steps taken to classify
defect images according to an embodiment of this invention.
[0020] FIG. 3 shows an example of a GUI screen displaying defect
images in classified form according to an embodiment of this
invention.
[0021] FIG. 4 shows an example of a GUI screen displaying defect
images in classified form according to an embodiment of this
invention.
[0022] FIG. 5 shows an example of a GUI screen displaying defect
images in the order of feature in classifying defect images
according to an embodiment of this invention.
[0023] FIG. 6 shows an example of similarity calculation in an
embodiment in classifying defect images according to an embodiment
of this invention.
[0024] FIG. 7 shows an example of a GUI screen displaying defect
images in classified form according to an embodiment of this
invention.
[0025] FIG. 8 shows an example of a GUI screen displaying defect
images in classified form according to an embodiment of this
invention.
[0026] FIG. 9 shows an example of estimation of categories
corresponding to defect images in classifying defect images
according to an embodiment of this invention.
[0027] FIG. 10 shows an example of a GUI screen displaying defect
images in association with features in classifying defect images
according to an embodiment of this invention.
[0028] FIG. 11 shows an example of a GUI screen displaying defect
images in association with features in classifying defect images
according to an embodiment of this invention.
[0029] FIG. 12 shows an example of a GUI screen displaying defect
images in association with features in classifying defect images
according to an embodiment of this invention.
[0030] FIG. 13 shows an example of a GUI screen displaying defect
images in association with features in classifying defect images
according to an embodiment of this invention.
[0031] FIG. 14 shows an example of a GUI screen displaying defect
images as divided into groups by size in classifying defect images
according to an embodiment of this invention.
[0032] FIG. 15 shows an example of a GUI screen displaying defect
images as divided into groups by size in classifying defect images
according to an embodiment of this invention.
[0033] FIG. 16 shows an example of a GUI screen displaying defect
images in classified form according to an embodiment of this
invention.
[0034] FIG. 17 shows an example of a GUI screen displaying defect
images whose sizes vary depending on the imaging magnification
ratio, in classifying defect images according to an embodiment of
this invention.
[0035] FIG. 18 shows an example of a GUI screen showing the result
of automatic classification of defect images according to an
embodiment of this invention.
[0036] FIG. 19 shows an example of a GUI screen showing the result
of automatic classification of defect images according to an
embodiment of this invention.
DESCRIPTION OF THE SPECIFIC EMBODIMENTS
[0037] Embodiments of this invention are described next referring
to the attached drawings.
[0038] FIG. 1 shows the role of defect image classifying apparatus
according to an embodiment of this invention, in the whole
semiconductor manufacturing process. For a wafer which has
undergone one or more processing steps, a contamination/defect
inspection is made using a defect inspection system such as model
I-890 from Hitachi, Ltd. and ADR is performed using an imaging
device such as an SEM; the image data obtained by ADR is sent to
defect classifying apparatus where ADC takes place to classify
defect images into categories. Next for illustration purposes, it
is assumed that a preliminary analysis has revealed that defects of
categories A, B, and C frequently occur in step A, step B, and step
C, respectively. A defect distribution by category, which gives the
locations of defects of, for example, category B may be generated.
If the defect distribution is abnormal, trouble in step B is
suspected and a warning is given to suggest that some corrective
measures for step B should be taken. In other words, in comparison
with a case in which no such defect classification is made and thus
the whole process is judged normal or abnormal from a decrease or
increase in the total number of defects, the introduction of a
defect classification step as mentioned above would provide the
user with more detailed information on defects. In addition, if a
new type of defect which does not belong to any preset category is
found, information on occurrence of a new type of trouble can be
obtained. Also, information to help find the possible cause of
processing step trouble can be given to the user by displaying
defect distribution tendencies category by category or size by
size.
[0039] FIG. 2 is a flowchart showing an example of a sequence for
automatic classification by the defect image classifying apparatus
according to an embodiment of this invention. The classifying
apparatus performs two phases of operation: (a) creation of
teaching data and (b) execution of automatic classification.
[0040] In (a) the phase of creation of teaching data consists of
the following steps. First, as for a wafer whose defect positions
are already known as a result of an inspection, defect images are
obtained by an imaging device such as an optical microscope or SEM
with the image pick-up position varying according to the defect
location data (step 100). Features of the obtained images are
calculated (step 101). Here, features refer to the size,
brightness, color, shape, texture and so on of a defect. In order
to calculate the features, a reference image which has the same
background as the defect image but has no defect may also be
produced (for example, a reference image with the same on-chip
relative position or an image taken at a different imaging angle
such as an overhead view).
[0041] The list of obtained defect images is shown to the user;
then the user classifies images into categories to teach the system
how to classify (step 102). FIG. 3A shows an example of Graphic
User Interface (hereinafter referred to as GUI), which is displayed
for the user during the teaching process, of one embodiment of the
present invention. This GUI is composed of several windows. Window
301 is an unclassified defect image window showing defect images
which the user has not classified yet. Windows 302, 303 and 304
show defect images classified into category A, category B and
category C, respectively. The three category windows in this
example, are for illustration purposes only; and the number of
category windows is not limited.
[0042] Classification is done by manually moving defect images
(data) in the unclassified image window into category windows. For
instance, if the user moves a certain image from the unclassified
image window 301 to the category A window 302, that image is
classified as an image of category A.
[0043] Then, teaching data as basic information for classification
is created to tell the system what feature of an image corresponds
to which category (step 103).
[0044] In the phase of teaching data creation shown in the
flowchart of FIG. 2, it is also possible to carry out step 101, in
which image features are calculated, after teaching step 102, in
which data is given to the automatic classification system. Also,
it is not necessary to carry out this teaching data creation phase
upon every imaging of a wafer to obtain a defect image (step 100);
it should be carried out as needed.
[0045] In (b) the phase of automatic classification, as in (a) the
phase of teaching data creation, as for a wafer whose defect
positions are already known as a result of an inspection, defect
images and reference images required for feature calculation are
obtained by an imaging device such as an optical microscope or SEM
with the image pick-up position varying according to the defect
location data (step 201). Then, features for the image group are
calculated (step 202). The calculated features are checked against
the teaching data and categorization takes place using a
classification algorithm such as a neural network and/or fuzzy
inference (step 203).
[0046] FIG. 3B illustrates a block diagram hardware description of
one embodiment of the present invention. FIG. 3B shows an Imaging
Device 320 coupled via a communications Network 330 to a Defect
Image Classifying Apparatus 340. The Defect Image Classifying
Apparatus 340 includes Memory 342, Processor 344, Network Interface
346, Output Interface 347, Input Interface 348, and Internal Bus
349, which couples the above components. The Network Interface
component 346 couples the Apparatus 340 to the Network 330. A
Display 350 is coupled to the Apparatus 340 via an Output Interface
347 and is used by Apparatus 340 to display the defect images and
feature information of embodiments of the present invention to a
user. The Input Interface 348 allows the user via Mouse and/or
Keyboard 360 to enter information and selections into the Apparatus
340. The processor may be, for example an Intel Pentium
microprocessor. The memory, includes both volatile and non-volatile
memory and stores the software as described in FIGS. 2, 3A and
other embodiments of the present invention.
[0047] Next, how images are categorized in the phase of teaching
data creation according to an embodiment of this invention is
explained.
[0048] For example, as shown in FIG. 4, the user picks up one or
more images which he or she considers as typical images of category
A and moves them to the category A window 402. Then, upon some user
action such as specifying the category A window 402 or pushing a
button after specifying it or choosing from a menu, images in the
unclassified image window 401 are rearranged according to the
degree of similarity to the category A window 402. FIG. 5 shows the
result of the re-arranged unclassified image window 501. The degree
of similarity is calculated, for example, by the following
method.
[0049] For each defect image, features for predetermined feature
items such as defect size, brightness, color, and shape are
calculated in advance. To simplify the explanation, it is assumed
here that two features are used. Regarding defect images which the
user has classified into category A, the position of each image is
plotted in the feature space 600 defined by two features, feature 1
601 and feature 2 602, as indicated in FIG. 6. It is supposed that
these images have similar features because the user has considered
them as belonging to the same category. Hence, distribution of
points plotted in the feature space 600 which represent the
features of defect images is expected to have a certain tendency.
Here, the center of gravity, G 605, in this distribution is
calculated. For unclassified defect images, the calculated features
of images 1 610, 2 611, and 3 612 are plotted in the same feature
space 600 as shown in FIG. 6. Then, their distances d1, d2, d3 from
the center of gravity G 605 are calculated; the shorter the
distance is, the more similar the image is. Thus, the images are
rearranged and redisplayed in the order of similarity.
[0050] This allows the user to pick up images to be classified into
a category by examining only images with high degrees of similarity
instead of all defect images of the group obtained as images for
teaching, thereby improving the working efficiency in teaching data
creation.
[0051] Here, as a method for specifying images whose attributes
correspond to black dots as plotted in FIG. 6, it is not always
necessary for GUI to provide a means to move images into category
windows. If GUI allows the user to specify an image which he or she
considers as belonging to the same category, it suffices. The
display order of images in a category window may also be changed
according to evaluations like the degree of similarity discussed
above.
[0052] With the arrangements, in mapping in the feature space, it
is possible to show the user which defect image is less coherent to
the typical category features than the other images of the same
category. The user can increase the feature coherence of teaching
data by removing images which he or she judges atypical or
inappropriate as teaching data.
[0053] Also, as shown in FIG. 7, user interface may be so designed
that when each category window contains one or more images, for
example 703, if an unclassified image 810 is chosen as shown in
FIG. 8, the category representing features closest to the feature
of the image may be made distinguishable from the other categories
(in the case of FIG. 8, the title of category C 803 is hatched).
The category closest in features can be determined as follows.
[0054] First, the quantities of predetermined features of the
defect of each defect image (size, brightness, color, shape and so
on) are calculated in advance. To simplify the explanation, it is
assumed here that two features are used. Regarding the images which
the user has classified into categories, the features of the images
are plotted in the feature space 910 defined by two features,
feature 1 912 and feature 2 914, in FIG. 9. This example assumes
that the images are classified into three categories, A 920, B 922,
and C 924. It is supposed that the images in each category are
similar in their features because the user has considered them as
belonging to the same category. Hence, distribution of points
plotted in the feature space 910 which represent the features of
defect images is expected to have a certain tendency. Here, the
center of gravity in this distribution, GA 930, GB 932, or GC 934,
is calculated for each category. Then, the features 940 of an image
chosen from among the unclassified defect images are plotted in the
feature space 910. Then, its distance dA, dB, or dC from the center
of gravity for each category, GA 930, GB 932, or GC 934, is
calculated. The category whose center of gravity is nearest to the
plotted point representing the features 940 of the chosen image is
suggested to the user.
[0055] In this way, the user is given a suggestion about a
theoretically closest category before he or she classifies an image
into a category, which saves the user time and labor in
classification work.
[0056] Explained next is an embodiment of a display method to show
the user correlations between images and their features in the
phase of teaching data creation according to an embodiment of this
invention.
[0057] An embodiment of a display method according to this
invention is shown in FIG. 10. Reference numeral 1002 represents a
window which displays images as classified into category A and 1005
represents a feature distribution window which shows histograms
indicating distributions of evaluation values of various features
of images classified into category A.
[0058] Images and histograms are displayed simultaneously in this
way, and if an image, for example 1110, is chosen in the category A
window 1002, the bars in the histograms corresponding to the
features of that image, for example, 1112, 1114, 1116, and 1118 are
identified by color. As an example the window for feature 1 1111,
may measure the value of brightness.
[0059] This tells the user where an image in a category is
positioned in the feature distribution of the category. In short,
the user can determine whether the feature of the chosen image is
positioned around the center of the feature distribution of the
category or in a marginal area or in a position atypical of the
category; this provides the user with information to improve
coherence in the features of the category by removing, from
teaching data, an image which is atypical of the category.
[0060] Conversely, as shown in FIG. 11, user interface may be
arranged so that if the user chooses a bar, for example 1110, in a
histogram 1111, in the feature distribution window 1115, the
images, for example, 1112, 1114, 1116, and 1118, whose features
corresponds to the chosen histogram bar 1110 are made
distinguishable from the other images in the category A window
1120.
[0061] This enables the user to know which image is atypical of the
category, from the feature distribution histogram; this provides
the user with information to improve coherence in the features of
the category, for example, by removing, from the teaching data, an
image which is atypical of the category.
[0062] Feature distributions may be presented in the form of
scatter diagrams in a feature scatter diagram window 1206 as shown
in FIG. 12, instead of histograms. In this case, two or three
features are selected and two-dimensional or three-dimensional
feature spaces are generated; then scatter diagrams representing
the features of a group of images related to a category are shown
in the feature spaces. It is also possible to arrange that if an
image, for example 1210, is chosen in the category A window 1202,
the points, 1212, 1214, and 1216, corresponding to that image 1210
is made identifiable by color in the feature scatter diagrams
window 1206. The user may select the axis for each feature. For
example in window 1206, feature 1 has the x-axis selected 1220,
feature 3 the Y-axis 1222, and feature 4 the Z-axis 1224. Two
dimensional scatter diagram 1230 has feature 4 (Z-axis) versus
feature 1 (X-axis). Two dimensional scatter diagram 1232 has
feature 4 (Z-axis) versus feature 3 (Y-axis). Three dimensional
scatter diagram 1234 has feature 4 (Z-axis) versus feature 1
(X-axis) and feature 3 (Y-axis). Note in this example, diagrams
1230 and 1232 are projections of diagram 1234.
[0063] Conversely, user interface may be arranged so that when a
point, for example 1300 in a feature space is chosen in the feature
scatter diagram window 1306, the image 1310 whose features
correspond to that point 1300 can be identified in the category A
window 1302.
[0064] Described next is how the sizes of defects as evaluation
values are classified into groups and categorization in each group
takes place according to an embodiment of this invention.
[0065] What feature items can be used in the user's evaluation of a
defect depends on the size of the defect. For instance, if the
defect is too small, it is difficult for the user to evaluate it
with regard to shape ("round," "square" and so on).
[0066] Also, when the texture of the area of the defect is to be
evaluated, if the defect is too small, the influence of noise may
be increased because of the smallness of the area to be analyzed,
thus resulting in a failure in correct texture evaluation. If the
texture evaluation data thus taken is used as teaching data,
coherence in the features of each category may deteriorate.
[0067] For these reasons, images are divided into two or more
defect size groups and categorization takes place in each size
group.
[0068] Switching from one size group to another is possible and the
user can observe the desired size group. Assuming that images are
divided into two defect size groups, an example of a larger defect
group is shown in FIG. 14 and one of a smaller defect group is
shown in FIG. 15.
[0069] For the larger defect size group, categories related to
texture details (e.g. "coarse" or "angular" texture, illustrated by
images, e.g., 1402 and 1403, in window 1401 of FIG. 14) are added,
while for the smaller defect size group, new category windows, for
example, 1507 and 1508 in FIG. 15 (for features such as "bright"
and "dark") that are not used for the larger size group are
provided to classify defects in a simpler manner than for the
larger size group. This means that teaching data is separately
prepared for the larger and smaller defect size groups. Before
starting automatic classification of images, which teaching data
should be used is determined depending on the defect size;
therefore, classification takes place using the more suitable
teaching data.
[0070] This reduces the influence of noise caused by the smallness
of a defect area, improves coherence in features for each category,
and thus increases the accuracy of automatic classification. Since
images are to be divided into defect size groups without using
teaching data prepared by the user, the user need not prepare
teaching data for different categories for different size groups,
thereby leading to improvement in teaching efficiency.
[0071] The number of category windows as mentioned above is not
limited; FIGS. 14 and 15 each show two category windows for
illustration purposes only.
[0072] Explained next is how images are displayed when the
magnification ratio in imaging is changed, according to an
embodiment of this invention.
[0073] If images with different defect sizes are shot at a single
magnification ratio, some of the obtained images may have the
problem that the defect image overflows the visual field or it is
too small to observe properly. There are a few ways to solve the
problem: the magnification ratio is varied freely according to
defect size information obtained from an inspection system or
defect size information based on calculation from defect images
produced by the imaging device; or several magnification ratios are
preset to suit different defect sizes and the imaging magnification
ratio is changed step by step according to the defect size so that
defect images of a virtually uniform size are produced regardless
of defect size. However, when images have been produced using
different magnification ratios in this way, it is difficult for the
user to properly compare the sizes of the defects in a defect image
window as shown in FIG. 16 because the images are equal in size
though the defects are enlarged at different ratios, for example,
the images in the category A window 1602.
[0074] As a solution, the image size is varied in inverse
proportion to the magnification ratio used for imaging as shown in
FIG. 17 so that they are displayed to an equal scale, for example,
the images in the category A window 1702 were image 1710 is small
compared to image 1714. This makes it easy for the operator to
compare the sizes of defects whose images have been produced using
different magnification ratios. To avoid mistaking defect sizes
because of display image size difference, dummy data may be added
to the contours of images whose sizes are reduced, so that the
displayed images (contours) appear to be equal in size.
[0075] Explained next are examples of how the result of automatic
classification is shown to the user according to an embodiment of
this invention.
[0076] FIGS. 18 and 19 show examples of screens showing the results
of automatic classification. Each screen has the following windows:
a wafer window 1809 which shows information on the inspected wafer
including the wafer type and inspection date/time; a wafer map
window 1810 which shows the defect distribution, i.e., locations of
defects detected on the wafer; a category window 1811 which
indicates the names of the categories into which the defects on the
wafer have been classified, and information on breakdown of
defects; and a size distribution window 1812 which shows a
histogram of defect size distribution. On the screen containing
these windows, when the user chooses one of the categories listed
in the category window 1811, plotted points on the wafer map window
1810 representing the defects corresponding to the chosen category
are made distinguishable from the other points by color or shape,
for example 1815 have a "O" and 1816 having an "X". In the size
distribution window 1812, a histogram of size distribution for the
chosen category in the category window 1811 is shown (in this case
category A). The histogram of size distribution for the chosen
category may be displayed in either of the following ways: it is
laid over the histogram of size distribution for all the defects,
or the histogram for all defects is switched to the one for the
chosen category. In an alternative embodiment, the user can choose
two or more categories at a time.
[0077] Further, as shown in FIG. 19, when the desired size range is
selected in the size distribution window 1912, for example,
horizontal range 1915 to 1919, the points on the wafer map in the
wafer map window 1910 corresponding to defects of the selected size
range are made identifiable by color or shape.
[0078] In another embodiment the identifiable points on the wafer
map in the wafer map window 1910, representing the defect
distribution, are narrowed down according to both category and size
information. Thus a filtered defect distribution is displayed, were
the filtering is by category and size. For example, if one category
is chosen in the category window 1911, the size or shape of points
corresponding to the category changes; if a defect size range is
selected in the size distribution window 1912, points corresponding
to only the selected size range among the ones whose sizes or
shapes have once changed, change their color or shape again. In an
alternative embodiment, among points representing defects of the
same category, only those not corresponding to the selected size
range may be made identifiable by color or shape.
[0079] In other embodiments, the size distribution may be expressed
not only as a histogram but in terms of cumulative frequencies or
in the form of a table showing quantities.
[0080] In an embodiment, if a certain area in the wafer map is
specified, concerning the defects within the specified area, the
total number of defects, the number of defects and percentage of
each category, and size distribution are displayed.
[0081] In addition, it is not necessary to simultaneously display
all the three types of information the wafer map, category listing
and defect size distribution. It is sufficient to display at least
two types of information out of the three simultaneously.
[0082] Besides, any of the three types of information need not be
continually displayed. It is acceptable that they are usually not
displayed but can be called to the screen by the user as
necessary.
[0083] The above-mentioned arrangements will help the user grasp
the tendency of distribution of defects of each category on the
wafer map, thus providing the user with useful information to find
the cause of occurrence of defects.
[0084] Though not shown graphically here, another embodiment
provides simultaneously or successively displaying wafer maps
concerning wafers from different production lots or wafer maps
concerning different processing steps for the same wafer, so that
the user can easily grasp how the condition of occurrence of
contaminations of each category changes between lots or steps.
[0085] As mentioned above, when the conditions of occurrence of
contaminations of each category concerning different wafer lots or
different processing steps for the same wafer are simultaneously or
successively displayed as wafer maps, the user can monitor the
condition of occurrence of contaminations of each category between
lots or steps. This implies that the semiconductor manufacturing
system can be properly serviced according to the condition of
occurrence of contaminations of each category so that semiconductor
devices can be efficiently produced with an improved yield rate in
processing of wafers.
[0086] As discussed above, one advantage of one embodiment of this
invention is to save the user time and labor in creating teaching
data because selection of similar defect images or displaying of an
appropriate category based on similarity is automatically performed
in advance and the user need not classify all images into
categories in order to create teaching data.
[0087] Another advantage of an embodiment of the invention is that
since a diagram prepared according to the feature distribution of a
defect image group is displayed in association with defect images
within the group, it is easy to determine whether the feature of a
certain defect image is typical or atypical of the category
specified by the user. Also, which image is atypical of the
category specified by the user can be indicated by displaying the
feature distribution for the category. Therefore, the user can
easily remove atypical images from the teaching data, so higher
quality teaching data with higher coherence in features can be
created.
[0088] A further advantage of another embodiment of the invention
is that since defect images are divided into two or more groups by
defect size and categorization is made within each group, the user
need not give teaching data for classification by size and the
influence on the feature of noise caused by the smallness of a
defect size can be reduced, thereby permitting automatic
classification with higher accuracy.
[0089] Another advantage of an embodiment of the invention is that
the image display size is varied depending on the imaging
magnification ratio so that images are displayed to an equal scale,
so the user can easily compare defect sizes relatively for teaching
data creation.
[0090] A further advantage of an embodiment of the invention is
that since it is possible to plot the condition of defect
distribution on the wafer map according to the category or defect
size, or both, the user can easily grasp the relations among
categories, sizes and defect locations and effectively provide
information for feedback to the processing steps.
[0091] A still further advantage of the invention is that since the
user can monitor the condition of occurrence of defects classified
into a specific category using teaching data, he or she can easily
determine the timing to service the system to correct the cause of
defects of that category, leading to higher wafer yield rates in
manufacture of semiconductor devices.
[0092] 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.
[0093] 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.
[0094] The preceding has been a description of the preferred
embodiment of the invention. It will be appreciated that deviations
and modifications can be made without departing from the scope of
the invention, which is defined by the appended claims.
* * * * *