JP2008281580A - Defect classification system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a defect classification system capable of performing classification similar to human judgement on defects such as resist coating irregularities, exposure failures, flaws, the adhesion of dust without having to register defect data in micro-inspections principally on semiconductor wafers and liquid crystal panels. <P>SOLUTION: The defect classification system for determining the types of defects detected on the basis of images of bodies to be inspected includes a characteristics extraction means 102 including a means for extracting characteristic amounts from images of detected defect regions and extracting characteristic amounts from the shape of arrangement of at least two detection regions; an inference determination means 106 for determining the type of defects on the basis of the characteristic amounts extracted by the characteristics extraction means 102; and a definite region removal means 107 for removing, from the images of the defect regions, regions of which the type of defects has been defined determined by the inference determination means 106. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to a defect classification apparatus, and more particularly to a type of defect type detected based on an image, and more particularly to a defect classification apparatus in an appearance inspection performed on a production line such as a semiconductor wafer or a liquid crystal panel.

  In the manufacture of semiconductor wafers, liquid crystal panels, and the like, a multilayer wiring pattern is formed by repeating a process such as film formation, resist coating, exposure, development, etching, and resist peeling a plurality of times.

  In order to improve the yield of products that are finally completed through several hundred manufacturing steps, early detection of defects and optimization of the manufacturing steps are essential items.

  For this reason, conventionally, before the etching process in which the actual pattern is formed, the appearance of the object is inspected, the defect is detected, the defective product is selected, the type of the defect is determined, and the information is provided in the manufacturing process. Efforts are being made to provide feedback.

  There are mainly two types of appearance inspections here: micro inspection and macro inspection.

  The micro inspection detects minute defects and pattern abnormalities with a resolution of about 5 μm, and the macro inspection has a resolution of about 100 μm with relatively large resist coating unevenness, poor exposure, scratches, and dust adhesion. This is an inspection for detecting defects.

  Macro inspection has advantages such as the ability to detect defects that cannot be seen locally by micro inspection, and high throughput due to low resolution.

  Regarding the defect classification device, learning to output the defect type to the input pattern of each defect information (= feature amount) to the “defect type determination device and process management system” described in Japanese Patent Application Laid-Open No. Hei 8-21803. A configuration of an apparatus for constructing a neural network and classifying defects by this is shown.

  Further, in the above-mentioned prior art, the problem relating to how to provide teaching data at the time of learning is pointed out, and the “teaching data creation method and defect classification method and apparatus” described in JP-A-11-344450 has been improved. Shows a configuration of an apparatus that classifies defects by statistical discriminant analysis by predetermining the distribution of each defect in a multidimensional space based on image feature values based on teaching data.

Furthermore, in the “defect type determination apparatus and process management system” described in Japanese Patent Application Laid-Open No. 8-94536, the measured values of a plurality of evaluation items (features) for detected defects are coded in stages, and the code A configuration (a so-called classification tree method) of an apparatus for classifying defects by referring to defect names set in advance based on columns is shown.
JP-A-8-21803 JP 11-344450 A JP-A-8-94536

  However, a defect classification apparatus using a neural network or a statistical method as in JP-A-8-21803 and JP-A-11-344450 needs to register a large number of defect samples in advance, and it takes a lot of time. There is a problem that it takes.

  In addition, since registration is usually required every time the product type and process changes, there is also a problem that a sufficient number of defective samples cannot be prepared with a small amount of many product types.

  On the other hand, in the apparatus described in Japanese Patent Laid-Open No. 8-94536, defects corresponding to each code string can be set based on human experience and knowledge, and the above problems do not occur. Cannot be processed vaguely, that is, there is a subject that is classified as a completely different result due to a subtle difference in the switching position, or a value that is clearly separated from a target that has a value in the vicinity of switching while having the same code string Problems such as only giving the same result to the subject arise.

  Furthermore, many of the feature quantities used in the conventional defect classification apparatuses are processed by treating the detected defect area as a single unit, and there is a one-to-one correspondence between the defect and the detection area recognized by humans as in micro inspection. However, as in macro inspection, defects recognized by humans do not correspond to detection areas on a one-to-one basis (as shown in FIG. 5A, defect division by pattern or image quantum). If one defect is divided and detected depending on the level of conversion), there is a problem that accurate classification is difficult.

  Until now, an apparatus for solving the above problems and effectively classifying defects has not been clarified.

  The present invention has been made paying attention to such problems, and the object of the present invention is mainly non-uniformity of application of resist, exposure failure, scratches, dust adhesion, etc. in macro inspection of semiconductor wafers and liquid crystal panels. It is an object of the present invention to provide a defect classification apparatus that does not require registration of defect data and enables classification closer to human judgment.

  In order to achieve the above object, a defect classification device according to a first aspect of the present invention is a defect classification device that determines the type of a defect detected based on an image of a subject. A feature amount extracting unit including a feature amount extracting unit that extracts a feature amount based on an arrangement shape of at least two or more detection regions, and a feature amount extracted by the feature amount extracting unit. A defect type discriminating unit for discriminating the defect type, and a confirmed region excluding unit for excluding an area in which the defect type discriminated by the defect type discriminating unit is determined from the image of the defect region.

  The defect classification apparatus according to the second aspect of the present invention is the defect classification apparatus according to the first aspect of the present invention, wherein the feature amount extraction means uses an adaptive structural element depending on the shape of each detection region. And a means for connecting at least two or more detection regions and a means for extracting a feature amount of the connection regions.

  The defect classification apparatus according to the third aspect of the present invention is the defect classification apparatus according to the second aspect of the present invention, wherein the connecting means includes a maximum and minimum ferret diameter ratio (≧ 1) and a main axis of each defect area. Means for calculating the angle, means for selecting the shape of the structural element used for the expansion process based on the calculated maximum / minimum ferret diameter ratio from circular or fine linear, and the defect region where the fine linear structural element is selected Means for aligning the fine linear direction with respect to the principal axis angle, and means for performing expansion processing on the defect area image using the structural elements determined for each defect area and connecting the overlapping areas.

  The defect classification apparatus according to the fourth aspect of the present invention is the defect classification apparatus according to the first to third aspects of the present invention, wherein the defect type discrimination means includes a feature value and a preset classification rule. Includes means for performing fuzzy inference based on.

  Moreover, the defect classification apparatus according to the fifth aspect of the present invention is the defect classification apparatus according to the first to fourth aspects of the present invention, wherein the defect is an exposure failure, a coating unevenness, or the like caused in a manufacturing process of a semiconductor wafer, a liquid crystal panel or the like. Scratches, dust, dust, etc.

  Moreover, the defect classification device according to the sixth aspect of the present invention is the defect classification device according to the fifth aspect of the present invention, wherein the feature amount extraction means uses the exposure section information in the exposure step to calculate the feature amount. Means for extracting.

  The defect classification apparatus according to the seventh aspect of the present invention is the defect classification apparatus according to the sixth aspect of the present invention, wherein the means for extracting the feature quantity using the exposure section information indicates an exposure section shape. Means for creating a template image; means for scanning the template image in accordance with the exposure section position of the image of the detected defect area; and calculating a correlation value of both images at each exposure section position; and calculating the correlation value. Means for adding the correlation value as a feature amount to the detected defect in the exposure section.

  The defect classification apparatus according to the eighth aspect of the present invention is the defect classification apparatus according to the sixth aspect of the present invention, wherein the means for extracting the feature amount using the exposure section information is a pixel of the detected defect area. Means for vertical or horizontal integration, means for Fourier transforming the outline curve of the integrated value to calculate the amplitude value at each frequency, and the amplitude value of the frequency corresponding to the exposure section period and the frequency corresponding to the other period. Means for calculating a ratio of average amplitude values, and means for adding the value of the ratio as a feature amount to the detected defect used in the integration.

  The defect classification apparatus according to the ninth aspect of the present invention is the defect classification apparatus according to the sixth aspect of the present invention, wherein the means for extracting a feature amount using the exposure section information is an image of a detected defect area. Means for creating a template image from an arbitrary exposure section in the above, means for calculating a correlation value by aligning the template image with exposure sections located above and below or on the left and right of the arbitrary exposure section, and calculating an average thereof, Means for adding the average value as a feature amount to the detected defect in an arbitrary exposure section.

  The defect classification apparatus according to the tenth aspect of the present invention is the defect classification apparatus according to the fifth aspect of the present invention, wherein the feature amount extraction means uses polar coordinates with respect to the wafer rotation center in the application step. Means for extracting the feature quantity;

  The defect classification apparatus according to an eleventh aspect of the present invention is the defect classification apparatus according to the tenth aspect of the present invention, wherein the feature amount extraction means is configured to apply the detected defect area to the wafer rotation center in the application step. Means for creating a polar coordinate image converted into a polar coordinate system; means for detecting a linear component in the polar coordinate image; and means for adding the length of the linear component as a feature quantity to a defect area constituting the linear component. Including.

  A defect classification device according to a twelfth aspect of the present invention is the defect classification device according to any one of the first to eleventh aspects of the present invention. In the defect classification device according to the first to eleventh aspects of the present invention, Means for clearly indicating the areas in which the defect types are determined, in the order in which the defect types are determined, thereby indicating to the user the progress of the determination of the defect types.

  According to the present invention, by extracting a feature based on an arrangement shape based on an image excluding a region where a defect type has been determined, a more accurate feature can be obtained and classification accuracy can be improved.

  Further, by extracting the features of the region obtained by performing the expansion (concatenation) process based on the image excluding the region where the defect type is determined, more accurate features can be obtained and the classification accuracy can be improved.

  Moreover, since the defect classification is performed according to the classification rule based on knowledge, a teaching sample is not required.

  In addition, it is possible to indicate a region where the feature indicating the defect type is remarkable and a region where the feature is not so as a difference in possibility of the defect type.

  Further, the classification accuracy can be improved by using not only the feature amount of the detection area alone but also the feature amount by arrangement.

  In addition, by performing connection using adaptive structural elements to the detection region, accurate connection region characteristics can be obtained, and classification accuracy can be improved.

  Embodiments of the present invention will be described below in detail with reference to the drawings. Prior to describing the embodiment of the present invention, the types of defects to be classified will be described. FIGS. 1A to 1H are diagrams illustrating main defect examples (target: semiconductor wafer) in macro inspection.

  In the resist coating process, for example, a coating material (liquid) is dropped onto the center of the wafer using a coater apparatus, and the wafer is rotated around the wafer to perform resist coating on the entire surface of the wafer. The following defects are generated in the same process.

  Uneven coating 1 (Fig. 1 (A)): If dripping does not go well and scatters, or if the viscosity of the coated material is not appropriate and does not stretch smoothly during rotation, the unevenness will be radially uneven with respect to the center of rotation. Arise.

  Coating unevenness 2 (FIG. 1B): When a foreign substance adheres to the wafer, the coating material does not flow smoothly from that portion to the outside, and the comet shape is uneven.

  Uneven coating 3 (FIG. 1C): When foreign matter adheres as in FIG. 1B, unevenness occurs spirally depending on the viscosity of the coating material and the rotation speed during coating.

  In the exposure step, transfer is performed in units of predetermined exposure sections in order to form a pattern on the resist-coated wafer. A typical device is a stepper. The following defects are generated in the same process.

  Exposure defect 1 (FIG. 1D): When the focus shifts in units of exposure sections, the section becomes a surface state different from the other sections (pattern is blurred).

  Exposure defect 2 (FIG. 1 (E)): The exposure machine is tilted, and the step is continued while the focus is partially deviated in the exposure section. As a result, a surface condition different from normal is generated in a step unit cycle.

  Exposure defect 3 (FIG. 1 (F)): A foreign material enters the back side of the resist film to be exposed, so that the resist film partially rises and the focus shifts at that part, resulting in a different surface state from the others. .

  In addition to the defects represented by the above steps, there are the following defects.

Scratches (Fig. 1 (G))
Dust and dust adhesion (Fig. 1 (H))
In appearance inspection using images, as shown in Japanese Patent Laid-Open No. 8-21803, the appearance of the subject is imaged by interference, diffraction images, etc., and compared with a reference non-defective wafer image, By comparing in units of sections, a portion (defective portion) different from the normal is detected.

  Hereinafter, embodiments of the present invention will be described in detail. FIG. 2 is a diagram showing a configuration of the defect classification apparatus according to the present embodiment. The defect classification apparatus according to the present embodiment includes a condition setting unit 101 that sets various conditions necessary for classification, a region unit feature extraction unit 103 that obtains features related to the shape and luminance of a detected region unit, and a detected region Each of the feature extraction units 104, 104, 105, 105, 105, 105, 105, 105, and 105, is a feature that extracts features of the detected regions from each other. Inferior discriminating means 106 that infers the possibility of each defect type based on the feature amount obtained by the means and the classification rule set by the condition setting means 101 and discriminates the defect, and the region in which the defect type is determined by the discrimination The fixed area excluding means 107 and the memory 108 used for each processing are included. The region single feature extraction unit 103, the region arrangement feature extraction unit 105, and the region connection unit 104 constitute a feature extraction unit 102.

  The defect classification apparatus of the present invention is a binary image (hereinafter referred to as a defect detection image) in which defective pixels detected by comparison with the above-mentioned non-defective wafer image or comparison in the same pattern section are bright and others are dark. And obtain the appearance (multi-value) image (hereinafter referred to as inspection image) of the object used for defect detection, and output the result of determining the defect type for each detection area It is the structure to do.

  The operation of the above configuration will be described below with reference to the flowcharts of FIGS. After acquiring the defect detection image and the inspection image (step S1), the defect classification device extracts the feature amount of the detection region alone in the region single feature extraction unit 103 (step S2). Next, the region arrangement feature extraction unit 105 extracts feature amounts based on the arrangement shape of a plurality of regions, and adds this feature amount to each detection region constituting the arrangement shape (step S3). Thereafter, based on the information obtained by the region single feature extraction unit 103 and the defect detection image, the region connection unit 104 performs an expansion process using a structural element suitable for the shape of each detection region, and images obtained by connecting the regions. (Hereinafter referred to as a connected region image) is created (step S4).

  The connected region image is sent again to the region single feature extracting unit 103, where the feature amount of the connected region is extracted, and this feature amount is added to each detection region constituting the connected region (step S5). Based on the feature values obtained in steps S2, S3, and S5 and the classification rule set by the condition setting unit 101, the inference determination unit 106 infers the possibility value of each defect type and determines the defect type ( Step S6). On the basis of the result of step S6, the confirmed region exclusion unit 107 excludes the region where the defect type is confirmed from the defect detection image (step S7). If all the regions are excluded as a result of step S7, the defect types of all the detection regions are determined, and the process ends.

  If an area remains in the defect detection image, steps S3 to S7 are repeated again based on this image.

  Since the images are different, the feature amount according to the arrangement shape and the feature amount of the connected region are different. That is, by removing the determined region, a more accurate feature amount can be extracted, and inference and discrimination can be performed based on this.

  Next, details of the processing in each of the above-described means will be described. The condition setting means 101 inputs various conditions necessary for classification. The input contents will be described along with the processing in each means described later.

  FIG. 4 is a flowchart for explaining details of processing in the region single feature extracting unit 103. The region single feature extraction unit 103 obtains the feature amount of each detection region according to the following procedure.

1. A plurality of detected areas in the defect detection image are labeled (step S10 in FIG. 4, FIG. 5B).

2. The position characteristics (center of gravity position, circumscribed rectangle position) of the detection area are calculated (step S11 in FIG. 4, FIG. 6 (A)).

3. The shape characteristics (area, circularity, maximum / minimum ferret diameter ratio (≧ 1), main shaft angle) of the detection region are calculated (step S12 in FIG. 4, FIG. 6A).

4). Luminance characteristics (average and variance) of the detection area are calculated (step S13 in FIG. 4, FIG. 6A).

5. A feature quantity list for each detection area is created (step S14 in FIG. 4, FIG. 6B).

  The series of processes indicated by the above 1, 2, 3, and 4 is a general image processing technique called particle analysis (or blob analysis).

  FIG. 7 is a flowchart for explaining details of the processing in the area connecting means 104. 8 and 9 are explanatory diagrams of region connection.

  The area connecting means 104 uses the features obtained by the area single feature extracting means 103 and creates a connected area image in which the detection areas are connected according to the following procedure.

1. Based on the maximum / minimum ferret diameter ratio of each region, a structural element for expansion processing for that region is selected. It is determined whether the maximum / minimum ferret diameter ratio is equal to or greater than a certain threshold value (either a fixed value as a result of trial or a value set by the condition setting means 101) (step S20 in FIG. 7). Considering the region, a fine linear structural element (FIG. 8A) is selected (step S21 in FIG. 7).

  If the determination in step S20 is NO, that is, if it is smaller than the threshold value, a structural element having a circular diameter (FIG. 8B) is selected (step S23 in FIG. 7). Note that the above selection may be performed using the circularity.

2. When a fine linear structural element is selected, the fine linear direction is further aligned with the principal axis direction of the region (step S22 in FIG. 7). FIG. 8C shows an example of the direction of the fine linear structural element. The direction is limited when the size of the structural element is small, but the direction can be subdivided by increasing the size. Note that the direction of the region may be estimated from the histogram distribution in the contour direction, and the direction may be the same.

3. Expansion processing (step S24 in FIG. 7) is performed using the selected structural element to create a connected region image. The expansion processing here is one of general image processing methods called morphology processing, and overlapping regions are connected as a result of expansion. The number of expansion processes is set by the condition setting means 101 based on the imaging system, various conditions of the subject, and the size of each structural element.

  If it is determined in step S25 in FIG. 7 that all the detection areas have been determined, the process ends.

  FIG. 8D shows the result of performing the expansion processing of this example on the detection region of FIG.

  By performing the expansion process with the structural element suitable for the shape of the detection region by the above procedure, the directional region can be expanded and connected in the longitudinal direction and the others in the radial direction. Further, if the structural elements are redetermined on the basis of the region shape after expansion and the contraction process is performed, better region connection is possible.

  In general expansion or expansion / contraction processing using the same structural element (circular shape) for all detection regions, the directionality of the detection region is not taken into consideration, and therefore FIGS. 8E, 9A, and 9B are used. As shown in FIG. 4, the scratch and the dust in the vicinity thereof cannot be connected in a good state, and subsequent feature extraction cannot be performed accurately.

  The connected area image created by the area connecting means 104 is sent again to the area single feature extracting means 103 and processed in the following procedure. The basic processing flow is the same as in FIG.

1. Each region after connection (referred to as a connection region) is labeled (FIG. 10A).

2. The position characteristics (center of gravity position, circumscribed rectangle position) of the connected area are calculated.

3. The shape characteristics (area, circularity, maximum / minimum ferret diameter ratio (≧ 1), spindle angle) of the connected region are calculated.

4). The calculated features are added to the detection regions that are the respective component parts, and a feature amount list for each detection region is created (FIG. 10B).

  The series of processes 1, 2, and 3 described above is based on particle analysis.

  The area arrangement feature extraction unit 105 obtains area arrangement characteristics based on the defect detection image before connection. The typical arrangement features are listed below, and the calculation method is shown.

1) A feature value indicating the degree of composition of the exposure section shape based on the detection area (used to determine exposure failure 1).

<Calculation method>
1. A template is created in which the portion of the exposure section shape is bright and the periphery thereof is dark (FIG. 11A).

2. Arranged so that the exposure section position in the defect detection image (binary image in which defective pixels are bright and others are dark) and the exposure section shape in the template completely overlap, and the sum of absolute differences between both images within the template range Is calculated (FIG. 11B).

The sum R of the absolute differences between both images is

It can ask for. Here, I (a, b) is a pixel value (1 or 0) at the coordinates (a, b) of the defect detection image. T (a, b) is a pixel value (1 or 0) at the coordinates (a, b) of the template.

3. The value calculated in 2 is added to the detection area in the target exposure section as the feature amount of 1).

4). The template is scanned (FIG. 12A), and the processes 2 and 3 are performed at all exposure division positions in the defect detection image.

  Note that a template as shown in FIG. 12B may be used to reduce the amount of calculation.

2) A feature amount indicating the periodicity of exposure block units in the horizontal and vertical directions (used to determine exposure failure 2).

<Calculation method 1>
1. The detection pixels (bright pixels) in the defect detection image are integrated in the vertical direction (FIG. 13A).

2. The curve obtained from the integration result is Fourier transformed (FIG. 13B).

3. A ratio between the power spectrum value X corresponding to the exposure section period and the power spectrum value a corresponding to another period (averaged in a portion other than the exposure section period) is calculated (FIG. 13B).

4). The value calculated in 3 is added to each detection area in the image as the feature amount (horizontal direction) in 2).

  In order to obtain a local feature amount, a plurality of strip regions as shown in FIG. 14 may be set, and the detection pixels in each strip region may be integrated to perform processing. In this case, the calculated feature amount is added to each detection area in the strip area.

  In the present embodiment, the integration in the vertical direction is shown, but by integrating in the horizontal direction, it is possible to obtain a feature amount that indicates the period of the vertical direction.

<Calculation method 2>
1. A template 52 is created from the detection area image in the target exposure section (FIG. 15A).

2. The template at 1 is arranged so as to completely overlap the exposure section adjacent to the left of the target exposure section (FIG. 15B), and the sum of absolute differences between both images is calculated within the template range.

3. The template at 1 is arranged so as to completely overlap the exposure section on the right side of the target exposure section (FIG. 15B), and the sum of absolute differences between both images is calculated within the template range.

4). The average value of the values obtained in 2 and 3 is added to each detection area in the target exposure section as a feature amount (horizontal direction) of 2).

5. The template is scanned (FIG. 12A), and processes 1 to 4 are performed at all exposure division positions in the defect detection image.

  When the target exposure section is located at the end and there is no left or right adjacent exposure section, the value of only one side is used.

  Further, in the present embodiment, the difference between the right and left neighbors in the horizontal direction is shown. However, by performing the difference between the vertical neighbors in the vertical direction, it is possible to obtain a feature amount indicating a vertical period.

3) A feature amount indicating the total amount of the radial component (relative to the rotation center at the time of application) in the entire image (used for determining application unevenness 1).

<Calculation method>
1. The edge of the detection area is extracted (FIG. 16A).

2. The extracted pixel of the edge portion is converted into a polar coordinate system image (horizontal axis: angle θ, vertical axis: distance r from the center) with respect to the rotation center at the time of application (FIG. 16B). The vertical line segment in the polar coordinate system image corresponds to the radial line segment in the original image. The vertical line segment is detected using an operator (template) for detecting the vertical line segment (FIG. 17A).

3. The total length (total area) of the detected vertical line segments is added to each detection area in the image as a feature amount 3).

  Note that performing connection processing with vertical thin line structure elements in a polar coordinate system image corresponds to performing radial direction connection in an original image. Furthermore, the vertical line segment as shown in FIG. 17B can also be detected by using contour tracking from the outer peripheral side together.

4) A feature amount indicating the amount of the radial component (relative to the rotation center at the time of application) in each detection region (used for determining application unevenness 2).

<Calculation method>
Processes 1 to 3 of 1.3) feature amount calculation are performed.

2. The length of the detected vertical line segment is added to each corresponding detection area as the feature amount 4).

  Note that the polar coordinate system image is connected with vertical thin line structure elements (corresponding to the radial direction connection in the original image), and then the length obtained by measuring the length of the vertical line segment is connected. You may add to each area | region used as the structure component of an area | region.

5) A feature amount indicating the ratio of the spiral component (with respect to the rotation center at the time of application) in each detection region (used for determining application unevenness 3).

<Calculation method>
The processing 1 and 2 of the feature amount calculation in 1.3) is performed. Note that the spiral component in the original image corresponds to a diagonal line segment in the polar coordinate system image.

2. The diagonal line segment is detected using an operator for detecting the diagonal line segment.

3. The length of the detected diagonal line segment is added to each corresponding detection area as the feature amount 5).

  Note that Hough transform may be used for line segment detection in the feature amount calculation of 3), 4), and 5).

  The inference determination unit 106 infers the possibility of each defect type based on the feature amounts obtained by the feature extraction units 103 and 105 and the classification rule set by the condition setting unit 101.

  Before showing the inference process, the classification rules are explained. If it can be said that “small and narrow areas are highly likely to be scratched” or “small and may be scratched if not smaller” than previous experience and knowledge, it can be used as an indicator of “size” The area can be ruled as follows by using the maximum / minimum ferret diameter ratio (the longer the value is, the larger the value is) as an index representing “strip length”.

(1): IF area = small AND Maximum and minimum ferret diameter ratio = large THEN Scratch probability = high (2): IF area = small AND maximum and minimum ferret diameter ratio = small THEN Scratch possibility = moderate (3) : IF area = large AND Maximum / minimum ferret diameter ratio = large The possibility of THEN scratch = moderate (4): IF area = large AND maximum / minimum ferret diameter ratio = small The possibility of THEN scratch = low Next used in rules For the expressions “small”, “large”, “high”, and “low”, the degree to which the numerical values on each feature axis match is indicated by a value of 0 to 1 (fitness). A change line (membership function) is set (FIGS. 18A to 18C).

  A process for inferring the possibility of a flaw in a certain detection region (area = x, maximum / minimum ferret diameter ratio = y) based on the above settings is shown in FIGS. This is a general reasoning method as a direct method in fuzzy reasoning ((Reference) Russell: Introduction to fuzzy theory for those who want to apply: by Kazuo Tanaka).

  In FIG. 20, the set obtained by the respective rules (1) to (4) in FIG. 19 is synthesized (ored) to obtain an intermediate solution (centroid Z). As a result, it corresponds to obtaining a complementary determination result by obtaining the possibility of scratches when considering the four rules from the viewpoint of each rule.

  In this example, only four rules are shown for scratches, but accuracy can be improved by setting more rules.

  Similarly, by setting a plurality of rules (& membership function) based on knowledge about other defects and processing those rules, the possibility values for all defect types can be calculated.

  After the possibility value of each defect type is inferred, the defect type is determined by the following processing.

1. The probability values of the defect types in the detection area to be determined are arranged in descending order.

2. When “the probability value of the highest defect type−the possibility value of the second highest defect type> threshold value”, it is determined that the defect type of the target detection region has been determined.

3. 1. And 2. This process is performed for all the detection areas in the defect detection image.

  In addition, 2. The threshold value at may be a fixed value as a result of the trial or a value set by the condition setting means 101. Further, it may be set so as to decrease as the number of the above-described repetitive processing (FIG. 3, steps S3 to S8) increases.

  The confirmed area exclusion means 107 excludes the area where the defect type is determined by changing the pixels in the detection area where the defect type is determined in the defect detection image from bright to dark.

It is a figure which shows the example of main defects in a macro test | inspection. It is a figure which shows the structure of the defect classification apparatus which concerns on this Embodiment. It is a flowchart for demonstrating the detail of a defect classification | category process. 10 is a flowchart for explaining details of processing in a region single feature extracting unit 103; It is explanatory drawing (the 1) of area | region single-piece | unit feature extraction. It is explanatory drawing (the 2) of area | region single-piece | unit feature extraction. 5 is a flowchart for explaining details of processing in an area connecting unit 104; It is explanatory drawing (the 1) of area | region connection. It is explanatory drawing (the 2) of area | region connection. It is explanatory drawing of a connected area single-piece | unit feature extraction. It is explanatory drawing of the feature-value which shows the structure of an exposure division shape, (A) shows the template which made the exposure division shape part bright and the periphery dark, (B) shows the exposure division part of a defect detection image. Show. It is explanatory drawing of the feature-value which shows the structure of exposure division shape, (A) shows the example of a template scan, (B) has shown the example of a template for calculation amount reduction. It is explanatory drawing (the 1) of the feature-value which shows the periodicity condition of exposure division size, (A) shows the integration to the vertical direction of a detection pixel, (B) is the Fourier transform of an integration curve, and exposure division size. The feature quantity indicating periodicity is shown. It is explanatory drawing (the 1) of the feature-value which shows the periodicity condition of exposure division size, and has shown the integration to the perpendicular direction in a strip area. It is explanatory drawing (the 2) of the feature-value which shows the periodicity condition of exposure division size, (A) shows preparation of the template based on the detection result image of object exposure area, (B) is the horizontal direction of object exposure area. The right and left exposure sections are shown. It is explanatory drawing of the feature-value which shows the quantity of the radial direction component in the whole image, (A) shows edge extraction of a detection area | region, (B) has shown the polar coordinate conversion with respect to the rotation center at the time of application | coating. It is explanatory drawing of the feature-value which shows the quantity of the radial direction component in the whole image, (A) shows the vertical line segment 1 in a polar coordinate image, (B) shows the vertical direction line segment detection 2 in a polar coordinate image. Show. It is explanatory drawing of calculation of the possibility of the defect by a fuzzy reasoning, and has shown the membership function with respect to the expression by a crack determination rule. It is explanatory drawing of calculation of the possibility of the defect by fuzzy reasoning, and shows the reasoning process (the 1). It is explanatory drawing of calculation of the possibility of the defect by fuzzy reasoning, and shows the reasoning process (the 2).

Explanation of symbols

101 condition setting unit 102 feature extraction unit 103 region single unit feature extraction unit 104 region connection unit 105 region arrangement feature extraction unit 106 inference determination unit 107 confirmed region exclusion unit 108 memory

Claims (12)

In the defect classification device for determining the type of defect detected based on the image of the subject,
Means for extracting a feature amount from an image of a detected defect area, and including a feature amount extraction means including means for extracting a feature quantity based on an arrangement shape of at least two or more detection areas;
A defect type discriminating means for discriminating a defect type based on the feature quantity extracted by the feature quantity extracting means;
A defect classification apparatus comprising: a defined area excluding unit that excludes an area in which the defect type determined by the defect type determining unit is determined from the image of the defect area. The feature amount extraction means includes a connection means for connecting at least two detection areas by means of expansion processing using an adaptive structural element depending on the shape of each detection area, and means for extracting the feature quantities of the connection areas. The defect classification apparatus according to claim 1, further comprising: The connecting means is a means for calculating the maximum / minimum ferret diameter ratio (≧ 1) and the main shaft angle of each defect area, and the shape of the structural element used for the expansion process based on the calculated maximum / minimum ferret diameter ratio is circular. Alternatively, a means for selecting from fine line, a means for aligning the direction of the fine line with the principal axis angle of the defect area for which the fine line structure element is selected, and a defect area using the structure element determined for each defect area 3. The defect classification apparatus according to claim 2, further comprising means for performing expansion processing on the image and connecting overlapping regions. 4. The defect classification apparatus according to claim 1, wherein the defect type determination unit includes a unit that performs fuzzy inference based on a feature value and a preset classification rule. The defect classification according to any one of claims 1 to 4, wherein the defect is an exposure failure, coating unevenness, scratches, dust, dust, or the like generated in a manufacturing process of a semiconductor wafer, a liquid crystal panel, or the like. apparatus. 6. The defect classification apparatus according to claim 5, wherein the feature amount extraction means includes means for extracting a feature amount using exposure section information in an exposure process. The means for extracting the feature amount using the exposure section information includes means for creating a template image indicating an exposure section shape, and scanning the template image in accordance with the exposure section position of the image of the detected defect area. 7. A means for calculating a correlation value between both images at a position and a means for adding the correlation value as a feature amount to a detected defect in the exposure section where the correlation value is calculated. Defect classification apparatus of description. Means for extracting the feature value using the exposure section information includes means for integrating the pixels of the detected defect area in the vertical or horizontal direction, and means for calculating an amplitude value at each frequency by Fourier transforming the outline curve of the integrated value. Means for calculating a ratio between an amplitude value of a frequency corresponding to an exposure section period and an average amplitude value of a frequency corresponding to another period, and adding the value of the ratio as a feature amount to the detected defect used in the integration The defect classification apparatus according to claim 6, further comprising: Means for extracting feature quantities using the exposure section information includes means for creating a template image from an arbitrary exposure section in the image of the detected defect area, and positions the template image above, below, or on the left and right of the arbitrary exposure section. A means for calculating a correlation value in accordance with an exposure section and calculating an average thereof, and means for adding the average value as a feature amount to the detected defect in the arbitrary exposure section. 6. The defect classification apparatus according to 6. 6. The defect classification apparatus according to claim 5, wherein the feature amount extraction means includes means for extracting feature amounts using polar coordinates with respect to the wafer rotation center in the coating step. The feature amount extraction unit includes a unit that creates a polar coordinate image obtained by converting the detected defect area into a polar coordinate system with respect to the wafer rotation center in the application step, a unit that detects a linear component in the polar coordinate image, and the linear component 11. The defect classification apparatus according to claim 10, further comprising means for adding the length of the linear component as a feature amount to a defect area to be processed. Display means for displaying an image of the detected defect area, and means for clearly indicating the areas in which the defect type is determined in the display means in the order of determination, thereby indicating to the user the progress of the defect type determined. The defect classification apparatus according to claim 1, wherein the defect classification apparatus is a defect classification apparatus.

Images