JPH0723845B2 - Defect detection method - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method of detecting a defect in a pattern to be inspected which is formed by repeating the same pattern, and inspects a semiconductor device and a mask used for manufacturing the same. The present invention relates to a suitable defect detection method.

[Background of the Invention]

Conventionally, as a method for automatically inspecting a fine and complicated pattern of a semiconductor element or the like, as seen in Japanese Patent Application No. 58-065420, by means of imaging a pattern to be inspected and converting it into a video signal, A defect inspection method is generally used in which two image signals of the same pattern are obtained and a defect of the pattern to be inspected is detected from the difference between these image signals. That is, two input images are subtracted from each other, and then binarized with a certain threshold,
This is a method of detecting a "1" area as a defect. However, such a simple two-pattern comparison method has a drawback that a defect that should be one originally is extracted by being divided into a plurality of regions.

A line pattern having a constant width as shown in FIG. 2 (a) and a pattern in which a part of the same line pattern as in FIG. 2 (a) is projected as shown in FIG. 2 (b). The case of comparison will be described as an example. It is assumed that the input image is obtained by illuminating the pattern to be inspected from directly above and capturing the reflected light, and assuming that the physical characteristics of the line pattern and the background are substantially the same, the input image is As shown in FIGS. 2C and 2D, the boundary portion of the line pattern appears as a dark line. Take the difference in brightness between these two images,
When an area with a large difference is detected, the image shown in FIG. 2 (e) is obtained. It is to be noted that FIGS. (C) to (e) schematically show the level (brightness) of a grayscale image in three stages. In these figures, a black-painted area is a dark area and a white area is white. The area of is a bright area, and the hatched area is an area having a brightness intermediate between them. As can be seen from this figure, the region having a large difference is detected by being divided into a region having a large negative value and a region having a large positive value.

As another drawback of the two-pattern comparison method, as can be seen from the example of FIG. 2 (e), only the boundary portion of the defect is detected as a region with a large difference, and the original defect region (indicated by a broken line in the figure). The enclosed area) may not be detected.

Due to the above problems, the conventional method cannot perform a precise defect inspection such as the determination of the type of defect and the criticality from the defect detection result.

[Object of the Invention]

An object of the present invention is to provide a method for detecting a defect existing on a pattern to be inspected with an accurate shape without the above problems.

[Outline of Invention]

In order to achieve this object, the present invention has a means for capturing an image of a pattern to be inspected and converting it into a video signal, and a means for comparing two video signals of the same type pattern. In a defect inspection method for detecting defects, threshold values are applied to a grayscale image obtained by comparison, and a binary value in which a positive threshold value or more is "1" and other areas are "0" An image is generated, and a binary image in which a region equal to or less than a separately set negative fixed threshold is “1” and the other region is “0” is generated. Further, the connection processing of the "1" area is performed between these two binary images to connect the defective area which should be originally one and is divided into a plurality of areas and extracted. Furthermore, "0" in "0" area surrounded by "1" area
By converting the value of 1 into “1”, the inner area, which is originally a part of the defect, is detected as a defect.

Example of Invention

Hereinafter, the present invention will be described with reference to examples. FIG. 3 is an overall configuration diagram of a semiconductor wafer pattern appearance inspection apparatus using the present invention. The wafer pattern 2 on the movable table 1 is magnified through the objective lens and forms an image on the image pickup surface of the TV camera 3. The image plane is electrically scanned by the TV camera 3 to convert the wafer pattern image into an analog image signal. This analog image signal is converted into a digital image signal by the A / D converter 4, and then stored in the image memory 5 as a digital image. Further, the computer 6 uses the pattern data stored in the image memory 5 as input data to perform defect detection processing.
Since the main object of the present invention is the defect detection processing by the computer 6, this portion will be described below in detail according to the processing procedure shown in FIG.

Step 101 ... By moving the movable table 1 in FIG. 3, a partial area on the surface of the pattern 2 to be inspected is imaged and stored in the image memory. The input image at this time is {f
_ij }. Here, the input image is an image in which the video signal from the TV camera is sampled in vertical i _max columns and horizontal j _max rows, and the brightness is quantized by a plurality of bits, and the pixel value in the i row and j column is f _ij.
The image that becomes is described as {f _ij }.

Step 102 ... By moving the movable table 1 in FIG. 3, a partial area of the pattern 2 to be inspected having the same pattern shape as the pattern obtained in Step 101 is imaged and stored in the image memory. To do. The input image at this time is {g _ij }.

Step 103 ... With respect to the two input images {f _ij }, {g _ij } to be compared, the subtraction process between the two images is performed by the following equation.

h _ij = g _ij −f _ij (1) Step 104 ... Binary image {h ^{B +} _ij } of the subtraction processed image {h _ij } according to the equation (2) using the threshold value t ₁ (> 0).
To generate.

This is an image in which the area where {f _ij } is particularly bright compared with the brightness of the input image {g _ij } is "1",
The "1" area is considered to be a defect existing area. The obtained binary image {h ^{B +} _ij } is tentatively referred to as a “normal” area image.

Step 105 ... Using the threshold value t ₂ (<0), the binary image {h ^B− _ij } of the subtraction processed image {h _ij } is calculated according to the equation (3).
To generate.

This is an image in which a region where the input image {f _ij } is particularly dark is “1” in comparison with the brightness of the input image {g _ij },
The "1" area is considered to be a defect existing area. The obtained binary image { ^hB- _ij } is tentatively called a "negative" region image.

The areas where "1" is present in the binary images {hB ⁺ _ij } and { ^hB- _ij } are both areas in which there is a large difference in brightness between the input images and are assumed to be defect existing areas. However, the defect area thus extracted has a problem that one defect is originally divided into a plurality of areas. In this state, the input image shown in FIG. 2 (c) obtained by imaging the pattern to be inspected shown in FIG. 2 (a) is _designated as {f _ij } and the inspection image shown in FIG. 2 (b) is obtained. A case will be described in which the input image shown in FIG. 2D obtained by imaging the pattern is {g _ij }. At this time, in the “correct” area image {h ^{B +} _ij }, as shown in FIG. 4A, a part of the abnormal protrusion of the pattern to be inspected is “1”.
The image is extracted as a region. The "negative" area image { ^hB- _ij } is an image in which a part of the abnormal protrusion of the pattern to be inspected is extracted as the "1" area, as shown in FIG. 4 (b). However, there is a problem in that the abnormal projection portion of the pattern to be inspected is divided into two regions and extracted only by performing the logical operation of the two images shown in FIGS. 4 (a) and 4 (b). Therefore, in order to correctly detect the number and shape of defects to be detected, it is necessary to connect the divided areas.

The connection process is divided regions, i.e., regions to be connected is extracted with always "positive" area image region extracted by {h ^B _ij} and "negative" area image {h ^B- _ij} By focusing on the fact that it is sandwiched between regions, it can be realized by steps 106 to 109.

Step 106: Enlargement processing is performed M times for the “normal” area {h ^{B +} _ij } to expand the “1” area by M pixels. The enlargement number M is set to a value corresponding to the minimum number of pixels required to connect the “positive” area image and the “negative” area. The specific method of the enlargement processing is as shown in the equation (4), and the enlargement processing is performed M times using {h ^{B (O) +} _ij } ≡ {h ^{B +} _ij } as the initial image, and the enlarged image { h ^{B (m) +}
_ij } (m = M). That is, for m = 1, ..., M, Step 107 ... Enlargement processing is performed M times on the "negative" area image {h ^B- _ij } to expand the "1" area by M pixels. The number of expansions M is the same as in step 106. The concrete method of the enlargement processing is as shown in the equation (5), and {h ^{B (O) −} _ij } ≡
Using {h ^B- _ij } as the initial image, this enlargement processing is performed M times to obtain an enlarged image {h ^{B (m)} _-ij } (m = M). That is, for m = 1, ..., M Step 108 ... "Positive" area image {h ^{B +} _ij }, "Negative"
As shown in the equation (6), the logical product is obtained between the enlarged image {hB ^{(m) +} _ij } and {hB ^(m) _-ij } (m = M) of the area image { ^hB- _ij }. Therefore, the connection area of the "positive" area image and the "negative" area image is set to "1".
Generate {hc ^B _ij } extracted as a region.

{Hc ^B _ij } = {h ^{B (m) +} _ij } ∧
{HB ^(m) _-ij }, m = M (6) Step 109 ... Fusion processing of the divided defects is performed using the "positive" area image, the "negative" area image, and the connection area image. That is, according to the equation (7), the "positive" region image and the "negative"
Performs the logical sum operation of the area image and the connected area image.

_{^{{H B ij} {hc B}} ij} ∨h B + ij ∨h B- ij} (7) Now, the processing of step 104 to step 109, 4 (a), "positive" as shown in (b) The area image and the “negative” area image will be described as an example with reference to FIGS. 4 (c) to 4 (f).
When the "positive" area image and the "negative" area image are enlarged several times, the images shown in FIGS. 4 (c) and 4 (d) are obtained. Also, by performing a logical product operation between these two images,
As shown in FIG. 4 (e), "positive" region image, "negative"
Image {hc extracted from the connection area of the area image as "1" area
^B _ij } is obtained. Further, logically ORing the "positive" area image shown in FIG. 4 (a), the "negative" area image shown in FIG. 4 (b), and the connection area image shown in FIG. 4 (e). Thus, the defect extraction image as shown in FIG. 4 (f) is obtained. The defective area here correctly reflects the abnormal protrusion of the pattern to be inspected. In addition, compared to the proposed defective area connection processing method, a simple scaling method (after taking the logical sum between the positive area image and the negative area image,
When the designated pixel is enlarged or reduced, (1) the detected defect area is deformed. (2) It may be connected to different defect areas. (3) Even if they are once connected in the enlargement processing, they may be divided again by the reduction processing. It has drawbacks such as. The proposed method solves such a problem.

On the other hand, a defect which is detected due to a portion where the difference in brightness between the defect and the background is small and the edge is dark is extracted as an annular defect even if the region connecting process is performed.
As can be seen from the example of FIG. 2, in many cases, the area inside the ring is also a part of the defect. That is, the defect shape can be correctly reproduced for the first time by filling the hole in the defect area. The filling processing is performed as follows.

Step 110 ... Performs the edge tracking processing on the defect extraction image {b ^B ij} to determine whether it is a hole boundary from the tracking direction of the defect boundary.
Identify if it is a background border. For example, A. Rosenfeld and ACKak: digital Picture Proc, "Digital Image Processing" (published by Academic Press Press) written by A. Rosenfeld and A.C. Kirk.
essing; Academic Press) Chapter 8 Here, the boundary belongs to the inside of the “1” area, and
It is a set of pixels adjacent to the outside of the "1" area. By performing the above tracking processing, the starting point coordinates and the direction code string of the defect area boundary can be obtained. The direction code indicates the direction of the neighboring pixel with respect to the pixel of interest, and has eight types of codes 0 to 7 as shown in FIG. Also,
The directional code string is a string of directional codes in which a series of boundary pixels is represented by a pixel corresponding to the start point coordinates as a starting point, and this is called a chain code. The boundary portion has a boundary portion with the background area and a boundary portion with the inside of the hole. According to the method described in the above literature, tracking is performed in the clockwise direction at the background boundary portion and at the inner boundary portion, the Tracking is done in a clockwise direction. This tracking direction is the first value of the chain code that can be easily identified from the chain code, that is, the direction code value from the pixel corresponding to the start point coordinates to the connected pixel is obtained. If this value is "3", The tracking direction is counterclockwise, and the tracking direction for other direction code values is clockwise. By using this background boundary chain code, the feature amount (position, area, length, etc.) of the detected defect area can be easily obtained.

Now, in order to perform the area filling processing, it is sufficient to ignore the hole boundary portion and paint the inside of the background boundary portion. For the method of filling the inside of the boundary from the chain code, see, for example, Li, et al .: Contour tracking and restoration of binary figures by intersection description method, IEICE Transactions (D) J65-D, 10,
pp1203-1210 (Sho 57-10). The {h ^B _ij } padded image {Z ^B _ij } thus obtained
Is the final defect extraction image, which is a faithful reproduction of the defect shape.

An example of the area to be filled in is shown in FIG. 4 (f),
An example of the hole-filling processing result image is shown in FIG. In the example shown here, it can be seen that the abnormal protrusion of the pattern to be inspected is accurately detected.

The defect area thus obtained is detected including the boundary portion of the defect. However, in some cases, it is important to extract the defect excluding the boundary portion. For example,
One of the important properties of defects is the brightness of the defects.
By knowing whether the defect is dark or bright, it is possible to estimate whether the defect is a metallic defect having a strong reflection property or a defect having a remarkable light absorption property. Furthermore, when there is no difference in brightness between the background area and the background area, it can be estimated that the defect has high transparency. Such brightness determination of the defect can be used as a clue to know the cause of the defect and the degree of fatality. Now, considering the nature of the defect, the brightness of the defect should be determined by the physical properties peculiar to the defect such as light reflection characteristics,
The light-dark change of the boundary area depending on the illumination direction is not essential. Explaining in the example of FIG. 2, the brightness of the defect itself in the comparative image (e) is the intermediate level brightness of the internal area, and the bright and dark areas of the boundary area are the brightness of the direct defect. It has nothing to do with. Therefore, the boundary portion is removed from the detected defect area. Then, a region that does not include this boundary portion is defined as a defect region. The method of removing the boundary area from the detected defect area is as follows.

Step 601 ... After the defect filling processing shown in FIG.
The reduction processing is performed N times on the defective area extraction image {Z ^B _ij } to reduce the “1” area by N pixels. The reduction number N is the number of pixels corresponding to the edge line width of the boundary area. A specific method of the reduction processing is {Z ^{B (O) as} shown in Expression (8 ^).
_ij } ≡ {Z ^B _ij } is used as an initial image to perform reduction processing N times to obtain a reduced image {Z ^{B (N)} _ij }.

For n = 1, ..., N, The reduced image thus obtained is used as a defect area extraction image {Z ^{B '}
_ij } (≡ {Z ^{B (N)} _ij }).

Step 602 ... Next, a method of obtaining the brightness of the defect based on the defect area extraction image will be described. First, a labeling process is performed in order to make it possible to distinguish defect areas (connected areas with a pixel value of "1") in the defect area extraction image into defect units, that is, an image in which a series of different pixel values is assigned to each defect area. Get {Zc _ij }. Here, when the number of defective areas is m, a value of 1 to m is assigned to each defective area unit. For the labeling process, refer to, for example, the Journal of the Association f.
or Computing Machinery) Volume 13 Issue 4 (1966.10) pp471
See -494.

Step 603 ... By referring to this labeled image {Zc _ij }, the defective area is cut out from the input image {f _ij }, and the average pixel value of this cut-out area is obtained to determine the defective area unit. Find the brightness of the defect. That is,
The brightness L _k (k = 1, ..., m) of each defect area is You can ask at.

Incidentally, when a defect is present in the other input image {g _ij} can be cut out a defective area from the input image {g _ij}.

〔The invention's effect〕

As described above, according to the present invention, it is possible to accurately detect a defect existing on the inspection object. Therefore, it is possible to accurately measure the characteristics of the detection area and realize a precise defect inspection.

[Brief description of drawings]

FIG. 1 is an explanatory view of a processing procedure of the defect detection method of the present invention,
FIG. 2 is an explanatory diagram of a conventional defect detection method, FIG. 3 is a configuration diagram of a semiconductor wafer appearance inspection apparatus adopting the present invention, FIG. 4 is an explanatory diagram of an image of a defect detection procedure, and FIG. Is
FIG. 6 is an explanatory diagram of the direction code, and FIG. 6 is an explanatory diagram of a processing procedure of the brightness determination method of the defective area. 1 ... Semiconductor wafer to be inspected, 2 ... Movable data stand, 3 ... TV camera, 4 ... A / D converter, 5 ... Image memory, 6 ... Computer.

Claims (2)

[Claims] 1. A defect inspection method for detecting a defect in a pattern to be inspected from a difference between the image signals, which has a means for capturing an image of the pattern to be inspected and converting the image signal into a video signal and a means for comparing two video signals of the same type pattern. , The binary signal is set to "1" when the difference signal is a positive constant value or more and "0" otherwise, and when the difference signal is another negative constant value or less, "1" is generated. Generates a binary image that is "0" in all other cases,
Furthermore, for these two binary images, after enlarging the "1" area, the logical product is taken between the two images, and the signal thus obtained and the two unenlarged binary images are ORed. In the signal, the defect detection method is characterized in that the "1" region is a defect. 2. The defect detection according to claim 1, wherein the value of "0" in the "0" area surrounded by the "1" area is converted into "1" to make the "1" area a defect. Method.