JP2001209798A - Method and device for inspecting outward appearance - Google Patents

Abstract

PROBLEM TO BE SOLVED: To realize a highly accurate and reliable outward appearance inspection by eliminating variable density nonuniformity due to light diffraction generated in a repeating pattern part. SOLUTION: A picked up image is smoothed to generate a reference image with the repeating pattern part as an object, an image obtained by fetching only a steep variable density change part with density nonuniformity eliminated is extracted by comparing the picked up image with the reference image, and the lightness value of the extracted image is binarized to decide the existence/ absence of a defective part. An appropriate binarization threshold is automatically set by adopting processing that a standard deviation in a repeating pattern area is measured from the extracted image as binarization processing, that a threshold responding to the standard deviation is calculated and binarized and that the standard deviation is subsequently measured again and the threshold is binarized with a pixel decided as a repeating pattern by the binarization as a true repeating pattern part.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a visual inspection method and an inspection apparatus for inspecting a panel-like object having a repetitive pattern such as a semiconductor wafer and a liquid crystal display panel for scratches.

[0002]

2. Description of the Related Art An example of an appearance inspection for inspecting a surface flaw of a semiconductor wafer will be described below.

[0003] First, a pattern is formed on the surface of a semiconductor wafer according to fine rules in sub-micron units.
In order to inspect the minute defect, a method (micro inspection) of observing and inspecting a minute area using a microscope having a magnification of 100 to 200 times is adopted. In addition, the surface of a semiconductor wafer may be damaged during the manufacturing and transport processes. For this inspection, a macroscopic appearance inspection (macro inspection) for observing and inspecting the entire wafer surface at once is applied. ing.

[0004] The macro inspection generally involves a method in which a skilled inspector visually observes a wafer from a plurality of angles to check for scratches, and judges pass / fail based on the inspection results based on the observation results. Has been adopted.

On the other hand, Japanese Patent Laid-Open No. Hei 10-144747 discloses an automatic wafer macro inspection apparatus for realizing automation of macro inspection. In the inspection apparatus described in this publication, the semiconductor wafer is illuminated obliquely, the semiconductor wafer is observed and imaged by a camera for imaging from a vertical direction, and an image of the inspected wafer and an image of the reference wafer are previously collected and acquired. The stored reference image is compared with the stored reference image by a pattern matching method in image processing to measure the degree of coincidence, and the presence or absence of a defective portion is determined according to the degree of coincidence.

[0006]

However, in the visual inspection, since the inspector's intuitive judgment is mainly performed, it is difficult to perform a quantitative inspection, and there is a problem that the judgment criteria differ depending on the skill level and physical condition. .

In recent years, as the pattern rules of wafers have become more and more precise, the influence of contamination by humans is great, so that complete unmanned processes are required, and the need for automatic inspection is increasing. However, as long as a visual inspection is performed, it cannot be realized.

On the other hand, according to the automatic wafer macro inspection apparatus disclosed in Japanese Patent Application Laid-Open No. 10-144747, a semiconductor wafer is irradiated with illumination light from an oblique direction and observed from above the wafer surface. When light is radiated from an oblique direction, in the repetitive pattern portion, an optical path difference occurs between the reflected light reflected on the upper surface of the pattern and the illumination light reflected on the groove between the patterns. The intensity of the diffracted light varies according to the distance from the illumination, and the angle of the irradiated light varies according to the distance from the illumination on the wafer surface. Therefore, if an image is picked up by a camera installed above the wafer, the picked-up image repeatedly has rainbow-like rainbow-like shades of violet to red on the pattern surface. Hereinafter, the density unevenness observed in the captured image is referred to as density unevenness due to diffraction.

The occurrence of density unevenness due to diffraction causes the position, width, etc. of the density unevenness due to diffraction to fluctuate slightly each time an inspection is performed due to a positional shift, warpage, bending, etc. when positioning the wafer. Therefore, as in the technique described in Japanese Patent Application Laid-Open No. H10-144747, simply measuring an image based on a normal wafer surface and comparing the image with a captured image of a wafer to be inspected results in shading unevenness due to diffraction. There is a problem that the influence of the fluctuation cannot be removed, and it is difficult to perform a highly accurate and reliable automatic inspection.

[0010] In order to solve this problem, it was experimentally examined to irradiate the illumination not from obliquely but from other angles to avoid the shading caused by diffraction. The results are described below.

First, when the illumination light is irradiated by the epi-illumination so that the elevation angle with respect to the wafer surface becomes 90 degrees, the density unevenness due to diffraction is eliminated, but the brightness of the repetitive pattern portion may be bright or dark. On the other hand, in the case of a wafer where the repeated pattern portion becomes dark, the scratch and the repeated pattern cannot be separated. Further, when illumination light is irradiated from the horizontal direction at an elevation angle of 0 °, the unevenness in density due to diffraction is eliminated, but the flaws and the repetitive pattern portions are darkened and cannot be similarly separated. From these examination results, it is necessary to irradiate the light obliquely to the wafer surface in order to obtain the largest difference in shading between the flaw and the non-flawed part. It can be seen that the influence of the density unevenness due to diffraction cannot be avoided.

The present invention has been made in view of such circumstances, and it is necessary to repeatedly inspect a defect such as a semiconductor wafer or a liquid crystal display panel by macroscopically observing a defect or a foreign substance using an optical method. It is an object of the present invention to provide an appearance inspection method and an inspection method capable of canceling the influence of shading unevenness due to diffraction of light generated in a pattern area and thereby detecting the presence or absence of a defective portion with high accuracy.

[0013]

SUMMARY OF THE INVENTION A visual inspection method according to the present invention is a method for optically inspecting an object having a repetitive pattern area for a defect, and excluding a density change due to the defect from a picked-up image of the repetitive pattern area. A process of generating a reference image by comparing the reference image with a captured image to obtain an image in which only a steep gradation change amount is extracted, and binarizing a brightness value of the extracted image to determine a defective portion. Is characterized by performing a process of determining the presence or absence of

In the appearance inspection method of the present invention, the reference image used for the comparison operation is preferably generated by a smoothing process of the picked-up image. The smoothing process includes, for example, a median filter, a minimum value filter, and an averaging filter. ,
Alternatively, it is preferable to use a rank extraction filter capable of arbitrarily setting the extraction rank.

In the appearance inspection method of the present invention, the standard deviation in the pattern area is repeatedly obtained from the extracted image from which only the steep gray-scale change amount is extracted, and then binarized by a threshold value corrected by a predetermined coefficient. A value obtained by correcting the standard deviation of the brightness value of the non-defective part in the repetitive pattern area by a predetermined correction coefficient, with a part determined as low brightness in the binarized image as a non-defective part (true repetitive pattern part). May be performed again, and in this case, an appropriate binarization threshold can be automatically set, and a highly reliable automatic inspection can be realized.

An appearance inspection apparatus according to the present invention is provided with an image pickup means for observing an object having a repetitive pattern in a direction perpendicular to an inspection surface of the object, an illumination system for irradiating the object with illumination light from an oblique direction, and A reference image is generated from the image by excluding a change in density due to a defect, and the reference image is compared with a captured image to obtain an image in which only a steep change in density is extracted. The brightness value of the extracted image is binary. It is characterized by having an image processing means for performing a conversion process.

[0017]

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 2 is a diagram schematically showing the configuration of an embodiment of the appearance inspection apparatus of the present invention.

The appearance inspection apparatus shown in FIG.
, A line-type illumination 2, and an imaging camera 6 disposed above (vertically) the wafer table 1.

The line type illumination 2 comprises a halogen light source 3, an optical fiber 4 and an irradiating section 5, and irradiates the light source light from the halogen light source 3 to the surface of the wafer W obliquely from above via the optical fiber 4 and the irradiating section 5. It is configured to be. The irradiating section 5 has a structure in which the optical fibers 4 are bundled in a line shape, and the light source light from the halogen light source 3 is emitted as a linear light beam.

The imaging camera 6 has a photoelectric conversion element C
A CD sensor is used to capture a monochrome grayscale image of the surface of the wafer W. The image data captured by the image capturing camera 6 is processed by the image processing device 7 and then taken into the control computer 8. The control computer 8 controls the operation of the entire apparatus, interfaces with peripheral process apparatuses, and interfaces with workers.

Next, this embodiment will be described more specifically.

First, the wafer W is transferred onto the wafer table 1 by an external transfer device (not shown). The wafer table 1 has a function of holding the transferred wafer W in a horizontal state, and a surface image of the wafer W held on the wafer table 1 is captured by the imaging camera 6.
The captured image data is input to the image processing device 7 and subjected to image processing described later.

In this embodiment, the images handled by the imaging camera 6 and the image processing device 7 are both black and white grayscale image data of 2048 horizontal pixels × 2048 vertical pixels × 8 bits (256) gradations. The entire surface of the 200 mm wafer W can be imaged with one image. Further, the size of one pixel is about 100 μm square, and the expected flaw width is assumed to be within 200 μm.

Also, as shown in FIG.
By irradiating the illumination light to the surface of the wafer W in an oblique direction, the brightness of the wafer surface can be made as dark as possible. The angle is set so that the difference between the brightness and the brightness of the wafer surface becomes large. Specifically, the line-type illumination 2 is arranged so that the elevation angle with respect to the wafer surface is 10 degrees near the center of the wafer W. The horizontal distance between the illumination unit 5 of the line type illumination 2 and the edge of the wafer W is 130 mm, and the vertical distance between the imaging camera 6 and the wafer surface is about 300 mm. However, these distances are appropriately determined by the focal length of the lens and the magnification of the image to be captured.

Further, since the non-repeated pattern portion of the wafer W is formed by intricately combining various horizontal patterns, the brightness is not uniform and the flaw and the pattern portion cannot be clearly separated. Have difficulty. Therefore, in the present invention, attention is paid to the fact that a uniform pattern is formed in the repeated pattern portion, and in order to perform a stable inspection, detection is performed on the scratches of the repeated pattern portion.

FIG. 3 is an example of an image captured by the image capturing camera 6. Repeated patterns 9... 9 on the surface of wafer W
Are formed in a matrix. The rays of the line illumination 2 applied to the surface of the wafer W have a large elevation angle on the wafer surface close to the illumination, and the elevation angle decreases as the distance from the illumination increases. For this reason, density unevenness due to diffraction occurs due to the difference in the irradiation angle, and the imaging camera 6 repeats patterns 9... 9 in the bright area 11 and the dark area 12 on the wafer surface.
A rainbow-like shading that leads to a continuous step is observed. Here, the image captured by the imaging camera 6 is a black-and-white grayscale image, and the color cannot be recognized. Therefore, the rainbow-like grayscale unevenness appears in the captured image as black-and-white grayscale unevenness. FIG. 3 shows an example in which two vertical scratches 10 extending slightly obliquely exist in the repeated pattern portion near the center of the wafer W.

FIG. 4 shows an example of a measurement result obtained by measuring the brightness of a repetitive pattern including a flaw 10 in the horizontal direction from a captured image of the wafer W shown in FIG. The result (brightness distribution) of the measurement of the horizontal arrow section crossing is indicated by a solid line. The brightness distribution 13 is a state in which a change in the brightness value due to shading unevenness appears largely as a whole, and includes two portions where the brightness value due to the scratches 10 is high. The left flaw is a flaw having a large width and a high brightness difference, and the right flaw is a flaw having a small width and a small brightness difference. In addition, random noise mixed at the time of imaging is mixed in the repetitive pattern portion, and there is a slight change in brightness as a whole.

Next, the flaw detection algorithm executed by the image processing apparatus 7 will be described below with reference to the flowchart of FIG. Note that the image processing device 7 used in this example is:
A plurality of processing target areas can be provided in one image, and when performing image processing, processing can be performed only on the target area. This processing target region is generally called a region of interest, and is also referred to as a region of interest in this description.

The specific procedure of the flaw detection algorithm shown in FIG. 1 is as follows.

Step S1: A region of interest in which only the repetitive pattern portion is to be processed is set in the image of the wafer W. Specifically, one region of interest is set for one region of the repeated pattern, and the region of interest is set for all the repeated pattern portions in the wafer W. The region of the repetitive pattern portion in the wafer W is registered in advance by a method in which the operator sets the coordinates of the captured image while viewing the screen. Note that all image processing by the flaw detection algorithm thereafter is performed on the region of interest.

Step S2: A smoothing filter process by a median filter is performed on the captured image. Hereinafter, the notation of the image is the image name (x, y). Where x = 0
2047, y = 0 to 2047. The median filter is an n × n size mesh-shaped filter, and is a smoothing filter for all the pixel data in the mesh space, which outputs a value in the middle of the numerical order as the output result.

In this example, considering that the width of the flaw 10 is within 2 pixels, the width of the flaw 10 is twice or more (x, y).
And a median filter with n = 5 width. If the maximum expected scratch width is one pixel or less, a median filter with n = 3 may be used. If there is a possibility that the maximum width of the flaw is 2 pixels or more, n
May be an odd value exceeding 5. However, it is important that the number of pixels in a high brightness portion due to a flaw in the filter space is less than 1/2 of the total number of pixels in the filter space.

FIG. 4 shows an example of the brightness distribution of an image obtained by the above-described smoothing process.

Although the median filter is a two-dimensional filter, FIG. 4 shows a processing result of one-dimensional data for simplifying the explanation.

As can be seen from the processing result 14 in FIG. 4, a high brightness value due to the scratch 10 can be canceled by the effect of the smoothing process, and an average brightness value near the scratch can be obtained.

Here, in the generally used smoothing method, there are many averaging filters in an n × n mesh. However, in the averaging, an output value becomes high in a portion where a flaw exists. Therefore, in the present invention, it is possible to eliminate the influence of the scratch by using a median filter having an appropriate space size.

Step S3: Image comparison between the captured image and the smoothed image is performed.

More specifically, the captured image is represented by I (x, y), the smoothed image is represented by M (x, y), and the comparison image D (x, y) is processed by the following equation as a method of image comparison. obtain.

D (x, y) = I (x, y) + OM (x, y) The image comparison is performed by the difference processing between the images. Only, not negative image brightness values,
The offset value O is uniformly added. Offset value O
Is set to a value larger than the maximum value of the distribution by previously measuring the brightness distribution of the repetitive pattern on a plurality of wafers. In this example, the offset value O is set to 10.

Step S4: A binary threshold value for separating a repetitive pattern portion and a non-repetitive pattern portion in the region of interest is obtained.

More specifically, a standard deviation 3σa is obtained for each region of interest in the comparative image D (x, y), and then the standard deviation 3σa is multiplied by a predetermined constant to obtain a separation 2.
A value threshold Ta is obtained.

Ta = 3σa × K + O Here, the constant K is set in consideration of the fact that the actual brightness distribution width is slightly wider than the distribution width obtained from the standard deviation amount of the repeated pattern. In this example, K = 1 is set in order to explain an example of detecting a flaw having a minute difference in brightness.

By the above processing, the standard deviation value can be measured for all the pixels in each region of interest, and a value including the pixel data of the flaw can be obtained. Therefore, as shown in FIG.
If there is a flaw, a standard deviation 15 larger than the standard deviation 16 of the pure repetitive pattern portion is obtained.

Step S5: Using the separation threshold,
The comparison image D (x, y) is binarized to separate images S (x, y).
y).

Specifically, the binarization process is performed by setting the pixel having the brightness equal to or higher than the threshold value to 1 and the pixel having the brightness lower than the threshold value to 0. As a result, the separated image S (x, y)
Within, the true repeated pattern portion is 0, and the non-repeated pattern portion is 1. However, since the threshold value measured in step S4 has an adverse effect due to the flaw, the processing in step S5 cannot properly separate the repetitive pattern portion from the non-repeated pattern portion. For example, FIG.
In the example shown in FIG. 5 and the scratch 10 having a small brightness difference on the right side in the drawing, the binarization threshold Ta for separation affected by the scratch 10 having a large brightness difference on the left side in the drawing is used. Cannot be separated exactly. Therefore, a process of performing binarization again after step S6 and determining a flaw is performed.

Step S6: A binarization threshold for flaw determination is obtained. For the comparison image D (x, y), a standard deviation 3σb is obtained for each region of interest. However, in this process, unlike step S4, the pixel whose separated image S (x, y) is 1 is not added to the measurement target. As a result, it is possible to exclude data on a flaw having a large brightness difference on the left side of the figure, which has had an adverse effect on the standard deviation 3σa.

The binarization threshold value Tb for flaw determination is set in step S
It is calculated by the following equation in the same way as 4.

Tb = 3σb × k + O Step S7: The comparison image D (x, y) is binarized using the flaw determination threshold value Tb to obtain a flaw image R (x, y).

More specifically, a pixel having a brightness equal to or higher than the threshold is set to 1 and a pixel having a brightness lower than the threshold is set to 0, and the binarization process is performed. As a result, in the flaw image R (x, y), a pixel having a value of 1 has a flaw.

Step S8: Image R resulting from detection of flaw
From (x, y), the area of the flaw pixel is calculated for each region of interest, and the presence / absence and area of the flaw for each region are supplied to the control computer 8 as a flaw inspection result. The control computer 8 displays and outputs the flaw inspection result.

With the above-described flaw detection algorithm, flaw inspection can be performed in which the adverse effect of shading unevenness due to light diffraction is canceled.

Here, in the above embodiment, step S
Although the median filter is used for the smoothing process of 2,
Instead of this, a minimum value filter, an average value filter, or a rank extraction filter capable of arbitrarily setting a rank may be used.

The minimum value filter is a filter in the form of a mesh of size n.times.n, which is a filter that outputs a value having the minimum numerical value for all pixel data in the mesh space, and is originally a filter for smoothing. Although it is rarely used as a filter, it is a filter that has a smoothing effect because it can exclude lightness higher than the surroundings in the sense of removing steep scratch-like changes with high lightness values, and removes scratched pixels. be able to. Further, in the minimum value filter, when the density unevenness greatly changes due to diffraction, and there is a large density change within the filter size, there is a possibility that a part of the influence of the density unevenness remains during the image comparison in step S3. Generally, in an image processing apparatus, the minimum value filter can perform processing at a higher speed than the median filter, has less shading due to diffraction, and is an effective means when used in a manufacturing process in which inspection time is particularly restricted.

The average filter is an n × n mesh filter having a smoothing effect of outputting an average value of numerical values for all pixel data in the mesh space. Can be removed. The average filter has a strong change in shading due to diffraction.
If there is a large change in shading within the filter size, the effect of shading may remain partially during the image comparison in step S3. However, in an image processing apparatus, in general, an average filter is faster than a median filter. This is an effective means when it is used in a manufacturing process where there is little shading due to diffraction and there is a special restriction on the inspection time. Further, the average value filter has a higher effect of canceling the shading unevenness than the minimum value filter, and high-speed inspection is possible even if the effect of shading unevenness due to diffraction is somewhat larger.

The rank extracting filter is a filter in the form of a mesh of size n.times.n, and is a filter in which the numerical rank is set as an output result with respect to all pixel data in the mesh space. For example, rank 10 by n = 5 filter
In the case of a normal median filter, the 13th numerical value from the minimum value is the 13th numerical value from the maximum value and is the value of the middle rank, but this filter outputs the 10th numerical value from the minimum value. Is done. In the median filter, when the width of the expected flaw increases, n needs to be increased, and the processing time increases in proportion to the square of n. On the other hand, in the order extraction filter, if the ratio of the number of defective pixels to the total number of pixels in the filter size is determined in advance, even if the ratio exceeds 1/2, the minimum number of pixels other than the defective pixels This can be dealt with by setting the number of pixels in the order of the filter, and n can be reduced. However, when a processing device equipped with an image processing LSI is used as an image processing apparatus, the LSI often does not support such a rank extraction filter. It is suitable for use in an image processing apparatus that performs the following. Then, by using the rank extraction filter, it is possible to minimize the adverse effect of shading unevenness due to diffraction and to perform a smoothing process capable of shortening the processing time due to the small n.

In the above embodiment, numerical examples of the image size, the wafer size, and various coefficients in the detection algorithm have been described. However, the present invention is not limited to these numerical values. Similarly, the resolution of the captured image,
It is not limited to the type of imaging camera, wafer size, and the like.

In the above embodiment, an example of inspection for a flaw on a semiconductor wafer surface has been described. However, the present invention is not limited to this. On the other hand, since the brightness value increases, detection can be performed in the same manner. further,
Not only semiconductor wafers, but also scratches / foreign matter of an object having a repetitive pattern with a height difference, such as a liquid crystal panel, can be similarly detected.

[0059]

As described above, according to the present invention,
An image in which only a steep gray-scale change amount that excludes gray-scale unevenness is extracted by removing a gray-scale change due to a defect from a picked-up image of a repetitive pattern area, generating a reference image, and comparing and calculating the reference image and the picked-up image. Then, the brightness value of the extracted image is binarized to determine the presence / absence of a defective portion, so that the influence of light and shade unevenness due to light diffraction can be canceled. As a result, a highly accurate and highly reliable macro inspection can be automatically performed. In addition, since the macro inspection is performed by an optical method, it is possible to perform a quantitative inspection in which the generation of dust is suppressed.

[Brief description of the drawings]

FIG. 1 is a flowchart of a defect detection algorithm according to an embodiment of the present invention.

FIG. 2 is a diagram schematically illustrating a configuration of an embodiment of a visual inspection device of the present invention.

FIG. 3 is a diagram illustrating an example of a captured image of a damaged wafer.

FIG. 4 is a graph showing an example of a brightness value near a flaw in a captured image and a data value after smoothing processing.

FIG. 5 is a graph showing an example of a brightness value near a flaw in a comparative image.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Wafer table 2 Line type illumination 3 Halogen light source 4 Optical fiber 5 Irradiation part 6 Imaging camera 7 Image processing device 8 Control computer 9 Repetition pattern 10 Scratches W Wafer

──────────────────────────────────────────────────続 き Continued on the front page (51) Int.Cl. ⁷ Identification symbol FI theme coat ゛ (Reference) G01B 11/24 K F term (Reference) 2F065 AA49 BB02 BB18 CC19 CC25 DD12 FF42 GG02 GG16 HH12 JJ03 JJ09 JJ26 LL03 PP11 QQ06 QQ08 QQ24 QQ25 QQ27 QQ33 QQ41 RR01 SS03 SS13 2G051 AA51 AA73 AB02 BA20 BB17 CA03 CA04 DA01 DA06 DA07 EA11 EB01 EB02 EC02 EC03 EC05 ED03 FA10 4M106 AA01 BA04 CA39 CA70 DB04 DB07 DB16 DB19 DJ12 DC20

Claims (5)

[Claims] 1. A method for optically inspecting an object having a repetitive pattern area for a defect, wherein a process of generating a reference image by excluding a density change due to the defect from a captured image of the repetition pattern area, and the reference image And a process for obtaining an image in which only a steep gradation change amount is extracted by performing a comparison operation with the captured image, and a process for binarizing the brightness value of the extracted image to determine the presence or absence of a defective portion. Appearance inspection method. 2. The appearance inspection method according to claim 1, wherein
An appearance inspection method, wherein a reference image is generated by performing a smoothing process on a captured image. 3. A smoothing process using one of a median filter, a minimum value filter, an averaging filter, and a rank extraction filter capable of arbitrarily setting an extraction rank. Visual inspection method described. 4. The visual inspection method according to claim 1, wherein
The standard deviation in the pattern area is repeatedly obtained from the extracted image in which only the steep gray-scale change amount is extracted, and then binarized by a threshold value corrected by a predetermined coefficient, and then the binary image is determined to have low brightness. The non-defective part is determined as a non-defective part, and the standard deviation of the brightness value of the non-defective part in the repetitive pattern area is binarized again with a value corrected by a predetermined correction coefficient, thereby determining the presence or absence of the defective part. Features a visual inspection method. 5. An image pickup means for observing an object having a repetitive pattern in a direction perpendicular to a surface to be inspected of an object, an illumination system for irradiating the object with illumination light from an oblique direction, and a change in shading due to a defect from the picked-up image. Image processing means for generating a reference image by excluding the reference image, comparing the reference image with the captured image, obtaining an image in which only a steep change in density is extracted, and binarizing the brightness value of the extracted image A visual inspection device comprising: