JPH11339040A - Macro-inspection method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a macro-inspection method for accurately detecting macro- abnormality by a simple processing. SOLUTION: At first, a semiconductor wafer image is picked up, and a picked up image is inputted (a step 11). The picked up image is second-dimensional Fourier transformed, and frequency components corresponding to a pattern cell and a noise are removed from the power spectrum, and a pre-processed image is generated by the second-dimensional inverse Fourier transformation (a step 12). A multi-stage resolution image is generated from the pre-processed image, and a range filter processing and a binarization processing is performed to each image so that density non-uniformity can be extracted (a step 13).

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an inspection method for automatically detecting a so-called macro abnormality such as an abnormal film thickness of an object to be inspected such as a semiconductor wafer or a defective exposure and development.

[0002]

2. Description of the Related Art In recent years, in the processing steps of semiconductor wafers, liquid crystal panels, and the like, the loading of substrates from cassettes and the handling of substrates have been automated for the purpose of saving labor and maintaining cleanliness. 5-
137799, JP-A-6-237206). In the inspection process of semiconductor wafers and the like, automation of micro inspection for inspecting minute defects is being studied.

Conventionally, however, automation of so-called macro abnormalities such as abnormal film thickness over the entire surface of a substrate such as a semiconductor wafer and poor exposure and development has been hardly advanced.

As a method of automating the macro inspection, there is known a method of treating the macro inspection as an extension of the defect size in the micro defect inspection (Japanese Patent Laid-Open No. 8-55218).

FIG. 8 is a diagram for explaining the macro inspection method. Here, the inspection pattern 10 is based on the result of imaging the inspection object with the detection optical system 101 having a microscope or the like.
2 is generated. The inspection pattern 102 is aligned with a CAD pattern 104 (reference pattern), which is CAD data 103 (positioning unit 105), and a difference from the CAD pattern 104 is obtained by a difference pattern extracting unit 106. The length of the pattern obtained by the difference is measured by the difference pattern length measuring means 107, and as a result of the measurement, if the length is equal to or larger than a predetermined value, the defect extracting means 108
Is extracted as a macro defect, and is output from the defect information output means 109.

[0006]

The inspection object to be automated by the present invention is equal to or more than a pattern cell which cannot be found by a microscopic inspection called a micro inspection, such as a film thickness abnormality or an exposure / development defect occurring at a wafer generation stage. Is a size abnormality, and is generally called a macro abnormality.

[0007] Such a macro abnormality cannot be detected even if it is treated as an extension of the defect size in the micro defect inspection as in the method described in the above-mentioned Japanese Patent Application Laid-Open No. 8-55218. This is because if the same detection optical system as that for micro inspection is used for sensing, the maximum detectable defect size is about the size of the microscope field of view, so large-sized defects (macro abnormalities) cannot be detected.

An object of the present invention is to solve the above-mentioned problem, and an object of the present invention is to provide a macro inspection method capable of accurately detecting a macro abnormality by a simple process.

[0009]

According to a first aspect of the present invention, there is provided a macro inspection method for detecting a macro abnormality of an inspection object by using the imaging data of the inspection object. A first step of obtaining a power spectrum of an original image by conversion, a second step of removing a periodic pattern component in the inspection object from the power spectrum, and performing an inverse Fourier transform on the power spectrum after removing the periodic pattern component And a fourth step of detecting a macro abnormality based on the distribution of shading of the preprocessed image.

A macro inspection method according to a second aspect of the present invention is the macro inspection method according to the first aspect, wherein a high frequency noise component is also removed from the power spectrum in the first step.

According to a third aspect of the present invention, in the macro inspection method according to the first or second aspect,
A fourth step of generating extracted data from the pre-processed image by extracting a shading unevenness using a range filter, and binarizing the extracted data using a preset shading threshold; A sixth step of detecting shading unevenness based on the binarization result.

According to a fourth aspect of the present invention, in the macro inspection method according to the first or second aspect,
The fourth step is a seventh step of creating a plurality of images having different resolutions from the pre-processed image, and creating extraction data in which shading unevenness is extracted from each of the images having different resolutions using a range filter. The fifth step,
A sixth step of binarizing the extracted data using a preset gray level threshold, integrating the respective binarized results, and detecting gray level unevenness.

The operation of the present invention will be described below.

According to the present invention, the whole or a part of the image to be inspected is acquired using an area sensor or a line sensor, and the power spectrum of the original image is obtained by Fourier transform. Then, using the cycle of the pattern cell obtained in advance from the reticle design or the like, the cycle component of the pattern cell is removed from the power spectrum, and then the two-dimensional inverse Fourier transform is performed to remove the pattern cell portion from the captured original image. Find the processed image. As a result, macro abnormalities over a wide range of the inspection object can be contained in one image, and only macro abnormalities that appear as shading unevenness can be extracted by removing the background pattern cells. Become.

Further, by performing range filter processing and binarization processing on the pre-processed image, it is possible to detect shading unevenness.

Further, by changing the resolution of the pre-processed image, that is, by performing the process on a multi-level resolution image, it is possible to detect all the shade unevenness having a small area to the uneven shade having a large area.

[0017]

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

FIG. 1 is a diagram for explaining the flow of processing in the macro inspection method according to the present embodiment. 2 to 7 are diagrams for explaining a specific example of the macro inspection method. Hereinafter, the macro inspection method according to the present embodiment will be described with reference to these drawings. Here, the macro inspection method of the semiconductor wafer 1 including M × N pattern cells and having unevenness due to distribution of film thickness or the like is present. An inspection will be described as an example.

(Step 11) First, an image of a semiconductor wafer to be inspected is taken by an image input means to obtain image data. The image input means may be an area sensor or a line sensor. Note that the imaging here is performed on the entire area of the semiconductor wafer 1 or at least an area including a plurality of chips therein. This is because the present macro inspection apparatus is different from a micro inspection apparatus that detects minute defects and is intended to detect a macro abnormality equal to or larger than the chip size in the semiconductor wafer 1.

Incidentally, the macro-abnormality is a defect caused by an abnormal film thickness in the wafer manufacturing process or a defective exposure / development, and is a type of defect that cannot be detected by a so-called micro inspection using a microscope. Note that, when at least a wide range of the semiconductor wafer 1 is imaged as described above for detecting a macro abnormality, micro abnormalities such as fine scratches do not appear as image data.

(Step 12) Next, two-dimensional Fourier transform processing of the image data is performed (first to first claims).
Third step). In general, when an image of a semiconductor wafer or the like is taken, macro abnormalities are observed as shading unevenness 3 and 4 together with the pattern cells 2 regularly arranged in the horizontal and vertical directions as shown in FIG. In the processing, the background pattern cell 2 is removed to extract only the macro abnormalities 3 and 4. That is, filtering of the specific frequency component corresponding to the pattern cell 2 is performed to generate a pre-processed image from which the pattern cell 2 is removed as shown in FIG. The specific processing here will be described later.

(Step 13) Next, macro abnormalities are detected using the preprocessed image shown in FIG. This detection is performed by detecting density unevenness in the preprocessed image (a fourth step in the claims). The detection of the shading unevenness is, for example,
There are a method of detection by range filter processing, a method of detection using the maximum and minimum values of density, a method of detection by calculating variance, and the like. The range filter processing will be described later in detail.

(Step 14) If there is shading in step 13, it is determined that there is a macro error,
If there is no shading, it is determined that there is no macro error, and the process is terminated.

As described above, in the present embodiment, the pattern cells 2 are removed by using the Fourier transform, so that a change in shading that is not related to the macro abnormality is eliminated. Abnormality can be detected. Therefore, it is not necessary to compare with the reference pattern as in the related art, and the processing can be simplified. Further, the conventional method cannot cope with a change in the background density of the image data. However, according to the present method, a macro abnormality can be accurately detected in such a case.

Next, the two-dimensional Fourier transform processing in the above (Step 12) will be described.

In this processing, first, a power spectrum of the image of FIG. 2 is obtained from the image data of FIG. 1 by a two-dimensional Fourier transform (first step in claims).
In the two-dimensional Fourier transform, a data sequence of an original image is converted to f _xy (x
= 0,..., M−1, y = 0,..., N−1), and the converted data sequence is represented by F _uv (u = 0,..., M−1, v = 0,.
When -1), the calculation is performed by the calculation shown in (Equation 1).

[0027]

(Equation 1)

The power spectrum is the square of the absolute value of each component of _Fuv , as shown in (Equation 2).

[0029]

(Equation 2)

This | F _uv | is a two-dimensional array of M × N. For convenience, it is moved by M / 2 in the U direction and by N / 2 in the V direction so that the origin is symmetric in the UV two-dimensional space. I do. That is, F _{u + M / 2} and _{v + N / 2} are obtained. As a result, the power spectrum of the image of FIG. 2 appears symmetrically at the origin in the UV two-dimensional space shown in FIG.

Then, in the UV two-dimensional space of FIG.
Filtering removal of a specific frequency component corresponding to the pattern cell 2 is performed (second step in claims). Here, noise components are also removed.

The periodic component of the pattern cell 2 depends on the reticle design used in the stepper, and can be determined by registering it as wafer information in advance. Also, since noise is a high-frequency component, it appears in the outermost rectangular area of the power spectrum. Therefore, in the UV two-dimensional space of FIG. 3, a specific frequency region (the rectangular region 7 in FIG. 3) corresponding to the periodic component of the pattern cell 2 and the outermost frequency region (the rectangular region 8 in FIG. 3) are masked. To remove. Thereby, the pattern cell 2
Can be removed. If the noise component is also removed in this way, erroneous macro abnormality detection in subsequent processing can be suppressed.

Here, in the two-dimensional space Fuv on which the power spectrum is generated, a two-dimensional data sequence obtained by masking a region corresponding to the masking region is Fuv ′. 1 of 2
By performing the dimensional inverse Fourier transform (third step in the claims), as shown in FIG. 4, compared with FIG. (Referred to as an image). Here, the two-dimensional inverse Fourier transform of FIG. 1 is given by (formula 3) as the inverse transform of (formula 1).

[0034]

(Equation 3)

As described above, in the present step 12, the periodic component and the noise component of the cell pattern 2 are removed by converting the image data using the Fourier transform. The flow of image data conversion in step 12 is summarized in (Equation 4).

[0036]

(Equation 4)

Next, a method of detecting the shading unevenness in the above (Step 13) will be described. here,
A case in which shading is detected by range filter processing using a 3 × 3 neighborhood range filter as shown in (Equation 5) will be described.

[0038]

(Equation 5)

The range filter represented by (Equation 5) is composed of each pixel f _xy and its eight neighboring pixels f _{x + 1} , _y−1 , f
_x , _y-1 , fx _-1 , _y-1 , fx _{+ 1} , _y , fx _-1 , _y ,
_{f x + 1, y + 1} , f x, in _{y + 1, f x-1} , y + 1, replacing the difference between the maximum and minimum gray value as g _xy. The operating principle of this range filter will be described using an example of the preprocessed image in FIG. In addition, here, the two-dimensional data series will be described as a one-dimensional data series for simplification.

FIG. 5A shows one-dimensional gray value series data in the preprocessed image, which has two different gray scales O and P. FIG. 5 shows a state in which this is quantized with a resolution M.
(B). FIG. 5C shows the result of performing the range filter operation (the fifth step in the claims) on the M data sequences.

As shown in FIG. 5C, in this method, even when the gradient range is small, the value after the range filter operation is large at the point where the gradation gradient is large (point O in FIG. 5).
It is easy to detect the position by binarizing with a predetermined threshold (the sixth step in the claims),
Although the gradient range is wide, the value after the range filter calculation is small at a point where the gradation gradient is small (point P in FIG. 5), and it is difficult to detect by binarization. That is, in this method (resolution M), if the density gradient is gentle, the density unevenness cannot be detected.

However, as shown in FIG. 5D, when the resolution is quantized to M / 2, which is half of the resolution described above, and a similar range filter operation is performed, the result becomes as shown in FIG. A wide portion can be detected by binarization. That is, by lowering the resolution, it is possible to detect light and shade unevenness having a wider gradient range. On the other hand, at this time, it is difficult to detect light and shade unevenness with a small gradient range (when the gradient range is smaller than point O in FIG. 5).

As described above, if the density unevenness is detected by the range filter, the location of the density unevenness (the location where the gradient of the density unevenness occurs) can be detected, and the macro abnormality can be detected. Further, in this case, by performing quantization at a multi-step resolution (seventh step in the claims) in consideration of the size of the gradient range, uneven shading can be detected even for different gradient ranges and gentle gradients. It is possible to do.

Using this principle, an image having the K-stage resolution shown in FIG. 1 is created with respect to the pre-processed image shown in FIG. By performing the binarization process based on the set shading threshold, and integrating the shading unevenness detected in the image of each resolution, if there is any shading unevenness (or if a predetermined number or more shading unevenness is detected). B) It is determined that there is a macro error. As a result, more accurate macro abnormality detection becomes possible. Resolution (M, N) for preprocessed image in FIG.
FIG. 6 shows an example of the range filter processed image in the image, and FIG. 7 shows the binarized image.

The method of obtaining the resolution at the K stage is, for example, (Equation 6), but another method may be used.

[0046]

(Equation 6)

In (Equation 6), the original image f _xy1 (resolution M
× N), the image in which the vertical and horizontal resolutions are reduced by half is f _xy2 (resolution M / 2 × N / 2), and the image whose resolution is further reduced in half is f _xy3 (resolution M / 4 × N /
4),... Indicate that an image obtained by repeating the above operation k times is represented by f _xyk (resolution M / 2k−1 × N / 2k−1). In (Equation 6), as a method for reducing the resolution by half, a method is used in which the average value of each 2 × 2 pixel area of the original image is replaced with one pixel as a representative value.

Here, the method of changing the resolution is performed by software processing, but a method of changing the resolution of hardware such as imaging means may be used.

In the above description, a macro inspection method for a semiconductor wafer has been described. However, the present invention is not limited to this, and can be used for a macro inspection in a liquid crystal display device such as a TFT liquid crystal. In this case, the image input in step 1 in FIG. 1 (that is, the image captured by the image input device) needs to capture the entire area excluding the driver portion of the liquid crystal display device or a wide area including at least a plurality of matrices.

[0050]

As described above, according to the present invention, a macro abnormality can be detected only by detecting the shading of an image from which pattern cells have been removed by using a Fourier transform, and it is not necessary to compare with a reference pattern. Simplification can be achieved. In addition, a macro error can be accurately detected regardless of a change in the background.

Further, by performing range filter processing and binarization processing on the pre-processed image, it is possible to detect shading unevenness.

If a multi-level resolution image is created from the pre-processed image, and then integrated after performing a range filter process and a binarization process, it is possible to detect all the shade unevenness from small to uneven shade. Becomes

[Brief description of the drawings]

FIG. 1 is a diagram illustrating a flow of processing of a macro abnormality detection method according to an embodiment of the present invention.

FIG. 2 is a diagram illustrating an example of a captured image of a semiconductor wafer to be subjected to macro inspection.

FIG. 3 is a diagram showing a power spectrum coordinate and a masking area of a captured image.

FIG. 4 is a diagram showing a preprocessed image obtained by removing a periodic component and a noise component of a pattern cell.

FIG. 5 is a diagram illustrating the operation principle of the range filter processing.

FIG. 6 is a diagram illustrating an example of an image after range filter processing.

FIG. 7 is a diagram illustrating an example of an image after binarization processing on the image illustrated in FIG. 6;

FIG. 8 is a diagram illustrating a conventional macro inspection method.

[Explanation of symbols]

 1 captured image 2 pattern cell 3, 4 shading unevenness (macro abnormal)

Claims (4)

[Claims] 1. A macro inspection method for detecting a macro abnormality of an inspection object by using imaging data of the inspection object, comprising: a first step of obtaining a power spectrum of an original image from the imaging data by Fourier transform; A second step of removing a periodic pattern component of the inspection object from the power spectrum; a third step of performing a reverse Fourier transform on the power spectrum after removing the periodic pattern component to obtain a pre-processed image; A fourth step of detecting a macro abnormality based on the distribution of shading of the macro. 2. The macro inspection method according to claim 1, wherein a high frequency noise component is also removed from the power spectrum in the first step. 3. The macro inspection method according to claim 1, wherein the fourth step is a step of creating, from the pre-processed image, extracted data obtained by extracting shading unevenness using a range filter; A sixth step of binarizing the extracted data using a preset gray level threshold and detecting gray level unevenness based on the binarization result. 4. The macro inspection method according to claim 1, wherein the fourth step is a step of creating a plurality of images having different resolutions with respect to the pre-processed image; A fifth step of creating extracted data from each of the images by extracting the shading unevenness using a range filter; and binarizing the extracted data using a preset shading threshold, each of the binarized results. And a sixth step of detecting shading unevenness.