JP2000285236A - Method and device for picture information processing and recording medium - Google Patents

Abstract

PROBLEM TO BE SOLVED: To obtain a picture information processing device which detects defects from an examination object picture without using a golden device or CAD data and without requiring much processing time like the FET method. SOLUTION: The picture information processing device which uses a picture to perform examination, recognition, and discrimination of an object is provided with a two-dimensional wavelet conversion means S1 which performs two-dimensional wavelet conversion of input digital picture, a binarization processing means S2 which applies threshold processing to longitudinal line detection components and lateral line detection components obtained by two-dimensional wavelet conversion to generate binary pictures of longitudinal line detection components and lateral line detection components, and a square Hough conversion means S3 which applies square Hough conversion for square detection to binary pictures obtained by the binarization processing means to obtain the position and the size of the object.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for recognizing defect information such as foreign matter generated on a semiconductor wafer or a printed circuit board.
The present invention relates to an image information processing device to be detected.

[0002]

2. Description of the Related Art Defects and foreign matters generated or mixed in a manufacturing process of a printed circuit board or a semiconductor wafer cause defective products. Therefore, it is necessary to detect the defect / contaminant as soon as it occurs.

[0003] In a photograph of a printed circuit board or an image obtained by photographing a semiconductor wafer with an optical microscope or the like,
Conventionally, there is an image information processing apparatus for detecting, recognizing, or judging the above-described defect or foreign matter. Most of these methods use digital image processing for handling images digitally, and perform image processing using a computer.

Conventionally, defect detection using image processing is mainly known as (a) comparison with a golden device, (b) comparison with CAD data, (c) die-to-die comparison, (d) FET method and the like. .

(A) In comparison with the golden device,
An image of an ideal semiconductor wafer having no defect (hereinafter referred to as a golden device) is prepared in advance, and a defect is detected by comparing the image of the golden device with the image of the inspection target. Comparing images means performing a process of creating a difference image of pixel values for each corresponding pixel of two images. If there is no defect in the inspection target image, the image becomes the same image as the image data of the golden device, so that the entire screen becomes a completely flat image having a pixel value of 0 by the difference processing. However, when there is a defect or foreign matter, pixels having a value other than 0 appear intensively in that part. Therefore, a defect or a foreign substance can be detected by extracting the group of pixels having a value other than 0 and obtaining the size and the center of gravity.

(B) In comparison with CAD data, CAD image data designed for a device to be inspected is compared with image data of a manufactured wafer.

As a result, (a) defects and foreign substances can be detected in the same manner as in comparison with the golden device.

(C) In the die-to-die comparison, the image data of adjacent dies is compared by utilizing the fact that a plurality of the same chips (hereinafter referred to as dies) are arranged on one wafer. As a result, defects and foreign substances can be detected in the same manner as in (a) and (b).

(D) In the FET method, the defect is detected by removing the wiring pattern by utilizing the property that the defect generated on the wafer is local while the wiring pattern is periodic. That is, the input wafer pattern image is converted to a two-dimensional FE
Fourier transform by T (fast Fourier transform) etc.
A defect can be detected by removing a specific spatial frequency component corresponding to the wiring pattern in the spatial frequency region using a band stop filter or the like and performing an inverse FET.

[0010]

(A) When utilizing a comparison with a golden device, it is necessary to prepare an image of the golden device in advance. To create an image of a golden device, you first need to find a defect-free sample. The task of finding this flawless sample needs to be done carefully by the human eye. Further, the patterns of semiconductor wafers are very diversified due to the current tendency of high-mix low-volume production, and there are many changes. It is necessary to create an image of the golden device for each of these types or design changes, which requires considerable effort.

(B) In comparison with CAD data, (a)
It is efficient because there is no need to search for a golden device like this. By the way, recently, with the advance of high integration of semiconductor chips, a conventional optical microscope cannot cope with obtaining a detailed image of a wafer pattern.
M; Scanning Electron Microscope) is being used. SEM images have more noise than optical microscope images. Therefore, noise-free CA
When a difference image between the D data and the SEM image is created, a large amount of noise other than a defect is detected.

Further, in both of the methods (a) and (b), it is necessary to first perform accurate alignment when comparing images.

(C) In the die-to-die comparison, adjacent dies are compared with each other, so there is basically no need to perform alignment. However, there is a disadvantage that if a similar defect is accidentally found in the same place, it cannot be detected. Further, similarly to (b), when an SEM image is applied, noise is detected.

(D) The FET method can detect defects even from SEM images. However, this method requires a very long processing time. For example, when a two-dimensional Fourier transform is applied to a digital image composed of 512 × 512 pixels, about 1.31 × (10 8) additions and subtractions are performed,
8.39 × (10 8) multiplications are required. Also, M
Complex wiring patterns such as PU cannot be accurately removed by filtering in the spatial frequency domain.

The first object of the present invention is to provide (a)
An object of the present invention is to provide an image information processing apparatus capable of detecting a defect even from an unclear SEM image in which the influence of noise is large and the defect cannot be detected accurately in the image comparison methods of (b) and (c). .

A second object of the present invention is to provide (a) and (b)
Golden device and CA needed for the image comparison method
An object of the present invention is to provide an image information processing apparatus capable of detecting a defect from a single image to be inspected without using D data.

A third object of the present invention is to provide an image information processing apparatus which can detect a defect from an image to be inspected at high speed without using a large amount of processing time as in the (d) FET method. is there.

[0018]

According to a first aspect of the present invention, there is provided an image information processing apparatus for inspecting, recognizing, and judging an object using an image, wherein two-dimensional wavelet transform is performed on input digital image data. A two-dimensional wavelet transform unit that performs threshold processing on a vertical line detection component and a horizontal line detection component obtained by the two-dimensional wavelet transform to create a binary image of a vertical line detection component and a horizontal line detection component And a square Huff for obtaining a position and a size of an object by applying a Hough transform of a square for detecting a square to the binary image obtained by the binarization processing means. (Hough) conversion means.

The two-dimensional wavelet transform means separates a wiring pattern having only vertical or horizontal components from a defect having components in various directions. Then, the detected components of the vertical line and the horizontal line are binarized by a binarization processing means to obtain a binary image, and a square Hough is added to the binary image.
The converting means applies a square Hough transform to detect a defect as a square. Note that a circular defect is also detected as a square.

Therefore, a defect can be detected from a single image to be inspected without using a golden device or CAD data, and a large amount of processing time is not required unlike the FET method. Does not take effect).

According to a second aspect of the present invention, in the image information processing apparatus according to the first aspect, the square Huff (Houg
h) a binarization processing means for applying threshold processing to the parameter space obtained by the conversion means to detect specific graphic information;

According to a third aspect of the present invention, in the image information processing apparatus according to the second aspect, a square Hough is provided.
Labeling means for grouping by attaching the same label to adjacent active pixels in a binary image of the parameter space obtained by binarizing the conversion result; and a centroid calculating means for calculating barycentric coordinates of each of the labels obtained by the labeling processing. ,
It has.

According to a fourth aspect of the present invention, in the image information processing apparatus according to any one of the first to third aspects,
There is provided a noise removing unit for removing local noise from the input digital image.

According to a fifth aspect of the present invention, in the image information processing apparatus according to the fourth aspect, the vertical line detection component and the horizontal line detection component after applying a binarization processing unit to a two-dimensional wavelet transform result. And an isolated point removing means for removing an isolated active pixel from the binary image.

According to a sixth aspect of the present invention, in the image information processing apparatus according to any one of the first to fifth aspects,
The square Hough transform means performs Hough transform for detecting the xy coordinates of the center of the square and the length of one side.

[0027] The invention described in claim 7 is the invention according to claims 1 to
7. The image information processing apparatus according to any one of 6.
The square Hough transform means converts the two binary images corresponding to the vertical line detection component and the horizontal line detection component of the two-dimensional wavelet transform into the same parameter space.
h) Means for conversion.

[0027] The invention according to claim 8 provides the invention according to claims 1 to
8. In the image information processing apparatus according to any one of items 7,
The square Hough transform means applies Hough to only one of the two binary images corresponding to the vertical line detection component and the horizontal line detection component of the two-dimensional wavelet transform.
h) Means for conversion.

According to a ninth aspect of the present invention, in the image information processing apparatus according to the eighth aspect, there is provided an energy calculating means for calculating energies of the vertical line detection component and the horizontal line detection component of the two-dimensional wavelet transform, Square Huff (Ho
ugh) The conversion means performs a square Hough transform on the binary image corresponding to the lower energy component.

According to a tenth aspect of the present invention, in the image information processing apparatus according to the sixth aspect, a square Hough is provided.
An image information processing apparatus wherein a threshold value of the binarization processing means for a parameter space obtained by conversion changes in accordance with a length parameter of one side of a square.

According to an eleventh aspect of the present invention, in the image information processing apparatus according to any one of the third to tenth aspects, when a plurality of objects overlap in position, the duplication detection and elimination for eliminating the duplication are performed. Means.

According to a twelfth aspect of the present invention, there is provided an image information processing method for inspecting, recognizing, and judging an object using an image. Applying a threshold process to the vertical line detection component and the horizontal line detection component obtained in the two-dimensional wavelet transform step to create a binary image of the vertical line detection component and the horizontal line detection component. A binarization processing step, and a square Huff for obtaining the position and size of the object by applying a square Hough transform for detecting a square to the binary image obtained in the binarization processing step (Hough) conversion step.

According to a thirteenth aspect of the present invention, there is provided a computer-readable recording medium having recorded thereon a program for causing a computer to execute image information processing for inspecting, recognizing, and judging an object using an image. ,
For input digital image data, two-dimensional wavelet transform processing for two-dimensional wavelet transform and applying threshold processing to vertical line detection components and horizontal line detection components obtained by the two-dimensional wavelet conversion processing A binarization process for creating a binary image of the vertical line detection component and the horizontal line detection component; and a square Hough (Houg) for detecting a square with respect to the binary image obtained by the binarization process.
h) A recording medium readable by a computer in which a program for causing a computer to execute a square Hough conversion process for determining the position and size of the object by applying a conversion is recorded.

[0033]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention can be broadly classified into two-dimensional wavelet transform means for separating a defect component from a wiring pattern of a wafer pattern image, and detecting a position and a size of a defect from a binary image obtained by separating the defect component. Huff
h) It is composed of conversion means, and other means are used supplementarily to improve the defect detection performance of the above means.

(A) Square Hough Transform Means Hough transform is described in "Basics of Image Recognition [Ι
Ι] ”(Mori, Sakakura, Ohmsha). Although described in detail in 3-19, the Hough transform processing will be briefly described here. Hough transform is image processing for detecting a specific figure from a binary image. Here, a method for detecting a square will be described.

On the xy plane, as shown in FIG.
A square inclined at 45 degrees is represented by the following equation, where (a, b) is the center coordinate of the square and r is the length of one side of the square.

| X−a | + | y−b | = r / √2 (1) That is, one square on the xy plane is (a, b, r)
In the parameter space, it is represented by one point. The Hough transform converts the xy plane into the parameter space. In this case, a square feature of the xy plane (a,
(b, r) space, so it is called a square Hough transform. The Hough transform of a square is realized by the following equation obtained by modifying the equation (1).

B = ± (r / √2- | x−a |) + y (2) A specific example of the Hough transform of a square will be described with reference to FIG. Each observation point α (x1, y1), β (x
(2, y2) and γ (x3, y3) are expressed as (a, b,
r) Map to space. The r-axis is orthogonal to the ab plane. FIG. 4B shows an ab plane sliced at a certain r0 (here, r0 = 24). For each plane at r, one Hough curve (square) centered on the observation point is obtained from one observation point. If the observation point is on a square whose half distance of one side (hereinafter, referred to as radius) is r0, the square (Hough) curve becomes a point (a0,
(b0, r0). That is, when a square exists on the xy plane, it can be detected as an intersection of many Hough curves in the parameter space. At the intersection, the brightness of the intersection increases in proportion to the number of intersecting Hough curves. Therefore, by finding a parameter having a large luminance value in the parameter space, a square in the xy plane can be detected. In FIG. 4B, only the parameter (a0, b0, 24) has the luminance value 3, and the other values are the luminance value 2 or less. By detecting a large luminance value (a0, b0, 24), the original x
A square having a radius of 24 with the center at (a0, b0) at three points on the -y plane is detected.

Note that a circle in a binary image can be detected as a square circumscribed by the Hough transform of a square. Digital image processing targets image data, that is, discrete data. As shown in FIG.
The condition that a square having the same center coordinates intersects or touches the circle of (hereinafter, referred to as “intersect”) is the radius r of the square
Is a section from r0 / √2 to r0. In this section, when the square circumscribes the circle, that is, when r = r0, it is sufficient to prove that the section where the circle and one side of the square intersect is the longest. This is proved using mathematical induction.

(1) When r = r0 Since the radius of the circle changes discretely, the radius longer than the radius r0 is r0 + 1. Since the square of r = r0 + 1 does not touch the circle of radius r0, the square of r = r0 is
The section that intersects the circle is longer than the square of r = r0 + 1. That is, assuming that a section where a square having a radius r intersects a circle with one side r0 is G (r), G (r0)> G (r0 + 1).

(2) When r = ri Assuming that G (ri) <G (ri + 1), r = ri−
Consider the case of 1. The relationship between a circle and a square when r = ri is as shown in FIG. Since the circle and the square intersect at eight places shown in FIG. 15, G (ri) = 8 (α
(Ri) -β (ri)).

[0041] ^{α (ri) 2 = r0 2} -ri 2 β (ri) 2 = r0 2 - (ri + 1) 2 ^{Thus, α (ri) 2 -β (} ri) 2 = 2ri + 1 In addition, the r = ri-1 Corresponding α (ri-1), β
When the same examination is performed for (ri-1), (α (ri-1)) ² − (β (ri−1)) ² = 2ri−
It becomes 1. Therefore, taking the squared difference and rearranging it, α (ri) ² −β (ri) ² -｛(α (ri−1)) ² −
(Β (ri−1)) ² ｝ = 2> 0 Therefore, α (ri) ² −β (ri) ² > (α (ri−1)) ² −
(Β (ri−1)) ² .

Α (ri)> 0, β (ri)> 0, α (r
i) -β (ri)> 0, α (ri-1) -β (ri-
1)> 0, α (ri) −β (ri)> (α (ri−1)) − (β
(Ri-1)).

G (ri) = 8 (α (ri) -β (r
i)) G (ri-1) = 8 (α (ri-1) -β (ri-
1)), G (ri)> G (ri-1).

(3) From (1) and (2),... <G (r0-2) <G (r0-1) <G (r0)> G
(R0 + 1) = G (r0 + 2) =..., And the section where the circle and the square intersect when G (r0) is the maximum value, that is, when r = r0, is the longest.

That is, when a square Hough transform is performed on this circle, the most Hough curves intersect in the ab parameter space of r = r0 and take a peak value. Therefore, when a square Hough transform is applied to a circle having a radius r0, a radius r0, that is, a square circumscribing the circle is detected. For example, FIG. 16B shows the result of applying the square Hough transform to a binary image of a particle defect on a semiconductor wafer as shown in FIG. r = 78
At the peak value, and detected as a square. Even when the Hough transform of a circle is applied to the same binary image, since a circle with r = 78 is detected, it can be seen that the square Hough transform correctly detects the circle.

(B) Wavelet Transform Means For the wavelet transform, see CHUI, “An Intr
oduction to WAVELETS ”Academic Press, 192
2, the wavelet transform processing will be briefly described here.

The wavelet transform for the image data is a two-dimensional wavelet transform. Since this is realized by combining the one-dimensional wavelet transform for the x-axis direction and the one-dimensional wavelet transform for the y-axis direction, the one-dimensional wavelet transform is performed. The conversion process will be described. There are many basis functions used for performing the wavelet transform. Here, Harr Wavelet, which has the simplest structure, will be described. For other wavelet basis functions, the output information is almost the same except for the shape of the function. The wavelet transform is composed of two orthogonal functions, a scaling function and a wavelet function. The scaling function is a function for outputting data smoothing information (= low-pass information), and the wavelet function is a function for outputting detailed data information (= high-pass information). In the case of Harr Wavelet, the scaling function is g0 = g1 = 1/2, and the wavelet functions are h0 = 1/2 and h1 = -1 / 2.

When there are input signals s0 to s15, Harr
Wavelet-converted output signals t0 to t15 are as follows.

T0 = g0 · s0 + g1 · s1, t8 = h0 · s0 + h1 · s1 t1 = g0 · s2 + g1 · s3, t9 = h0 · s2 + h1 · s3 t2 = g0 · s4 + g1 · s5, t10 = h0 · s4 · h1 T7 = g0.s14 + g1.s15, t15 = h0.s14 + h1.s15 A specific signal is input to the Harr Wavelet transform. For example, the following input signal s is given.

S (n) = {0,0,0,0,0,0,0,2,2,2,2,2,2,2,2} (3) The signal greatly changes from 0 to 2. A place where such a signal greatly changes is called an edge. An edge having a large signal value as in (3) is called an edge rising, and an edge having a small signal value is called a falling edge. (3) Harr Wavelet
When converted, the following result t is obtained.

N → 0,1,2,3,4,5,6,7; 8,9,10,11,12,13,14,15 t (N) = {0,0,0,1 ,, 2, 2, 2, 2; 0, 0, 0, -1, 0, 0, 0, 0} (4) (low-pass component) (high-pass component) The result of the wavelet transform is called a wavelet coefficient. An edge of the input signal s is detected in the wavelet coefficient t (11) of the high-pass component. Thus, the wavelet transform can detect the edge component of the input signal.

This wavelet transform can be applied to two-dimensional image data such as an SEM image of a wafer pattern. FIGS. 5 and 6 show examples in which a wavelet transform is specifically applied to image data. In this example, the original image of FIG. 5A is 512 × 512 digital data. This image is first subjected to a one-dimensional wavelet transform in the x-axis direction. That is, the wavelet transform for 512 signals in the x-axis (horizontal axis) direction is repeated 512 times in the y-axis (vertical axis) direction. Thereby, the image of FIG. 5B is obtained. FIG. 5B is a diagram in which the original image of FIG. 5A is vertically divided into two parts, and the left side contains low-pass information (L) and the right side contains high-pass information (H).

The same wavelet transform is applied to FIG. 5B in the y-axis direction. As a result, the image shown in FIG. 6 is obtained. FIG. 6 is an image in which FIG. 5B is vertically divided into two parts, and low-pass information (L) is placed on the upper side and high-pass information (H) is placed on the lower side. Accordingly, FIG. 6 divides the original image of FIG. 5A into four parts, and combines low-pass information (LL component) in both the x and y directions in the upper left, and information combining high-pass information in the x-axis and low-pass information in the y-axis in the upper right. (HL component), information (LH component) combining low-pass information in the x-axis direction and high-pass information in the y-axis direction is stored in the lower left, and high-pass information (HH component) in both the x and y axes directions is stored in the lower right. Will be.
That is, the upper right portion is a line component (vertical line component) in the Y-axis direction existing in the original image, and the lower left portion is an X component existing in the original image.
An axial line component (horizontal line component) is shown, and a lower right portion shows an oblique line component. Also, here, the conversion processing is first performed in the x-axis direction, and then the conversion processing is performed in the y-axis direction to obtain FIG. 6. However, even if the x and y conversion orders are exchanged, the final two-dimensional wavelet After the conversion, the same image of FIG. 6 is obtained.

When the two-dimensional wavelet transform having the above properties is applied to a semiconductor wafer image having a granular defect, the defect and the background wiring pattern can be separated. FIG. 7 simulates an image of a semiconductor wafer. Horizontal lines running horizontally correspond to wiring patterns, and black circles at the center correspond to defects. As described above, in the semiconductor wafer, the wiring patterns are arranged in the horizontal direction or the vertical direction. When a two-dimensional wavelet transform is applied to such an image, the result of FIG. 8 is obtained. Focusing on the vertical line detection component and the horizontal line detection component, the wiring pattern appears only in the LH component that detects the lower left horizontal line.
On the other hand, defect components having various directional components are detected as an HL component at the upper right and an LH component at the lower left. As described above, by using the two-dimensional wavelet transform, it is possible to separate a wiring pattern having only vertical or horizontal components from a defect having components in various directions.

By combining the two-dimensional wavelet transform described above and the Hough transform, it is possible to detect the position and size of a defect existing in a semiconductor wafer image.

Hereinafter, embodiments of the present invention will be described with reference to the drawings. First, the configuration of the image information processing apparatus according to the embodiment of the present invention will be described with reference to FIG. The computer 101 includes a CPU 31 that performs input / output and calculations, and a cache 3 that holds frequently used instructions and data to speed up processing.
2. Floating point arithmetic unit 33, RAM 34 for storing user input and data, ROM for storing system programs
36, a display device 35 for displaying a user's selection menu and calculation results, a keyboard and a pointing device (mouse or the like) input device 37 for inputting parameters and instructions,
A magnetic disk 38 for storing calculation results and the like is provided. These devices are connected by a CPU bus 30.

DUT such as semiconductor wafer to be inspected
The (device under test) 41 is loaded / unloaded to / from the stage 45 by the loader / unloader 46.
The loaded DUT is image-input by an image input device 42 such as an electron microscope, is converted into digital data by an A / D converter 43, and is stored in a frame memory 44.
Alternatively, it is transferred to the RAM 34 on the computer. This digital image is, for example, a grayscale image with a resolution of 512 × 512 pixels and 256 gradations. The frame memory 44 is connected to the CPU bus 30 via the I / O bus 40. Further, the control device 47 includes a stage 4
5. The operation and the like are controlled by the loader / unloader 46. The control device 47 is connected to the computer 101 via the I / O bus 40.

The wavelet transform means 51, the Hough transform means 53, the binarization processing means 54, the labeling processing means 55, the noise removal means 56, the isolated point removal means 57, the center-of-gravity calculation means 58, and the overlap detection removal means 59 The frame memories 61, 63, 64, 65, 66, 67, 68, 69 are connected to the CPU bus 30. The energy calculating means 52 is connected to the frame memory 61 and the CPU bus 30. The wavelet transform means 51, the energy calculation means 52, the square Hough transform means 53, the binarization processing means 54, the labeling processing means 55, and the noise removal means 5
6. Operations of the isolated point removing means 57, the center-of-gravity calculating means 58, and the duplication detection removing means 59 will be described in the description of the operation of the embodiment.

Next, the operation of the first embodiment will be described with reference to FIG.
A will be described with reference to FIG. Frame memory 44 or R
First, the image data stored in the AM 34 is input to the frame memory 61 as input image data, and two-dimensional wavelet transform is performed by applying the wavelet transform means 51 (S1). A vertical line detection component and a horizontal line detection component are obtained by two-dimensional wavelet transform. These components have a wavelet coefficient value having a large absolute value in the edge portion, and the wavelet coefficients in the other portions have a value of 0 or a value close to 0.

Next, the absolute value of the wavelet coefficient is transferred to the frame memory 64, and binarization processing is performed by the binarization processing means 54 (S2). That is, the binarization processing means 54
Generates a binary image in which an appropriate threshold value is given to the vertical line detection component and the horizontal line detection component and binarization processing is performed, and the coefficient value of the edge portion is 1, and the other values are 0. Here, the threshold value was empirically set to ± 5.

Finally, the binary image is transferred to the frame memory 63, and the square Hough transform is performed by the square Hough transform means 53 (S3). The parameter space has three dimensions of (a, b, r). By adding a process after the first embodiment, it is possible to automate the specification of the position and the size of the defect.

According to the first embodiment, the following effects can be obtained. Although the absolute value of the wavelet coefficient at the edge portion is large, the sign may be positive or negative. Since the negative coefficient is 0 in the normal binarization processing, for example, FIG.
Performing normal threshold processing on the vertical line detection component at the upper right of
As shown in FIG. 9, only half of the edge is detected. But,
Hough transform can be detected even if the target graphic is missing. In contrast to FIG. 9, when the Hough transform is used, the luminance of a parameter corresponding to a defect increases in a parameter space in a parameter space in a square component existing in an original image, that is, in an image of a wafer pattern. By observing the parameter space, the position and size of the defect can be specified.

The operation of the second embodiment will be described with reference to FIG. 1B. The data in the parameter space after the Hough transform is sent to the frame memory 64, and the binarization processing means 54
Performs a binarization process (S4). The threshold value for binarization is changed by r. The center (a
0, b0) and a square with a radius r0 exist in a perfect form, the luminance of the parameter (a0, b0, r0) becomes 8r
Since it becomes 0, for example, if the threshold is set to 4r0,
Even if half of the square of radius r0 is missing, the luminance of the parameter exceeds the threshold. When the threshold processing is performed with this threshold value, one group of luminance 1, that is, one white pixel appears.

The operation of the third embodiment will be described with reference to FIG. 1C. The result of the binarization processing (S4) is
The data is transferred to the frame memory 65. Then, the labeling processing unit 65 performs a labeling process for grouping pixels having a luminance value of 1 adjacent to the result of the binarization process (S4) (S5). In the parameter space after the Hough transform, the brightness value of the parameter in which the detected Taisho graphic exists is the maximum value, but the brightness value of the parameter adjacent thereto is also high. Therefore, as a result of the binarization processing, the luminance value 1 near the parameter corresponding to the figure (for example, a square) to be detected.
Parameters exist intensively. By the labeling process (S5), it is regarded as a feature parameter for one figure in a portion where the parameter of the luminance value 1 is concentrated.

Then, the obtained a-coordinate and b-coordinate of the parameter belonging to each label are sent to the frame memory 68, and
The center-of-gravity calculating means 58 calculates the average value (a ', b') of the a-coordinate and b-coordinate of the parameter as the center of gravity (S6). Thus, the position and size of the detection target graphic can be automatically specified from the input image.

The operation of the fourth embodiment will be described with reference to FIG. 2A. Two-dimensional wavelet transform (S1)
Before input, the input digital image is sent to a frame memory 66, and noise removing means 56 such as a median filter is used.
Removes noise.

If spike-like noise is present in the input image, it is detected as an edge by wavelet transform. If a large number of noises other than the detection target graphic are subjected to the Hough transform, the noise other than the detection target is erroneously detected. By applying a noise removal filter to the original image, erroneous detection can be prevented.
Further, the Hough transform requires a longer processing time in proportion to the number of pixels having a pixel value of 1 in the binary image. The noise removal unit 56 reduces the number of wavelet coefficients of the pixel value 1 generated by the wavelet transformation unit + binarization processing unit, thereby improving the processing speed.

The operation of the fifth embodiment will be described with reference to FIG. 2B. For the purpose of preventing erroneous detection and improving the processing speed, the image after the wavelet transform means + binary processing means is sent to the frame memory 67 and the isolated point removing means 5
7 to remove an isolated point (S2A). This means that in the binary image, all the adjacent pixels having a pixel value of 1 have a pixel value of 0
Is regarded as an isolated point, and the pixel value of the pixel is set to 0
The image processing means is replaced with.

In the binary image after the wavelet transform, the wavelet coefficient having the coefficient value 1 corresponding to the figure to be detected exists adjacently. Therefore, if the coefficient value is 1 and the surroundings are all 0, the coefficient is regarded as noise and the coefficient value is set to 0. This prevents erroneous detection and improves the processing speed.

The operation of the sixth embodiment will be described with reference to FIG. 2C. The binarization process (S2) in the fifth embodiment is replaced with an absolute value binarization process (S2). This is to perform a binarization process on the absolute value of the wavelet coefficient instead of the normal binarization process, considering that the coefficient value of the edge portion may be negative due to the wavelet transform. .

When this absolute value binarization process is applied to the vertical line detection component at the upper right of FIG. 8, the entire edge can be detected as shown in FIG. By applying the Hough transform to this, the luminance of the parameter corresponding to the defect becomes larger than when ordinary binarization processing is used, and detection becomes easier.

The operation of the seventh embodiment will be described with reference to FIG. 3A. This is different from the fifth embodiment in the square Hough transform (S3). That is, the square Hough transform means 53 performs the Hough transform on both the binary image corresponding to the vertical line detection component and the horizontal line detection component of the wavelet transform result sent to the frame memory 63, but converts the two Hough transforms into the same parameter space. Convert over the top. Since the graphic to be detected appears in both the vertical line component and the horizontal line component, it is possible to reliably detect the graphic by this method.

FIG. 3A also shows the operation of the eighth embodiment.
This will be described with reference to FIG. This is different from the fifth embodiment in the square Hough transform (S3). That is, the square Hough transform unit 53 performs a Hough transform on one of the binary images corresponding to the vertical line detection component and the horizontal line detection component of the wavelet transform result sent to the frame memory 63.

As shown in FIGS. 7 and 8, the wiring patterns of the semiconductor wafer image are aligned in the horizontal or vertical direction, and appear only in either the vertical line component or the horizontal line component.
By performing the Hough transform only on the component in which the wiring pattern is not appeared, only the object to be detected can be subjected to the Hough transform and the detection can be reliably performed. There is no erroneous detection of the wiring pattern as a detection target.

FIG. 3B shows the operation of the ninth embodiment.
This will be described with reference to FIG. This is obtained by adding means for automatically determining which of the vertical line component and the horizontal line component is to be Hough-transformed to the eighth embodiment. That is, the energy calculating means 52 calculates the energy based on the wavelet coefficients sent to the frame memory 61 (S1A), and sets the direction in which the energy is small as the direction in which no wiring pattern is detected, and performs the Hough transform only in that direction. .

In FIG. 8, only the wiring pattern + defect appears in the horizontal line detection component at the lower right of the screen, and only the defect appears in the vertical line detection component at the upper right of the screen. In the part where the edge is detected, the absolute value of the wavelet coefficient becomes large,
Has a coefficient value close to. Therefore, the degree of detection of the edge component of the vertical line component and the horizontal line component is quantified and compared, and the one with a lower degree of detection is regarded as a component in which no wiring pattern is detected, and the Huff transform is performed.
In the case of, only the vertical line component at the upper right of the screen is Hough-transformed, and the detection target graphic can be detected. As a method of quantifying the degree of edge detection, the energy of wavelet coefficients is defined. Each wavelet coefficient value is W (s, t), the number of two-dimensional wavelet coefficients is M in the x-axis direction, and the y-axis The direction is defined by equation (5) as N.

[0077] E = (1 / MN) Σ s = 1 M Σ t = 1 N W (s, t) (5) The operation of the tenth embodiment will be described with reference to Figure 3A. This is different from the fifth embodiment in the binarization process (S4). That is, the threshold value of the binarization process (S4) is variable depending on the value of r. For example, in the original image, the center (a, b), the square C1 with r = 10, and the center (c,
d) Assume that a square C2 with r = 100 exists. In the parameter space, the parameters (a,
The luminance value of (b, 10) is 4 × 10 = 40. The luminance value of the parameter (c, d, 100) corresponding to C2 is approximately 4 × 100 = 400. As described above, the luminance value of the parameter greatly changes depending on r. By changing the threshold value for binarization in proportion to r, it is possible to detect a square of an arbitrary size.

The operation of the eleventh embodiment is described with reference to FIG.
This will be described with reference to C. This is one in which the calculation of the center of gravity (S6) in the ninth embodiment is followed by the duplicate detection and removal (S7). This is to check whether a plurality of squares are detected on the xy plane when a plurality of squares are detected by the center of gravity calculation (S6).

For example, there is a case where a noise component is added to a part of an arc of a square existing on an xy plane, and a square slightly smaller than the arc is detected. It is determined that it is an erroneous detection to detect a plurality of defects at similar positions, and only the larger square is detected, and even if a small square overlapping the larger square is detected, it is excluded. Specifically, when multiple squares are detected, check if the center of the smaller square is not in the larger square,
This is realized by adding a process of removing the object if it is within the square.

In the embodiment of the present invention, a dedicated frame memory is provided for each processing means.
By sharing the frame memory 61 as shown in (1), it is also possible to save the memory and increase the efficiency.

As shown in FIG. 13, it is also possible to replace the devices 41 to 47 with a general-purpose electron microscope system 102 and to replace the units 51 to 59 and their attached frame memories with a computer 101 such as a workstation.

[0082]

According to the present invention, claims 1 to 3 and claims 6 to 11
Corresponds to the second object of the present invention. As a result, it is possible to detect a defect without a golden device or CAD data required in the conventional image comparison method.

Claims 4 to 5 and 7 to 11
Corresponds to the first object of the present invention. In the case of an image such as the SEM image described in the embodiment in which the pixel value is largely changed over time, a large amount of noise is detected by the conventional image comparison method. On the other hand, according to the present invention, it is possible to detect a defect with almost no detection of noise.

Claims 1, 4, 5, 8 and 9 correspond to the third object of the present invention. In the case of the conventional FET method, in order to remove a periodic wiring pattern from an image of 512 × 512 pixels, addition and subtraction of 1.31 × (10 8) times and 8.39 × (10 8) times However, in the case where the square Hough transform of the present invention is replaced by the circle Hough transform, addition and subtraction of about 5.5 × (10 to the fifth power) times and 1.1 × (10
6) times of multiplication. Other image processing also requires a calculation amount of 10 7 or less except for Hough transform. F
Even in the ET method, processing such as Hough transform is required to automatically detect the position and size of an image, so that the present invention is not disadvantageous.

The circle Hough transform includes a square process and a square root operation which are not necessary for the square Hough transform. In the present invention, since the Hough transform is limited to the square Hough transform, the calculation can be performed at a higher speed than the circular Hough transform. For example, Ultr
aSparc S-7 / 300U Workstation (170MHz)
In the C compiler, the circle Hough transform takes 4650 msec, whereas the square Hough transform takes only 3441 msec. Therefore, the operation can be performed at a higher speed than the FET method.

The amount of calculation of the Hough transform is proportional to the number of pixels of the binary image to be processed. Therefore, the smaller the number of pixels subjected to the Hough transform, the smaller the amount of calculation. Therefore, by performing the noise elimination and the isolated point elimination processing described in claims 4 and 5, the calculation speed can be further improved.

As described in detail above, the present invention provides an apparatus that combines image processing centered on Hough transform and wavelet transform on an image having complicated background information such as a semiconductor wafer pattern. This makes it possible to automatically detect the position and size of a defect, and the effect is great both in terms of defect detection performance and processing speed.

[Brief description of the drawings]

FIG. 1 is a flowchart showing operations of first, second and third embodiments, respectively.

FIGS. 2A, 2B, and 2C are flowcharts illustrating operations of fourth, fifth, and sixth embodiments, respectively.

FIG. 3A shows the operation of the seventh, eighth and tenth embodiments, and FIG.
19 is a flowchart showing the operation of the ninth embodiment, and C is a flowchart showing the operation of the eleventh embodiment.

FIG. 4 is an explanatory diagram of a square Hough transform, where A is (x,
y) Square in image space, B is Hough curve in (a, b, r) parameter space.

5A is an original image of a sample “house”, and FIG. 5B is an image obtained by performing a wavelet transform on the image of A in the x-axis direction.

FIG. 6 is an image obtained by performing a wavelet transform in the y-axis direction on the image of FIG. 5B.

FIG. 7 is a sample image imitating a semiconductor wafer image.

8 is an image obtained by performing a two-dimensional wavelet transform on the image of FIG. 7;

9 is an image obtained by applying a binarization process to the vertical line detection component (upper right portion) in FIG.

10 is an image obtained by applying an absolute value binarization process to the vertical line detection component (upper right portion) in FIG.

FIG. 11 is a block diagram showing a configuration of an image information processing apparatus according to an embodiment of the present invention.

FIG. 12 is a block diagram showing another configuration of the image information processing apparatus according to the embodiment of the present invention.

FIG. 13 is a block diagram showing still another configuration of the image information processing apparatus according to the embodiment of the present invention.

FIG. 14 is a diagram showing a relationship between a circle and a square when the radius is r0.

FIG. 15 is a diagram showing a relationship between a circle and a square when the radius r = ri.

FIG. 16 is a diagram showing a binary image of a particle defect on a semiconductor wafer and its square Hough transform result.

[Explanation of symbols]

 Reference Signs List 51 wavelet transform means 52 energy calculating means 53 Hough transform means 54 binarization processing means 55 labeling processing means 56 noise removing means 57 isolated point removing means 58 centroid calculating means 59 duplication detection removing means

Claims (13)

[Claims] An image information processing apparatus for inspecting, recognizing, and judging an object using an image, comprising: a two-dimensional wavelet transform unit for performing two-dimensional wavelet transform on input digital image data; Binarization processing means for applying a threshold value process to the vertical line detection component and the horizontal line detection component obtained as described above to create a binary image of the vertical line detection component and the horizontal line detection component; Square Huff transform means for applying a square Hough transform for detecting a square to the binary image obtained by the value processing means to determine the position and size of the object; Image information processing device provided. 2. The image information processing apparatus according to claim 1, wherein a threshold value process is applied to the parameter space obtained by said square Hough transform means to detect specific graphic information. An image information processing apparatus provided with means. 3. An image information processing apparatus according to claim 2, wherein adjacent active pixels are grouped with the same label in a binary image of a parameter space obtained by binarizing a square Hough transform result. An image information processing apparatus, comprising: a labeling unit that performs a labeling process; 4. The image information processing apparatus according to claim 1, further comprising noise removing means for locally removing noise from an input digital image. 5. The image information processing apparatus according to claim 4, wherein the binary image of the vertical line detection component and the horizontal line detection component after applying a binarization processing unit to a two-dimensional wavelet transform result. And an isolated information removing device for removing an isolated active pixel. 6. The image information processing apparatus according to claim 1, wherein the square Hough conversion means detects the xy coordinates of the center of the square and the length of one side. An image information processing device that performs conversion. 7. The image information processing apparatus according to claim 1, wherein the square Hough transform means corresponds to the vertical line detection component and the horizontal line detection component of the two-dimensional wavelet transform. Houg (Houg) in the same parameter space
h) An image information processing device as a means for converting. 8. The image information processing apparatus according to claim 1, wherein the square Hough transform means corresponds to the vertical line detection component and the horizontal line detection component of the two-dimensional wavelet transform. Houg only one of the two binary images
h) An image information processing device as a means for converting. 9. An image information processing apparatus according to claim 8, further comprising: energy calculating means for calculating energies of said vertical line detection component and said horizontal line detection component of two-dimensional wavelet transform, wherein a square Hough conversion means is provided. Is an image information processing apparatus that performs a square Hough transform on the binary image corresponding to the lower energy component. 10. The image information processing apparatus according to claim 6, wherein the threshold value of the binarization processing means for the parameter space by the square Hough transform is set to the magnitude of the length parameter of one side of the square. An image information processing device that changes correspondingly. 11. The image information processing apparatus according to claim 3, further comprising: a duplication detection / removal unit that removes duplication when a plurality of objects overlap in position. Processing equipment. 12. An image information processing method for inspecting, recognizing, and judging an object using an image, comprising: a two-dimensional wavelet transform step of performing two-dimensional wavelet transform on input digital image data; Obtained by the process,
A threshold value process for a vertical line detection component and a horizontal line detection component to generate a binary image of the vertical line detection component and the horizontal line detection component; and a binarization process, A square Hough transform step of applying a square Hough transform for detecting a square to the obtained binary image to obtain the position and size of the object. Processing method. 13. A computer-readable recording medium storing a program for causing a computer to execute image information processing for inspecting, recognizing, and judging an object using an image, wherein the computer-readable recording medium stores A two-dimensional wavelet transform process for two-dimensional wavelet transform, and obtained by the two-dimensional wavelet transform process,
A vertical line detection component, a binarization process of applying a threshold value process to a horizontal line detection component to generate a binary image of the vertical line detection component and the horizontal line detection component, and a binarization process obtained by the binarization process. A square Hough (Houg) for calculating the position and size of the object by applying a square Hough transform for detecting a square to the binary image
h) A computer-readable recording medium recording a program for causing a computer to execute the conversion process.