JP2926118B2 - Defect detection method - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a defect detection method for comparing patterns of, for example, LSI wafers and TFTs and recognizing pattern defects and the like.

[Conventional technology]

An example in which pattern recognition is applied to pattern defect inspection will be described.

As described in Japanese Patent Application Laid-Open No. 57-196377, a conventional method includes a means for detecting a pattern, a means for storing a detected pattern, and a pattern detected and stored immediately before. Means for aligning the detected pattern with each pixel, and detecting and detecting an error between the two aligned patterns.
Means for comparison are provided so that the pattern defect can be recognized by comparison. Here, the recognition target is a pattern such as a memory LSI as shown in FIG. 2, a TFT (Thin Film Transistor) pattern, a printed wiring board pattern, a ceramic substrate pattern, and a process of manufacturing them. , Such as a mask and a reticle. Here, a pattern on a semiconductor wafer will be described as an example, but the same holds for other patterns. The pattern of the semiconductor wafer is
Dozens of chips that are finally separated and become individual products are mounted on a single wafer, and these chips have the same pattern.

The principle of recognizing such a pattern defect will be described with reference to FIG. In other words, paying attention to the fact that each chip has exactly the same pattern, detect and store a certain pattern, then detect another pattern that should be the same, and align it in pixel units. (Calculate the sum of the difference of each pixel between the detected pattern and the stored pattern over the entire image. This is performed while shifting the stored pattern with respect to the detected pattern on a pixel-by-pixel basis. Is adjusted to a position where the sum of the two patterns is small, that is, a position where the two images are the same.), And an error between the two aligned patterns is extracted and compared.
When there is no defect in any of the patterns, there is almost no difference between the patterns, but when there is a defect in any of the patterns, there is a difference in the pattern at the defective portion. A pattern defect can be recognized by detecting a place where an error occurs.

At this time, if there is a difference in comparison, it can be said that any of the patterns has a defect, but it cannot be determined which of the patterns has a defect. There are various means for judging this, but they are not specifically described here because they have no direct relation to the technology according to the present invention.

[Problems to be solved by the invention]

The normal part pattern of the wafer to be inspected differs depending on the location due to various error factors. For this reason, even if there is a difference in the pattern, it is not necessarily a defect, and it becomes difficult to distinguish a small defect from an error in a normal part.
This will be described below.

First, the difference between the defect and the error of the normal part is defined. It is assumed that the error of the normal part has the same error in the same pattern even in the vicinity of the point of interest. On the other hand, it is assumed that the defect is a local difference, and the same pattern in the vicinity has no error. The allowable dimensional error of the normal part is
It should be larger than the minimum defect size to be recognized. Of course, there are exceptions, but they are out of scope.

In the case of the above definition, if there is a difference between the patterns, the method of determining a defect has a problem as shown in FIG. That is, FIG. 3 shows a binary image for simplicity of description.
FIG. 3 (a) shows a comparison between a normal part having an allowable dimensional error between a storage pattern and a detection pattern, and FIG. 3 (b) shows a defect having a minimum size to be recognized between a normal pattern and a detection pattern having a defect. (C) and (d) show the differences between the patterns of FIGS. (A) and (b), respectively, when comparing the patterns with. Comparing FIGS. 3 (c) and 3 (d), the different areas of the normal part patterns are larger than the different areas of the defective part patterns, making it difficult to identify defects.

In the above example, there is a need for a method that facilitates the discrimination between a normal part pattern having an error and a defect. Also, when recognizing pattern characteristics such as pattern shift and deformation amount, the problem can be solved by the same method as described in the section of the embodiment.

An object of the present invention is to provide a defect detection method that can solve these problems. Note that only the defect inspection will be described in the section of the means for solving the problems described below and the section of operation, but it goes without saying that the same applies to other examples.

[Means for solving the problem]

In order to achieve the above object, a first pattern of a plurality of patterns originally formed to be the same is captured to obtain a first image, and a second pattern of the plurality of patterns is captured. To obtain a second image, and the first image and the second image are relatively displaced from each other, and
A plurality of difference images between the first image and the second image are obtained, and a defect is detected based on the plurality of difference images having different obtained shift amounts.

Further, the relative shift amount between the first image and the second image is determined by the shift amount at which the absolute value of the difference between the first image and the second image is minimized. And the difference between the second image and the second image.

Further, when each of the difference images having different shift amounts is equal to or more than a predetermined value, a defect candidate is obtained from the difference image,
The feature amount of the obtained defect candidate is extracted to make a defect determination.

FIG. 1 shows a principle diagram of the defect detection method of the present invention. That is, the detecting means 2 for detecting the pattern of the wafer 1 to be inspected, the storing means 3 for storing the detected pattern, and the pattern previously detected and stored (first
) And a detected pattern (second image), and a shift amount that minimizes the absolute value of the difference between the aligned first image and second image. And alignment candidate selection means 5 for selecting, as the defect candidates (alignment candidates), all of the difference images having different shift amounts, which are equal to or more than a predetermined value (threshold), from all the selected candidates. An error image extracting unit 6 for calculating an error image by using a candidate value and a value near the pixel for each pixel, and a defect recognizing unit 7 for recognizing a pattern defect from the error image.

Further, the error image extracting means 6 has a method of taking the minimum value of candidate values, for example.

[Action]

The operation of the configuration of FIG. 1 will be described with reference to FIGS. 3 (a) and 3 (b).
An example in which the present invention is applied to will be described.

First, in the case of FIG. 3 (a), a plurality of alignment shift candidates are searched for within an image shift range of ± 2 pixels near the best alignment position. Table 1 shows an example of the sum of the differences in the entire image at each shift position.

Here, the difference between (0,0) and (1,0) has a small sum and becomes a candidate for the alignment shift, and the difference images of the two candidates are as shown in FIGS. 4 (a) and 4 (b), respectively. And the minimum error of the two candidates is 0 for all pixels. On the contrary,
In the case of FIG. 3 (b), an example of the sum of the differences is shown in Table 2, but the sum of the differences other than (0,0) is large, and only (0,0) is a candidate for alignment. The error image is the same as FIG. 3 (d). From these, the defect can be recognized as a place where the area of the difference is large.

As shown in the above example, in the case of a normal part having the same error near the point of interest, the position where the error is eliminated is a candidate for the alignment shift, and there is no error near the point of interest. Is not a candidate for the alignment shift,
Only a defect appears in the error image.

As a method of determining the alignment shift candidate, an image having a high degree of coincidence is selected, for example, by using a position having a sum of differences smaller than twice the sum of differences at the best alignment position. Further, the positioning means 4 can be omitted, and the positioning candidate selection means 5 for selecting a plurality of positioning shift candidates can simultaneously search for the best positioning position.

〔Example〕

Hereinafter, a first embodiment of the present invention will be described with reference to FIG. The present embodiment is an example in which the defect detection method of the present invention is applied to a defect inspection of an LSI wafer pattern, but it is needless to say that the present invention can be applied to a pattern such as a TFT. FIG. 5 is a configuration diagram of an LSI wafer defect detection apparatus (hereinafter, abbreviated as an apparatus) used in the embodiment. The apparatus comprises an XY stage 8 for scanning the wafer 1, an illumination light source 9 and an illumination optical system 10 for illuminating the wafer, an objective lens 11 for detecting an optical image of the illuminated wafer, and a detection unit comprising a one-dimensional image sensor 12. 2A , and an image input unit 15 including an A / D converter 13 and an image memory unit 14 for digitizing and storing a signal of the one-dimensional image sensor 12, and a detected image 16 input to the image input unit 15. Comparative image 17 from image memory unit 14
Thus, a matching unit 18 that calculates a matching value represented by Expression (1) described later, an operation unit 19 that obtains a best alignment position and a plurality of alignment candidate shifts from the matching value from the matching unit 18, An error image extraction unit 21 that extracts an error image 20 from the detected image 16 and the comparison image 17 using the alignment candidate shift amount, binarizes the error image 20, and extracts various feature amounts of a place where a difference exists, the image processing unit 23 composed of defect determination unit 22 for determining the defect, and was composed of X-Y stage 8 computer to perform the management of the storage and display and the entire sequence of the defect information output from the control image processing part 23 of the It comprises an overall control unit 24.

Each part of the apparatus operates as follows to detect a pattern defect. That is, after initializing each unit according to a command from the overall control unit 24, the pattern of the wafer 1 illuminated by the illumination light source 9 is one-dimensionally passed through the objective lens 11 in synchronization with the scanning of the XY stage 8. A two-dimensional pattern is detected by photoelectric conversion by the image sensor 12, and A / D
As a two-dimensional detection image 16 digitized by the D converter 13,
The obtained detected image is stored in the image memory unit 14. From the detected image 16 and the comparison image 17 stored in the image memory unit 14, the matching unit 18 obtains a matching value. This operation will be described below with reference to FIG.

FIG. 6 is a diagram showing the operation principle of the matching unit 18. From the detected image 16 and the comparative image 17, the comparative image 17 is represented by ΔX,
± δ pixels of the permissible displacement in the ΔY direction (in this embodiment, δ pixels
= 1, but this is a value determined by the dimensional accuracy of the detection target and the positioning accuracy of the defect detection device, and a necessary value is appropriately set. ), The image difference between the detected image 16 and the comparison image 17 is calculated by equation (1), and the image difference corresponding to each shift amount is output as the matching value S1.

S1 (Δi, Δj) = ΣΣ | f (i, j) −g (i + Δi, j + Δj) | (1) where f (i, j) is a value at the pixel (i, j) of the detected image 16 , G (i, j) are values at the pixel (i, j) of the comparative image 17, and S1 (Δi, Δj) is the difference between the images at the pixel shift amount (Δi, Δj). ΣΣ is the addition over the entire image for calculating the displacement, and Δi, Δj are −1 to 1 in the case of FIG.

The arithmetic unit 19, S1 (Δi, Δj) obtaining the minimum value S _min of taking the minimum value (.DELTA.i, .DELTA.j) best registration position is a set of (Δi _m, Δj _m). Next, a plurality of alignment candidate shifts (Δi _s , Δj _s ), s = 1, 2,... Are obtained as a set of (Δi, Δj) satisfying the expression (2).

S1 (Δi, Δj) <in S _min × Th ...... (2) where, Th is a threshold value set in advance, 2
Take the value of the degree.

The error image extracting unit 21 obtains the error image 20 from Expression (3) from the plurality of alignment candidate shifts (Δi _s , Δj _s ) from the arithmetic unit 19.

h (i, j) = M _in (s = 1, 2,...) ｛| fi, j) −g (i + Δi _s , j + Δj _s ) |｝ = 0 (3) 20 is the threshold value for defect judgment
_Binarization is performed using _Vth , and various feature amounts such as the area, width, and projection length of the location where the difference exists are extracted to perform defect determination.

According to the present embodiment, it is possible to easily distinguish between a normal pattern having an error and a defect, so that there is an effect that a defect smaller than the normal part allowable error dimension can be recognized.

The following are modifications of the present embodiment. That is, first, instead of S _min × Th in equation (2), S _min × S _min ×
Some functions are S _min functions such as Th. According to this modification, the degree of freedom can be increased, and in this example, the threshold is set to be low when the S _min is small and the degree of coincidence is high, and to be high when the converse is high, and a threshold is set according to the image quality. Can be.

Secondly, there is a method in which a storage pattern and a detection pattern are first aligned, and then a plurality of alignment candidates are obtained in a range narrower than the alignment range. In this case, since the range for obtaining the alignment candidate shift is narrow, the size of the circuit in this part is small. In addition, alignment is generally performed, and there is an effect that the method of the present invention can be introduced with a minimum addition to such an apparatus.

Third, in Expression (3), Expression (4) is used instead of obtaining the minimum value. That is, if ｛f (i,
j) −g (i + Δi _s , j + Δj _s )} if there is a different sign h (i, j) = M _in (s = 1,2,...) ｛| (f (i, j) −g (I + Δi _s , j + Δj _s ) |｝ (4) In this case, the fact that there is a different sign means that there is a value of 0 somewhere in the unit of a pixel or less, which means that this is correct. There is an effect that can be evaluated.

Fourth, instead of detecting the two-dimensional pattern by photoelectrically converting the one-dimensional image sensor 12 in synchronization with the scanning of the XY stage 8, the XY stage 8 is step-moved and the TV camera performs photoelectric conversion. In some cases, two-dimensional patterns are detected by conversion. Alternatively, instead of the one-dimensional image sensor 12, another type of sensor such as a point type sensor such as a photomultiplier and a scanning mechanism may be used.

Fifth, instead of calculating the matching value between the detected image 16 and the comparison image 17 by the equation (1), the detection image 16 and the comparison image 17 are respectively filtered to extract edges, and the matching is performed on the edge images. In some cases, the value is calculated by equation (1). Alternatively, the detection image 16 and the comparison image 17 are respectively filtered to convert the edge into a binary value, and a matching value is calculated for the edge binary image by Expression (1). According to this modified example, since the edge is used, there is an advantage that it is hardly affected by a difference in brightness between the pattern of the detected image and the pattern of the comparative image.

Next, a second embodiment of the present invention will be described with reference to FIG. The present embodiment is an example in which the defect detection method of the present invention is applied to the measurement of the amount of deviation of the pattern of an LSI wafer, but it is needless to say that the method can be applied to a pattern such as a TFT. Seventh
FIG. 1 is a configuration diagram of an LSI wafer defect detection apparatus (hereinafter, abbreviated as an apparatus) used in the embodiment. The apparatus comprises an XY stage 8 for scanning the wafer 1, an illumination light source 9 and an illumination optical system 10 for illuminating the wafer, a detection unit 2 comprising an objective lens 11 for detecting an optical image of the illuminated wafer, and a TV camera 25.
B , and an image input unit 15 including an A / D converter 13 and an image memory unit 14 for digitizing and storing signals of the TV camera 25, and a detected image 16 and an image memory input to the image input unit 15. A matching unit 18 for calculating a matching value represented by the formula (1) from the comparison image 17 from the unit 14, an operation unit 19 for obtaining a best alignment position and a plurality of alignment candidate shifts from the matching value from the matching unit 18, An image processing unit 23A comprising a shift amount extracting unit 26 for obtaining a shift amount,
And an overall control unit 24 composed of a computer for controlling the XY stage 8, storing and displaying the shift amount information output from the image processing unit 23A, and managing the overall sequence.

Each unit of the apparatus operates as follows to measure the amount of pattern displacement. That is, after performing initialization of each unit according to a command from the overall control unit 24, the XY stage 8 is positioned, and the wafer 1 illuminated by the illumination light source 9 is positioned.
The two-dimensional pattern is detected by photoelectrically converting the pattern with the TV camera 25 via the objective lens 11, and the two-dimensional detection image 16 digitized by the A / D converter 13 is obtained. It is stored in the image memory unit 14. From the detected image 16 and the comparison image 17 stored in the image memory unit 14, the matching unit 18 obtains a matching value. The operation for obtaining the matching value is the same as that described in the first embodiment, and is performed using the above-described equation (1).

In the arithmetic unit 19, as in the first embodiment, the expression (1)
Take the minimum value S _{min of} S1 (Δi, Δj) and its minimum value (Δ
i, Δj), the best alignment position (Δi _m , Δj _m )
Ask for. Next, the above expression (2) is satisfied (Δi, Δ
j), a plurality of alignment candidate shifts (Δi _s , Δj
_s ), s = 1, 2,.

The shift amount extraction unit 26 outputs the S1 (Δi, Δj) in ascending order.

According to this embodiment, there is an effect that the properties of the patterns such as the displacement and the deformation amount of a plurality of patterns having different overlapping displacement amounts can be recognized without using information such as the segments.

〔The invention's effect〕

According to the present invention, it is possible to easily discriminate a normal part pattern having an error from a defect, and to recognize a pattern property such as a pattern shift and a deformation amount.

[Brief description of the drawings]

FIG. 1 is a principle diagram of a defect detection method of the present invention, FIG. 2 is a principle diagram of a pattern defect detection method by a general pattern comparison method, (a) is a storage pattern, and (b) is a detection pattern having a defect. , (C) show patterns obtained by taking the difference between (a) and (b). FIG. 3 shows an example of a pattern, in which (a) shows a normal part having an error in allowable dimensions, (b) shows a defective part,
(C) shows the difference image of (a), and (d) shows the difference image of (b). FIG. 4 shows the difference image at the shift position of the alignment candidate,
FIG. 5 is a configuration diagram of an apparatus used in the first embodiment of the present invention,
FIG. 6 is a diagram illustrating a difference image at a shift position for explaining a matching unit, and FIG. 7 is a configuration diagram of an apparatus used in a second embodiment of the present invention. DESCRIPTION OF SYMBOLS 1... Wafer 2... Detecting means 2 A and 2 B ... Detecting section 3... Storage means 4... Positioning means 5. ... Defect recognition means 8 ... XY stage, 9 ... Illumination light source 10 ... Illumination optical system, 11 ... Objective lens 12 ... One-dimensional image sensor 13 ... A / D converter, 14 ... Image memory Unit 15 Image input unit 16, Detection image 17 Comparison image 18, Matching unit 19 Computing unit 20, Error image 21 Error image extraction unit 22, Defect determination unit 23 , 23A … Image processing unit, 24… Overall control unit 25… TV camera, 26… Deviation amount extraction unit

──────────────────────────────────────────────────続 き Continuation of the front page (72) Inventor Hitoshi Kubota 292 Yoshida-cho, Totsuka-ku, Yokohama-shi, Kanagawa Prefecture Inside Hitachi, Ltd. Production Engineering Laboratory (56) References ) Surveyed field (Int.Cl. ⁶ , DB name) H01L 21/66 G01N 21/88

Claims (9)

(57) [Claims] An image of a first pattern of a plurality of patterns originally formed to be the same is obtained to obtain a first image, and an image of a second pattern of the plurality of patterns is obtained. A second image is obtained, the first image and the second image are relatively shifted, and a plurality of difference images between the first image and the second image having different shift amounts are obtained. A defect detection method, wherein a defect is detected based on a plurality of difference images having different amounts. 2. The relative shift amount between the first image and the second image is determined by the shift amount at which the absolute value of the difference between the first image and the second image is minimized. The defect detection method according to claim 1, wherein the defect is determined based on a difference between the first image and the second image. 3. When each of the difference images having different shift amounts is equal to or larger than a predetermined value, a defect candidate is obtained from the difference image, and a feature amount of the obtained defect candidate is extracted to perform defect judgment. The defect detection method according to claim 1, wherein: 4. A method of imaging a first pattern of a plurality of patterns originally formed to be identical to obtain a first image, and imaging of a second pattern of the plurality of patterns. Obtaining a second image, obtaining an image of the edge of the first pattern from the first image, obtaining an image of an edge of the second pattern from the second image, obtaining an edge of the first pattern; The image of the edge of the second pattern and the image of the edge of the second pattern are relatively shifted, and a plurality of difference images between the image of the edge of the first pattern and the image of the edge of the second pattern having different shift amounts are obtained. A defect detection method characterized by detecting a defect based on a plurality of difference images having different shift amounts. 5. An image of an edge of the first pattern and a second image of an edge of the first pattern.
The relative shift amount between the image of the edge of the pattern and the image of the edge of the first pattern and the image of the edge of the second pattern are the shift amount at which the absolute value of the difference between the image of the edge of the first pattern and the image of the edge of the second pattern is the smallest. 5. The defect detection method according to claim 4, wherein the determination is performed based on a difference between an image of an edge of the first pattern and an image of an edge of the second pattern. 6. When each of the difference images having different shift amounts is equal to or more than a predetermined value, a defect candidate is obtained from the difference image, and a feature amount of the obtained defect candidate is extracted to perform defect determination. The defect detection method according to claim 4, wherein: 7. The defect detection method according to claim 1, wherein the defect to be detected includes a defect smaller than an allowable error dimension of the plurality of patterns. 8. The defect detection method according to claim 1, wherein the first pattern and the second pattern are imaged by a one-dimensional image sensor. 9. The defect detection method according to claim 1, wherein the first pattern and the second pattern are imaged by a TV camera.