JP2003083908A - Defect inspection method and apparatus - Google Patents

Abstract

(57) [Summary] [PROBLEMS] To provide a highly accurate position shift amount detection method that does not depend on the density or shape of a pattern in a defect inspection that compares an inspection target image with a reference image and detects a defect from the difference. It is in. It is another object of the present invention to provide a highly sensitive defect inspection method and apparatus that reduce erroneous detections (false reports) due to misregistration detection errors. Further, it is possible to provide a defect inspection method and apparatus by comparing the image with high sensitivity in a wider range by lowering the inspection sensitivity only when the displacement amount detection fails, thereby minimizing the decrease in the detection sensitivity. is there. A detected image and a reference image of an object to be inspected are each divided into a plurality of regions, positional deviation information is calculated between the divided images, and a highly reliable positional deviation among the plurality of positional deviation information is calculated. By determining the amount of misregistration of the entire image using only information, it is possible to detect position amounts with high accuracy regardless of the density and shape of the pattern, the magnitude of the luminance difference between the images, and the magnitude of the luminance unevenness in the image. Become. In addition, the detection sensitivity is adjusted as needed by monitoring the positional deviation amount detection accuracy.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention provides a detection image of an object to be inspected such as a semiconductor wafer, a TFT, a photomask obtained by using a beam of light, a laser or charged particles, and a reference stored in advance. The present invention relates to a defect inspection method and apparatus for comparing reference images and inspecting for fine pattern defects, foreign matters, and the like based on the difference. In particular, the present invention relates to a defect inspection method and apparatus suitable for performing visual inspection of semiconductor wafers.

[0002]

2. Description of the Related Art As a conventional technique for detecting a defect by comparing a detected image of an object to be inspected with a reference image serving as a reference,
The method described in Japanese Patent Laid-Open No. 05-264467 is known.

In this method, a sample to be inspected in which repetitive patterns are regularly arranged is sequentially picked up by a line sensor, compared with an image with a time delay of the repetitive pattern pitch, and the non-matching portion is detected as a pattern defect. It is a thing. However, in reality, there are vibrations of the stage, inclination of the target object, and the like, and the positions of the two images are not necessarily aligned. Therefore, the image captured by the sensor and the image delayed by the repeated pattern pitch It is necessary to find the amount of displacement. Then, after aligning the two images based on the obtained positional deviation amount, the difference between the images is taken, and when the difference is larger than a specified threshold value, it is a defect, and when the difference is smaller than a non-defect. judge.

As a general method of aligning two images, there is a method of collectively calculating the amount of alignment using the information of the entire target image and performing the alignment.

[0005]

In the detection of the positional deviation amount between images in the defect inspection by image comparison, the edge pattern in the image is used as one piece of information for detecting the positional deviation amount, and the deviation of the edge pattern between the images is minimized. It is general to calculate such a positional shift amount. In practice, various methods such as a method using normalized cross-correlation, a sum of residuals, and a sum of squares of residuals have been proposed. However, when the edge information is extremely small, that is, when the ratio of the edge portion to the entire displacement detection area (hereinafter referred to as pattern density) is small (FIG. 2A), when the number of edges in the specific direction is extremely large ( 2B), when the number of small-pitch patterns is extremely large (FIG. 2C), the optimum position cannot be uniquely determined by the method of calculating the positional deviation amount from the information of the entire image.
There is a high possibility that a positional deviation amount calculation error will occur. Further, when there is unevenness in brightness in the images or unevenness in brightness between the images, the probability becomes even higher.

On the other hand, in the inspection by image comparison, the positional deviation between the inspection image and the reference image is detected → positioning → image comparison, that is, the difference calculation at the corresponding positions of the two images → defect extraction is performed. Although the defect inspection is performed, if the correct positional deviation amount is not obtained at the time of detecting the positional deviation, the images are compared at the wrong position. Figure 3 (a)
(B) and (c) are difference images between images when comparison is performed at the wrong position in the patterns of FIGS. 2 (a), (b) and (c). The difference image is an image in which a portion having a large difference is displayed brightly and a portion having a small difference is displayed darkly according to the difference value at each corresponding position of the inspection image and the reference image. ) (B) and (c) around the irregular patterns 21a, 21b, and 21c (non-defects existing in both images to be compared), the difference is large because the comparison is performed when the positions are displaced ( 31a, 31b, 31 of FIG.
c). When a region having a difference value larger than the specific threshold value TH is a defect, 31a, 31b, 31c will be detected, but this should not be originally detected as a defect. In other words, it is false information. Conventionally, as a method for avoiding the occurrence of false information as shown in FIG. 3, there is a method of increasing the threshold value TH to avoid the occurrence of false information. However, this lowers the sensitivity and is equal to or less than the position deviation occurrence portion. Defects in the difference value cannot be detected.

An object of the present invention is to reduce the sensitivity by solving the problems of the conventional inspection technique by calculating and aligning a highly accurate positional deviation amount that does not depend on the density and shape of the pattern. It is another object of the present invention to provide a defect inspection method and apparatus by highly sensitive image comparison, which reduces false reports due to the displacement of the positions of the comparison images.

Another object of the present invention is to monitor the accuracy of the positional deviation amount detection and reduce the inspection sensitivity only when the positional deviation detection fails, so that the deterioration of the detection sensitivity is minimized. It is to provide a sensitive defect inspection method and apparatus.

[0009]

In order to achieve the above object, the present invention provides a pattern density and shape in an inspection in which a defect is detected by comparing two images (detection image and reference image) with each other. , The two images are each divided into a plurality of regions in order to detect the positional deviation amount with high accuracy regardless of the size of the brightness unevenness between the images and the size of the brightness unevenness in the images. Positional deviation information is calculated between the detected image and the reference image in each area, and among the plurality of positional deviation information, a plurality of highly reliable one or only one positional deviation information is used to detect the detected image and the It is characterized in that the amount of positional deviation of the entire image from the reference image is determined.

Further, the present invention evaluates the reliability of the positional deviation amount of the entire image determined when the highly reliable positional deviation information is not obtained, and when the reliability is low, the past (same object) is detected. The position of the entire image between the detected image and the reference image is defined as the highly reliable position deviation amount at the time when the position deviation amount is detected on the inspection object before the time point at which the position deviation amount is detected). It is characterized in that it is a shift amount.

Further, the present invention evaluates the reliability of the positional deviation amount of the entire image which is determined when highly reliable positional deviation information is not obtained. The positional deviation amount predicted from the highly reliable positional deviation amount is used as the positional deviation amount of the entire image between the detected image and the reference image.

The present invention also evaluates the reliability of the determined positional deviation amount, and when the reliability is high, the reference for the positional deviation calculation after the next time (including the time when the next positional deviation amount is calculated according to the scanning). It is characterized in that the current positional displacement amount of the image is held as data.

Further, according to the present invention, in order to realize highly sensitive inspection irrespective of pattern density, in the case of continuous inspection, past reliable positional deviation amounts are collected and compared with them to obtain the current image. A means for monitoring the positional deviation amount detection accuracy of is provided.

Further, the present invention is provided with means for lowering the detection sensitivity so as to prevent erroneous detection due to the positional deviation when the reliability of the positional deviation determined as a result of the monitor is low.

Further, according to the present invention, even when the reliability of the determined positional deviation amount is low as a result of monitoring, if the positional deviation is not fatal to the image to which it is applied, the detection sensitivity is not reduced. Is equipped with.

Further, in order to provide a defect inspection apparatus by highly sensitive image comparison, the present invention provides means for dividing two images to be inspected and positional deviation information of each divided image in parallel. , A means for evaluating the reliability of the positional deviation information calculated from each divided image, a means for calculating the positional deviation amount of the entire image from the plural positional deviation information, and a calculated positional deviation amount. And a means for detecting a defect from the difference and a means for monitoring the accuracy of detecting the amount of positional deviation of the divided images are provided in, for example, the positional deviation amount detection unit. To be done.

[0017]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described in detail below with reference to FIGS. 1 to 13.

As an embodiment according to the present invention, a defect inspection method in an optical appearance inspection apparatus in which a semiconductor wafer is an object to be inspected is taken. FIG. 1 shows an embodiment of the configuration of the appearance inspection apparatus. Is.

The appearance inspection apparatus has a stage 12 on which a sample (object to be inspected such as a semiconductor wafer) 11 is mounted and moved.
A light source 201, an illumination optical system 202 that collects light emitted from the light source 201 and illuminates the sample 11 as illumination light,
The objective lens 203 that forms an optical image obtained by reflection from the sample 11 and the detection unit 13 that includes the image sensor 204 that receives the formed optical image and converts it into an image signal according to brightness. Composed.

The image processing unit 15 calculates a defect candidate in the wafer, which is a sample, from the image detected by the detection unit 13. The image processing unit 15 includes an AD conversion unit 14 that converts an input signal from the detection unit 13 into a digital signal, a preprocessing unit 205 that performs image correction such as shading correction and dark level correction from the digital signal, and a digital signal of a chip to be compared. A delay memory 206 that stores the signal as a reference image signal, a position shift detection unit 207 that detects a position shift amount between the digital signal (detected image signal) detected by the detection unit 13 and the reference image signal of the delay memory 206, An image comparison unit 208 that compares the image signals of the detected image and the reference image by using the calculated positional deviation amount, and outputs a portion where the mismatched portion is larger than a specific threshold as a defect candidate, and the coordinates of the defect candidate. And a feature extraction unit 209 that calculates a feature amount and the like.

The overall control unit 16 has a user interface having a display unit and an input unit for accepting a change in inspection parameters (threshold value used in image comparison) from a user and displaying detected defect information. Part 21
0, a storage device 211 that stores the detected feature amount of a defect candidate, an image, and the like, and a CPU that performs various controls. The mechanical controller 212 drives and controls the stage 12 based on a control command from the overall control unit. The image processing unit 15, the detection unit 13 and the like are also driven by a command from the overall control unit 16.

In the position shift detector 207 according to the present invention,
First, the detected image f (x, input from the preprocessing unit 205
y) and the reference image g (x, obtained from the delay memory 206)
y) and y) are each divided into n small areas, and the positional deviation detection is performed in parallel for each small area. 207-1, 207-2, ...
Reference numeral 207-N denotes a positional deviation amount calculation unit in a divided image that detects positional deviation information from each of the N small area images, which operate in parallel. Reference numeral 213 denotes a divided image positional deviation information integrated CPU which detects the positional deviation amount of the entire image from the total N positional deviation information detected by the positional deviation amount calculation unit in each divided image.

The semiconductor wafer 11 to be inspected is shown in FIG.
As shown in, a large number of chips having the same pattern are regularly arranged. In the inspection apparatus of FIG. 1, in the image comparison unit 208, the two chips have the same position, for example, the area 41 of FIG.
And the images in the region 42 and the like are compared, and the difference is detected as a defect. The operation will be described. In the overall control unit 16, the semiconductor wafer 11 as a sample is continuously moved by the stage 12. In synchronization with this, the images of the chips are sequentially captured by the detection unit 13. The image sensor 204 of the detection unit 13 is divided into a plurality of channels in a direction perpendicular to the traveling direction of the stage, and outputs the input signal to the image processing unit 15 as a plurality of divided signals. In the image processing unit 15, first, the analog signal input after being divided into a plurality of pieces is converted into a digital signal by the AD conversion unit 14, and the preprocessing unit 205 performs shading correction, dark level correction, S /
N improvement processing is performed. The misregistration detection unit 207 outputs the image signal (detected image signal) of the inspection target chip output from the preprocessing unit 205 and the image output from the delay memory 206 delayed by the time required for the stage to move by the chip interval. A signal, that is, an image signal (reference image signal) of the chip immediately preceding the chip currently input as the inspection target is input as a set. The image signals of these two chips, which are sequentially input in synchronization with the movement of the stage, do not become signals at exactly the same location when the stage is vibrated or the wafer set on the stage is tilted.

For this reason, the positional deviation detecting unit 207 calculates the positional deviation amount between two images which are successively input as shown in FIG. 5 with a specific length for each divided image as one processing unit. To do. At this time, the detected image signal f (x, y) and the reference image signal g (x, y) are continuously input, but the position shift amount is calculated by setting a specific length as one processing unit for each divided image. And sequentially. Reference numerals 51 to 5N in FIG. 5 indicate one processing area in the case where an area having a length D (pixel) of one processing unit is divided into N processing. The following processing is also performed for each processing unit.

In the image comparison unit 208, the positional deviation detection unit 2
The detected image and the reference image are aligned (positional deviation correction) using the positional deviation amount calculated in 07, and comparison is performed to obtain a difference or difference image, and the difference value or difference image is specified. The position coordinates and the image thereof are output with the area larger than the threshold value of 1 as the defect candidate. In the feature extraction unit 209, for each of the plurality of defect candidates, based on the position coordinates and the image, editing such as deleting a small one as noise or merging neighboring defect candidates as one defect is performed. The feature quantities such as the position, area, size, etc. are calculated and output as a final defect. These pieces of information are stored in the storage device 211. It is also presented to the user via the user interface unit 210.

Next, an embodiment of the method for detecting the amount of displacement performed by the displacement detection unit 207 of the image processing unit 15 will be described. In the detection of the positional deviation amount between the detected image and the reference image, it is general to calculate the positional deviation amount so that the edge deviation between the images is minimized. However, when the ratio of the edge portion to the displacement detection area, that is, the edge density is small,
If the number of edges in a specific direction is extremely large, or if the number of repetitive patterns of fine pitches is extremely large, etc., the position displacement amount is not uniquely determined by the conventional method of calculating the position displacement amount from the information of the entire image. There is a high possibility that a misalignment amount calculation error will occur. If such an image includes an irregular pattern, the possibility of false alarm due to misdetection of the amount of misregistration increases at that portion. An example in which a positional deviation amount calculation error occurs due to a pattern will be described below.

There are various methods for calculating the amount of positional deviation,
Here, a method using normalized cross-correlation will be taken as an example of an example of the positional deviation amount detection method. First, as shown in FIG. 12, two images whose position shift amounts are to be calculated are relatively shifted in the X direction by −m to + m pixels and in the y direction by −n to + n pixels, and a total of M images M = ( A correlation value of (2 × m + 1) × ((2 × n + 1)) is calculated. Here, m = n = 3.
At this time, as shown by 1201 in FIG. 12, a total of 49 correlation values are calculated according to the shift amount (shift amount) of the image. Below, the correlation value 1 arranged according to the shift amount of the image
201 is described as a correlation map. Then, the shift amount when the correlation value becomes maximum in the correlation map is the positional shift amount between the images. In the correlation map 1201, the center (the shift amount is X
Since the maximum value is obtained in both the direction and the Y direction (0), the amount of positional deviation of the image is 0 pixels in both the X and Y directions. Here, when there are sufficient patterns in the image as shown in FIG. 12A, that is, when the pattern density is high, the correlation map is 1201. The figure (a ') is a three-dimensional display of this distribution, and the number of peaks is one, and its correlation value is also large.

However, when there is no pattern as shown in FIG. 12B, there is no peak in the correlation map and the value is small (b '). Further, when there is only a pattern in a specific direction as shown in FIG. 7C, the correlation map also has a large correlation value in the pattern direction, and a ridge is formed instead of a peak (c ′).

On the other hand, as shown in FIG. 12D, in a repeating dot pattern with a fine pitch less than the image shift range, a plurality of high peaks are formed on the correlation map (d '). As described above, in the patterns of FIGS. 9B, 9C and 9D, the amount of positional deviation is not uniquely determined, and it is difficult to find the correct amount of positional deviation. At this time, there is no problem if the entire comparison area has the same pattern, but if there are overwhelmingly many such patterns and very slight irregular patterns are mixed, it is a false report due to the shift of the comparison position in that part. Occurs in the image comparison unit 208.

Reference numerals 131 and 132 shown in FIG. 13 are examples of detected images and reference images, and 133 is a correlation map calculated from the entire image area. Since the dot pattern occupies most of this image, the correlation map 133 is affected by the dot pattern.
Has a plurality of peaks, and there is almost no difference in the correlation value at the peak position. Therefore, there is a high possibility of false detection.

Therefore, in the present invention, as shown in FIG.
One image signal (detected image) f (x, y) of the current chip input from the detection unit 13 and one image signal (reference image) g (x, y) of the previous chip input via the delay memory 206 The image is divided into the above plurality (here, N pieces), and the correlation map is individually obtained in these divided images. These processes are performed by 207-1 to 207-n. Hereinafter, the individual processing systems 207-1 to 207-N for each of the divided images will be described as channels.

In practice, as shown in FIG. 7, the inspection target image f (x, y) 71 and the reference image g (x, y) 72 are divided and input to the respective channels 207-1 to 207-N. To do. Then, the above-described correlation map 1201 is calculated for each of the channels 207-1 to 207-N, and the maximum correlation value position is detected as the position shift amount within the channel.

The divided image positional deviation information integration CPU 213 aggregates the positional deviation information calculated for each channel 207-1 to 207-N, determines the positional deviation amount common to all channels, and determines each channel 207-1. ~ 207-N. That is, the CPU 213 determines the positional deviation amount common to all channels and provides it to the image comparison unit 208 so as to apply it to each channel 207-1 to 207-N.
As a result, the image comparison unit 208 corrects the positional deviation between the detected image and the reference image based on the determined positional deviation amount, and obtains a difference between the detected image and the reference image, a difference image, or the like. Then, comparison is made, and if this difference value or difference image exceeds a predetermined judgment threshold value, it is judged as a defect candidate.

In the step 213 of collecting and integrating the positional deviation information of all channels, the N correlation maps calculated in each channel are examined and only the correlation maps of the highly reliable channels are selected. There are many ways to check the reliability by checking how likely the correlation value is. As one example thereof, it is assumed that a channel whose peak value of the correlation map is larger than the threshold value TH1 has high reliability. As another example, it is assumed that the channel having only the correlation value equal to or higher than the threshold value TH2 in the correlation map has high reliability. Further, as another example, it is assumed that a channel in which the difference between the peak value in the correlation map and the second correlation value is the threshold value TH3 or more has high reliability. Three examples of these evaluation methods (Equation 1)-
(Equation 3)

Max (Cor (i, j))> TH1 (i = −m to + m, j = −n to + n) (Equation 1) Num (Cor (i, j)) ≧ TH2) = 1 Num is ( (Formula 2) Max (Cor (i, j)) − Second (Cor (i, j)) ≧ TH3 (Formula 3) where the possible correlation value is −1.0. .About.1.0, and when the images completely match, the value is 1.0. Therefore, if only the channel with higher reliability is selected, the thresholds TH1 and TH2 should be 0.9 or more. Is desirable.

Divided image position shift information integration CPU 213
In this way, one or a plurality of correlation maps are selected in this way, the positional deviation amount common to all channels is determined from this, and applied to each channel.

An example of the method of determining the positional deviation amount common to all channels will be described. Addition of a plurality of selected correlation maps, that is, summation of correlation values at corresponding positions of each correlation map (equation 4). Hereinafter, this is referred to as an addition correlation map. Then, the peak position in the additive correlation map is set as the positional deviation amount common to all channels (Equation 5).

CorA (i, j) = Cor0 (i, j) + ... + CorN (i, j) + ... (Equation 4) CorA (i, j): Addition correlation map of (i, j) Correlation value Cor0 (i, j), ..., CorN (i, j), ...: Correlation value at high reliability channel (i, j) (i, j): Relative shift between detected image and reference image Amount of positional deviation (i, j) = Max (CorA (i, j)) (i, j) (Equation 5) When the positional deviation amount of 131 and 132 shown in FIG. Although it was previously described that the possibility of erroneous detection is high from the area, the above-described present invention makes it possible to detect the correct positional deviation amount.

The correlation map 134 shown in FIG. 13B is
Images 131-N and 132- of the N-th channel surrounded by thick lines when 131 and 132 are divided into small regions, respectively
It is calculated from N. In the Nth image, the influence of the dot pattern is small, and the correct positional deviation amount can be calculated.

Reference numeral 135 shown in FIG. 13C is an addition correlation map obtained from the correlation map of the channel selected by evaluating the reliability according to (Equation 1).
The correct peak can be calculated by including the information of.

As described above, in the correlation map of the entire area, even for a pattern in which a plurality of peaks are formed and there is a high possibility of false detection, by calculating the correlation map for each small area and adding them, The correct amount of positional deviation can be calculated. As a matter of course, without using the addition correlation map, one channel having the highest reliability may be selected and applied to all the channels. In this case, if the correlation maps of all channels are evaluated by (Equation 2) or (Equation 3), the correlation map 134 will be selected in the image shown in FIG.

Although the positional deviation amount is detected by the normalized cross-correlation value and the reliability is also evaluated by the correlation value here,
The amount of positional deviation is calculated for each channel by another method such as the sum of the grayscale differences between images, and a reliability index is provided accordingly, and the amount of positional deviation is determined from the information of multiple channels or a single channel. I don't mind.

Another example of the specific effect will be described with reference to the image shown in FIG. In the two images to be inspected, the pattern in the specific direction is included only in the two channels, channel 81 and channel 82, where there is almost no pattern. The directionality of these two patterns is different. Although edge information is basically used to detect the amount of positional deviation, in such an image, when trying to calculate the amount of positional deviation from the entire image as in the conventional art, the pattern density becomes extremely small. Even if there is slight brightness unevenness or brightness unevenness between images, the probability of calculating an incorrect position shift amount increases. Further, if one of the most reliable positional deviations is selected from the positional deviations detected in each channel and it is attempted to apply it to all the channels, in the image of FIG. The position shift amount of is selected. However, since both have only edge information in a specific direction, there is a high possibility that a calculation error will occur in the direction of the arrow in FIG. As a result, if the positional shift amount of one channel is applied to all channels, erroneous detection (false alarm) will occur in the channel including the other pattern.

On the other hand, in the position shift amount detecting method according to the present invention by the CPU 213, in the image of FIG.
The positional shift amount of the entire image is comprehensively calculated only from the positional shift information of the channel 81 and the channel 82 having the pattern, without using the positional shift information of the channel having low reliability such as the channel of the plain image. Actually, since the positional deviation amounts of the channels 81 and 82 are highly reliable in the direction of the arrow in FIG. 9B, the CPU 213 obtains the peak position of the addition correlation map obtained by adding the correlation maps of the two channels, Alternatively, by adopting and integrating only the positional deviation amount in the highly reliable direction from each of the two channels, x
It is possible to calculate the correct amount of misalignment in both the direction and the y direction. This is because the information of the channel having no pattern is removed, and the same effect as that of detecting the positional deviation amount from the image having a large pattern density can be obtained.

As described above, according to the present invention, it is possible to substantially eliminate the occurrence of a false alarm due to the misdetection of the positional deviation amount depending on the density and the directionality of the pattern. In addition, the CPU 213
Is to calculate statistics necessary for positional deviation detection, such as the normalized cross-correlation and the sum of grayscale differences in a small area.
The influence of uneven brightness in images and uneven brightness between images is also reduced, and misregistration amount detection errors are reduced.

As in the present invention, the target image (detected image f
(X, y) and the reference image g (x, y)) are divided into N small areas, the positional deviation information is obtained in a channel unit, that is, in the small area, and the image is obtained using only highly reliable channel information. The amount of positional deviation of the entire image is calculated, and the calculated amount of positional deviation of the entire image is provided to the image comparison unit 208. When the image comparison unit 208 compares the detected image with the reference image, the detected image and the reference image are compared with each other. By correcting the amount of misalignment, it is possible to reduce the false alarm rate due to misregistration detection error to a specified rate or less regardless of the density or shape of the pattern. The ideal false detection rate is, of course, 0%, but the comparison image may be deteriorated due to image capturing conditions and the like. According to the method of the present invention, when a good image is obtained, the misregistration detection error does not occur at all regardless of the density or shape of the pattern, but even if the image is deteriorated due to lack of focus, 0 It is possible to be less than 1%.

Further, according to the present invention, the positional deviation detecting unit 207
The divided image positional deviation information integration CPU 213 evaluates the reliability of the positional deviation amount of the entire image calculated using only the highly reliable channel information, and when the reliability is low, If none of the calculated positional displacement amounts has high reliability, the positional displacement amount is not determined from the current image, and, for example, the storage device 2 is used.
11 or once stored in a memory or the like provided in the positional deviation detection unit 207 (one time means one processing unit D for calculating the positional deviation amount shown in FIG. 5) and past (in FIG. 5) The positional shift amount of one processing unit before the present scanned as shown by the arrow: naturally one processing unit in an adjacent chip even within a chip) is used. For example, when the CPU 213 examines the image for each small area shown in FIG. 8 and the pattern density is extremely low, there is a possibility that no highly reliable positional deviation amount can be found from all channels. In such a case, the divided image positional deviation information integration CPU 213 of the positional deviation detection unit 207 calculates the positional deviation calculated from the past image having a high pattern density stored in the storage device 211 or the memory of the positional deviation detection unit 207. See the quantity and apply it. On the other hand, if the current image to be inspected has sufficient patterns and the calculated displacement amount is highly reliable, the reference will be made with the image from which only the next unreliable displacement amount was obtained. As data, for example, the overall control unit 16
It is held in the storage device 211 via the. As a matter of course, the memory provided in the positional deviation detecting unit 207 may store the positional deviation amount as highly reliable reference data.

As an example of the reliability evaluation method of the determined positional deviation amount in the CPU 213, the relationship between the past positional deviation amount loci stored in the storage device 211 and the calculated positional deviation amount is examined. Do that. For example, FIG. 10A is an example in which past positional deviation amounts are plotted in time series, but if the positional deviation amounts periodically change due to the vibration cycle of the device, the CPU 213 Following the trajectory of the positional displacement amount so far, the positional displacement amount of this time is obtained in advance by extrapolation (10 in FIG. 10).
1). Then, the CPU 213 determines that the positional deviation amount is highly reliable if the actually calculated positional deviation amount is close to the predicted value obtained by extrapolation. For example, the case where the actually calculated positional deviation amount is within 5% of the predicted positional deviation amount is regarded as reliable. As described above, by checking the reliability of the positional deviation amount by comparing it with the past accumulated positional deviation amount, the calculated positional deviation amount can be evaluated more accurately. Here, although it has been described that the predicted value is calculated by extrapolation assuming that the amount of positional deviation periodically changes, other methods can be used as the method of calculating the predicted value. For example, it is possible to collect only the most recent positional deviation amount and use the average value as the predicted value (102 in FIG. 10). However, no matter which method is used to calculate the predicted value, the past positional deviation amount collected must be highly reliable. for that reason,
If the positional deviation amount calculated from the current image to be inspected is highly reliable, the past data for calculating the predicted value for the next time (after one processing unit D that appears according to the scan indicated by the arrow in FIG. 5) For example, it is held in the storage device 211.

Here, when the reliability of the calculated positional deviation amount is low, or when there is no highly reliable positional deviation information calculated from all channels,
The CPU 213 can also apply the predicted value obtained for the reliability evaluation.

As another embodiment of the reliability evaluation method, the size of the pattern density in the target image f (x, y) obtained from the preprocessing unit 205 can be mentioned. For example, if the number of patterns in each of the x direction and the y direction is sufficiently large, the probability of misregistration detection is low, and thus the reliability is high. For example, as an example of the pattern density measuring method performed by the CPU 213, first, in the x direction in each pixel of the entire image obtained from the preprocessing unit 205 or the divided small area,
Calculate the differential value in the y direction. There are generally various operators for calculating the differential value, and an example thereof is shown in FIG. 11, and for a certain target pixel E, a value in the vicinity thereof is used, and a differential value in the x direction at E = (B + H− 2 × E), the differential value in the y direction at E = (D + F-2 × E), and what percentage of the total number of pixels has the value larger than a specific threshold value (for example, (10% or more)
Determine with.

Further, the CPU 213 can also evaluate the reliability by combining a plurality of conditions, such as comparison with the past locus of the positional deviation amount and evaluation from both the magnitude of the pattern density of the current image. . By checking the reliability in this way, the positional deviation amount detection accuracy can be monitored,
If it is determined that the accuracy is low, the image comparison unit 20
It is also possible that the detection sensitivity for the image comparison unit 208 is lowered in advance, assuming that there is a high possibility of erroneous detection (false alarm) due to the shift of the comparison position in 8. In this way, the means for monitoring the positional deviation amount detection accuracy is provided.

Further, when the CPU 213 monitors the positional deviation detection accuracy from the pattern density, for example, it is naturally judged that an image with a low pattern density has a low positional deviation amount reliability. However, if there is no pattern in the image, there is a high possibility that the correct amount of misalignment cannot be obtained, but no false alarm will occur. Similarly, as shown in FIG.
Even if only the pattern in the specific direction as shown in (c) is included, no false alarm occurs. Therefore, in the present invention, the CPU 21
If it is determined in 3 that the positional deviation amount detection accuracy is poor,
Further, for the detected image f (x, y), it is checked whether comparison at a shifted position is fatal, that is, whether erroneous detection occurs, and if not fatal, the detection sensitivity is not lowered.

In this way, the CPU 213 can prevent the detection sensitivity from being unnecessarily lowered by checking the fatality for the detected image. By accurately monitoring the positional deviation amount detection accuracy in this way and checking the criticality due to the comparative positional deviation of the current image, the detection sensitivity is reliably lowered only when necessary, and false detections (false alarms) are reduced.
Further, it is possible to minimize the decrease in detection sensitivity and realize a highly sensitive inspection in a wider range.

As a result, a detection sensitivity of 100 nm can usually be achieved in the optical visual inspection apparatus. Further, by using the present invention, it is possible to obtain a detection sensitivity of 50 nm when the image quality is good.

The overall control unit 16 evaluates the reliability of the positional deviation amount, and the CPU 21 outputs the result.
May be provided to 3.

The embodiment of the present invention has been described above by taking the positional deviation amount detecting method in the optical appearance inspection apparatus for the semiconductor wafer as an example. However, the positional deviation in the electron beam pattern inspection or the defect inspection by DUV illumination is described. It is also applicable to quantity detection. In this case, detection sensitivity of 30 to 70 nm can be achieved. Further, the inspection target is not limited to the semiconductor wafer, and if the defect detection is performed by comparing the images, for example, a TFT substrate, a photomask,
It is also applicable to printed boards.

[0057]

According to the present invention, in a defect inspection in which a detected image of an object to be inspected is compared with a reference image and a defect is detected from the difference, the two target images are each divided into a plurality of regions, The positional deviation amount of the entire image is calculated using only highly reliable positional deviation information calculated from each of the small areas that have been calculated, so highly accurate positional deviation amount detection that does not depend on the density or shape of the pattern can be performed. It has the effect that can be realized.

Further, according to the present invention, it is possible to provide a highly sensitive defect inspection method and apparatus in which erroneous detection (false alarm) due to the shift of the comparison position between images is reduced as described above.

Further, according to the present invention, the positional deviation amount detection accuracy is further monitored, and the inspection sensitivity is lowered only when the detection is unsuccessful. A sensitive defect inspection device can be provided.

Specifically, the positional deviation detection accuracy is 0.1 pixel or less, and the rate of defect erroneous detection due to positional deviation detection error is
Even if the comparative image is deteriorated, it can be 0.1% or less regardless of the density and shape of the pattern. When applied to an optical appearance inspection device, it is 100 to 50.
A defect detection sensitivity of nm can be achieved.

[Brief description of drawings]

FIG. 1 is a diagram showing an embodiment of a schematic configuration of an optical visual inspection apparatus according to the present invention.

FIG. 2 is a diagram showing an example of an image in which misregistration detection easily fails according to a conventional technique.

FIG. 3 is a diagram showing an example of a difference image when a misregistration detection fails according to a conventional technique.

FIG. 4 is a view showing a wafer as an example of a defect inspection target according to the present invention.

FIG. 5 is a diagram showing an example of a processing unit processed by a positional deviation amount detection unit of the image processing unit according to the present invention.

FIG. 6 is a diagram showing an example of a processing configuration in a positional deviation amount detection unit of the image processing unit according to the present invention.

FIG. 7 is a diagram showing an embodiment of a position shift amount calculating method according to the present invention.

FIG. 8 is a diagram showing an example of an image specifically showing the effect of the present invention.

FIG. 9 is a diagram showing an example of divided images specifically showing the effect of the present invention.

FIG. 10 is a diagram showing an embodiment of a method of predicting a positional deviation amount of a current image from a past positional deviation amount trajectory according to the present invention.

FIG. 11 is a diagram showing an example of a method of calculating a differential value used for edge density according to the present invention.

FIG. 12 is a diagram showing an example of a correlation map according to the present invention and an example of a difference in the correlation map depending on the pattern shape of the target area.

FIG. 13 is a diagram showing an example of an image specifically showing the effect of the present invention.

[Explanation of symbols]

11 ... Sample, 12 ... Stage, 13 ... Detection part, 15 ... Image processing part, 14 ... AD conversion part, 201 ... Light source, 202 ...
Illumination optical system, 203 ... Objective lens, 204 ... Image sensor, 205 ... Preprocessing unit, 206 ... Delay memory, 207
... misregistration amount detection unit, 208 ... image comparison unit, 209 ... feature extraction unit, 16 ... overall control unit, 210 ... user interface unit, 211 ... storage device, 212 ... mechanical controller, 213 ... divided image misregistration information integrated CP
U, 1201 ... Correlation map.

   ─────────────────────────────────────────────────── ─── Continued front page    (72) Inventor Takashi Okabe             292 Yoshida-cho, Totsuka-ku, Yokohama-shi, Kanagawa             Inside the Hitachi, Ltd. production technology laboratory F term (reference) 2G051 AA51 AB02 CA03 CB01 DA06                       EA11 EA14 EB02 EB09                 4M106 AA01 BA04 CA39 CA50 DB20                       DJ11 DJ18 DJ21 DJ23                 5L096 BA03 EA35 FA06 FA34 FA69                       GA04 HA07

Claims (10)

[Claims] 1. An input detection image of a sample to be inspected and a reference reference image are each divided into a plurality of small areas, and between the divided small area detection image and the reference image. Position misalignment information is calculated by using the calculated misalignment information for each of the plurality of small areas, and the misalignment amount determination process for determining the misalignment amount between the detected image and the reference image, Using the positional shift amount determined in the determination process, the detected image and the reference image are aligned and compared,
And a defect detection process for detecting a defect or a defect candidate from the difference. 2. In the step of determining the amount of displacement, a plurality of highly reliable ones are used when determining the amount of displacement between the detected image and the reference image from the displacement information for each of the plurality of small areas. 2. The defect inspection method according to claim 1, further comprising: selecting the positional deviation information of 1. and determining the positional deviation amount between the detected image and the reference image from the selected plurality of positional deviation information. 3. A position having the highest reliability in determining the amount of positional deviation between the detected image and the reference image from the positional deviation information for each of the plurality of small areas in the positional deviation amount determination process. 2. The defect inspection method according to claim 1, wherein one piece of shift information is selected, and the selected shift amount is determined as a shift amount between the detected image and the reference image. 4. An input detection image of a sample to be inspected and a reference reference image are each divided into a plurality of small regions, and the divided detection region and the reference image are divided between the divided small regions. If the positional deviation information for each of the plurality of calculated small areas does not include the positional deviation information whose reliability is higher than a predetermined reference, the positional deviation information with high reliability is calculated in the past. A positional deviation amount determining step of determining an amount as a positional deviation amount between the detected image and the reference image,
The detected image and the reference image are aligned and compared using the amount of displacement determined in the amount of displacement determination process,
And a defect detection process for detecting a defect or a defect candidate from the difference. 5. The input detected image of a sample to be inspected and a reference reference image are each divided into a plurality of small areas, and between the divided small area detected image and the reference image. The positional deviation information is calculated at, and the positional deviation amount between the detected image and the reference image is evaluated based on the calculated positional deviation information for each of the plurality of small areas, and the reliability is higher than a predetermined standard. When it is low, the positional deviation amount determining step of determining a highly reliable positional deviation amount in the past as the positional deviation amount between the detected image and the reference image, and the positional deviation amount determined in the positional deviation amount determining step are A defect inspecting method, comprising: aligning and comparing the detected image and the reference image with each other, and detecting a defect or a defect candidate from the difference. 6. The input detected image of the sample to be inspected and the input reference image are each divided into a plurality of small areas, and between the divided small area detected image and the reference image. If the positional deviation information for each of the plurality of calculated small areas does not include the positional deviation amount whose reliability is higher than a predetermined reference value, the plural positional information with high reliability in the past is calculated. The position shift amount determining process of determining the position shift amount predicted from the shift amount as the position shift amount between the detected image and the reference image, and the detection using the position shift amount determined in the position shift amount determining process. A defect inspection method comprising: aligning and comparing an image and the reference image, and detecting a defect or a defect candidate from the difference. 7. The positional deviation amount determining step further evaluates the reliability of the positional deviation amount between the determined detected image and the reference image, and when the reliability is lower than a predetermined reference, The position shift amount predicted from a plurality of highly reliable position shift amounts in the past is determined as the position shift amount between the detected image and the reference image. Defect inspection method. 8. A stage for mounting a sample to be inspected, an irradiation optical system for irradiating the sample with a beam, a detector for detecting a beam obtained from the sample and converting it into a signal, and the detector. A output signal is converted into a detected digital image
A D conversion circuit, a memory for storing the detected digital image, a positional deviation amount determining unit for determining an amount of positional deviation between the detected digital image and a reference digital image serving as a standard image, and the positional deviation amount determining unit. An image comparison unit that corrects and compares the detected digital image and the reference digital image based on the determined positional deviation amount, and outputs defect candidate information as a comparison result, and a comparison result in the image comparison unit An image processing unit having a feature extraction unit for detecting defects from the defect inspection apparatus. 9. The defect inspection apparatus according to claim 8, further comprising an output monitor for displaying an inspection result detected by the image processing unit. 10. A misregistration amount determining unit in the image processing unit divides each of the detected digital image and the reference digital image into a plurality of small areas, and the detected image and the reference image for each of the divided small areas. And the positional shift information between the detected image and the reference image is determined from the calculated positional shift information for each of the plurality of small areas. Claim 8 or 9
The described defect inspection device.