JP5011348B2 - Pattern inspection method and apparatus - Google Patents

Description

  The present invention compares an image of an object obtained using light or a laser with a reference image, and relates to the inspection of fine pattern defects and foreign matters from the difference, and in particular, the appearance of semiconductor wafers, TFTs, photomasks, etc. The present invention relates to a pattern inspection apparatus and method suitable for inspection.

  As a conventional technique for detecting a defect by comparing an inspection object image with a reference image, a 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 imaged by a line sensor, compared with an image with a time delay corresponding to the repetitive pattern pitch, and the mismatched portion is detected as a pattern defect. . Such a conventional inspection method will be described by taking an appearance inspection of a semiconductor wafer as an example. As shown in FIG. 6, a large number of chips having the same pattern are regularly arranged on the semiconductor wafer to be inspected. Each chip can be roughly divided into a memory mat portion 71 and a peripheral circuit portion 72 as shown in FIG. The memory mat 71 has a small repetitive pattern (
The peripheral circuit unit 72 is basically a set of random patterns. In general, the memory mat portion 71 has a high pattern density, and an image obtained by the bright field illumination optical system becomes dark. On the other hand, the peripheral circuit portion 72 has a low pattern density, and the obtained image becomes bright.

In the conventional appearance inspection, images at the same position of two adjacent chips, for example, images in the region 61 and the region 62 in FIG. 6 are compared, and the difference is detected as a defect. At this time, because there is vibration of the stage, tilt of the target, etc., the positions of the two images are not necessarily aligned, so the positional deviation between the image captured by the sensor and the image delayed by the repeated pattern pitch After obtaining the amount and aligning the two images based on the obtained misregistration amount, the difference between the images is taken, and when the difference is larger than the prescribed threshold value, the defect is detected. Judge as a defect.

Japanese Patent Laid-Open No. 05-264467

In the alignment of two images in the comparative inspection, the edge portion in the image is used as one piece of information for calculating the displacement amount, and the displacement amount is calculated so that the displacement between the patterns in the image is minimized. It is common. In practice, methods using normalized cross-correlation, methods using residual summation, and the like have been proposed.

In addition, in a semiconductor wafer to be inspected, a subtle difference in film thickness occurs in the pattern due to flattening such as CMP, and there is a local brightness difference in the image between chips. For example, FIG.
41 in FIG. 4B is an example of a reference image, and reference numeral 4a in FIG.
As shown in 4b of FIG. 4B, a difference in brightness occurs in the same pattern of the inspection target image and the reference image. Further, the inspection target image 41 in FIG. 4A has a defect 4d. The difference image in this case is as shown in FIG. The difference image is an image displayed as a difference in density according to a difference at each corresponding position between the inspection target image and the reference image. The difference waveform at positions 1D-1D ′ is shown in FIG.
It is like (d). For such an image, if a portion where the difference value is equal to or greater than a specific threshold TH is determined as a defect as in the conventional method, the difference value 4c between the patterns 4a and 4b having different brightness is detected as a defect. Will be. This should not be detected as a defect. In other words, it is a false report. Conventionally, as one method for avoiding the generation of false alarms such as 4c in FIG. 4 (c), the threshold value TH is increased (FIG. 4 (d) TH → TH2).
However, this lowers the sensitivity, and the defect 4d having a difference value equal to or less than the same level cannot be detected.

Further, the difference in brightness due to the difference in film thickness may occur only between specific chips in the wafer among the arranged chips shown in FIG. 6, or may occur only in a specific pattern in the chip. If the threshold value TH is adjusted to the local area, the overall inspection sensitivity is remarkably lowered.

Furthermore, as shown in FIG. 4, when the difference in brightness due to thin film interference is large in some patterns, the correct misregistration amount is obtained both in the method using normalized cross-correlation and the method for obtaining the sum of residuals. There is a possibility that it cannot be calculated.

On the other hand, the defect may not be manifested by a combination of factors depending on the object such as the material, surface roughness, size, and depth and factors depending on the detection system such as illumination conditions.

The object of the present invention is to solve such problems of the conventional inspection technique and compare the images of the corresponding regions of the two patterns formed so as to be the same pattern, and determine the mismatched portion of the image as a defect. An object of the pattern inspection apparatus is to reduce the influence of uneven brightness of a pattern caused by a difference in film thickness and to realize a highly sensitive pattern inspection.

In addition, it has a plurality of detection system units that can be switched, selects a detection system according to the target and the defect of interest, and inspects it according to the corresponding comparative inspection method. It is to realize pattern inspection.

In order to achieve the above object, in the present invention, it is possible to switch between pattern inspection apparatuses that compare images of corresponding areas of two patterns formed to have the same pattern and determine a mismatched portion of the image as a defect. And a plurality of different detection systems, a plurality of image comparison processing methods, and a plurality of defect classification methods. One or more detection systems are selected in consideration of the detection performance of the corresponding image comparison processing method and defect classification method. This makes it possible to select the optimum conditions and detect various types of defects.

Further, the pattern inspection apparatus is provided with means for converting the gradation of the image signal between the comparison images in a plurality of different processing units. As a result, the inspection target is a semiconductor wafer, and brightness differences occur in the same pattern between images due to differences in film thickness within the wafer, fluctuations in the amount of illumination light, variation in sensitivity for each pixel of the image sensor, unevenness in the light amount accumulation time, etc. Even if it is, the defect can be detected correctly.

Further, the apparatus includes a means for sequentially matching the positional shift and the brightness shift between the comparison images, and a means for simultaneously matching. As a result, even when a strong brightness difference is caused by a specific pattern between images due to a difference in film thickness within the wafer, a positional shift can be detected with high accuracy. In addition, even when a position shift occurs between images, the brightness shift can be detected with high accuracy.

According to the present invention, by having a plurality of image comparison processes and classification processes corresponding to a plurality of detection systems, it is possible to realize a high-sensitivity inspection and to detect various defects.

In addition, the brightness of images to be compared is caused by differences in brightness (color unevenness) between chips caused by various factors such as differences in film thickness between chips, differences in accumulated light quantity due to uneven stage speed, and illumination fluctuations. By combining the brightness with a plurality of different methods, it becomes possible to reveal and detect the defect of the weak signal buried in the strong brightness unevenness.

Furthermore, by calculating the amount of misregistration that is robust against color unevenness, a more sensitive inspection can be performed.

  Hereinafter, an embodiment of the present invention will be described in detail with reference to FIGS.

As an example, taking a defect inspection method in an optical appearance inspection apparatus for a semiconductor wafer as an example, FIG. 1 shows an example of the configuration of the apparatus. Reference numeral 11 denotes a sample (inspection object such as a semiconductor wafer), 12 denotes a stage on which the sample 11 is mounted and moved, and 13 denotes a detection unit.
The detection unit 13 illuminates and reflects the sample 11 with the light source 101 that irradiates the sample 11, the illumination optical system 102 that condenses the light emitted from the light source 101, and the illumination light collected by the illumination optical system 102. An objective lens 103 that forms an optical image to be obtained, an image sensor 104 that receives the formed optical image and converts it into an image signal corresponding to brightness, and an AD that converts an input signal from the image sensor 104 into a digital signal The conversion unit 105 is configured.

Here, in the example shown in FIG. 1, a lamp is used as the light source 101, but a laser may be used. The wavelength of the light emitted from the light source 101 may be a short wavelength, or may be a light having a broad wavelength (white light). When using short wavelength light,
In order to increase the resolution of the detected image (to detect minute defects), light in the ultraviolet region (
Ultra Violet Light (UV light) can also be used. When a laser is used as a light source, if it is a single wavelength laser, it is necessary to provide means for reducing coherence, not shown.

Further, a time delay integration type image sensor (Time Delay Integration) configured by arranging a plurality of one-dimensional image sensors in the image sensor 104 in a two-dimensional manner.
By adopting Image Sensor (TDI image sensor), the signals detected by each one-dimensional image sensor in synchronization with the movement of the stage 13 are transferred to the next-stage one-dimensional image sensor and added, so that it is relatively fast and high. Sensitivity can be detected.

An image editing unit 14 stores a digital signal of a reference image to be compared, a pre-processing unit 106 that performs image correction such as shading correction and dark level correction on the digital signal of the image detected by the detection unit 13. The image memory 107 is provided.

Reference numeral 15 denotes an image comparison processing unit that calculates defect candidates in a wafer as a sample. The image comparison unit 15 compares two images (detected image and reference image) stored in the image memory 107 of the image editing unit 14 and compares them. The portion where is larger than the threshold is defined as a defect. First, digital signals of the detected image and the reference image stored in the image memory 107 are read out, the correction amount calculation unit 108 calculates the position and the correction amount for matching the brightness, and the position calculated by the image comparison unit 109 And the brightness correction amount are used to compare the brightness at the corresponding positions of the image signal of the detected image and the reference image, and a portion where the difference value is larger than a specific threshold value is output as a defect candidate. The parameter setting unit 110 sets image processing parameters such as a threshold value when extracting defect candidates from the difference value, and supplies the image processing unit 109 with the image processing parameters. Then, the defect classification unit 111 extracts true defects from the feature amounts of the defect candidates and classifies them.

Reference numeral 16 denotes an overall control unit, which is a user interface unit 112 having a display means and an input means for accepting changes in inspection parameters (threshold values used in image comparison, etc.) from the user and displaying detected defect information. The storage device 113 stores the feature amounts and images of the detected defect candidates, and the CPU (internal to the overall control unit 16) performs various controls. 114
Is a mechanical controller that drives the stage 12 based on a control command from the overall control unit 16. Note that the image comparison processing unit 15, the detection unit 13, and the like are also driven by commands from the overall control unit 16.

The semiconductor wafer 11 to be inspected has many chips with the same pattern as shown in FIG.
They are lined up regularly. In the inspection apparatus of FIG. 1, the overall control unit 16 continuously moves the semiconductor wafer 11 as a sample by the stage 12, and in synchronization with this, sequentially captures the chip image from the detection unit 13, The same position of the chip, for example, region 61 and region 6 in FIG.
2 are compared with each other as a detected image and a reference image in the above-described procedure, and a defect is detected.

Here, the detection unit 13 of the inspection apparatus of the present embodiment has a plurality of switchable detection systems. FIG. 2 is a diagram showing details of the detection unit 13.

The light emitted from the light source 101 passes through the quasi-continuous optical system 3210 and the light amount is temporally averaged or leveled, and enters the optical path branching optical system 23. The light source 101 is a laser, and emitted light is ultraviolet light (ultra violet light: UV light, deep ultra violet light: DUV light, etc.) or visible light. The polarization state of the light incident on the optical path branching optical system 23 is adjusted by the polarization unit 2302, and two optical paths 2601 and 2602 are obtained by the polarization beam splitter 2301.
It branches to. The polarizing means 2302 is composed of, for example, a rotatable half-wave plate, and the ratio of the amount of light of P-polarized light and S-polarized light that is transmitted according to the rotation angle of the half-wave plate is 1: 0.
To 0: 1. The light branched into the optical path 2601 is the beam forming optical system 20.
1 is adjusted, the beam diameter is adjusted and the illuminance distribution is adjusted, bent by the mirror 202 and incident on the coherence reduction optical system 203, and the temporal and spatial coherence is reduced. The illumination light emitted from the coherence reduction optical system 203 is deformed by the modified illumination optical system 20 in the illuminance distribution at the pupil position of the objective lens 10, and the S polarization component is reflected by the polarization beam splitter 27 toward the objective lens 103. The wafer 11 is irradiated through the light modulation unit 21 and the objective lens 103. Hereinafter, the illumination light irradiated in this optical path, that is, the illumination light irradiated through the objective lens 103 is referred to as bright field illumination.

In this way, the light branched into the optical path 2601 is converted into the objective lens 103 by the modified illumination optical system 20.
By changing the illuminance distribution of illumination light at multiple pupil positions to multiple types, illumination that can be applied to various process wafers is performed. This may be, for example, a filter in which the light transmittance is changed in the cross section of the optical axis, an optical element that forms four or eight light beams arranged symmetrically about the optical axis, and an element that can oscillate the beam. May be used to change the beam position. The element that can oscillate the beam is, for example, a galvanometer mirror or a semiconductor resonant mirror. These can be switched.

On the other hand, the light branched into the optical path 2602 by the beam splitter 2301 passes through the coherence reduction optical system 203, then enters the polarization dark field illumination optical system A24, and is branched into two optical paths by the partial mirror 2401, and one of them is an optical element. 2403 and 2405 are transmitted into the polarization dark field illumination optical system B25, and the other is reflected by the total reflection mirror 2402 to be optical elements 2403 and 240.
5 enters the polarization dark field illumination optical system B25. Respective lights entering the polarization dark field illumination optical system B25 are transmitted through the optical elements 2501 and 2502, and mirrors 2503 and 2504.
And the surface of the wafer 11 is irradiated from an oblique direction.

Of the reflected light of each light irradiated on the wafer 11 through the optical paths 2601 and 2602, the light condensed by the objective lens 103 passes through the light modulation unit 21, the polarization beam splitter 27, and the light modulation unit 22. Imaged on the detection surface of the image sensor 104,
The formed optical image is detected by the image sensor 104, and the detection signal is an A / D converter 105.
Is converted into a digital signal and output from the detector 13. Here, the image sensor 104
The plurality of detection signals are output in parallel, and the plurality of detection signals output in parallel by the A / D converter 105 are A / D converted and output in parallel.

Here, the light modulation unit 21 is for controlling the amount and phase of illumination light and reflected light from the wafer 11. For example, the contrast of the circuit pattern signal detected by the image sensor 104 is improved by adjusting the light quantity ratio between the 0th order diffracted light and the higher order diffracted light reflected from the wafer 11. Alternatively, the contrast of the circuit pattern is improved by polarization differential interference.
Here, in order to adjust the light quantity ratio between the 0th-order diffracted light and the higher-order diffracted light, the 1/2 wavelength plate and 1 /
This can be realized by changing the vibration direction of light by combining four wavelength plates. Polarization differential interference can be realized by using a birefringent prism. 1 Nomarski prism
The physical phenomenon by the polarization differential interference optical system used is the same as that of a general differential interference microscope. These can also be switched.

The light modulation unit 22 is installed at a position conjugate with the pupil position of the objective lens 103, and performs optical modulation at the pupil position. For example, an object having a dielectric film deposited on the center of a transparent substrate such as quartz is installed, and the light detected by the image sensor 104 is modulated by changing the transmittance of the dielectric film portion. Instead of the dielectric film, a unit shielded from light with metal or the like may be used. These can also be switched.

Further, the polarization dark field illumination optical system 24 and the polarization dark field illumination optical system 25 are provided with the objective lens 10.
3 is irradiated with illumination light from the outside. The optical path 2602 that has entered the polarization dark field illumination optical system 24 includes a partial mirror 2401 that branches off the optical path and a total reflection mirror 2402. On the other hand, the illumination light branched to the optical path 2601 by the optical path branching optical system 23, that is, the illumination light irradiated through the objective lens 103 is called bright field illumination.

The optical modulation unit, the modified illumination, and the dark field illumination as described above have functions such as simultaneous illumination with the bright field illumination, so that the optimum optical system can be selected for the inspection object and the optimum inspection can be performed.

  The image comparison processing unit 15 is also configured by a plurality of processing methods corresponding to the detection system.

FIG. 3 shows the configuration of the image comparison processing unit 15. The image comparison processing unit 15 has a plurality of methods (1501 to 1503 in the example shown in FIG. 3) for each function such as image editing, position correction coefficient, and brightness correction coefficient calculation method. When the combination of
The optimum combination of each function corresponding to it and its processing order are determined. For example, in the image comparison processing method 1501, defect candidates are extracted at 108-1 and 109-1, and defects are detected and classified at 111-1. At this time, the value corresponding to the image processing parameter is set to 110-1.

In FIG. 26, as an example of the optimal combination of each function and the processing order, calculation of position correction coefficient by correlation coefficient → edge height direction smoothing (subtle difference in height direction of sample 11 exists in image In this example, comparison processing is performed in the order of image editing by smoothing the signal amount in the case of image processing → calculation of brightness correction coefficients using contrast and brightness. There are various combinations and calculation orders such as batch calculation of coefficients and brightness correction coefficients. The same applies to the classification.

In this way, high-sensitivity inspection is performed by imaging the specimen under the optimal detection conditions selected from multiple detection systems and performing image comparison based on the corresponding image comparison method and defect classification method. As well as realizing various defects.

Here, the selection method of the optimum detection system will be described. First, as shown by reference numeral 801 in FIG. In the case of a semiconductor wafer, a process, a target area (for example, a memory mat portion), and a defect to be found are known as a target defect and its coordinates are taught. Then, the detection system is switched to capture the image of the area of interest and the defect of interest, and the contrast of the image, brightness value, pattern direction, pattern density, difference in brightness value between the defect and its surroundings (S / N), etc. Calculate and display them side by side on the screen as shown in FIG. FIG. 9A is an example showing an image, contrast, contrast difference, luminance distribution, and S / N by each detection system of the designated defect of interest. FIG.
b) is an example showing an image, contrast, brightness, differential value, and luminance distribution by each detection system of the area of interest. Thereby, one or more conditions are selected. The condition can be selected by the user looking at the displayed image or evaluation value as shown in FIG. 9, but can also be automatically selected from the evaluation value. Then, according to the selected detection system, an inspection is performed by a comparison method and a classification method corresponding to each detection system as indicated by 802 in FIG.

FIG. 10 is a flow of a procedure for selecting a detection system from a more detailed evaluation. First, a detection system is set (S101), an image is captured (S102), an image comparison method corresponding to the detection system is set (S103), and a test inspection is performed (104). This is done with several types (or all types) of detection systems. If the defect to be detected is found in any of the detection systems, the coordinates are designated (S105). The user also designates the coordinates of an area to be inspected with high sensitivity, such as a memory mat portion (S106). If the defect that is the detection target is known in advance,
This test inspection need not be performed. Then, images of defects and regions designated by each detection system are acquired (S107 to S109), the quantitative evaluation values shown in FIG. 9 are calculated, and the results are displayed (S
110). On the other hand, the test inspection is performed by the image comparison method corresponding to the set detection system, and the result is displayed (S111). And both, that is, quantitative evaluation values of images by different detection systems,
One or more optimum conditions are selected in consideration of detection performance based on different image comparison methods (S112).
). As a selection criterion, the detection results that include the quantitative evaluation value of the image and the performance of the image comparison method may have the same weight, or one of them may be weighted. Thereby, an optimal detection system is selected (S113), and an image comparison method and a classification method corresponding to the detection system are selected (S11).
4) The inspection is performed under all the selected inspection conditions (S115), and when all the selected conditions are completed (S116), the result is displayed in a map format on the display screen (S117).

FIG. 11 shows one or more selected detection systems and the corresponding detection processing and classification results according to the comparison processing method and the classification method. Test results obtained by a plurality of detection methods can be individually displayed for each detection method as in 1101, and logical products and logical sums of detection results obtained in the respective detection methods are obtained as in 1102. It is also possible to synthesize and display the results. The detection result may be displayed on the map as it is or not, but by displaying the classification result on the map as shown in 1101 and 1102, the user can best detect the target defect under any condition. You can see at a glance.

Next, an example of processing in the image comparison processing unit 15 will be described in detail with reference to FIG. First,
A detection image signal and a reference image signal that are continuously input to the memory 107 in synchronization with the movement of the stage are read out. These two-chip image signals do not become signals at exactly the same location due to the vibration of the stage and the inclination of the wafer set on the stage. For this reason, the correction amount calculation unit 108 normally calculates the amount of misalignment between two images (S1081). The calculation of the amount of displacement is performed sequentially with a specific length in the direction of stage movement as one processing unit. 51, 52 of FIG.
... Each processing area when the length D (pixel) is set as one processing unit. Hereinafter, the processing area of this unit is described as a unit.

In this way, the amount of positional deviation is calculated by the unit 51 and the unit of the adjacent chip corresponding thereto, and then the amount of positional deviation is calculated by the unit 52 and the unit of the adjacent chip corresponding thereto.
As described above, the positional deviation amount is sequentially calculated for each input image in units. There are various methods for calculating the misregistration amount, such as a normalized cross-correlation between images, a sum of shade differences between images, a square sum of shade differences between images, and the like. Then, based on the calculated displacement amount, the two images are aligned in units (S1082). Next, for two images that have been aligned, a correction amount for adjusting the brightness deviation is calculated in two stages. Factors that cause the brightness deviation include a subtle difference in film thickness between chips of a semiconductor wafer and fluctuations in the amount of illumination light. Brightness shifts due to subtle differences in film thickness occur depending on the pattern of the semiconductor wafer. Therefore, in the present invention, first, a correction amount for uneven brightness that occurs in a spatially continuous area depending on the pattern is calculated (S1083), and then a correction amount for uneven brightness that occurs randomly is calculated. (S1084). Image comparison is performed using the correction amounts calculated hierarchically and image processing parameters such as threshold values set according to each detection condition by the parameter setting unit 110 (109).

FIG. 12 shows the flow of processing for adjusting the difference in brightness generated in units of regions. A spatially continuous region is extracted from the detected image 1210 after alignment and the reference image 1220 (1
201) A brightness correction amount is calculated for each extracted area (1202), and brightness correction is performed for each area based on the calculated result (1203).

FIG. 28 shows an example of the behavior when the brightness is corrected hierarchically according to the present invention. The detected image (a) has two defects 2701 (a part surrounded by a circle), but the defect 2701 is brighter than the band-shaped background pattern and can be distinguished from the background pattern. However, since the corresponding pattern 2702 of the reference image (b) has the same brightness as the defect, if the difference between the two images is simply taken, the difference between the defective portions becomes small as in the difference image (c). If a scatter diagram (d) of the brightness of the detected image and the reference image is created, it cannot be distinguished from a region without uneven brightness. Scatter chart (
In d), the brightness is plotted with the detected image on the vertical axis and the reference image on the horizontal axis. On the other hand, in the present invention, a partial image is extracted from the target image according to the pattern or area as in 2702, and the brightness correction amount is calculated for each area (FIG. 12-1220). An example of a correction amount calculation method is shown in (Equation 1).

Here, F (i, j) and G (i, j) are the detected image after alignment and the position (i, j) of the reference image, respectively.
It is the brightness at. Then, the correction amount is calculated as in (Equation 2),

The region is corrected as shown in (Equation 3) with respect to the reference image (1203).

Here, as shown at 1202 in FIG. 28, a histogram of shade differences is obtained for each region (120
2-1) An offset amount such that the peak position of the histogram is 0 may be calculated and corrected as a correction amount for the region (1202-2).

Next, a correction amount for adjusting a difference in brightness that occurs randomly depending on the pattern of the semiconductor wafer is calculated for each pixel. An example of the procedure is shown in FIGS.

First, FIG. 14 shows a processing flow. For the image whose brightness has been corrected for each region described in step 1203 of FIG. 12, the feature value of each pixel is calculated and mapped to an N-dimensional feature space with the feature value as an axis (14- 1) ((f) in FIG. 28) Next, the feature space is divided into segments (14-2), and a correction amount is calculated for each divided segment (14-3).
.

Next, FIG. 15 shows processing based on the processing flow shown in FIG. 1501, 15 of (a)
Reference numeral 02 denotes a reference image and a detection image corrected by 1203, but uneven brightness occurs due to repeated dot patterns. Thereby, as shown in the scatter diagram (b), the distribution of data is widened, and defects cannot be detected. For these images, the feature quantity of each pixel is calculated, and FIG.
) To the feature space (corresponding to 14-1 in FIG. 14). FIG. 15C shows an example of a two-dimensional feature space. FIG. 28F shows an example in which each pixel is mapped to an N-dimensional feature space. The feature amount may be anything that indicates the feature of the pixel, such as the contrast, brightness, second-order differential value, contrast difference between corresponding pixels, and a variance value using neighboring pixels.
Mapping may be performed in a space using all feature quantities, or feature quantities effective for defect determination may be selected and mapped. Next, the feature space is divided into a plurality of segments (14 in FIG. 14).
-2) For each segment, the correction amount is calculated using the statistics of the pixels belonging to the segment (14-3 in FIG. 14). 15D and 15E are scatter diagrams generated by extracting the pixels included in the segment divided in the feature space from the scatter diagram of the entire image of FIG. 15B, and (
d) The correction amount is calculated from the scatter diagram of each segment as in (e). As a result,
As shown in the density difference image in FIG. 28E, the brightness unevenness of the dot pattern is corrected, and only the defect can be detected.

FIG. 16 shows an example of a dividing method for dividing the feature space of FIG. 15C into segments. In the present invention, segment division is automatically performed according to the target image. The upper graph in FIG. 16A is an example of a feature space based on the difference between brightness and lightness, and the lower graph is a histogram showing the frequency of each brightness (luminance value). It is an example determined from the brightness histogram of the target image.

FIG. 17 shows an example of the processing procedure. First, a histogram of luminance values in the target area is calculated (17-1). This may be calculated from the detected image or the reference image, or may be calculated from the average value of the two images. Next, the luminance histogram is smoothed, small peaks and the like are removed (17-2), and a differential value of the smoothed histogram is calculated (17-3). Next, looking at the differential value from the one with the smallest brightness, the luminance value that becomes positive becomes Start, and the luminance value that becomes negative next becomes End (17-4). Then, the luminance value having the maximum differential value within the range from Start to End is set as a threshold value for division (17-5). Thereby, as shown in FIG. 16A, the division is performed at the valley portion of the histogram. This indicates that segment division is performed according to the pattern in the target region. In this way, it is possible to divide according to the pattern in the image, but it is also possible to divide by a fixed value set by the user as shown in FIG.

FIG. 27 shows an example of the behavior of the correction amount calculation processing for each pixel shown in FIG. 14 using a scatter diagram. There is a large brightness difference between the detected image (a) and the reference image (b), and the difference becomes large as shown in the difference image (c). (D) is a scatter diagram according to (a) and (b). By disassembling this as a feature quantity as shown in (e), calculating a correction amount for each scatter diagram, and performing correction, the scatter diagrams in (a) and (b) have a data spread as shown in (f). Is suppressed, and the difference becomes smaller as shown in the corrected difference image (g).

As shown in FIGS. 15 (d) and 15 (e), the correction amount of each scatter diagram decomposed by segmental division is obtained by a least square approximation in the scatter diagram, and the slope and y intercept are used as the correction amount. To do. Further, the correction amount may be obtained from the pixels belonging to each segment as in the above (Equation 1) and (Equation 2). In addition, the area for forming the feature space can be arbitrarily set to a minimum of 1 × 1 pixel or more. However, if correction is performed with 1 × 1 pixel, which is the highest frequency, defects are also combined, so a slightly larger area is set.

As described above, (1) by correcting the brightness for each region, the defect buried in the uneven brightness of the background is revealed, and (2) the brightness generated randomly by correcting the brightness for each pixel. Only defects can be detected by hierarchical brightness correction that suppresses unevenness (FIG. 2).
8- (e)).

FIG. 18 shows the brightness correction result by this method. (A) is a difference image after alignment. There is a defect in the circled area. (D) is a luminance waveform of two images after alignment of 1D-1D ′ including a defective portion and a difference value at that time. The detected image has a defect, and the portion is brighter than the surroundings, but the reference image is brighter as a whole, and the difference value of the defective portion is smaller than the surroundings. On the other hand, (b) is a difference image after correcting the difference in brightness for each region. Band-shaped brightness unevenness is extracted as a region and corrected. As a result,(
As shown in e), the luminance value is adjusted, and the defect buried in the background color irregularity becomes obvious. However, depending on the repetitive pattern, the brightness difference that occurs randomly cannot be corrected.

(C) is a difference image obtained by calculating a correction amount for each pixel from the statistics of the entire image area of (b) in this example and correcting a difference in brightness. In this way, by calculating the brightness correction amount in a hierarchical manner using two or more different processing methods, the difference in brightness with different occurrences is corrected, and the defect of the weak signal buried in the strong brightness unevenness Can be realized and detected.

The present invention can be applied even when there is a stronger brightness unevenness. FIG. 19A is a detected image, and FIG. 19B is a reference image, but the background 1901 has uneven brightness, and a difference image (c)
Then, the difference of the part corresponding to 1901 is large. The distribution of the scatter diagram data according to (a) and (b) spreads as shown in (d) in a part of the brightness region. On the other hand, the positional deviation is shown in FIG.
As shown in Fig. 5, normally, the correction is performed before the brightness unevenness is corrected. Therefore, if the brightness shift amount is large, there is a possibility that the correct position shift amount cannot be calculated. On the other hand, if there is a position shift, pixels cannot be matched between images, and a correct brightness correction amount cannot be calculated. For this reason, the present invention has means for simultaneously adjusting the positional deviation and the brightness deviation.

FIG. 20 is an example in which there is a further uneven brightness compared to FIG. Base 2001 is (a)
Is brighter, but the pattern 2002 there is almost no difference in brightness. Further, in the upper pattern 2003, a brighter one and a darker one are mixed in (a), and light and dark are reversed with respect to (b). Such a scatter diagram of images (a) and (b) has a positive correlation and a negative correlation as shown in (d). On the other hand, the correct misregistration amount cannot be calculated by the method of calculating the sum of squares of the shading difference that calculates the misregistration amount so that the difference in brightness is minimized. Even if the normalized correlation is robust against uneven brightness, the correct misregistration amount cannot be calculated for an image in which a positive correlation and a negative correlation are mixed. FIG. 20 (e) shows the correlation coefficient between (a) and (b), but the value is small as a whole, and it is impossible to calculate the amount of misalignment uniquely.

For this reason, in the present invention, as shown in FIG. 21, the position correction amount is shifted to the periphery ((αi, βi)) around the position shift amount (α, β) of the previous unit. , Position correction (S211) → Brightness unevenness correction amount calculation (S212) → Brightness adjustment (S213) → Scatter chart evaluation (S214) And scatter diagram data (a-1), (a
The inspection is performed by applying the one with the smallest variation (a-i) among -i) and (a-n). Scatter chart data variation is calculated as the sum of squares of differences between pixels.

At this time, as shown in FIG. 29, first, as shown in FIG. 15, the scatter diagram can be decomposed from the feature amount of each pixel, and the processing of FIG. 21 can be repeated for each scatter diagram. That is, while correcting the positional deviation amount for each scatter diagram ((αi, βi)), position correction (2901) → brightness unevenness correction amount calculation (2902) for the corresponding pixel → brightness correction (2903) → scatter diagram evaluation (2904) ) Is repeated, and the amount of positional deviation ((αi, βi) and brightness correction amount when the spread of the scatter diagram data becomes the smallest, that is, the sum of the differences of the corresponding pixels between images is minimized is applied (2905).
).

As another method, as shown in FIG. 22, a differential filter is applied to the detected image 2201 and the reference image 2202 that are target images (S221), and the edge of the pattern is detected from the image (S222), which corresponds to the edge. The position correction amount (α, β) is calculated using only the pixels to be processed (S2
23) The position is corrected based on this (S224), and the image F ′ (2203) and the image G ′ (2
204). As a result, it is possible to calculate the amount of misregistration excluding the influence of uneven brightness, and the brightness village correction amount can be calculated and obtained using this result (S225).
When detecting an edge in (S222), a pixel having a steep brightness gradient between adjacent pixels, that is, a pixel having a large primary differential value, a zero-crossing pixel having a secondary differential value, or the like is selected.

Here, the selection of a pixel effective for calculating the positional deviation amount is not limited to the edge portion having a large brightness gradient. As shown in FIG. 32, for each pixel of the detected image 2201 and the reference image 2202, a difference from an adjacent pixel is calculated (S321), and a code array of the calculated difference values is calculated (S322).
. Here, if the image data is considered as a one-dimensional array for the sake of simplicity, the differences df (i) and dg (i) from the adjacent pixels of the pixels f (i) and g (i) of the detected image and the reference image are expressed as ( (4)

The array of codes cf (i) and cg (i) is obtained.

if df (i) ≧ 0 then cf (i) = 1 else cf (i) = 0
if dg (i) ≧ 0 then cg (i) = 1 else cg (i) = 0
Then, a pixel whose code arrangement is close (originally because there is a positional shift between images) is selected and a position correction amount (α, β) is calculated (S323), and position correction is performed (S323). S324
). As a result, it is possible to perform position correction by eliminating a pattern in which the brightness is partially inverted.

Further, as an index of pixel selection effective for calculating the positional deviation amount, as shown in FIG.
201, for each pixel f, g of the reference image 2202, the n neighboring pixels f (i), g (i) are calculated from the variance values fs, gs as in (Equation 5) (S331),

It is also possible to select pixels that are close to each other, calculate the position correction amount (α, β) (S333), and perform position correction (S334). here,

Is the average luminance of n neighboring pixels.

Although the example of the pixel selection method for calculating the positional deviation amount has been described above, other selection indices may be used. The present invention realizes highly accurate calculation of a positional deviation amount by selecting a pixel by any index, in other words, by eliminating pixels that adversely affect the calculation of the positional deviation amount.

As another method, as shown in FIG. 23, a detected image 2301, a reference image 2302 are provided.
Each region is extracted (S231), and a position correction amount is calculated for each corresponding region (S232).
, Position correction (S233), brightness unevenness correction amount calculation (S234), brightness unevenness correction (S23)
5) is performed. As a result, it is possible to calculate the correction amount of the position / brightness deviation by dividing into a portion having a positive correlation and a portion having a negative correlation.

As another method, as shown in FIG. 24, a detected image 2401 and a reference image 2402 are used.
Each region is extracted (S241), the position correction amount and the brightness correction amount are assigned as parameters for each corresponding region, and the position correction amount and the brightness correction amount that minimize the square sum of the differences are applied (
S242).

As another method, as shown in FIG. 30, a detected image 3001 and a reference image 3002 are used.
Each region is extracted and the scatter diagram is decomposed by the feature amount for each pixel (S301). Then, the position correction amount and the brightness correction amount are assigned as parameters for each scatter diagram of the image decomposed by the region information and the feature amount, and the position correction amount and the brightness correction amount that minimize the square sum of the differences are applied (S30).
2).

As another method, as shown in FIG. 25, the scatter diagram (a) is Hough transformed to separate the distribution data into a negative correlation and a positive correlation (b) (c) (d). Thus, the position correction amount and the brightness correction amount are calculated so that the dispersion of the scatter diagram data is minimized for each separated distribution data.

As described above, the effect of performing multi-stage brightness correction on an inspection image can be achieved by, for example, performing a CMP process on a semiconductor wafer whose surface is optically transparent and covered with a flat insulating film. A case of inspecting the formed pattern will be described. Detection unit 1 detects wafer after CMP
The image obtained by imaging 3 is affected by the variation in the thickness of the insulating film within the wafer surface, the distribution of the amount of reflected light caused by the density of the pattern in the chip, etc., and the brightness varies depending on the location of the wafer. The image will vary. By correcting the brightness of an image with this brightness variation in multiple steps in different units using the method described above, the effect of brightness variation between images is reduced to reveal defects. Therefore, the defect detection rate can be improved.

In addition, brightness is adjusted by using different methods to compensate for differences in brightness (color unevenness) between chips due to various factors such as differences in film thickness between chips, differences in accumulated light intensity due to uneven stage speed, and illumination fluctuations. By calculating the correction amount for each of a plurality of different areas (ie, calculating the correction amount), it becomes possible to reveal and detect the defect of the weak signal buried in the strong brightness unevenness.

The processing of the image comparison processing unit 15 according to the present invention described so far is realized by software processing by the CPU, but the core calculation part such as normalization correlation calculation and feature space formation is made hard processing by LSI or the like. Is also possible. As a result, high speed can be realized. CM
Even if there is a subtle difference in the film thickness of the pattern after the flattening process such as P, or a large difference in brightness between the comparison dies due to the shorter wavelength of the illumination light, the present invention can detect defects of 20 nm to 90 nm. It becomes.

Furthermore, inorganic insulating films such as SiO ₂ , SiOF, BSG, SiOB, porous Syria film, methyl group-containing SiO ₂ , MSQ, polyimide-based film, parelin-based film, Teflon (registered trademark) -based film, amorphous In the inspection of a low-k film such as an organic insulating film such as a carbon film, even if there is a local brightness difference due to intra-film variation in the refractive index distribution, the present invention can detect a 20 nm to 90 nm defect.

Next, FIG. 3 shows an example of a semiconductor device manufacturing system using the pattern inspection apparatus according to the present invention.
1 will be used for explanation. A circuit pattern on the wafer is sequentially formed by a plurality of apparatuses A, B,... E (3101). Then, by performing inspection and defect classification after the process by each manufacturing apparatus using the pattern inspection apparatus according to the present invention, it becomes possible to trace what kind of defect occurs through each apparatus. For example, the wafer that has undergone the process by the apparatus B is inspected, and the detected defect is composed of a carry-in defect from the previous apparatus A and a defect generated in the apparatus B.
By comparing with defect data obtained by inspection after the process by the apparatus A, it can be determined whether the defect has occurred in the apparatus B or a defect brought in from the apparatus A (3102). Here, among the carry-in defects from the previous apparatus A, there are some foreign objects that do not cause a shape defect or discoloration in the apparatus B. However, the pattern with non-fatal foreign matter attached does not become a shape defect even after the following equipment,
It is a good product. Accordingly, the coordinates of defects classified as foreign substances by the pattern inspection apparatus having this configuration are stored, and after passing through the next apparatus, the defect detection and classification are performed again so that the foreign substances are fatal. It becomes possible to determine whether or not. This makes it possible to grasp the state of the manufacturing apparatus more accurately.

As described above, the embodiment of the present invention has been described by taking the comparative inspection image in the optical appearance inspection apparatus for the semiconductor wafer as an example, but it can also be applied to the comparative image in the electron beam pattern inspection. Further, the inspection object is not limited to a semiconductor wafer, and a TFT substrate, a photomask, a printed board, or the like can be applied as long as defect detection is performed by comparing images.

An example of a structure of an inspection apparatus. An example of a multiple detection system. An example of a multiple image comparison method. An example of an inspection object image when there is uneven brightness between comparison chips and a conventional threshold value setting method. Example of image comparison processing unit Semiconductor wafer to be inspected. An example of the structure of a chip. Example of detection system and comparison processing method optimization. An example of image quantitative value display of each detection system. An example of a detection system setting flow. An example of test result display using multiple detection systems, comparison, and classification methods. An example of a procedure for adjusting brightness unevenness generated in units of areas. An example of the processing procedure of the correction amount calculating unit. An example of a procedure for adjusting brightness unevenness that occurs at random. Example of adjusting brightness irregularities that occur randomly. An example of segmentation of a feature space. An example of the segmentation process flow of feature space. An example of processing for adjusting the difference in brightness. An example of strong brightness unevenness. An example of strong brightness unevenness. An example of alignment and brightness adjustment for strong uneven brightness. An example of alignment and brightness adjustment for strong uneven brightness. An example of alignment and brightness adjustment for strong uneven brightness. An example of alignment and brightness adjustment for strong uneven brightness. An example of the process which divides | segments an image by Hough transformation from a scatter diagram. An example of image comparison method optimization. An example of a procedure for adjusting brightness unevenness for each pixel. An example of a procedure for adjusting stepwise brightness unevenness. An example of alignment and brightness adjustment for strong uneven brightness. An example of alignment and brightness adjustment for strong uneven brightness. The figure which shows the semiconductor manufacturing system by this invention. An example of alignment and brightness adjustment for strong uneven brightness. An example of alignment and brightness adjustment for strong uneven brightness.

DESCRIPTION OF SYMBOLS 11 ... Sample, 12 ... Stage, 13 ... Detection part, 101 ... Light source, 102 ... Illumination optical system, 10
DESCRIPTION OF SYMBOLS 3 ... Objective lens, 104 ... Image sensor, 105 ... AD conversion part, 14 ... Image editing part, 1
06 ... Pre-processing unit, 107 ... Image memory, 15 ... Image comparison processing unit, 108 ... Correction amount calculation unit,
DESCRIPTION OF SYMBOLS 109 ... Image comparison part 110 ... Parameter setting part 111 ... Defect classification part 16 ... Overall control part 112 ... User interface part 113 ... Memory | storage device 114 ... Mechanical controller 70 ... Chip | tip 72 ... Peripheral circuit part, 71 ... Memory mat part, 41 ... Detection image, 4
2 ... Reference image

Claims (6)

In a plurality of inspection conditions in which the setting of the illumination light quantity ratio between the illumination light from the oblique direction and the illumination light from above is different in the corresponding areas of the two patterns formed to be the same shape pattern on the sample Illuminating and capturing a plurality of images by imaging an illumination area for each illumination under the plurality of conditions , processing the plurality of images using an image processing method corresponding to each inspection condition , and processing the plurality of processed classifying detected defects by comparison of images, a Rupa turn inspection method to display the results of the classification on a screen,
The plurality of image processing methods use a plurality of image comparison processing methods corresponding to the illumination light quantity ratio and a plurality of defect classification methods in combination, respectively. The pattern inspection method according to claim 1,
The pattern inspection method is characterized in that the image processing method is processed using a method selected from a plurality of preset image processing methods. The pattern inspection method according to claim 1,
A pattern inspection method, wherein the illumination light quantity ratio is adjusted according to a sample type, a pattern density, a pattern direction, or a type of defect to be detected. A pattern inspection method according to any one of claims 1 to 3 ,
A pattern inspection method, wherein a combination of the image comparison processing method and the defect classification method is selected . The pattern inspection method according to claim 1,
A pattern inspection method comprising adjusting an illumination light amount ratio between an illumination light amount from the oblique direction and an illumination light amount from above in a range of 1: 0 to 0: 1. An oblique illumination unit that irradiates a corresponding region of two patterns formed to be the same pattern on the sample from an oblique direction, an epi-illumination unit that illuminates from above, and the oblique illumination unit and the epi-illumination unit. A lighting means having a light quantity ratio adjustment unit for adjusting the illumination light quantity ratio;
Imaging optical system means for forming an image of reflected light from the area illuminated by the illumination means;
An imaging unit that captures an image formed by being illuminated by each of the illumination units of the illumination unit;
An image picked up by the image pickup means is processed using an image processing method corresponding to the light amount ratio adjusted by the illumination light amount ratio adjustment unit of the illumination means from among the plurality of image processing methods. Image processing means for detecting and classifying defects,
Display means for displaying a result processed by the image processing means,
The image processing means processes a plurality of images obtained by illuminating under a plurality of conditions with different settings of the illumination light quantity ratio using the image processing method corresponding to each inspection condition, and the processed plurality of images a pattern inspection apparatus characterized in that it comprises means for comparing the image,
The plurality of image processing methods use a plurality of image comparison processing methods corresponding to the illumination light quantity ratio and a plurality of defect classification methods in combination .

Images