Pattern inspection method and pattern inspection device

The present invention relates to an inspection for comparing an image of an object obtained by using light or a laser with a reference image and detecting fine pattern defects, foreign substances, etc. based on the comparison result, particularly a semiconductor wafer. The present invention relates to a pattern inspection device suitable for performing visual inspection of TFTs, photomasks, etc., and a method thereof.

As a conventional technique for detecting a defect by comparing an image to be inspected with a reference image, the method described in Japanese Patent Application Laid-Open No. 05-264467 (Patent Document 1) is known.

In this method, a sample to be inspected in which repeated patterns are regularly arranged is sequentially imaged by a line sensor, compared with an image delayed by the time of the repeated pattern pitch, and the inconsistent portion is detected as a pattern defect. .. Such a conventional inspection method will be described by taking the appearance inspection of a semiconductor wafer as an example. As shown in Fig. 2 (a), a large number of chips with the same pattern are regularly arranged on the semiconductor wafer to be inspected. As shown in FIG. 2B, each chip can be roughly divided into a memory mat section 201 and a peripheral circuit section 202. The memory mat unit 201 is a set of small repeating patterns (cells), and the peripheral circuit unit 202 is basically a set of random patterns. Generally, the memory mat portion 201 has a high pattern density, and the image obtained by the bright field illumination optical system becomes dark. On the other hand, the peripheral circuit unit 202 has a low pattern density, and the obtained image becomes bright.

In the conventional visual inspection, the peripheral circuit unit 202 compares the images in the same position of the adjacent chips, for example, the region 21 and the region 22 in FIG. 2, and detects the portion where the brightness difference is larger than the threshold value as a defect. To do. Hereinafter, such an inspection will be referred to as a chip comparison. The memory matte unit 201 compares the images of adjacent cells, and similarly detects a portion where the brightness difference is larger than the threshold value as a defect. Hereinafter, such an inspection will be referred to as a cell comparison.

<patcit num="1"><text>Japanese Patent Application Laid-Open No. 05-264467</text></patcit>

<p> In the semiconductor wafer to be inspected, the flattening of CMP or the like causes a slight difference in the film thickness of the pattern, and the images between the chips have a local difference in brightness. If a portion where the brightness difference is equal to or greater than a specific threshold value TH is defined as a defect as in the conventional method, a region having different brightness due to such a difference in film thickness is also detected as a defect. This should not be detected as a defect by nature. In other words, it is false news, but in the past, as one method to avoid the occurrence of false news, the threshold value for defect detection was increased. However, this lowers the sensitivity, and defects with a difference value of the same or less cannot be detected. In addition, the difference in brightness due to the difference in film thickness may occur only between specific chips in the wafer among the array chips shown in FIG. 2, or may occur only in a specific pattern in the chip. If the threshold is set to the local area of, the overall inspection sensitivity will be significantly reduced.</p><p> In addition, as a factor that hinders the sensitivity, there is a difference in brightness between chips due to a variation in the thickness of the pattern edge. Figure 6 shows the cross-sectional waveform of the brightness of the two chips to be compared at the same position. Brightness varies between chips. In the conventional comparative inspection based on brightness, if there is such a brightness variation, it becomes noise at the time of inspection.</p><p> On the other hand, there are various types of defects, which can be roughly divided into defects that do not need to be detected (those that can be regarded as noise) and defects that should be detected. In the visual inspection, it is required to extract only the defects desired by the user from a huge number of defects, but it is difficult to realize by comparing the above luminance difference with the threshold value. On the other hand, the appearance often changes depending on the type of defect depending on the combination of factors such as material, surface roughness, size, and depth that depend on the inspection target and factors that depend on the detection system such as lighting conditions.</p><p> An object of the present invention is a pattern inspection that solves such a problem of the conventional inspection technique, compares images of corresponding regions of patterns formed to have the same pattern, and determines a non-matching portion of the image as a defect. In the device, it is not necessary to detect noise or detection by reducing the uneven brightness between comparative images caused by the difference in film thickness and the thickness of pattern edges, and by changing the sensitivity according to the type of defect. The purpose is to realize a pattern inspection that detects defects buried in defects and desired by the user with high sensitivity.</p>

<p> In order to achieve the above object, in the present invention, there is a difference in film thickness in a pattern inspection device that compares images of corresponding regions of patterns formed to have the same pattern and determines a mismatched portion of the image as a defect. The effect of uneven brightness between comparative images caused by differences in the thickness of patterns and pattern edges has been reduced, enabling highly sensitive pattern inspection.</p><p> Further, in the present invention, in the pattern inspection apparatus, the feature amount of each pixel is calculated between the comparative images, and the outliers in the feature space are set as defect candidates, so that various defect types can be dealt with. Made it possible to perform sensitive pattern inspection. Further, in the present invention, the defect species to be detected can be adjusted by forming a feature space from the feature quantities selected from a plurality of feature quantities.</p><p> Furthermore, the user teaches outliers that he does not want to detect so that similar outliers are not detected. As a result, even if there is a difference in brightness in the same pattern between images due to a difference in pattern line width, it is possible to detect only a desired defect type from various defect types with high sensitivity. I did.</p><p> In addition, by instructing the user that there are no defects, the threshold value for detecting outliers is automatically set so as to include all distribution points in the feature space. This simplifies the setting of inspection conditions and makes it possible to detect defects other than those taught with high sensitivity.</p><p> In addition, by increasing the teaching, the threshold value was optimized and the sensitivity could be easily adjusted automatically.</p><p> According to the above invention, even when the inspection target is a semiconductor wafer and the difference in brightness occurs in the same pattern between images due to the difference in film thickness in the wafer, fatal defects are highly sensitive. Made it possible to detect.</p>

<p> According to the present invention, by interactively selecting the optimum feature quantity from a plurality of feature quantities to detect the defect type desired by the user, the desired defect can be highly sensitive from various defect types and noise. It becomes possible to detect.</p><p> Further, by teaching the defect type desired by the user and the pattern that the user does not want to detect, it is possible to easily set the sensitivity according to the defect type and the pattern.</p><p> Furthermore, noise due to variations in brightness can be tolerated by using the value calculated by converting the image to low bits as a part of the feature amount.</p>

Hereinafter, an embodiment of the present invention will be described in detail with reference to FIGS. 1 to 17.

As an example, a defect inspection method in an optical visual inspection apparatus for a semiconductor wafer will be described as an example. Figure 1 shows an example of the configuration of a bright-field illumination type optical visual inspection device. 11 is a sample (object to be inspected such as a semiconductor wafer), 12 is a stage on which the sample 11 is mounted and can move and rotate in the XY plane and move in the Z direction, and 13 is a detection unit. The detection unit 13 is focused by an illumination optical system 102 and an illumination optical system 102 including a light source 101 that irradiates the sample 11, a lens system 1021 that collects the light emitted from the light source 101, and a beam splitter 1022 that converts the light path. An objective lens 103 that illuminates the sample 11 with the emitted illumination light and forms an optical image obtained by reflecting the sample 11, an image that receives the formed optical image and converts it into an image signal according to the brightness. It is composed of an AD conversion unit 105 that converts an input signal from the sensor 104 and an image sensor 104 into a digital signal.

Here, as the light source 101, in the example shown in FIG. 1, a case where a lamp is used is shown, but a laser may be used. Further, the wavelength of the light emitted from the light source 101 may be a short wavelength, or may be light having a wide band wavelength (white light). When short wavelength light is used, light with a wavelength in the ultraviolet region (Ultra Violet Light: UV light) can also be used in order to improve the resolution of the image to be detected (detect fine defects). When a laser is used as a light source, if it is a single wavelength laser, a means for reducing coherence (not shown) is provided inside the illumination optical system 102 or between the light source 101 and the illumination optical system 102. There is a need.

In addition, a time delay integration type image sensor (TDI image sensor), which is composed of a plurality of one-dimensional image sensors arranged in two dimensions on the image sensor 104, is adopted and synchronized with the movement of the stage 12. By transferring the signals detected by each one-dimensional image sensor to the next-stage one-dimensional image sensor and adding them, it is possible to detect them at a relatively high speed and with high sensitivity. By using a parallel output type sensor equipped with multiple output taps as this TDI image sensor, the output from the sensor can be processed in parallel, enabling faster detection. Further, when a light source 101 that emits UV light is adopted, if a back-illuminated sensor is used for the image sensor 104, the detection efficiency can be improved as compared with the case where the front-illuminated sensor is used. ..

14 is an image editing unit, which stores a preprocessing 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, and a digital signal of the corrected image. It is composed of image memory 107.

Reference numeral 15 denotes an image comparison processing unit that calculates defect candidates in the wafer, which is a sample. The images in the corresponding areas stored in the image memory 107 of the image editing unit 14 are compared, and outliers are extracted by statistical processing. However, it is regarded as a defect. First, the digital signal of the image of the area to be inspected (hereinafter referred to as the detected image) stored in the image memory 107 and the image of the area corresponding to the image (hereinafter referred to as the reference image) is read out, and the position shift detection unit 108 determines the position. The correction amount for matching is calculated, the statistical processing unit 109 aligns the detected image and the reference image using the calculated position correction amount, and uses the feature amount of the corresponding pixel to obtain the statistical deviation value. Is output as a defect candidate. The parameter setting unit 110 sets image processing parameters such as a feature amount and a threshold value when extracting defect candidates, and gives them to the statistical processing unit 109. Then, the defect classification unit 111 extracts true defects from the feature quantities of each defect candidate and classifies them.

16 is an overall control unit, which is equipped with a CPU (built-in to the overall control unit 16) that performs various controls, and accepts and detects changes in inspection parameters (features used for extraction of outliers, threshold values, etc.) from the user. It is connected to a user interface unit 112 having a display means for displaying the detected defect information and an input means, and a storage device 113 for storing the feature amount and the image of the detected defect candidate. Reference numeral 114 denotes a mechanical controller that drives the stage 12 based on a control command from the overall control unit 16. The image comparison processing unit 15, the detection unit 13, and the like are also driven by commands from the overall control unit 16.

As shown in FIG. 2 (c), the semiconductor wafer 11 to be inspected has a large number of chips 200 having the same pattern including the peripheral circuit portion 202 and the memory mat portion 201, as shown in FIG. 2 (a). Lined up in. In the overall control unit 16, the semiconductor wafer 11 as a sample is continuously moved by the stage 12, and in synchronization with this, the images of the chips are sequentially captured from the detection unit 13 and regularly arranged with respect to the detected images. Pixels that are statistically outliers when the digital image signals of regions 21, 22, 24, and 25 are compared with the same position of the chip, for example, region 23 of the detected image in FIG. 2 (b) as a reference image in the above procedure. Is detected as a defect candidate.

FIG. 3 shows an example of the processing flow of the image comparison processing unit 15 for the image 23 of the chip to be inspected shown in FIG. 2 (b). First, the displacement of the reference image 32 (here, the image 22 of the adjacent chip) corresponding to the image of the chip to be inspected (detection image 31) is detected and the alignment is performed (303). .. Next, for each pixel of the detected image 31 that has been aligned, a plurality of feature quantities are calculated with the corresponding pixel of the reference image 32 (304). The feature amount may be any feature as long as it represents the feature of the pixel. Examples of this are (1) brightness, (2) contrast, (3) shading difference, (4) brightness variance value of nearby pixels, (5) correlation coefficient, and (6) brightness with nearby pixels. Increase / decrease, (7) Second-order differential value, etc. An example of these features is expressed by the following equation, where the brightness of each point of the detected image is f (x, y) and the brightness of the corresponding reference image is g (x, y). Brightness; f (x, y), or {f (x, y) + g (x, y)} / 2 (number 1) Contrast; max {f (x, y), f (x + 1, y), f (x, y + 1), f (x + 1, y + 1)}- --min {f (x, y), f (x + 1, y), f (x, y + 1), f (x + 1, y + 1)} (Equation 2) Shading difference; f (x, y) -g (x, y) (Equation 3)

<maths num="4"><img file="JP5028014B2_D0001.tif" /></maths>Then, a feature space is formed by plotting each pixel in a space centered on two or more of these features (305). Then, the pixels plotted outside the data distribution in this feature space, that is, the pixels having characteristic outliers, are detected as defect candidates (306).

FIG. 4 is a diagram showing an example of the feature space formation (306) in the flow diagram shown in FIG. 3, and FIG. 4 (a) shows the steps up to the feature space formation (306) in FIG. In b), as the feature space 40 as a result of mapping the feature amount, the feature amount is calculated from the pixels at the corresponding positions of the detected image 31 and the reference image 32, and the feature amount A and the feature amount B are used as axes. An example of a feature space formed by plotting each point in a two-dimensional space is shown. In the feature space 40, the points surrounded by the broken line are outside the dense data distribution and indicate the pixels with outliers. FIG. 4 (c) is a difference image showing the difference in brightness of each pixel of the detected image 31 and the reference image 32 with a value from 0 to 255 (256 gradations). Pixels with a small difference are displayed darker, and pixels with a large difference are displayed brighter. In Fig. 4 (c), in addition to the defect, the difference is large in the part of the normal pattern where the brightness differs between the two images (shown as uneven brightness). In the conventional method of detecting a portion where the difference in brightness between images is larger than the threshold value as a defect, these uneven brightness are also detected together with the defect. FIG. 4 (d) is a distance image showing the distance from the center of the dense data distribution in the feature space 40 with a value from 0 to 255 (256 gradations). In the feature space, only the defects that are outliers are brightened, and the uneven brightness is suppressed, indicating that only the defects are detected.

In this embodiment, it is possible to make the feature space N-dimensional with three or more dimensions. Figure 5 shows an example. FIG. 5 is a diagram showing another example of the feature space formation (306) in the flow diagram shown in FIG. 3, and FIG. 5 (a) shows the steps up to the feature space formation (306) in FIG. In 5 (b), as the feature space 50 as a result of mapping the feature quantities, the feature quantities are calculated from the pixels at the corresponding positions of the detected image 31 and the reference image 32, and N of the feature quantities are used as axes. An image diagram of a feature space formed by plotting each point in an N-dimensional space is shown. By detecting outliers in the N-dimensional feature space 50, it is possible to detect defects from a variety of noises more characteristically. FIG. 5 (c) is a difference image between the detected image 31 and the reference image 32, and the place where the difference between the defect and the one shown is large is a normal pattern such as uneven brightness. FIG. 5 (d) is a distance image in the feature space 50. By detecting characteristic deviation values in a space consisting of a plurality of feature quantities in this way, it is shown that noise of various normal patterns can be suppressed and only defects can be detected.

As described above, in this embodiment, N pieces are selected from a plurality of feature quantities to form a feature space, and characteristic outliers are detected as defect candidates, but the feature quantity is the feature or detection of noise that is desired to be suppressed. Select the most suitable one according to the characteristics of the defect type you want. An example of this is shown in (Equation 1) to (Equation 4), but as another example of the feature quantity, it has a feature quantity obtained by converting the brightness data into lower bits. Fig. 6 (a) is the difference image between the detected image and the reference image, (b) is the waveform of the brightness of the detected image at the position shown by the broken line in (a), and (c) is the position shown by the broken line in (a). It is a waveform of the brightness of the reference image of. (d) shows the waveforms of (b) and (c) in which the parts having different peak positions are superimposed. As can be seen from the waveform shown in (d), the brightness of a specific pattern differs between images, so the difference becomes large and it is detected as a defect.

FIG. 7 shows an example of applying a feature amount converted to a low bit to such an image. 71 in FIG. 7 (a) shows the brightness of a total of 9 pixels, that is, the pixel of interest and the pixel in the vicinity of 8. Figure 7 (b) of Fig. 7 (b) shows that the magnitude relationship of the brightness with the vicinity of 8 is converted to 0 or 1, that is, 1 bit data if it is brighter than the pixel of interest and 0 if it is darker than the pixel of interest. .. Then, the array of data converted in the vicinity of 8, that is, the value in which the codes are arranged clockwise from the 12 o'clock position, 11000110 is taken as the feature amount of the pixel of interest. In Fig. 7 (c) 73, the array of 1-bit × 8 elements is calculated for all pixels of the detected image and the reference image, and the features calculated by comparing the arrays with the corresponding pixels are shown on the vertical axis. It is a two-dimensional feature space with the feature on the horizontal axis. The part surrounded by is a defective pixel, but the pixel whose difference is large due to the variation in brightness is plotted in the part where the data distribution is dense.

FIG. 8 is an example of the processing flow of the image comparison processing unit 15 to which the feature amount calculated by the low bit conversion is added. After aligning the detected image 31 and the reference image 32 (303), the feature amount is calculated (304) and both images are subjected to low bit conversion (801). Then, the feature amount is calculated from the low bit value of the corresponding pixel (802). A feature space is formed by selecting a feature amount according to the defect type to be detected and the noise to be suppressed from a plurality of feature amounts calculated from a low bit value and a plurality of feature amounts calculated from the original luminance value (803). , Outliers are detected as defect candidates (804).

Figure 9 shows an example of the procedure for selecting the feature amount according to the type of defect to be detected and the type of noise to be suppressed. First, as a test inspection, outliers are detected by executing the flow described in Fig. 8 using the default features set in advance (901). The user confirms the image of the peripheral area of the pixel detected as an outlier on the monitor of the user interface unit 112 (902). At this time, the images in the corresponding areas of the reference image are also displayed side by side, so that the user can compare them. If it is determined that the detected outliers are the defects desired by the user, the selection of the feature amount is completed and the inspection is performed (903). If it is determined that the desired defect type has not been detected, the change in the feature amount from the user is accepted (904), and the formation of the feature space by the changed feature amount and the outlier detection are performed (901). Hereinafter, until the result is satisfactory for the user, the detection of outliers by reforming the feature space, the display of the result, and the change of the feature amount are repeated.

Fig. 10 shows the procedure for displaying the outlier detection result on the monitor of the user interface unit 112, confirming by the user (902), and selecting the feature amount. 1000 in Fig. 10 (a) is a part of the detection result screen displayed on the monitor. 1001 in it is a defect map showing the location of defects on the wafer. The inspected chips are shown brightly (here the central 5 chips) and the detected defects are plotted on it. 1002 is a feature space based on the peripheral area where defects are detected. On the feature space, the pixels judged to be outliers from the normal pixels are displayed in different colors, and the outlier area is displayed as a curved surface (curve if the feature space is two-dimensional) (1003). In addition, a list of defects such as the dimensions of the detected defects and all features is displayed (1004). When any of the defects in the defect map 1001, the feature space 1002, and the defect list 1004 is individually specified with the mouse, the reference image corresponding to the image around the outlier and the list of the feature amount are displayed (1005). The user selects the condition setting button 1006 with the mouse if the outliers are not the desired defect type. When the condition setting button 1006 is pressed, A list of features as shown in 1010 in Fig. 10 (b) is displayed, and each feature space axis can be selected and changed. In addition, as shown in Fig. 10 (c), the feature space can change the position of the viewpoint (that is, rotate the feature space), and the local area can be enlarged or reduced at the changed viewpoint position. (1020). When the feature amount is changed in 1010 in Fig. 10 (b), the rediscovery result is displayed.

1100 in Fig. 11 (a) is an example of a monitor screen that displays the feature space with the changed features and the detected defect map. It can be confirmed that different defect species are detected. In this way, the user can select a feature amount capable of detecting a desired defect type with high sensitivity while checking the detected defect type and the degree of separation of outliers from normal pixels. In other words, in the present invention, various defect types can be detected by detecting outliers by changing various feature quantities. Furthermore, the sensitivity for detecting outliers can also be changed while the user confirms the detected outliers. In the present invention, the data other than the outliers can be confirmed by designating the image and the feature amount list in the feature space with the mouse as in the case of the defect. For example, in the feature space displayed on the monitor screen 1100, data other than the outliers can be confirmed, and if it is a defect to be detected, the outlier area can be expanded to include the data. The monitor screen 1110 shown in FIG. 11 (b) is an example. Check the image of the data in the normal area, and if it is a defect to be detected, teach that it is a defect. In the present invention, the area of outliers is changed so that the data taught thereby becomes outliers. The monitor screen 1120 in FIG. 11 (c) displays the result of expanding only the threshold value, which is the outlier region, without changing the feature amount. As the outlier region expands, so does the number of defects, which is reflected in the defect map. Similarly, check the pixels that have outliers, and if they are non-defective, teach that they are normal from the menu of 1110. This makes it possible to narrow the outlier area so that the taught data does not become outliers. In this way, it is possible to change the sensitivity by the user confirming the image and the feature amount and instructing whether or not the image should be detected.

Since the image used for the test inspection is stored in the memory after the first image acquisition, it is not necessary to acquire the image every time the feature amount is changed. If the memory capacity is small or the area for test inspection is large and all images cannot be stored in the memory, the acquired images are temporarily saved in a storage medium such as a hard disk. It is also possible to select several sets of feature quantity combinations in advance, detect outliers for each feature space at once, and display the detection results (1000 in Fig. 10 and 1100 in Fig. 11) side by side. ..

In this embodiment, when the defect to be detected is known, such as the image of the desired defect type detected in the past inspection is left, the user teaches it to select the feature amount and set the outlier area. Can be done automatically. 1200 in FIG. 12A is a partial example of the monitor display of the user interface unit 112 before the inspection. Here, the user selects the teaching button with the mouse, specifies the folder in which the image and the feature amount are saved, and specifies the defect image and the feature amount. The inspection apparatus according to the present invention reads these and displays a list of defect images and reference images side by side as shown in 1201 of FIG. 12 (b). The user specifies the defect part to be detected with a rectangle and teaches that it is a defect. After this teaching is sequentially performed, when the user selects the condition setting 1202, the feature amount is selected so that the specified defective pixel is farthest from the data distribution of the non-defective pixel in the feature space, and the outlier area. Is set automatically. At this time, the axis scaling is also performed automatically. As a result, the condition of 1220 in FIG. 12 (d) is automatically selected from the default condition 1210 in FIG. 12 (c).

On the other hand, even if it is an image of an actual defect, if the defect does not need to be detected by the user, the defect part is specified by a rectangle as shown in 1300 in Fig. 13 (a), and it is taught that it is a normal pixel. To do. As a result, the present invention automatically selects the feature amount, sets the outlier region, scales the axis, etc. so that the specified defective pixel is closest to the data distribution of the non-selected pixel in the feature space. Do. The condition of 1310 in Fig. 13 (c) is automatically selected against the default condition 1210 in Fig. 13 (b).

Similarly, for pixels that do not need to be detected, such as noise and areas that you want to be non-inspected areas, specify those areas and teach them that they are normal. 1400 in Fig. 14 (a) shows an example in which the condition that only the pixel designated as a defect has an outlier cannot be found in the feature space, and the pixel designated as normal also has an outlier. In this case, in this embodiment, it is possible to set a plurality of normal regions (distribution surrounded by the broken line of 1410 in FIG. 14 (b)) and set only the data outside the regions as outliers.

On the other hand, if there is no known defect information, it is possible to automatically set the conditions by teaching only the normal part. FIG. 15 shows an example. First, an image of the normal pattern of the sample is taken, and the entire area is designated as normal as shown in 1500 in Fig. 15 (a). Do this in several places and select automatic condition setting. In this embodiment, when there is no teaching of the defective part in this way, as shown in the feature space of 1501 in FIG. 15 (b), the envelope of the distribution of all the teaching pixels (the minimum area surrounding the normal distribution) 1502. The outside is the outlier area. Then, a test inspection is performed, and the outlier region is optimized by further teaching what is plotted and detected outside the envelope 1502. 1510 in Fig. 15 (c) is an image with outliers in the test inspection. If this is not a defect, the user teaches the entire area to be normal. In this example, as shown in FIG. 15 (d), when the data of the image of 1510 is plotted in the feature space 1501, the envelope is expanded to include the data (1503). Similarly, 1520 in FIG. 15 (e) is an image plotted outside the envelope 1502 in the test inspection and outliers. If this is a defect, as described above, when the user specifies the defective pixel and teaches it as a defect, in this embodiment, the data of the defective portion of 1520 is in the feature space 1501 as shown in FIG. 15 (f). The envelope is plotted and the envelope is set so that the defective pixels are outliers (1504).

As described above, according to the inspection apparatus described in each embodiment of the present invention, it is possible to detect defects buried in noise with high sensitivity by detecting outliers in the feature space. In addition, there are various defects that are important to the user, and each defect has factors such as the type, material, surface roughness, size, depth, pattern density, and pattern direction of the sample to be inspected, and the lighting conditions. It has various features depending on the combination with factors that depend on the optical system, such as, but as explained in each example, multiple types of features are prepared, and among them, interactive according to the defect type that the user wants to detect. By making it selectable, various defects can be detected with high sensitivity. Similarly, by interactively teaching the features of noises and patterns that do not need to be detected by the user, it is possible to easily adjust the sensitivity corresponding to various noises and patterns.

In this example, the feature amount was calculated using the reference image as an image of adjacent chips (22 in Fig. 2), but the reference image is the average value of multiple chips (21, 22, 24, 25 in Fig. 2), etc. It is okay to generate one from, perform one-to-one comparisons in multiple areas, such as 23 and 21, 23 and 22, ..., 23 and 25, and statistically process all comparison results. Detecting defects is also within the scope of the present invention.

Up to now, the chip comparison process has been described as an example, but cell comparison performed in the memory mat section when the peripheral circuit section and the memory mat section are mixed in the chip to be inspected as shown in Fig. 2 (c) is also possible. This is the scope of the present invention. FIG. 16 is a diagram illustrating application to cell comparison. The memory mat portion consists of a set of small repeating patterns (cells) shown in FIG. 16 (a). Cell comparison compares adjacent cells, that is, pixels adjacent to each other by the distance between cells, and detects a portion where the brightness difference is larger than the threshold value as a defect. Is. On the other hand, in the method according to the present invention, as shown in FIG. 16B, a plurality of corresponding reference pixels (integer multiples of the cell pitch, pixels separated from each other) are set as reference pixels for the inspection target pixel, and the target pixel and the reference pixel are used. The feature amount is calculated between. In the subsequent processing, the outliers in the feature space are extracted as defect candidates in the same manner as the chip comparison described so far.

The processing of the image comparison processing unit 15 according to the examples described so far is realized by software processing by the CPU, but the core calculation part such as the normalization correlation calculation for positioning deviation detection and the feature amount calculation is an LSI or the like. It is also possible to perform hard processing by. As a result, high speed can be realized. Further, even if there is a slight difference in the film thickness of the pattern after the flattening process such as CMP or a large difference in brightness between the chips to be compared due to the shortening of the wavelength of the illumination light, according to the present invention, it is about 20 nm to 90 nm. It is possible to detect defects of size.

In addition, SiO₂Inorganic insulating films such as SiOF, BSG, SiOB, porous Syrian films, and methyl group-containing SiO₂, MSQ, polyimide film, parerin film, Teflon (Teflon is a registered trademark) film, organic insulating film such as amorphous carbon film, etc. However, according to the present invention, it is possible to detect defects of 20 nm to 90 nm.

Although one embodiment of the present invention has been described by taking a comparative inspection image in an optical visual inspection apparatus for a semiconductor wafer as an example, it can also be applied to a comparative image in an electron beam pattern inspection. It can also be applied to defect inspection of dark field illumination.

FIG. 17 shows an example when the present invention is applied to a defect inspection apparatus using dark field illumination. The defect inspection device shown in FIG. 17 has an XYZ-θ stage 1712, an XYZ-θ controller 1716, and a light source 1770 that can move in three axial directions on which a sample (semiconductor inspected substrate) 1711 to be inspected is placed. , Diagonal lighting system 1771, Upward detection system 1772, Oblique detection system 1773, Oblique detection system image comparison processing unit 1715, Upward detection system image comparison processing unit 1715', Overall control unit 1716, User interface unit 17112, It is configured to include a storage device unit 17113.

With the above configuration, illumination light such as a laser emitted from the light source 1770 is irradiated to the sample 1711 mounted on the XYZ-θ stage 1712 via the illumination optical system 1771, and the scattered light from the sample 1711 is emitted by the upper detection system 1772. It collects light, detects it with a photoelectric converter 17104, and performs photoelectric conversion. On the other hand, the orthorhombic detection system 1773 also collects the scattered light from the sample 1711, detects it with the photoelectric converter 17105, and performs photoelectric conversion. At this time, by detecting the scattered light from the sample 1711 while moving the XYZ-θ stage 1712 in the horizontal direction, the detection result can be obtained as a two-dimensional image.

The obtained images are input to the image comparison processing units 1715 and 1715', respectively. The image comparison processing units 1715 and 1715'are the misalignment detection unit 108, the statistical processing unit 109, the parameter setting unit 110, and the defect in the image comparison processing unit 15 of the bright field type optical visual inspection apparatus described with reference to FIG. 1, respectively. The position shift detection unit 17108, 17108'corresponding to the classification unit 111, the statistical processing unit 17109, 17109', the parameter setting unit 17110, 17110', and the defect classification unit 17111, 17111' are provided, and the bright field method described above is provided. The images obtained in the same manner as in the examples of the optical visual inspection apparatus of No. 1 are compared, and defects are detected.

Further, the inspection target is not limited to semiconductor wafers, and can be applied to, for example, TFT substrates, photomasks, printed boards, etc., as long as defects are detected by comparing images.

<figref num="1">An example of the configuration of an inspection device.</figref><figref num="2">An example of chip configuration and information collection for multiple chips.</figref><figref num="3">An example of the defect candidate extraction processing flow.</figref><figref num="4">An example of out-of-pixel detection in a two-dimensional feature space.</figref><figref num="5">An example of out-of-pixel detection in N-dimensional feature space.</figref><figref num="6">An example of pattern brightness variation between chips to be compared.</figref><figref num="7">An example of low bit conversion and its effect.</figref><figref num="8">An example of an outlier detection flow in which a low bit conversion value is a part of a feature amount.</figref><figref num="9">An example of the feature quantity selection flow.</figref><figref num="10">An example of the condition setting screen</figref><figref num="11">Another example of the condition setting screen.</figref><figref num="12">An example of a defect information teaching screen.</figref><figref num="13">An example of a teaching screen that eliminates unnecessary information.</figref><figref num="14">An example of threshold setting to eliminate unnecessary information.</figref><figref num="15">An example of a threshold setting that detects only critical defects.</figref><figref num="16">An example of feature calculation in cell comparison.</figref><figref num="17">Another example of an inspection device configuration.</figref>

Code description

11 ... sample, 12 ... stage, 13 ... detector, 12 ... stage, 13 ... detector, 101 ... light source, 102 ... illumination optics, 103 ... Objective lens, 104 ... image sensor, 105 ... AD conversion unit, 14 ... image editing unit, 106 ... preprocessing unit, 107 ... image memory, 15 ... image comparison processing unit, 108 ... misalignment detection unit, 109 ... statistical processing unit, 110 ... parameter setting unit, 111 ... defect classification unit, 16 ... overall control unit, 112 ... user interface unit, 113 ... storage device, 114 ... mechanical controller, 200 ... chip, 202 ... peripheral circuit section, 201 ... memory mat section