JP2006319351A - Pattern inspection method and pattern inspection device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a pattern edge shape inspection method and a pattern inspection device which quantitatively analyze, by the nondestructive inspection, line edge shapes of a micropattern featuring materials, processes and exposure optical systems, and find out faults in manufacturing processes and image distortions. <P>SOLUTION: The pattern inspection method involves inspecting a pattern shape from the two-dimensional distribution information (42) of a secondary electron intensity or a reflected electron intensity obtained by the observation (41) of the pattern formed on a substrate by a scanning microscope using charged particle beams, and comprises a step (45) of detecting a set of edge points expressing the edge points of the pattern in a two-dimensional surface by the threshold value method, a step (48) of obtaining an approximate line for the set of the detected edge points, and a step of determining an edge roughness shape and characteristic by calculating (49) a difference between the set of the edge points and the approximate line, wherein as the threshold value used in the threshold method, a plurality of values are used. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a pattern inspection technique, and more particularly to a method and apparatus for observing a fine pattern using a scanning microscope and inspecting its edge shape.

  In recent LSI processes, in particular, microfabrication processes after ArF lithography, pattern edge roughness (unevenness of pattern end portions) is a major problem as patterns become finer.

  The cause of the roughness is the nature of the material itself, the exposure apparatus, the base substrate, and the observation apparatus itself. In mass production processes, the magnitude of roughness greatly affects product performance. Even if the roughness is not abnormally large, the appearance of a characteristic roughness often reflects a decline in the performance of the manufacturing apparatus, which may lead to a future product failure. There is an urgent need to develop a system that observes the roughness shape of pattern edges and identifies the cause of occurrence based on the features. In consideration of use in mass production processes, the inspection method must be non-destructive inspection.

  Conventionally, information has been obtained through experience mainly from visual observation of an observation image with a scanning electron microscope. For example, when the resist line pattern is observed, it can be seen that the left and right edge fluctuations are synchronized. In this case, for example, the line is thin and the top of the pattern fluctuates during development, the light intensity distribution during exposure was originally distorted, or the observed image itself is distorted. In addition, the vicinity of the surface may have relatively small unevenness, while the pattern bottom may appear to have large unevenness. From such a phenomenon, there is a possibility that the residue remains because the compatibility with the base is bad.

  However, since the criteria for these judgments are not quantitative, the conclusion varies depending on the observer. In order to analyze the cause of roughness systematically without depending on the observer, it is necessary to discuss the roughness shape quantitatively.

  As an attempt to quantitatively represent the feature of the pattern edge shape that has been performed so far, an example described in JP-A-8-14836 can be given. In this method, the taper shape of the line pattern edge is digitized from the SEM image. As a result, information on the edge portion in the cross-sectional shape can be obtained to some extent, but the feature of the edge in the direction along the line edge cannot be obtained. The obtained value is an index of the inclination of the edge side face, and the roughness cannot be evaluated.

  As a method of detecting the roughness of the pattern edge, a method is generally used in which several deviations from the linear approximation of the edge position are obtained and a standard deviation σ or a value three times as large as σ in their distribution is calculated. However, the roughness here is, as an example, disclosed in Japanese Patent Application Laid-Open No. 2000-58410, and is merely a dimensional error and does not serve as an index for edge shape evaluation.

JP-A-8-14836 JP 2000-58410 A

  As described above, there has never been a quantitative evaluation method of the edge shape. Further, the visual judgment depends on the observer even though the three-dimensional shape of the edge can be seen.

  An object of the present invention is to provide a pattern inspection method capable of systematically identifying the cause of roughness and systematically performing quantitative and high-accuracy rapid evaluation of the shape of an edge that has been performed by visual observation of an observation image. It is to provide a pattern inspection apparatus.

  In the present invention, a data is directly processed from a two-dimensional distribution image of secondary electrons or reflected electrons obtained by observation with a scanning microscope using a charged particle beam such as an electron beam, ionizing radiation, or an ion particle beam. The edge point position was obtained by the threshold method while maintaining the accuracy of the above, and the deviation of the edge point position from the approximate line, that is, the edge position fluctuation was obtained. A set of edge position fluctuations obtained for edge points belonging to one edge represents a two-dimensional feature of the edge roughness shape. When this is performed on different threshold values to obtain a set of a plurality of edge position fluctuations, these represent the three-dimensional edge roughness shape of the original image.

  Next, using the result, a step of calculating and displaying the spatial frequency distribution of the edge position fluctuation and the threshold value dependency thereof was provided. As a result, it is possible to find a spatial period having a particularly strong intensity, that is, a characteristic period of roughness.

  Further, by providing a step of calculating and displaying the threshold dependence of the standard deviation of the edge position fluctuation, it is possible to distinguish between the case where the edge roughness is large near the surface and the case where the edge roughness is large at the bottom. In the former case, it can be estimated that there is a cause of roughness in the atmosphere of the pattern forming process, and in the latter case, the consistency with the underlayer.

  In addition, a processing step for calculating and graphing a correlation graph, a correlation coefficient, and a threshold dependence of the correlation coefficient of left and right edge position fluctuations of one line is provided. As a result, the direction of the roughness at the line pattern edge is the same (1) on the left and right (FIG. 1), (2) the opposite direction on the left and right (FIG. 2), and (3) irrelevant on the left and right. It is possible to clarify whether the situation changes in the depth direction.

  FIG. 1 and FIG. 2 show examples of edge fluctuations when one line pattern exists in the vertical direction in an image. In the figure, 1 and 3 are the left edge of the line, and 2 and 4 are the right edge.

  FIG. 1 shows a case in which the line width is constant and the line itself is wavy, and FIG. 2 shows a case in which the left and right edges of the line are synchronized but fluctuate in the opposite direction to FIG. When there is a tendency of (1) above, the correlation between the left and right edge position fluctuations becomes positive. When there is a tendency of (2) above, the correlation between the left and right edge position fluctuations is negative. There is no correlation when the left and right edges fluctuate independently. Specific calculation of the correlation coefficient and details of the determination criteria will be described later.

  These calculation results are input to a determination function, and a function for selecting and displaying a pattern formation process candidate that is considered to be the cause of roughness is provided. In addition, by using a system that can transmit this signal to the corresponding device, it is possible to respond quickly to the appearance of a defect and reduce loss compared to the conventional system.

  In addition, when it is pointed out that the cause of edge roughness may be in the observation device itself, in order to examine the observation device, the line-shaped standard sample is observed, and the observation position is set in a direction parallel to the line pattern. The image was integrated. In the obtained two-dimensional data, the randomly generated roughness is averaged, but the distortion of the observed image remains without being removed. By storing the amount of distortion as data, it is possible to remove the distortion in subsequent observations and obtain an image with a smaller error.

  In this way, the pattern inspection method according to the present invention is based on the two-dimensional distribution information of secondary electron intensity or reflected electron intensity obtained by observing a pattern formed on a substrate with a scanning microscope using a charged particle beam. A method for inspecting a pattern shape, comprising: a step of detecting a set of edge points representing a position of an edge of the pattern in a two-dimensional plane from the two-dimensional distribution information by a threshold method; The method includes a step of obtaining an approximate line for a set of edge points to which the image belongs, and a step of obtaining an edge roughness shape by calculating a difference between the set of edge points and the approximate line.

  In addition, the pattern inspection method according to the present invention obtains a pattern shape from two-dimensional distribution information of secondary electron intensity or reflected electron intensity obtained by observing a pattern formed on a substrate with a scanning microscope using a charged particle beam. An inspection method comprising: detecting a set of edge points representing a position of a line edge of the pattern in a two-dimensional plane from the two-dimensional distribution information; and the set of edge points detected for each line edge Obtaining an approximate line for the line by the method of least squares, calculating an edge roughness shape by calculating a difference between the set of edge points belonging to each line edge and the approximate line, and edge roughness of different line edges And a step of displaying a correlation between shapes.

  Further, the present invention is characterized in that, in the above configuration, a plurality of values are used as threshold values used in the threshold method.

  Further, the present invention is characterized in that, in the above configuration, the step of calculating a spatial frequency distribution of the obtained edge roughness shape is included.

  In the configuration described above, the present invention calculates a standard deviation represented by a mean square root of a square of a difference between the set of edge points obtained for the plurality of threshold values and the approximate line. Thus, the method includes a step of obtaining the magnitude of the edge roughness.

  Further, the present invention is characterized in that, in the configuration described above, a step of selecting and displaying a pattern formation process candidate of the substrate that causes the occurrence of roughness from the obtained edge roughness shape.

  Furthermore, the present invention includes a step of placing a sample processed into a line pattern shape with a predetermined pitch on a scanning microscope, observing the sample to obtain a two-dimensional intensity distribution of secondary electrons or reflected electrons, There is provided a pattern inspection method comprising a step of calculating an edge roughness shape of the line pattern from an intensity distribution and a step of storing the obtained edge roughness shape as image distortion information.

  Furthermore, the present invention includes a step of placing a sample processed into a line pattern shape with a predetermined pitch on a scanning microscope, observing the sample to obtain a first two-dimensional intensity distribution of secondary electrons or reflected electrons, A step of moving the observation position in the direction of the side of the line pattern by a certain amount to obtain a second two-dimensional intensity distribution of secondary electrons or reflected electrons; the first two-dimensional intensity distribution; A step of obtaining a sum of two two-dimensional intensity distributions, a step of calculating an edge roughness shape of the line pattern from the sum data, and a step of storing the obtained edge roughness shape as image distortion information. A pattern inspection method is provided. Furthermore, the present invention is obtained as a result of observing an arbitrary sample by calculating an image offset amount in a direction perpendicular to the edge of the line pattern in the observation area from the obtained image distortion information in the pattern inspection method. There is provided a pattern inspection method comprising a step of correcting a third two-dimensional intensity distribution of secondary electrons or reflected electrons, or a pattern edge position obtained from the third two-dimensional intensity distribution.

  Furthermore, the present invention provides a charged particle optical system, a charged particle optical system that irradiates a sample with a charged particle beam emitted from the charged particle source through a converging lens, a deflector, and an objective lens, and deflects and scans the sample. A stage to be mounted; a detector for detecting the intensity of secondary electrons or reflected electrons emitted from the sample by the charged particle beam irradiation; and a control system for controlling the deflection / scanning. There is provided a pattern inspection apparatus comprising signal processing means for obtaining an edge roughness shape and characteristics of the pattern based on a threshold method from the two-dimensional distribution of the secondary electrons or reflected electron intensity.

  According to the present invention, a pattern inspection method capable of systematically identifying the cause of occurrence of roughness systematically by quantitatively measuring the shape of an edge that has been performed by visual observation of an observation image so far and performing quantitatively with high accuracy and speed. And a device can be realized. Furthermore, it can be expected to improve the yield and throughput by managing the ultra-fine processing process by using the control of the manufacturing process or manufacturing apparatus that is the cause of the occurrence.

  Embodiments of the present invention will be described below with reference to the drawings.

(Example 1)
A first embodiment of the present invention will be described with reference to FIGS. FIG. 3 is a schematic diagram showing the apparatus configuration of the present embodiment, FIG. 4 is a flowchart showing the procedure of the present embodiment, FIG. 5 is a schematic diagram of an image of data used for evaluation, and FIG. 6 is a threshold based on the data. It is a figure showing the edge of the line pattern detected by the value ratio 0.5.

  Using the apparatus shown in FIG. 3, the pattern was inspected according to the flow shown in FIG.

  First, as shown in step 41, an operation was performed from the control system 15 of the scanning electron microscope, and the sample 11 placed on the stage 12 in the microscope casing 5 was observed. The electron beam 7 emitted from the electron gun 6 of the scanning electron microscope is irradiated to the sample 11 on the stage 12 through the converging lens 8, the deflector 9, and the objective lens 10, and secondary electrons 13 emitted from the sample 11. Was detected by detector 14.

  Sample 11 had a structure having a resist line pattern formed by electron beam drawing on a silicon substrate. When there is no measurement result regarding the spatial period of the edge roughness of the pattern, or when there is no particular requirement for the size of the observation area, it is desirable to observe at a magnification of 100,000 to 300,000 times. In this example, observation was performed at 200,000 times. The sample 11 was placed on the stage 12 so that the line pattern was almost perpendicular to the scanning direction.

  Next, proceeding to step 42, scanning was performed a plurality of times, and the measurement results of the secondary electron intensity emitted from the pattern were integrated to calculate an average value. In order to obtain an image with less noise, it is desirable that the number of data accumulations is 64 times or more. In this example, the integration was performed 128 times.

  The two-dimensional electron intensity distribution thus obtained is data to be analyzed. Further, the obtained electron intensity distribution data was converted to a gray scale density and displayed as an image on the terminal screen of the computer 16. A schematic diagram of this image is shown in FIG.

  The image data was composed of 512 pixels in both the horizontal and vertical directions. When it is desired to inspect the edge roughness shape for a longer region in the vertical direction, it is possible to set the interval between scanning lines corresponding to one horizontal row of data to an integral multiple of the horizontal pixel interval. For example, a region of 675 nm in the horizontal direction and 2700 nm in the vertical direction can be represented on a screen of 512 pixels in the vertical and horizontal directions. In this example, the observation area included in the image is 675 nm both vertically and horizontally, and represents an area of about 1.318 nm square per pixel. Hereinafter, the upper left of the image is the origin, the distance in the right direction is x, and the distance in the lower direction is y. The pixel numbers in the x and y directions are represented by m and n. In practice, an image having a light and shade corresponding to the secondary electron intensity appears. In FIG. 5, the area where the secondary electron intensity is particularly large, that is, the region where the edge can exist is represented by white, and the small area is represented by oblique lines. Yes. The coordinates shown in FIG. 5 represent the pixel numbers of the image. After stopping the subsequent irradiation of the electron beam to the wafer, the image data is transferred from the control system 15 to the adjacent computer 16 to end the step 42 and proceed to the step 43 to execute the shape analysis according to the present invention. Was executed.

  As a result, the program averaged and smoothed the data as follows to reduce noise. First, the data was divided into 512 sets of secondary electron intensity sets or profiles arranged in a line in the x direction. Each profile is the x dependence of the secondary electron intensity when y is constant, and the profile has a total number of pixels in the y direction, that is, 512.

With respect to this data, noise was reduced by the following procedure. First, an averaging parameter k ₁ (natural number) and a smoothing parameter k ₂ (odd number) were given. When k ₁ is an even number, k ₁ ′ = k _1/2 and when it is an odd number, k ₁ ′ = (k ₁ −1) / 2. Further, k ₂ ′ = (k ₂ −1) / 2. The average of N-k ₁ 'th to k ₁ profiles was taken, and this was taken as the n-th profile after averaging. Next, smoothing using a hamming window is performed for the area from pixel number m−k ₂ ′ to m + k ₂ ′ in the profile after averaging obtained in the above step, and a new m-th smoothing is performed. The value of In the case of data in which one pixel corresponds to a length from 0.8 nm to 2 nm, the averaging parameter k ₁ is preferably 4 or more and 11 or less, and the smoothing parameter k ₂ is preferably 3 or more and 11 or less. If both values are smaller than these values, the noise cannot be sufficiently reduced, and if it is larger, the edge roughness of a fine spatial period cannot be detected. Here, k ₁ = k ₂ = 7.

  Next, in order to detect the edge of the line, an area for searching for an edge point was input. First, for the left edge, the pixel number in the x direction of the search region was determined from m = 210 to m = 250 from the position of the region 18 in FIG. Similarly, regarding the right edge, m = 280 to m = 320 as measured from the position of the region 19. Designation of these calculation areas can be performed by two methods: (1) numeric input and (2) screen input on the screen of FIG. Here, numerical input was performed.

Next, as shown in steps 44 to 47, edge points were detected. The threshold ratio p for performing the detection is changed by changing the set Δp from the minimum value p ₁ to the maximum value p ₂ of the set p. In this example, for each value of p ₁ = 0.2, p ₂ = 0.9, and Δp = 0.1, the calculation is performed for a total of 256 profiles in which the pixel number n in the y direction is given by an even number 2n ′. It was. The threshold method used here is generally known, and is given by (I _max −I _min ) × p + I _min from p, the secondary electron intensity maximum value I _max, and the minimum value I _min. The threshold value is calculated, and the point where the secondary electron intensity becomes the threshold value on the profile is used as an edge.

The detected x-coordinates of the left and right edge points were x _L (2n ′) and x _R (2n ′). The profile is a set of a set of numerical values I (x, I (x)) that gives the secondary electron intensity at that position with respect to the x coordinate given by an integral multiple of the length of 1.318 nm for one pixel in the x direction. The points that meet at the time of calculation were connected by a straight line, and the intersection of this broken line and the threshold value was obtained. The y coordinate of the edge point thus obtained is 2n ′ × 1.318 in units of nm.

Through the above steps, 256 (x _L , y) coordinates are obtained for the left edge for one value of p. Similarly, a set of 256 points is obtained for the right edge. The procedure of giving 0.2 as the value of p (step 44), detecting the edge, and increasing the value of p by 0.1 was repeated until p = 0.9 (steps 45 to 47). As an example, a set of edge points obtained for the case of p = 0.5 is shown in FIG.

  Next, as shown in step 48, the total of 512 points on the left and right sides was approximated using the method of least squares. Although it was possible to approximate with a curve representing an arbitrary function form, approximation was performed with two parallel straight lines x = ay + b and x = ay + b + w.

Next, proceeding to step 49, for the profile whose y coordinate is given by an even number 2n ', the intersection of the x coordinate x _L (2n') of the left edge point and the approximate line and the profile, that is, the x coordinate a x 2n of the approximate point. The difference from '+ b' was calculated as edge point fluctuation Δx _L (2n '). This was performed for profiles from n ′ = 1 to n ′ = 256 to obtain 256 left edge point fluctuations. In performing the calculation here, the position of the point is represented by x, y, that is, the unit of length nm, but the processing may be performed by representing the pixel number m, n. In the latter case, n is a positive integer, so less storage capacity is required. Similar calculations were performed for the right edge to obtain 256 right edge point fluctuations. In this way, sets of edge position fluctuations (Δx _L (2n ′), 2n ′), (Δx _R (2n ′), 2n ′) (where n ′ = 1, 2) representing the shape of edge roughness. ... 256) could be obtained. The program also calculated these values for each threshold ratio. These results were stocked in a storage area within the computer.

  As a result, the shape of edge roughness can be extracted as digital data from a microphotograph represented by grayscale shading, and can be displayed as a set of points on a graph. Therefore, it was easy to see when showing the edge shape. In addition, it is now possible to analyze the pattern shape by performing further analysis using these data.

(Example 2)
A second embodiment of the present invention will be described with reference to FIGS.

  The edge position was detected from the image data of the resist pattern shown in FIG. 5 by the method shown in Example 1, and data representing the edge roughness shape was obtained as the fluctuation of the edge position.

Next, a set of edge position fluctuation data (Δx _L (2n ′), 2n ′) obtained for each threshold ratio (or a set of (Δx _R (2n ′), 2n ′)) is obtained. The distribution of the period was calculated by regarding the superposition of periodic functions in the y direction. That is, Fourier transform is performed on the data series {Δx _L (2), Δx _L (4), Δx _L (6), Δx _L (8)...}, And the absolute value of the Fourier coefficient with respect to the y-direction spatial frequency f, that is, the intensity A (f) was obtained.

  An example of the frequency distribution obtained as a result is shown in FIG. It is the result of having performed Fourier transformation about the line left edge illustrated in FIG. The spatial frequency f on the horizontal axis is the ratio of the corresponding spatial period to the length of the image area of 675 nm. For example, f = 10 corresponds to a spatial period of 675 nm / 10, that is, a period of 67.5 nm.

Next, a spatial frequency characteristic of this frequency distribution was extracted by the following procedure. 15 <intensity in the region of f <128 A (f) is by using the method of least squares approximated with the function A ₀ × 1 / f, the function obtained by substituting Ficoll Te' I ing parameter A ₀ obtained A ₀ × 1 / f Are plotted on the same graph in all regions where f <128. The curve represented by the thick solid line in FIG. 7 is the approximate curve thus obtained. Next, A (f) obtained from the actual measurement value in the region of f designated in advance and the approximate value A ₀ × 1 / f were compared, and the value of f in which the former was larger than the latter was picked up. In this example, the designated area is 3 <f <20.

  When this analysis was performed for the left and right edge roughness of all threshold ratios, it was found that the spatial frequency components of f = 5 and f = 7 greatly contributed to the roughness.

  As one of the features that characterize the roughness of the edge being observed in this way, it was possible to extract the spatial period that can be said to be the characteristic roughness of the roughness. As described above, numerical values of f = 5 and f = 7 were obtained from the results obtained by inspecting the pattern shown in FIG. In this way, when roughness with a constant frequency is observed regardless of the threshold value, when the pattern is distorted from the bottom to the vicinity of the surface, and when the frequency component is distortion of the observed image You could think so. The former is caused by the resist exposure apparatus, and the latter is caused by the observation apparatus. In this way, by performing this inspection, it was possible to select process candidates that cause roughness.

(Example 3)
A third embodiment of the present invention will be described with reference to FIGS.

  The edge position was detected from the image data of the resist pattern shown in FIG. 5 by the method shown in Example 1, and data representing the edge roughness shape was obtained as the fluctuation of the edge position.

  Next, from the edge position fluctuation data obtained with respect to each threshold ratio, the standard deviation in the distribution represented by the following (Equation 1), that is, the fluctuation distribution is calculated, and three times, ie, 3σ is calculated as edge roughness. It is defined as the size of.

ここで、添字kは、k＝Lあるいはk＝Rである。この計算を各しきい値比率に対して行って、ひとつのラインエッジについてしきい値比率ｐとラフネスの大きさ３σとの関係が得られる。図５に概略図を示した例について上記の計算を行った結果を図８に示す。ラフネスの大きさがしきい値比率に殆ど依存せず、ほぼ一定であることが分かる。

Here, the subscript k is k = L or k = R. This calculation is performed for each threshold ratio to obtain a relationship between the threshold ratio p and the roughness magnitude 3σ for one line edge. FIG. 8 shows the result of the above calculation for the example schematically shown in FIG. It can be seen that the magnitude of the roughness hardly depends on the threshold ratio and is almost constant.

The graph of the dependence of 3σ on p could be quantitatively analyzed as follows. The graph of 3σ vs. threshold ratio is approximated by a linear function y = ax + b by the method of least squares. Here, y is a value of 3σ (unit is nm), and x is a threshold ratio. Then, as compared with the value alpha ₁ for a value has been set in advance of the fitting parameters a obtained, 3 [sigma] If a> alpha ₁ increases with increasing p, the roughness is larger and more in other words near the surface Judge. Further, as compared with the value alpha ₂ that have preset values of a, 3 [sigma] If a <alpha ₂ decreases with increasing p, that is, more determined that the roughness is large at the vicinity of the substrate.

From the inspection results regarding the line pattern of the conventional resist, 4 and -4 are standard as the set values of α ₁ and α ₂ , respectively. Although it is possible for the observer to set a different value, in this example, the inspection was performed using this standard value.

  When this method is applied to the result shown in FIG. 8, a = 0.02, and it can be seen that the roughness is almost the same from the bottom of the pattern to the vicinity of the surface.

  Further, an image obtained by observing another resist pattern was inspected, and threshold dependence as shown in FIG. 9 was obtained. In this graph, the value of a was 6.62, and it was determined that the roughness was large near the surface. Since this resist was a chemically amplified negative resist, it was pointed out that the acid on the resist surface was deactivated in an alkaline atmosphere, and the amine concentration in the atmosphere was measured. It was confirmed that it was higher than that. As described above, the roughness generation process candidates can be selected from the threshold ratio dependency of the magnitude of roughness.

(Example 4)
A fourth embodiment of the present invention will be described with reference to FIGS. 5 and 10 to 12.

  The edge position was detected from the image data of the resist pattern shown in FIG. 5 by the method shown in Example 1, and data representing the edge roughness shape was obtained as the fluctuation of the edge position.

Next, from the set of edge position fluctuation data (Δx _L (2n ′), 2n ′) and the set of (Δx _R (2n ′), 2n ′) obtained for each threshold ratio, the same y coordinate is used. By combining the fluctuation of the position of the left edge with the fluctuation of the right edge position, 256 points (Δx _L (2n ′), Δx _R (2n ′)) were obtained. This point is shown in the graph as shown in FIG. This is the case when p = 0.5.

  From this result, it can be seen that there is a positive correlation between the roughness of the left edge and the roughness of the right edge. Further, the correlation coefficient ρ of the left edge position fluctuation and the right edge position fluctuation at p = 0.5 was calculated from this data according to (Equation 2). However, the numerator on the right side of (Formula 2) is the amount represented by (Formula 3).

また、σ_L、σ_Rはそれぞれ左エッジ位置および右エッジのゆらぎの分布における標準偏差であり、ラフネスの大きさの１/３に相当する。ρは０.６４となった。

Also, σ _L and σ _R are standard deviations in the fluctuation distribution of the left edge position and the right edge, respectively, and correspond to 1/3 of the magnitude of roughness. ρ was 0.64.

From the value of ρ, the roughness type can be classified as follows. Compared to a preset reference value ρ _th of absolute value of ρ, if ρ> ρ _th , the type of FIG. 1 and if ρ <−ρ _th , the type of FIG. Judge that there is no. In this example, ρ _th = 0.4 is set based on accumulation of inspection result data regarding the resist pattern observation image. This value is standard, but the observer can enter another value. The line pattern shown in FIG. 5 has the roughness of the type shown in FIG. 1 when p = 0.5.

  Further, for the line pattern shown in FIG. 5, p was varied from 0.1 to 0.9 in increments of 0.1 to calculate the value of ρ. The result is shown in FIG. It was found that the dependence of ρ on p was small.

  This graph of dependence of ρ on p could be quantitatively analyzed as follows. A graph of ρ vs. threshold ratio p is approximated by a linear function y = cx + d by the method of least squares. Here, y is ρ, and x is the value of the threshold ratio p.

And the resulting compared to fitting parameter value gamma ₁ which has been set in advance the value of c, is ρ if c> gamma ₁ increases with increasing p, that is, the more lateral edges near the surface Judge that the correlation of fluctuation is large. In addition, when c <γ ₂ , the value of c decreases as p increases, that is, the correlation between the fluctuations of the left and right edges is greater in the vicinity of the substrate than c = γ _2. Judge.

From the inspection results relating to the line pattern of the conventional resist, 0.4 and −0.4 are standard as set values of γ ₁ and γ ₂ , respectively. Although it is possible for the observer to set a different value, in this example, the inspection was performed using this standard value.

  When this method is applied to the result shown in FIG. 11, c = 0.15, and it can be seen that the correlation between the left and right edge fluctuations is constant from the bottom of the pattern to the vicinity of the surface.

  Further, the image obtained by observing another resist pattern was inspected, and the threshold dependence shown in FIG. 12 was obtained. In this graph, the value of c is 0.57, and it can be seen that the tendency for the left and right edges to fluctuate together becomes closer as it approaches the surface. Since there is no correlation between left and right edge fluctuations at the bottom, and fluctuations are maintained with the width kept constant even near the surface, it is assumed that the pattern once formed is distorted during development or post-development baking due to insufficient strength It was done. Thus, the candidate for the roughness generation step could be selected from the dependency of ρ on the threshold ratio.

(Example 5)
A fifth embodiment of the present invention will be described with reference to FIGS. 1 to 3, 5 to 8, 10, 11, 13 and 14.

  First, an outline of the procedure will be described with reference to FIGS. 13 and 14. FIG. 14 describes in detail a part of step 139 in the flow shown in FIG.

  First, a line pattern is observed and data is captured by a scanning microscope in the same procedure as in the first embodiment (step 131, step 132). After noise reduction is performed on the obtained two-dimensional data by the method described in the first embodiment (step 133), an edge roughness shape is obtained using a standard value of p (usually 0.5). (Step 134). Further, the roughness magnitude (3σ) given in (Equation 1) is calculated for all the edges in the image data from the edge roughness shape data (step 135).

  Next, the process proceeds to step 136, where the quality of the wafer is determined. This means that only when the measured 3σ of all edges is smaller than the reference value, the substrate with a pattern to be observed is judged as a non-defective product and passed to the next process. Regardless of the result of this determination, it is possible to select whether or not to perform shape analysis after step 139 (steps 137 and 138). If not, the inspection substrate inspection is terminated, the non-defective product is transferred to the next process, and the defective product is lotted out.

  In the case of performing shape analysis, the process proceeds to step 139. First, a target line is selected from two-dimensional data with reduced noise, and a plurality of threshold ratios are used one after another as shown in the first embodiment. Edge detection is performed to obtain edge roughness shape data for each threshold ratio.

  After obtaining this, data processing is performed according to the flow shown in FIG. The data processing includes three types of inspections for expressing the feature of the edge shape. Among these, items to be executed are selected (step 147). In order to obtain a more reliable conclusion, it is desirable to select all processes, but two or less items may be executed in order to reduce time. Details of steps 148 to 150 will be described below.

  First, calculation of the characteristic spatial frequency of the edge roughness shape. This is obtained by a method of picking up a characteristic spatial frequency common to the edge roughness-shaped spatial frequency distribution for all threshold ratios after obtaining the spatial frequency distribution by the method shown in the second embodiment (step 148). ).

  Second, calculation of the threshold ratio dependence of roughness. This is determined by the method shown in Example 3 (step 149).

  Third, a graph showing the correlation between the left and right edge point fluctuations belonging to one line at each threshold ratio, and the calculation of the threshold ratio dependence of the correlation coefficient of the left and right edge point fluctuations. This is obtained by the method shown in Example 4 (Steps 150 and 151).

  Analysis results are displayed for these items (step 152). Thereafter, it is selected whether or not the roughness generation step is automatically specified as shown in step 140 of FIG. If not, the person will consider the above results if necessary, and the inspection substrate inspection will be completed. The wafer is processed according to the result of the good / bad determination (step 144). If the roughness generation process is automatically specified, the program compares the above result with the reference in step 141 to determine which of the pattern formation processes may be causing the roughness, and determines the result. Output. Further, when the manufacturing apparatus in the pattern forming process is controlled from this inspection apparatus, as shown in steps 142 and 143, a signal is sent to the manufacturing apparatus according to the result, and the inspection substrate inspection is completed. The wafer is then processed according to the good / bad determination (steps 144 to 146). The specific contents of the inspection performed in this example are shown below.

  In this example, the line pattern image of the electron beam resist whose schematic diagram is shown in FIG. 5 was inspected using the apparatus of FIG. The resist used for this pattern production was a negative type.

  First, two-dimensional data representing an image was processed using the method and parameters described in Example 1, and edge roughness shapes of left and right edges at a threshold ratio p = 0.5 were obtained. This is as shown in FIG. Next, the magnitude of the roughness of the left and right edges was calculated using the data and displayed with the image. Moreover, as a result of performing the quality determination of the observation sample, it was determined to be a non-defective product. The reference value for the roughness of good products in this inspection was set to 6 nm. If the product is determined to be defective, a warning sound is emitted, and a roughness value larger than the reference value is displayed as a red number on the image. Numbers below the reference value are displayed in white or black.

  Here, it was determined that the sample was a non-defective product, but the analysis on the roughness shape was continued. First, the spatial frequency is analyzed using the method and parameters described in the second embodiment, and the spatial frequency distribution from the threshold ratio p = 0.2 to p = 0.9 is obtained. We found f = 5 and f = 7 as frequencies. The case of p = 0.5 is shown in FIG.

  Next, the dependence of the roughness level 3σ on the threshold ratio p is calculated using the method and parameters described in the third embodiment, and at the same time the graph shown in FIG. 8 is displayed, 3σ is almost independent of p. The result was obtained.

  Next, using the method and parameters described in Example 4, the correlation between the left and right edge roughness shapes was calculated as a correlation coefficient, and the dependency on the threshold ratio p was obtained. As a result, the graphs shown in FIGS. 10 and 11 are obtained, and the correlation coefficient is positive and larger than the reference value, that is, the type shown in FIG. 1, and this tendency does not depend on the threshold ratio p. I understood.

The observer displayed these results and then operated an automatic determination function for determining the roughness generation process. The following describes the candidate narrowing-down procedure for the resist roughness generation process of a general automatic determination program. However, in addition to the general criteria used in this example, the user can also store data for setting values such as α ₁ and α ₂ used in these judgment methods, as well as for setting criteria and narrowing criteria for narrowing down candidates. Useful. It is also possible to inspect the roughness of patterns other than resist. In this example, a storage device is provided in the computer for data storage.

  First, as candidates for the cause of roughness, (1) resist properties, (2) exposure apparatus, (3) developer, (4) atmosphere, (5) foundation surface, (6) foundation pattern, and (7) observation apparatus , Is given.

  Specifically, (2) indicates the unevenness of the edge of the reticle pattern and the fluctuation of the beam position or intensity at the time of drawing. (3) is the distortion of the entire line due to swelling or vortex of the developer because the developer concentration does not match, and (4) is the erosion of the pattern surface portion by the amine or acid in the atmosphere. (5) is a footing due to chemical properties such as insufficient undersurface treatment, (6) is an uneven reflectance due to a lower layer pattern, and (7) is a screen distortion due to electrical noise or vibration.

  Among the inspection items described above, in the first frequency distribution calculation, when no characteristic frequency is observed at p = 0.5, among the above causes (1), (2), (4), (5) (3), (6), and (7) are excluded. If a characteristic frequency is seen, the possibility of (5) is eliminated. When the frequency is 20 or more, (7) is excluded. In addition, when the characteristic frequency is converted to a period of 0.5 μm or less, (3) is excluded. Further, when p = 0.2 and 0.3, no characteristic frequency is observed, and when p = 0.8 and 0.9, it is (3). More specifically, the physical strength of the resist It is considered that the vicinity of the surface was bent by an external force generated after the formation of a pattern such as a vortex of the developing solution in a weak place.

  From the dependency of the roughness level, which is the second inspection item, on the threshold ratio, it is determined as follows. When it is determined that the roughness is larger near the surface from the graph showing the edge roughness from p = 0.2 to 0.9 using the method described in the third embodiment, (1), (4 ) And the others are excluded. Conversely, if it is determined that the roughness is large near the ground, there is a possibility of (1) and (5).

  If it is determined that the type shown in FIG. 1 is the correlation between the left and right edge roughness at the third inspection item p = 0.5, the possibilities of (2), (3), and (7) remain. In the case of the type shown in FIG. 2, there are possibilities (2) and (6). Also, if the correlation is strong only in the vicinity of the surface as a result of calculating the threshold ratio dependency of the correlation coefficient, in addition to the roughness of the above-described FIG. 1 or FIG. Roughness with no correlation has occurred, and it is considered that the roughness with strong left and right correlations is rare at the bottom. On the contrary, when it is determined that the correlation is strong only at the bottom, in addition to the roughness of the above-described FIG. 1 or 2 type, the roughness near the surface is greatly eroded due to the roughness with no correlation between left and right caused by (4). It is thought that there is.

  According to these determination criteria, the edge roughness shown in FIG. 5 was determined as (2) or (7). The observed wafer was sent to the next process as a non-defective product, and the inspection was once completed.

  Next, when the resist pattern formed using a different exposure apparatus was inspected, the same conclusion was obtained from the above results. Therefore, the observer determined that (7) was more likely to be the cause than (2), and when the scanning electron microscope was inspected, the screen of the present observation device was affected by the magnetic field generated from the peripheral device. I found it distorted. Shielding with a thorough magnetic field eliminates distortion and allows more accurate measurements.

(Example 6)
Next, a sixth embodiment of the present invention will be described with reference to FIG. 3, FIG. 15, and FIG. FIG. 15 is a conceptual diagram of the imaged data used for evaluation, and FIG. 16 is a graph showing the correlation between the left and right edge position fluctuations of the observed line pattern.

  The pattern shape evaluation and determination were performed using the apparatus shown in FIG. 3 according to the same flow as in Example 5.

  First, an operation was performed from the control system 15 of the scanning electron microscope having a length measuring function, and the line pattern of the positive ArF resist formed on the silicon wafer was observed by ArF lithography on the sample 11. For the purpose of observing edge roughness with a large spatial period, the magnification is desirably 100,000 times or less. In this example, the magnification was 100,000 times. The pattern to be observed was placed so that the line pattern was in a direction substantially perpendicular to the scan. The observation area was 1.35 μm in the direction perpendicular to the line pattern, 5.40 μm in the parallel direction, and the distance between adjacent scanning lines was 10.55 nm. For the purpose of observing the presence / absence of edge position fluctuation with a large spatial period for a thin line pattern, it is desirable that the aspect ratio of the observation area is 2: 1 or more. In this example, since the region to be observed was 6 μm in the vertical direction, it was set to 4: 1. The measurement results of the secondary electron intensity emitted from the pattern were accumulated by the scan performed 64 times, and displayed as an image on the screen of the control system 15 with the average value as the gray scale density.

  A schematic diagram of an image appearing on the screen is shown in FIG. This image data is composed of 512 pixels in both the horizontal and vertical directions. Hereinafter, the upper left of the image is the origin, the distance in the right direction is x, and the distance in the lower direction is y. The numbers of pixels in the x and y directions are represented by m and n. For each pixel, the x direction represents a region of 2.637 nm and the y direction represents a region of 10.55 nm. Actually, an image having a light and shade corresponding to the secondary electron intensity appears, but in FIG. 15, the area where the secondary electron intensity is particularly large, that is, the region where the edge can exist is represented by white, and the small area is represented by oblique lines. Yes. The coordinates shown in FIG. 15 represent the pixel numbers of the image.

  Next, after the irradiation of the wafer with the electron beam was stopped, the image data was transferred from the control system 15 to the adjacent computer 16. A program for performing an inspection according to the present invention was executed from the terminal of the computer 16. As a result, the program processes the image file converted into 512 × 512 numerical data using the threshold method shown in the first embodiment, and the two lines existing in the image, a total of four edges, are processed. The edge point coordinates were detected. However, the averaging parameter was set to 4 and the smoothing parameter was set to 3 from the balance of noise reduction and accuracy. The calculation is performed for all profiles, that is, 512 profiles, the threshold ratio p is 0.5, and the input edge search area is determined by visual observation from the position of the area 20 at the left edge of the first line m = From 170 to m = 200, the right edge of the first line is determined from the position of the region 21 and m = 230 to m = 270, and the left edge of the second line is determined from the position of the region 22 m = From 340 to m = 380, the right edge of the second line was determined from the position of the region 23, and m = 410 to m = 450.

Next, a set of points representing these four edges is approximated by a least square method with four parallel lines x = ay + b ₁ , x = ay + b ₁ + w ₁ , x = ay + b ₂ , x = ay + b ₂ + w ₂ Then, the edge point fluctuation was obtained by the same method as in Example 1. The fluctuation of the left edge point of the first line obtained for a profile whose y coordinate is given by an integer n is expressed as Δx _1L (n), the fluctuation of the right edge point is expressed as Δx _1R (n), and the like. Fluctuations were calculated for all profiles from n = 1 to n = 512.

  Next, when it was determined whether or not this sample was a non-defective product, all line edge roughness values exceeded 6 nm, and a warning sound was emitted. In order to investigate the cause of the large roughness, shape analysis was further performed, and only the first and third inspection items described in Example 5 were performed.

Spatial frequency analysis, which is the first inspection item, was performed on a total of four edges on both the left and right edges of the first and second lines. The obtained graph was displayed on the screen of the computer 16. Next, the intensity A (f) in the region of 15 <f <256 is approximated by a function A ₀ × 1 / f in the same manner as in the first embodiment, and the obtained fitting parameter A ₀ is substituted. The function A ₀ × 1 / f was plotted on the same graph in all regions where f <256. As a result of calculation for all edges, in all cases, the intensity of the actually measured value exceeded the approximate value at f = 6, 7, 13, 14, 19, 20, 27, and 34. This means that the line width changes at a constant cycle, and the cycle is about 1/7 to 1/6 of the length of the image processed data of 5.40 μm.

  Next, when a correlation coefficient between a set of left and right edge points for one line is obtained, the correlation coefficient is -0.52 for the first line for any value of p. It was ± 0.12, and the second line was in the region of −0.45 ± 0.14, and it was found that there was a strong negative correlation. A graph of the correlation between the left and right edges in the first line when p = 0.5 is shown in FIG.

  When the roughness generation process determination function is executed here, the lithography process that the substrate has undergone is temporarily stopped, together with a display that warns of the abnormality of the underlying pattern that is the cause of roughness generation (6) described in the fifth embodiment, I was instructed to check. When the details of the warning are displayed, it has been pointed out that a periodic pattern having a pitch between 0.7 and 0.9 μm exists on the substrate base, which may cause the reflectance distribution.

  In accordance with the warning, a signal is sent from the computer 16 to the pattern forming apparatus 17 to stop the lithography process and temporarily stop the processes before lithography. When the history of the substrate is referred to, the substrate has a direction perpendicular to the observed line pattern. It was confirmed that a metal line pattern was present at a pitch of 0.8 μm. It was presumed that the antireflection layer was incomplete in the region corresponding to the upper layer of the metal layer, so that the line pattern was thinned. When thorough antireflection was performed based on this, such a phenomenon disappeared and the yield was improved. In addition, by temporarily stopping the process due to a warning, it was possible to minimize the number of wafers on which the antireflection film was formed again.

(Example 7)
A seventh embodiment of the present invention will be described with reference to FIGS. 3 and 17 to 20. FIG. 17 is a flow chart for acquiring image distortion data, FIG. 18 is a schematic diagram of a cross section of a sample structure used, FIG. 19 is a schematic diagram of a microscope screen showing a sample arrangement at the time of observation, and FIG. 20 is a graph of the obtained image distortion amount. .

  Distortion detection and correction data acquisition were performed as follows using the apparatus of FIG. 3 according to the flow shown in FIG.

  First, an operation was performed from the control system 15 of the scanning electron microscope having a length measuring function, and a standard sample made of silicon provided on the stage 12 of the electron beam microscope was observed (step 171). The structure of the sample is shown in a cross-sectional view in FIG. 18 and an upper observation view in FIG. The line-and-space pattern formed on the standard sample is preferable because a magnification of 100,000 times or more is desirable in the measurement of the edge roughness of the line pattern, and in this method, it is necessary to observe the edges of at least two line patterns. The pitch is preferably 0.5 μm or less. The ratio of the line width to the space width is preferably 1 or less. In this example, a sample having a pitch of 0.24 μm and a line width of 0.10 μm was used, and the observation magnification was 200,000 times. The observed area was a square area of 675 nm both vertically and horizontally. As a result of observation, that is, the intensity distribution of the detected secondary electrons was divided into 512 pixels arranged in the region and displayed as gray scale density on the corresponding pixels.

  The initial position was set so that the center of the space part 25 between the two line parts 24 substantially coincided with the vertical axis 26 representing the center of the observation region. At this time, the edge direction of the line pattern was visually parallel to the vertical axis 26.

  Next, a data integration program for strain detection was executed.

  First, the storage area for taking in image data in the control system 15 of the scanning electron microscope was initialized to all zero values. Next, the following first and second procedures were repeated (corresponding to the operations of returning from step 172 to 174 and returning to step 171 in the figure).

  First, the scanning electron microscope scan is performed 8 times, the secondary electron intensity emitted from the sample is integrated and the average value is calculated, and then added to the storage area of the control system (step 172). Secondly, the irradiation of the electron beam is stopped, and it is checked whether or not the set number of repetitions has been reached (step 173). If it has reached, the process proceeds to the next step 175. If not, the scanning position is moved upward in the screen. Move 8 pixels, or 10.55 nm (step 174). Here, the number of repetitions was set to 128. It took about 40 seconds to accumulate the average image data of 8 scans for 128 images. The number of scans in the first procedure is desirably performed at least four times or more for noise reduction. In addition, since it is desirable that the observation area does not overlap at the beginning and the end, the product of the movement distance in the second procedure and the number of times the first and second procedures are repeated is the vertical length of the area that can be observed at one time. It is good to make it more than this. Hereinafter, the upper left of the image is the origin, the distance in the right direction is x, and the distance in the lower direction is y. The numbers of pixels in the x and y directions are represented by m and n.

  The above steps were completed, and the 512 × 512 secondary electron intensity distribution data stored in the storage area of the control system 15 was divided by the number of repetitions to obtain an average value per observation. Next, the process proceeds to step 175, where the obtained 512 × 512 two-dimensional data array is handled as one piece of image data, and noise reduction (step 175) and threshold method are performed by the method described in the first embodiment. Edge detection and approximate line calculation (step 176). Detection was performed on the left and right edges of the first line and the second line in the image. Both the averaging parameter and the smoothing parameter were set to 11. The threshold ratio used was 0.5. Detection was performed for all profiles, and 512 edge points were calculated for one edge. A set of edge point fluctuations was obtained from these data (step 177), and the roughness of the edge, that is, 3σ, was obtained according to (Equation 1) as a reference. It is possible to proceed to the next step 178 without obtaining 3σ.

  The purpose of this example is to detect image distortion mainly due to the influence of an apparatus having a power supply or a power cable installed in the vicinity of a scanning microscope. These image distortions appear in a region having a spatial frequency of 20 or less with respect to the image. Since the spatial frequency 20 is about 25 pixels when converted to a spatial period, it is necessary to use numerical values of 25 or less for the averaging parameter and the smoothing parameter. A larger value of these parameters can reduce noise more, but if it is too large, the image is averaged too much. It is desirable to use a value of 7 to 15 from these balances.

  As a result of the above process, four edge data were obtained. One edge data is composed of position coordinates of 512 edge points. This edge is not an edge that actually exists, but is obtained by averaging the actual edge data in the y direction by the above method. Therefore, the roughness that occurs randomly in the line being observed is eliminated by averaging.

  However, the roughness level 3σ of the edge data actually shows a value of about 3 to 4 nm. This value is large as noise, and the microscope image itself may be distorted. From this data, the correlation coefficient of the left and right edge roughness was determined by the method shown in Example 4 and found to be 0.68. Also, the correlation coefficient of the right and left edge roughness of the second line was as high as 0.55.

  Also, between edges belonging to different lines, that is, (1) the left edge of the first and second lines, (2) the left edge of the first line and the right edge of the second line, (3) the first edge When the edge roughness correlation coefficient is obtained by combining the right edge of the line and the left edge of the second line and (4) the right edge of the first and second lines, both values are 0.5 or more. there were. This indicates that the entire image has an image distortion such that a part of the profile appears to translate in the x direction. Therefore, it was determined in step 179 that the image needs to be corrected.

  Next, the four edge roughness values were averaged for each profile number, and this was used as the image distortion of the microscope itself. A graph of the image distortion amount Δx (n) with respect to the obtained profile number n is shown in FIG. Note that the roughness of the edge close to the center may be regarded as the image distortion amount without averaging the data of the four edges. The image distortion amount data thus obtained was recorded as a file (step 180).

  Subsequently, an arbitrary sample was observed at the same magnification, and an offset of −Δx (n) was applied to the obtained secondary electron intensity profile to correct image distortion. If the amount of distortion is large, the same effect can be obtained by calculating the offset time by dividing the offset amount −Δx (n) by the scanning speed for each profile and shifting the scanning start timing of each profile by the offset time. can get.

  Further, when observing with different observation magnifications, a file of image distortion amount data Δx (n) at each magnification was created by the above procedure, and the image distortion was corrected by the above method using the file.

  As a result, the image distortion of the scanning electron microscope can be removed and the inspection can be performed with high accuracy by an inexpensive and simple method, without the need for thorough hardware repair as in the prior art.

  Although all of the above-described embodiments have been described for the observation of the two-dimensional distribution of secondary electrons with a scanning microscope using an electron beam, the secondary electrons emitted from the sample, such as reflected electrons, have been described. The present invention is applicable even when using a two-dimensional distribution of various particles, and even when using a scanning microscope using an ion particle beam, a charged particle beam such as ionizing radiation, and light. Applicable.

  As described above, according to the present invention, the three-dimensional shape of the pattern edge can be represented as numerical data by scanning microscope observation of a fine pattern, that is, non-destructive inspection. Further, it is possible to quantitatively represent the roughness in the direction along the edge roughness line, the appearance of one line undulating, and the difference in roughness shape between the pattern bottom and the vicinity of the surface.

  Furthermore, by analyzing these results, it is possible to select process candidates that are the main cause of roughness, and to control the manufacturing process of the semiconductor device and the micromachine. In addition, image distortion of the microscope itself used for observation can be extracted, and distortion of the microscope image can be removed from an arbitrary image by a simple and inexpensive method.

Schematic of an edge for explaining a first type of line edge roughness. Schematic of the edge explaining the 2nd type of line edge roughness. The conceptual diagram showing the structure of the apparatus for implementing this invention. The flowchart showing the procedure of the 1st Example of this invention. The schematic of the observation image evaluated in the 1st Example of this invention. The figure showing the line edge obtained by the 1st Example of this invention. The figure which shows the spatial frequency distribution of the line edge roughness obtained by the 2nd Example of this invention. The figure showing the threshold value dependence (1) of the roughness amount of the line edge obtained by the 3rd Example of this invention. The figure showing the threshold value dependence (2) of the roughness amount of the line edge obtained by the 3rd Example of this invention. The figure showing the correlation of the left edge roughness and the right edge roughness in one line obtained by the 4th Example of this invention. The figure showing the threshold value ratio dependence (1) of the correlation coefficient of the left edge roughness and the right edge roughness in one line obtained by the 4th Example of this invention. The figure showing the threshold value ratio dependence (2) of the correlation coefficient of the left edge roughness and the right edge roughness in one line obtained by the 4th Example of this invention. The flowchart explaining the procedure of the 5th Example of this invention. 14 is a flowchart for explaining a roughness generation cause analysis step in the flow shown in FIG. 13. Schematic of the observation image evaluated in the 5th Example of this invention. The figure showing the threshold value ratio dependence of the correlation coefficient of the left edge roughness and the right edge roughness in one line obtained by the 5th Example of this invention. The flowchart showing the procedure of the 6th Example of this invention. The schematic of the structure of the sample observed in the 6th Example of this invention. The schematic diagram of the observation image evaluated in the 6th Example of this invention. The figure which shows the image distortion amount obtained by the 6th Example of this invention.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1, 3 ... Left edge of line, 2, 4 ... Right edge of line, 5 ... Housing of scanning electron microscope, 6 ... Electron gun, 7 ... Electron beam, 8 ... Converging lens, 9 ... Deflector, 10 ... Objective lens, 11 ... sample, 12 ... stage, 13 ... secondary electron, 14 ... detector, 15 ... control system of scanning electron microscope, 16 ... computer for inspection, 17 ... pattern forming device, 18 ... image A region indicating the vicinity of the left edge of the line, 19... A region indicating the vicinity of the right edge of the line in the image, 20... A region indicating the vicinity of the left edge of the first line in the image, 21. A region indicating the vicinity of the right edge of the line, 22... A region indicating the vicinity of the left edge of the second line in the image, 23... A region indicating the vicinity of the right edge of the second line in the image, and 24. Line and space pattern In part, the space part of the line-and-space pattern consisting of 25 ... silicon, the vertical axis representing the center position in the lateral direction of 26 ... image.

Claims (8)

A data processing apparatus for processing two-dimensional electron distribution data obtained by irradiating a predetermined region on a pattern formed on a semiconductor substrate with an electron beam and detecting generated secondary electrons or reflected electrons,
A storage device storing a program for the processing;
Processing means for processing the program,
The processing means includes
A set of edge points corresponding to the edge portion of the pattern is calculated from the two-dimensional electron distribution data,
Calculating an approximate line of the pattern edge;
Calculate edge roughness data determined by the difference between the approximate line of the pattern edge part and the set of edge points and the approximate line,
A data processing apparatus for calculating a correlation or a correlation or a correlation coefficient between the edge roughness data of the first portion and the edge roughness data of the second portion on the pattern. A data processing apparatus for processing two-dimensional electron distribution data of secondary electrons or reflected electrons obtained from a predetermined region on a pattern formed on a semiconductor substrate obtained by a scanning electron microscope,
A storage device storing a program for the processing;
Processing means for processing the program,
The processing means includes
From the two-dimensional electron distribution data, the edge roughness data in the first region on the pattern and the edge roughness data in the second region are calculated,
A data processing apparatus for calculating a correlation or a correlation or a correlation coefficient between the edge roughness data of the first part and the edge roughness data of the second part. A data processing apparatus for processing two-dimensional electron distribution data obtained by irradiating a predetermined region on a pattern formed on a semiconductor substrate with an electron beam and detecting generated secondary electrons or reflected electrons,
A storage device storing a program for the processing;
Processing means for processing the program,
The processing means includes
From the two-dimensional electron distribution data, the edge roughness data in the first region on the pattern and the edge roughness data in the second region are calculated,
A data processing apparatus for calculating a correlation or a correlation or a correlation coefficient between the edge roughness data of the first part and the edge roughness data of the second part. The data processing device according to claim 1, wherein
The pattern is a line pattern;
The data processing apparatus is characterized in that the processing means calculates a correlation or correlation coefficient between the first edge roughness data and the second edge roughness data corresponding to the left and right edge points of the line pattern. In the data processing device according to any one of claims 1 to 3,
The sample is a sample in which a plurality of line patterns are formed,
The processing means calculates a correlation or correlation coefficient between edge roughness data at an arbitrary position of one line pattern and edge roughness data at an arbitrary position of a line pattern different from the one line pattern. Data processing device. In the data processing device according to any one of claims 1 to 5,
A data processing apparatus comprising display means for displaying the obtained correlation or correlation coefficient. In the data processing device according to any one of claims 1 to 6,
The data processing apparatus, wherein the processing means calculates a spatial frequency distribution of the edge roughness data. In the data processing device according to any one of claims 1 to 6,
The data processing apparatus, wherein the processing means uses a standard deviation as an index of the edge roughness.

Images