JP4193760B2 - Image pattern correction method, simulated image generation method using the same, and pattern appearance inspection method - Google Patents

Description

  The present invention mainly relates to an image pattern correction method for correcting a pattern in a design image, a simulated image generation method to which the image pattern correction method is applied, and a pattern appearance inspection method.

  Conventionally, in the appearance inspection of a mask used in a semiconductor integrated circuit such as a reticle or a photomask, higher precision is required as the pattern becomes finer. These masks include a transmission surface that transmits light and a light-shielding surface that does not transmit light during exposure, and a pattern on the mask can be expressed in a binary manner. In the design data of the pattern on the mask, a binary image having a transmissive surface as a white pixel and a light-shielding surface as a black pixel is used as a design image. An image in which the position of the edge of the pattern to be expressed is expressed by multi-gradation pixels with subpixel accuracy is used as a design image.

  FIG. 8 is a schematic view illustrating a conventional design image expressed in multiple gradations. In this design image, one cell is one pixel, the pixel value of a pixel existing in the transmissive surface region (region indicated by pixel p3 in FIG. 8) is 1, and the light shielding surface region (indicated by pixel p2 in FIG. 8). When the pixel value of the pixel existing in the region is 0 and the boundary between the transmission surface and the light shielding surface is at the center of one pixel, the pixel of the pixel on the boundary line indicated by the pixel p1 in FIG. It shows that the value is a column of 0.5. The above-described mask for a semiconductor integrated circuit can be regarded as a processed pattern created by performing processing such as drawing and etching on a semiconductor substrate based on such a design image. Create a reference image from the design image corresponding to the actual image collected by irradiating the processed pattern with laser light or charged particle beam, and compare the actual image with the reference image Visual inspection can be performed.

  In other words, the reference image represents an image collected when the pattern can be processed according to the design data, and if there is a defect such as a missing pattern or a difference in dimensions on the actual image side, the difference between them is It will appear as a difference. Therefore, in order to perform a high-accuracy visual inspection, it is necessary that the normal pattern on the actual image and the corresponding pattern on the reference image exactly match each other. It is necessary to reflect pattern deformation that occurs during processing and image degradation such as image blurring in image capturing performed using an optical system.

  In general, pattern deformation in a process of processing design data into a mask includes deformation in a pattern drawing process as described above and deformation in a chemical process such as etching. The image blur is a linear deterioration process represented by a product-sum operation on a processed pattern of a point spread function that expresses the intensity distribution of a beam to be scanned in a normal inspection. An electron beam is often used for pattern drawing. In this case, pattern deformation in the drawing process is a drawing process response function that is a response function of the semiconductor substrate in the drawing process with the irradiation dose of the beam used for drawing as an input. Can be approximated by the output of

That is, the intensity distribution of the electron beam is expressed as f (r) = A ₁ exp (−r ² / B ₁ ² ) + A ₂ exp (−r ² / B ₂ ) as a function f (r) of the distance r from the beam center. ² ) The accumulated charge amount x at an arbitrary point on the mask expressed by the relational expression 1 is calculated by the product-sum operation of the region centered on that point and the distribution of the relational expression 1. The output value of the drawing process response function using the accumulated charge amount x as an input represents the pattern drawing accuracy at that point after the drawing process. It has been reported that the shape of the drawing process response function is as shown in FIG. 9 (see Non-Patent Document 1).

  In the drawing process response function shown in FIG. 9, the abscissa represents the amount of accumulated charge, and the ordinate represents the amount of light shielding material to be removed by drawing. On the design image, the drawing electron beam is not irradiated in the light shielding region with the pixel value 0, and therefore the amount of the removed light shielding material is zero. In the transmission region with the pixel value 1, the drawing electron beam is sufficiently irradiated and the accumulated charge amount increases. As a result, the amount of light shielding material is completely removed. Pattern deformation in the drawing process occurs particularly at the corners of the image pattern. The reason is that, even if the corner portion is expressed by 90 degrees on the design image, the electron beam used for drawing has a beam aperture of a finite size. Because it ends up. Accordingly, the corner portion in the actual image is observed to be rounded, and in order to create a reference image having a high degree of coincidence with the actual image, the corner portion of the design image also needs to be corrected according to the actual image. .

  Thus, several known techniques are known as techniques for correcting corner portions in an image pattern in a design image when a reference image similar to a real image is created from the design image.

  Specifically, in the case of the system according to the known example 1, pattern shape discontinuity (for one pixel) caused by a quantization error generated in a corner portion of a pattern when a multi-tone design image is binarized. In order to eliminate the occurrence of unevenness), a corner portion is detected from the multi-tone design image by template matching, and a template of a binary corner image prepared in advance is applied to the corner portion (see Patent Document 1).

  In the case of the system according to the known example 2, correction processing for the corner portion of the pattern is realized by deleting the bit at the corner portion of the pattern in the binary design image (see Patent Document 2). FIG. 10 is a schematic diagram for explaining a technique for correcting a corner portion of a pattern in a design image according to Patent Document 2, and FIG. 10 (a-1) relates to a two-dimensional map of a triangular pattern. (A-2) relates to the bit representation, FIG. (B) relates to a bit string scanned and projected, and (c-1) relates to a two-dimensional map obtained by deforming a corner portion of a triangular pattern. The figure (c-2) relates to the bit representation. That is, here, when there is a triangular pattern as shown in FIG. 10 (a-1), the bit representation is as shown in FIG. 10 (a-2). ) To scan in the x and y directions and set “1” if there is at least one image, and “0” otherwise, to obtain a projection bit string in which the pattern is projected in the x and y directions. In addition, the bit of the pixel at the corner corresponding to the position of “1” indicated by a circle in FIG. 10B that is the boundary between “0” and “1” of the projection bit string is set to 0. As a result, the image has a deformed corner as shown in FIG. 10C-1 and a two-dimensional map as shown in FIG. 10C-2.

  In the case of the system according to the known example 3, the pixel value of the design image is calculated for each pixel of the design image by filter processing using a filter such as a circle or a ring, and the corner portion is set to the pixel value of the design image after correction. (See Patent Document 3). FIG. 11 is a schematic diagram for explaining a technique for correcting a corner portion of a pattern in a design image according to Patent Document 3. FIG. 11A relates to a coefficient value of a filter. b) relates to the coefficient sum region of the filter. That is, here, when a filter as shown in FIG. 11A is given, the filter coefficient sum in one region of the two regions divided by the column passing through the filter center is M1, and the filter coefficient sum in the column passing through the center. If M2 is M2, M1 and M2 respectively calculate the coefficient sum in the region shown in FIG. 11B, and M1 = 26 and M2 = 22. When the product-sum operation result of this filter and the design image is C, the pixel value L (C) after correcting the corner portion is 0 when C <M1, and when (M1 + M2) ≧ C ≧ M1 ( When C−M1) / M2 and C> (M1 + M2), the calculation is performed using the relational expression 2 which is 1.

  Therefore, the operation of the function L in the design image in which the linear boundary between the transmission surface represented by the pixel value 1 and the light shielding surface represented by the pixel value 0 as shown in FIG. . However, in FIG. 8, one square represents one pixel. Since the result C of the product-sum operation at the pixel p1 in FIG. 8 is C = M1 × 1 + M2 × 0.5, L = (M1 + M2 × 0.5−M1) /M2=0.5 from the relational expression 2. , The pixel value is the same as before the function L is applied. Since the result C of the product-sum operation at the pixel p2 in FIG. 8 is smaller than M1, L = 0, and since the result C of the product-sum operation at the pixel p3 in FIG. 8 is larger than (M1 + M2), L = 1. It becomes. That is, the function L is a function that changes the pixel value of only the pixels constituting the corner portion without changing the pixel value of the pixel of the design image on the linear pattern.

  Similarly, in the case of the system according to the known example 4, a technique for correcting a corner portion of a pattern in a design image is disclosed (see Patent Document 4). FIG. 12 is a schematic diagram for explaining a technique for correcting a corner portion of a pattern in a design image according to Patent Document 4, and FIG. 12A shows a corner of a main pattern of an optical proximity effect correction mask. (B) is related to the state where the corner portion is surrounded by a broken line with respect to the detection of the line, and (c) is the line detected by the template matching. FIG. 4D relates to a state surrounded by a broken line, and FIG. That is, here, a template as shown in FIG. 12B is used as a correction process for an auxiliary pattern called a serif added to the corner portion of the main pattern of the optical proximity effect correction mask as shown in FIG. 12 is detected by the matching process using the, and the appropriate correction according to the size, shape, and drawing characteristics is applied to the detected serif. FIG. 12D shows that a simple correction process is performed on the corner as shown in FIG.

  In the case of the system according to the well-known example 5, a circle, an ellipse, a straight line, or a curve described by another method is applied to the corner portion of the design image so as to be close to the shape of the corner portion of the actual pattern. It is corrected (see Patent Document 5).

  In the case of the system according to the well-known example 6, the corner portion of the pattern in the design image is determined by determining the degree of correction of the corner portion according to the unevenness of the corner portion of the pattern and the angles of 90 degrees, 45 degrees, and 135 degrees. Is corrected (see Patent Document 6).

Y. C. Patty et al., “An error major doses collection in e-beam nanolithography”, Journal of Vacuum Science Technology B8 (6), 1990, pp. 1882-1888 (YC Pati et al, “an error measurement for dose correction in e-beam nanography”, J. Vac. Sci. Technol. B 8 (6), Nov / Dec 1990p. 18p. .1888) Japanese Patent Laid-Open No. 5-205046 (Summary) JP-A-5-346405 (Abstract) Japanese Patent Laid-Open No. 11-304453 (Summary, page 5, FIGS. 8 to 10) Japanese Unexamined Patent Publication No. 2000-106336 (Summary, pages 6 to 8, FIGS. 4 and 7) JP 2001-338304 A (summary, pages 4 to 6, FIG. 7) JP2003-107669 (Abstract, Page 8)

  In the case of the technique for correcting the corner portion in the pattern in the design image according to Patent Document 1 to Patent Document 6 described above, the correction processing parameter used for the pattern correction processing in the corner portion of the design image is a discrete amount. As a result, there is a drawback that it is difficult to optimize the correction processing parameter based on the statistical learning rule.

  In general, a statistical learning rule such as a neural network uses the real image as a teacher signal and the error between the real image and the reference image as an evaluation function, thereby minimizing the evaluation function value and making corrections necessary to generate the reference image. An optimum value of the processing parameter can be calculated. Since the statistical learning rule requires partial differential calculation related to the correction processing parameter of the evaluation function, the reference image must be expressed by a continuous amount of correction processing parameter. Since the degradation of the real image can be expressed by the product-sum operation of the point spread function representing the intensity distribution of the scanning beam, which is a linear operation, and the design image, the reference image is expressed by a continuous amount of correction processing parameters. This has the same value as the pattern correction processing for the pattern in the image controlled by a continuous amount of correction processing parameters.

  Then, if the point of the problem in the technique which concerns on patent document 1-patent document 6 is demonstrated, in the case of patent document 1, since the correction process of the corner part in a pattern is performed by template matching, a correction process parameter is a template. In the case of Patent Document 2, correction processing parameters for corner portions in an image pattern by bit processing, which require learning, cannot be used. Does not exist. In the case of Patent Document 3, one of the filters numbered in advance is selected to correct a corner portion of the pattern, and the correction processing parameter is a discrete quantity called a filter number. In the case of Patent Document 4, the serif detection process using the template is performed before the corner correction process in the pattern. However, since the template is a discrete expression, Can't use genetic learning. Further, in the case of Patent Document 5, there is no description about the calculation method of the correction processing parameter, and it is necessary to detect the corner portion of the image pattern when correcting the corner portion of the pattern. In the case of Patent Document 6, there is no description about the calculation method of the correction processing parameter, and in addition to detecting the corner from the design image, the determination of the unevenness of the corner in the pattern is not possible. Calculation of the corner angle in the pattern and correction processing of the corner portion in the pattern for each angle discretely expressed as 90 degrees, 45 degrees, and 135 degrees are necessary, and thus the correction processing parameters are calculated by statistical learning. Can not do it.

  The present invention has been made to solve such problems, and its technical problem is that it can optimize the correction processing parameters based on the statistical learning rules, and can easily accurately correct the corners in the pattern in the design image. An object of the present invention is to provide an image pattern correction method that can be well corrected, a simulated image generation method to which the image pattern correction method is applied, and a pattern appearance inspection method.

  According to the present invention, in an image pattern correction method for correcting a pattern in a design image, an image pattern correction method can be obtained in which pattern correction is controlled by a continuous amount of correction processing parameters.

  According to the present invention, in the image pattern correction method, an image pattern correction method in which the correction processing parameter is calculated by a statistical learning rule can be obtained.

  On the other hand, according to the present invention, in a simulated image generation method for generating a simulated image by inputting a design image and performing an image degradation process on an output image obtained by performing the pattern correction process in the design image, the pattern correction is performed. As a process, a simulated image generation method employing any one of the above image pattern correction methods is obtained.

  On the other hand, according to the present invention, an actual image obtained by beam-scanning a processed pattern and converting it into an electrical signal, and a pattern processing process and optical at the time of acquisition of the actual image from a design image corresponding to the processed pattern In a pattern appearance inspection method for inputting a reference image obtained in consideration of an image degradation process by a system and comparing the actual image and the reference image to output a determination result of the presence or absence of a defect in the processed pattern A pattern appearance inspection method to which any one of the above image pattern correction methods is applied as the pattern correction processing for the pattern in the design image is obtained.

  Furthermore, according to the present invention, in the pattern appearance inspection method, the image correction method is applied to the pattern correction processing, and the correction processing parameter is a pattern calculated by a statistical learning rule using an actual image as a teacher signal. An appearance inspection method is obtained.

  In addition, according to the present invention, in the pattern appearance inspection method described above, the pattern correction processing is performed by inputting a processing by one of the filters controlled by a continuous amount of correction processing filter parameters, and a result of the processing of the filter. Thus, a pattern appearance inspection method performed by one of the response functions controlled by the continuous amount of correction processing response function parameters is obtained. In this pattern appearance inspection method, the correction processing filter parameter and the correction processing response function parameter are calculated by a statistical learning rule using an actual image as a teacher signal, and the correction processing filter parameter is a statistical processing using an actual image as a teacher signal. It is preferable that the correction processing response function parameter is calculated by a statistical learning rule using a real image as a teacher signal.

  Further, according to the present invention, in the pattern appearance inspection method, the pattern correction processing is performed using a plurality of filters controlled by the correction processing filter parameters and a processing result of the plurality of filters as inputs. A pattern appearance inspection method performed by integrating output values of a plurality of response functions controlled by function parameters is obtained. In this pattern appearance inspection method, the correction processing filter parameter and the correction processing response function parameter are calculated by a statistical learning rule using an actual image as a teacher signal, and the correction processing filter parameter is a statistical processing using an actual image as a teacher signal. It is preferable that the correction processing response function parameter is calculated by a statistical learning rule using a real image as a teacher signal.

  Furthermore, according to the present invention, in the pattern appearance inspection method, the reference image is obtained using the integrated filter generated by integrating the filter represented by the correction processing filter parameter and the filter for reproducing the image deterioration process. A pattern appearance inspection method for generating In this pattern appearance inspection method, it is preferable that the correction processing filter parameter and the correction processing response function parameter are calculated by a statistical learning rule using an actual image as a teacher signal.

  In the case of the image pattern correction method of the present invention, the pattern correction is basically performed by controlling with a continuous amount of correction processing parameters, and the correction processing parameters are calculated by a statistical learning rule. Is controlled by a continuous amount of correction processing parameters, and correction processing parameters can be calculated using a statistical learning rule such as a neural network. The portion can be easily and accurately corrected. In the case of the present invention, the pattern correction processing is also applied to a simulated image generation method for generating a simulated image by performing image degradation processing on an output image obtained by inputting a design image and performing pattern correction processing in the design image. Similarly, by controlling the pattern correction using the continuous correction parameter calculated by the statistical learning rule, a highly accurate simulated image can be easily generated. Further, in the case of the present invention, an actual image obtained by beam scanning the processed pattern and converting it into an electrical signal, and an image by the optical system at the time of acquiring the pattern processing process and the actual image from the design image corresponding to the processed pattern A pattern appearance inspection method that outputs a determination result of the presence or absence of a defect in a processed pattern by inputting a reference image obtained in consideration of the deterioration process and comparing the actual image and the reference image is also included in the design image. By controlling pattern correction with a continuous amount of correction processing parameters obtained by a statistical learning rule similar to pattern correction processing for patterns, the difference between the processed pattern and the corresponding pattern in the design image This is eliminated by correcting the pattern inside, and the pattern matches the corner of the pattern observed in the real image. It is possible to realize a corner portion in over down on the reference image, resulting becomes possible to perform pattern appearance inspection with high accuracy. In particular, in the pattern appearance inspection method of the present invention, a partial differential value related to the correction processing parameter of the reference image can be calculated. Therefore, the partial differential value is used to statistically minimize the square difference between the actual image and the reference image. By executing the learning rule, it is possible to calculate the optimum correction processing parameter that minimizes the error from the actual image, so that it is possible to perform the pattern appearance inspection with higher accuracy than ever before.

  In the image pattern correction method according to the best mode of the present invention, when a pattern in a design image is corrected, the pattern correction is controlled by a continuous amount of correction processing parameters. However, it is assumed that the correction processing parameter is calculated by a statistical learning rule. As a result, pattern correction processing is controlled by a continuous amount of correction processing parameters, and correction processing parameters can be calculated using a statistical learning rule such as a neural network. The corner portion of the pattern in the image can be easily and accurately corrected.

  That is, in the image pattern correction method of the present invention, a filter representing the intensity distribution of the drawing electron beam is drawn as a technique for correcting pattern deformation in the drawing process, which is one of the pattern processing processes, in order to solve the problems of the prior art. 9 is regarded as a machining process characteristic filter representing the characteristics of the process, and the drawing process response function shown in FIG. 9 is regarded as a machining process response function representing the response of the substrate in the drawing process, and the product sum of the machining process characteristic filter and the design image. An accumulated charge amount in each pixel is calculated by calculation, and the obtained accumulated charge amount is input to a machining process response function to output a corrected pattern.

  An image pattern correction apparatus for carrying out this image pattern correction method has a design image correction execution means constituted by a machining process characteristic filter processing execution means and a machining process response function calculation means.

In the processing process characteristic filter processing execution means, a product-sum operation (F * G) with a processing process characteristic filter F having a coefficient equivalent to the drawing electron beam intensity distribution represented by the relational expression 1 for each pixel in the design image G. In the machining process response function calculation means, the drawing process response function shown in FIG. 9 is set as a sigmoid function S (x) as shown in FIG. 8, and S (x) = {1 + exp (−βx)}. ^-1 is approximated by the relational expression 3. That is, FIG. 1 illustrates a sigmoid function approximated to a drawing process response function used in the image pattern correction method of the present invention.

  Further, the machining process response function is a sigmoid function S (x) of the relational expression 3, and the bias α is added to the value of the product-sum operation (F * G) of each pixel calculated by the machining process characteristic filter processing execution means. The feature amount C is obtained by the relational expression 4 C = (F * G) + α. Thus, if the relational expression 4 is input to the relational expression 3 of the sigmoid function S (x) that is a machining process response function, the image H can be calculated by the relational expression 5 of H = S (C). 5 is output as a design image after pattern correction processing.

  By the way, the image pattern correction method according to the present invention is a method of performing pattern processing and actual image acquisition from a real image obtained by beam scanning a processed pattern and converting it into an electrical signal, and a design image corresponding to the processed pattern. It is applied to a pattern appearance inspection method that outputs a judgment result of the presence or absence of a defect in a processed pattern by inputting a reference image obtained by taking into account the image degradation process by the optical system and comparing the actual image and the reference image. The pattern correction is controlled by a continuous amount of correction processing parameters calculated by a statistical learning rule using a real image similar to the pattern correction processing for the pattern in the design image as a teacher signal. The corner part of the pattern that matches the corner part of the pattern observed in can be realized on the reference image and processed The difference between the turn and the pattern in the design image corresponding thereto is to be solved by the correction of the pattern in the design image, a pattern can be appearance inspection with high accuracy as a result.

  However, in the pattern appearance inspection method of the present invention, the correction processing parameters are calculated by a statistical learning rule using an actual image as a teacher signal. However, the pattern correction processing is controlled by a continuous amount of correction processing filter parameters. The processing by one of the filters to be processed, and the result of the processing of the filter is input, and the processing is performed by one of the response functions controlled by the continuous amount of correction processing response function parameters, or the correction processing Performed by integrating the processing by a plurality of filters controlled by the filter parameters and the output values of the plurality of response functions controlled by the correction processing response function parameters with the processing results of the plurality of filters as inputs. It can be. In any case, it is preferable that the correction process filter parameter and the correction process response function parameter are selectively calculated by a statistical learning rule using a real image as a teacher signal. In addition, when generating a reference image, a reference image can be generated using an integrated filter generated by integrating a filter expressed by a correction processing filter parameter and a filter for reproducing an image degradation process. . Even in this case, it is preferable that the correction process filter parameter and the correction process response function parameter are calculated by a statistical learning rule using an actual image as a teacher signal.

In any case, the present invention has a correction processing filter parameter expressing the filter processing and a correction processing response function parameter expressing the response function as the correction processing parameters, and A1, A2, B1, B2 is a correction processing filter parameter, α in relational expression 4 and β in relational expression 3 are correction processing response function parameters. Assuming that the image spread process reproduction filter P is used as a point spread function representing a beam to be scanned in the pattern appearance inspection, the reference image Ref is Ref = H * using a function H of a correction processing parameter that is a continuous amount. It can be expressed as P. Accordingly, the partial differentiation of the reference image Ref with respect to the correction processing parameters A1, A2, B1, B2, α, β is ∂Ref / ∂A ₁ = (∂H / ∂A ₁ ) * P = {(∂C / ∂A ₁ ) × [βexp (−βC) / {1 + exp (−βC)} ² ]} P = [{exp (−r ² / B ₁ ² ) * G} × βexp (−βC) / {1 + exp (−βC) } ² ] * P, ∂Ref / ∂A ₂ = (∂H / ∂A ₂ ) * P = {(∂C / ∂A ₂ ) × [βexp (−βC) / {1 + exp (−βC)} ² ] } * P = [{exp (−r ² / B ₂ ² ) * G} × βexp (−βC) / {1 + exp (−βC)} ² ] * P, ∂Ref / ∂B ₁ = (∂H / ∂ _{B 1) * P = {(} ∂C / ∂B 1) × [βexp (-βC) / {1 + exp (-βC)} 2]} * P = [{(2A 1 r 2 / B 1 3) e ^{_{p (-r 2 / B 1 2}} ) * G} × βexp (-βC) / {1 + exp (-βC)} 2] * P, ∂Ref / ∂B 2 = (∂H / ∂B 2) * P = {(∂C / ∂B ₂ ) × [βexp (−βC) / {1 + exp (−βC)} ² ]} * P = [{(2A ₂ r ² / B ₂ ³ ) exp (−r ² / B ₂ ² ) * G} × βexp (−βC) / {1 + exp (−βC)} ² ] * P, ２Ref / ∂α = (∂H / ∂α) * P = {(∂C / ∂α) × [ βexp (−βC) / {1 + exp (−βC)} ² ]} * P = [βexp (−βC) / {1 + exp (−βC)} ² ] * P, ∂Ref / ∂β = (∂H / ∂β ) * P = [Cexp (−βC) / {1 + exp (−βC)} ² ]} * P, and using these partial differential values, the square difference between the real image Real and the reference image Ref By executing the statistical learning rule so as to minimize Diff = (Real-Ref) 2 that is Diff, it is possible to calculate the optimum correction processing parameter that minimizes the error from the real image Real.

  Hereinafter, an image pattern correction apparatus for carrying out the image pattern correction method (applicable to the pattern appearance inspection method) of the present invention will be specifically described with reference to some examples.

  FIG. 2 is a block diagram illustrating basic functions of the image pattern correction apparatus according to the first embodiment of the present invention.

  The image pattern correction apparatus according to the first embodiment includes a processed pattern scanning actual image input unit 1 that forms a real image by scanning a processed pattern with a laser beam, a charged particle beam, or the like and converting it into an electrical signal. Design data input means 2 for taking in design data as an image, design image correction processing filter coefficient calculation means 3 for calculating the shape of a machining process characteristic filter used for design image correction processing and the shape of a machining process response function, and an image deterioration reproduction filter The image deterioration reproduction filter coefficient calculation means 4 for calculating the image deterioration reproduction filter shape used for the processing, the design image correction processing execution means 5 for correcting the pattern in the design image according to the actual image, and the optical system used for the scanning Filter processing that reproduces the deterioration of the actual image caused by the image processing is performed on the output image of the design image correction processing execution means 5 to obtain a reference image. The pattern appearance inspection is performed by comparing the image degradation reproduction filter processing execution means 6 to be output, the actual image obtained by the processed pattern scanning actual image input means 1 and the reference image obtained by the image degradation reproduction filter processing execution means 6. And an inspection result output means 8 for outputting a pattern appearance inspection result as a comparison result in the image comparison means 7.

  In this image pattern correction apparatus, a precondition that the design image correction processing filter coefficient calculation means 3 uses correction processing parameters calculated by a statistical learning rule, and a design image surrounded by a broken line in FIG. The correction execution means 5 is a new functional part. FIG. 3 is a block diagram showing detailed functions of the design image correction execution means 5.

  Referring to FIG. 3, the design image correction execution unit 5 includes a processing process characteristic filter processing execution unit 11 that performs a filtering process using a processing process characteristic filter on the design image that is an output of the design data input unit 2. The machining process response function calculation means 12 is provided for calculating the output value by inputting the execution result of the filter process into the machining process response function. However, the processing process characteristic filter processing execution means 11 here is controlled by a continuous amount of correction processing filter coefficients to perform filtering on the design image, and the processing process response function calculation means 12 is a continuous amount correction process. Designed after pattern correction processing by controlling the response function parameters (all correction processing parameters are calculated by a known statistical learning rule) and inputting the output result into the machining process response function to calculate the output value. It is designed to output as an image.

  The overall operation of the image pattern correction apparatus will be described below with reference to FIGS. In the following description, attention is paid to a drawing process as a processing process, and an electron beam is used as a drawing beam.

  In this image pattern correction apparatus, the processed pattern scanning actual image input means 1 obtains an actual image by scanning the processed pattern with a laser beam or a charged particle beam and converting it into an electrical signal, and sends it to the image comparison means 7. Output.

  The design data input means 2 takes in design data corresponding to the processed pattern scanned by the processed pattern scanning actual image input means 1 and outputs it as a design image to the design image correction processing execution means 5.

In the design image correction processing filter coefficient calculation means 3, the drawing electron beam of the relational expression 1 calculated by using a known statistical learning rule such as a neural network, the Levenberg-Muckart method, and the conjugate gradient method with the actual image as a teacher signal. Intensity distribution parameters A ₁ , A ₂ , B ₁ , and B ₂ are used as correction processing filter parameters, and α in relational expression 4 and β in relational expression 3 are read as correction processing response function parameters to represent the characteristics of the drawing process. A filter having a coefficient equivalent to the drawing electron beam intensity distribution of the relational expression 1 is used as a machining process characteristic filter, and the response function of the relational expression 3 is calculated as a machining process response function and output to the design image correction processing execution means 5. .

  The image degradation reproduction filter coefficient calculation means 4 reads, as a parameter, the spread of the blur function for reproducing the blur in the actual image caused by the optical system used for scanning the processed pattern, and calculates the shape of the image degradation reproduction filter. And output to the image deterioration reproduction filter processing execution means 6.

  The design image correction processing execution means 5 uses the machining process characteristic filter and the machining process response function calculated by the design image correction processing filter coefficient calculation means 3 for the design image output from the design data input means 2. A pattern correction process is performed, and a corrected image of the pattern is output to the image deterioration reproduction filter process execution means 6.

  In the image deterioration reproduction filter processing execution means 6, the correction processing image of the pattern output from the design image correction processing execution means 5 is applied according to the shape of the image deterioration reproduction filter calculated by the image deterioration reproduction filter coefficient calculation means 4. The image deterioration reproduction process is performed, and the result is output to the image comparison means 7 as a reference image.

  In the image comparison means 7, the actual image obtained by the processed pattern scanning actual image input means 1 and the reference image obtained by the image deterioration reproduction filter processing execution means 6 are compared on an appropriate basis, and the comparison result is compared with the inspection result. Output to the output means 8. The inspection result output means 8 outputs the pattern appearance inspection result, which is the comparison result output from the image comparison means 7.

  By the way, referring to FIG. 3, in the design image correction processing execution means 5, the product-sum operation using the processing process characteristic filter calculated by the design image correction processing filter coefficient calculation means 3 by the processing process characteristic filter processing execution means 11. For each pixel of the inputted design image, and the processing process response function calculation means 12 calculates the relational expression to the product-sum operation value calculated by the processing process characteristic filter processing execution means 11 for each pixel in the design image. 4 is added to the machining process response function calculated by the design image correction processing filter coefficient calculation means 3, and the output value is used as the pixel value (pattern correction processing image) of the design image after the pattern correction processing. To do.

  FIG. 4 is a flowchart showing the basic operation of the image pattern correction apparatus according to the first embodiment.

  In the case of this image pattern correction apparatus, referring to the block diagrams of FIGS. 2 and 3, in contrast to the configuration of FIG. 2, step A1 is processed pattern scanning actual image input means 1 in FIG. 2, and step A2 is design data. Input means 2, steps A3 and A4 are design image correction processing filter coefficient calculation means 3, step A5 is image deterioration reproduction filter coefficient calculation means 4, steps A6 to A13 are design image correction processing execution means 5, and step A14 is image deterioration reproduction. The filter processing execution means 6, step A15 represents the operation of the image comparison means 7, step A16 represents the operation of the inspection result output means 8, and in contrast to the configuration of FIG. Step A9 represents each operation of the machining process response function calculation means 12.

Specifically, in step A1, the processed pattern is scanned with a laser or a charged particle beam to convert it into an electrical signal, and an A × A array RealImg [A] [A] as an A × A real image. ]. In step A2, design data corresponding to the actual image (processed pattern) captured in step A1 is imaged and stored in an A × A array Design [A] [A]. In step A3, the size of the process characteristic filter is set to M × M, and A ₁ , A ₂ , B ₁ , B ₂ in relational expression ₁ , β in relational expression 3 and α in relational expression 4 are set. A value is given. In step A4, a filter F [M] [M] having the same shape as the distribution of relational expression 1 is calculated using A ₁ , A ₂ , B ₁ , B ₂ input in step A3 and stored in an array. . In step A5, the size of the image degradation reproduction filter is set to N × N, and image blur is expressed by reading the spread of the pattern scanning beam as a parameter as image degradation caused by the optical system used for scanning the processed pattern. A blur filter P [N] [N] for reproducing the image degradation process is calculated and stored in the array.

  In steps A6 and A7, array subscripts j = 1 and i = 1 are initialized. In step A8, a product-sum operation using the filter (array) F is performed in an M × M region centered on the pixel (i, j) in the design image Design, and the result is stored as a product-sum operation value C. Is done. In step A9, the pixel value C ′ after the correction processing is calculated from the product-sum operation value C obtained in step A8 and the parameters α and β obtained in step A3, and C ′ = 1 / (1 + exp {−β (C + α )}) Is used as an image value after the pattern correction processing of the pixel (i, j), and is stored in the array Design ′ as Design ′ [i] [j] = C ′. In step A10, it is determined whether i = A. If i = A, it is subsequently determined in step A11 whether j = A. If i = A, step A12 is not performed. If i = i + 1, the process returns after i = 1 in step A7. If j = A, the process returns to j = j + 1 in step A13 and then returns after j = 1 in step A6. Until the correction is made, the processes of steps A8 and A9 are repeated.

  In step A14 in the case of j = A, the product-sum operation process using the filter (array) P set in step A5 is performed on the array Design ′ which is the design image after the correction process, The designed image after the pattern correction processing is stored in the array RefImg [A] [A] as a reference image. In step A15, the actual image RealImg obtained in step A1 is compared with the reference image RefImg obtained in step A14, and a pattern appearance inspection is executed according to an appropriate standard. In step A16, the inspection result performed in step A15 is output.

  In the image pattern correction apparatus according to the first embodiment, the case where an electron beam is used to obtain the drawing beam intensity distribution has been described. However, the drawing beam intensity distribution is controlled by a continuous amount of correction processing filter parameters. If so, the shape may be changed according to the type of drawing beam to be used. In addition, as a response function in the drawing process, the sigmoid function assuming that the drawing beam is an electron beam has been described, but when the drawing beam is another beam, it is controlled by a continuous correction filter parameter corresponding thereto. A drawing process response function may be used. Furthermore, although the drawing process has been described as an example of the pattern processing process, it is controlled by a characteristic filter controlled by a continuous correction filter parameter expressing a chemical process such as etching and a continuous correction process response function parameter. It is also possible to correct the design image using the response function.

  FIG. 5 is a block diagram showing detailed functions of the design image correction execution means 5 ′ provided in the image pattern correction apparatus according to the second embodiment of the present invention.

  The image pattern correction apparatus according to the second embodiment includes n filters (n ≧ 1) used for the machining process characteristic filter process and m response functions (m ≧ 2) used for calculating the machining process response function. The design image correction process is executed.

In the case of the design image correction processing execution means 5 'here, 1, 2,..., N (where n.gtoreq.1) machining process characteristic filter processes are executed on the design image output from the design data input means 2. Filter processes by the machining process characteristic filters ₁ to _n are respectively executed in the means 11 _{1 to} 11 _n , and the output results of the respective machining process characteristic filter processing execution means 11 _{1 to} 11 _n are respectively 1, 2,..., M (≧ 2 ) The machining process response function calculation means 12 _{1 to} 12 _m are inputted to the machining process response function calculation means 12 _{1 to} 12 _m to calculate the output values of the machining process response functions ₁ to _m , respectively. output to the output value integrating means 13, in the course of processing the response function output value integrating means 13 integrates the output of each machining process response function calculating means 12 ₁ to 12 _m and design image after pattern correction To output.

That is, when this design image correction processing means 5 'is compared with the design image correction processing means 5 shown in FIG. 3, there are provided n (≧ 1) machining process characteristic filter processing execution means 11 in the configuration, and the machining process M (≧ 2) response function calculating means 12 are provided, and machining process response function output value integrating means 13 for integrating the outputs of the machining process response function calculating means 12 _{1 to} 12 _m is added. The point is different.

  The overall operation of this image pattern correction apparatus will be described below with reference to FIG. However, in order to simplify the description here, correction processing is performed only at the corners of the pattern using one machining process characteristic filter and two machining process response functions, and correction processing is performed at other portions. The case where there is not will be described.

Here, the function L of the relational expression 2 is a polygonal line function related to the product-sum operation value C, but the relational expression L is related by regarding the polygonal line function as a rough approximation of the response function of the drawing process shown in FIG. The design image pattern correction process is executed as a second response function together with the sigmoid function of Expression 3. However, when using the response function L, the filter coefficient sum in one of the two regions divided by the column passing through the filter center as shown in FIG. A filter coefficient sum in a column passing through the filter center is M2. Since the drawing electron beam intensity distribution in relational expression 1 is a symmetric distribution, the same result is obtained even when the coefficient sum in the left region of FIG. 11B is M1. The pixel value of the pixel Pix on the design image before the pattern correction processing is q, the product-sum operation value (product-sum operation result) C of the processing process characteristic filter and the design image in the pixel Pix, α in the relational expression 4, and the relational expression 3 and the filter coefficient sums M1 and M2, the function S of the relational expression 3 and the function L of the relational expression 2 are calculated, and {1-exp (− {q−L (C)} ² / γ ² }) } S (C) + exp (− {q−L (C)} ² / γ ² ) The linear combination of the relational expression 6 as expressed by L (C) is used as the pixel value of the pixel Pix after the pattern correction processing. To do.

  Using this relational expression 6, by adjusting the value of γ, the coefficient of the response function L can be set to 1 in the pattern other than the corner portion of the pattern, and the coefficient of the response function S can be set to 1 in the corner portion of the pattern. The correction process that reflects the drawing process characteristics is performed only at the corners of the pattern, and the correction is not performed on the linear pattern.

In the process characteristic filter processing execution means 11 ₁ in FIG. 5, the relational expression 1 calculated from the correction processing parameters A ₁ , A ₂ , B ₁ , B _{2 in the} design image correction processing filter coefficient calculation means 3 shown in FIG. The product-sum operation is performed on the design image using the machining process characteristic filter 1 having a coefficient equivalent to the drawing electron beam intensity distribution. In one th machining process response function calculating unit 12 _1, as a processing step response function, entered in the course of processing characteristic filter processing executing means 11 ₁ of the product-sum operation result design image correction processing filter coefficient calculation means 3 shown in FIG. 2 The value of the function of the relational expression 3 is calculated for each pixel by using the β of the sigmoid function of the relational expression 3 of the corrected correction parameters and the α of the relational expression 4.

In 2 -th machining process response function calculating unit 12 _2, the processing step response function, calculating the values of the function L for each pixel from the processing process characteristic filter processing executing means 11 ₁ of the product-sum operation result as equation 2 which . The processing process response function output value integration unit 13 linearly combines the values of the function S and the function L in each pixel according to the relational expression 6 and outputs the result as a design image after pattern correction processing.

  FIG. 6 is a flowchart showing the basic operation of the image pattern correction apparatus according to the second embodiment.

In the case of this image pattern correction apparatus, referring to the block diagrams of FIGS. 2 and 5, in contrast to the configuration of FIG. 2, step B1 is processed pattern scanning actual image input means 1, and step B2 is design data input means 2. Steps B3 to B5 are design image correction processing filter coefficient calculation means 3, Step B6 is image deterioration reproduction filter coefficient calculation means 4, Steps B7 to B16 are design image correction processing execution means 5, and Step B17 is image deterioration reproduction filter processing execution. Means 6 and step B18 represent the operations of the image comparison means 7 and step B19 represent the respective operations of the inspection result output means 8, and step B9 is a processing process characteristic filter processing execution means 11 ₁ and step B10 in contrast to the configuration of FIG. , B11 machining process response function calculating unit ₁₂ 1, 12 _2, step B12 is processed process response function output value integration It represents the operation of each of the stages 13.

  Note that steps B1 and B2 in FIG. 6 are steps A1 and A2 in FIG. 4, respectively, steps B4 and B6 in FIG. 6 are steps A4 and A5 in FIG. 4, and steps B7 to B9 in FIG. Steps A6 to A8 in FIG. 4 and Steps B13 to B19 in FIG. 6 perform the same operations as Steps A10 to A16 in FIG.

Will describe the different part, in step B3, the size of the machining process characteristic filter and M × M, _A 1 in relation _{1, A 2, B 1,} B 2, β in relation 3, equation 4 And the value of γ in relational expression 6 used for integrating the response function calculation results. In step B5, as shown in FIG. 11B for the filter (array) F, one coefficient sum of two regions of the filter divided by the column passing through the filter center is M1, and the filter coefficient in the column passing through the filter center The sum is calculated as M2. In step B10, the output of the response function S is calculated as C1 ′ = 1 / (1 + exp {−β (C + α)}) according to the relational expression 5 from the product-sum operation values C and α. In step B11, the output C2 ′ of the response function L is calculated from the filter coefficient sums M1 and M2 and the product-sum operation value C according to the relational expression 2. In Step B12, δ = C2′-Design [i] [j] is set, and C ′ = (1−exp {−δ × δ / (γ × γ)}) × C1 ′ + exp according to the relational expression 6 using γ. C ′ obtained by {−δ × δ / (γ × γ)} × C2 ′ is linearly integrated with pixel values (C1 ′, C2 ′) after pattern correction processing in the pixel (i, j), and the result is The pixel value of the design image after the pattern correction processing is stored), and the design is stored as an array Design '[i] [j] = C'.

In the case of the image pattern correction apparatus according to the second embodiment, one of the processing process characteristic filter processing execution unit 11 ₁ comprises a machining process characteristic filter 1, two machining processes response function calculating unit 12 _1, 12 ₂ However, the number of machining process characteristic filter processing execution means (machining process characteristic filter) and machining process response function calculation means (machining process response function) is limited to this. It can be set arbitrarily. Also, here, the case of linear combination is described as an example of the method of integrating the output values of the characteristic process response function, but instead of this, it is also possible to integrate by using an average image or using a difference image It is. Further, here, as a method for integrating the output values of the characteristic process response function, a configuration is described in which one design image after pattern correction processing is output. However, according to a case where a plurality of filters are used in image degradation reproduction filter processing, etc. It is also possible to output a plurality of design images after correction processing.

  FIG. 7 is a block diagram illustrating basic functions of the image pattern correction apparatus according to the third embodiment of the present invention.

  The image pattern correction apparatus according to the third embodiment includes a processing process characteristic filter and an image deterioration reproduction filter when the processing process response function is a linear function and the image deterioration reproduction filter process is a linear process such as a product-sum operation. By integrating, the design image correction process and the image deterioration reproduction filter process are performed together.

  That is, when this image pattern correction apparatus is compared with the image pattern correction apparatus shown in FIG. 2, the filter used in the process characteristic filter process in the design image correction process and the filter used in the image deterioration reproduction filter process are integrated on the configuration. The difference is that there is a filter integration means 20 and an integrated filter processing execution means 21 in which the design image correction processing execution means 5 and the image degradation reproduction filter processing execution means 6 are integrated.

  The overall operation of the image pattern correction apparatus will be described below with reference to FIG. However, here, attention is paid to the drawing process as the processing process, and an electron beam is used as the drawing beam. Further, an identity function that outputs the product-sum operation value of the drawing electron beam as it is is assumed as a linear machining process response function. Therefore, a reference image generation process in which a product-sum operation of the processing process characteristic filter F having a coefficient equivalent to the drawing electron beam intensity distribution is performed on the design image G, and a product-sum operation by the image deterioration reproduction filter P is performed thereafter is P *. Relational expression 7 is obtained as (F * G) = (P * F) * G. As a result, the processing is equivalent to executing the product-sum operation of the filter integrated with the relational expression 8 of P * F and the design image G. The processed pattern scanning actual image input means 1, the design data input means 2, the design image correction processing filter coefficient calculation means 3, the image deterioration reproduction filter coefficient calculation means 4, the image comparison means 7, the inspection result output means 8, Are the same as those provided in the image pattern correction apparatus shown in FIG.

  In the case of this image pattern correction apparatus, the filter integration means 20 calculates the processing process characteristic filter F and the image deterioration reproduction filter coefficient having coefficients equivalent to the drawing electron beam intensity distribution set by the design image correction processing filter coefficient calculation means 3. The sum-of-products calculation process represented by the relational expression 8 is performed on the image deterioration reproduction filter P set by the means 4, and the integrated filter Q is created. The integrated filter process execution means 21 executes a product-sum operation process using the integrated filter Q on the design image, and outputs it as a design image after the integrated filter process.

  In the case of the image pattern correction method of the present invention, a simulated image generation method is also effective in which a simulated image is generated by performing image degradation processing on an output image obtained by inputting a design image and performing pattern correction processing in the design image. Similarly to the pattern correction processing here, if the pattern correction is controlled by the continuous correction parameter calculated by the statistical learning rule, a highly accurate simulated image can be easily generated.

6 illustrates a sigmoid function approximated to a drawing process response function used in the image pattern correction method of the present invention. It is the block diagram which showed the basic function of the image pattern correction apparatus which concerns on Example 1 of this invention. It is the block diagram which showed the detailed function of the design image correction execution means with which the image pattern correction apparatus shown in FIG. 2 is equipped. 3 is a flowchart showing a basic operation of the image pattern correction apparatus shown in FIG. It is the block diagram which showed the detailed function of the design image correction execution means with which the image pattern correction apparatus which concerns on Example 2 of this invention is equipped. 6 is a flowchart showing a basic operation of the image pattern correction apparatus described in FIG. 5. It is the block diagram which showed the basic function of the image pattern correction apparatus which concerns on Example 3 of this invention. It is the schematic diagram which illustrated the design image expressed by the conventional multi-gradation. It is the schematic diagram which illustrated the form of the reference drawing process response function shown by the conventional nonpatent literature 1. It is the schematic diagram shown in order to demonstrate the technique which correct | amends the corner part of the pattern in the design image based on the conventional patent document 2, (a-1) is related with the two-dimensional map of a triangular pattern, (a- 2) relates to the bit representation, (b) relates to the scanned and projected bit string, (c-1) relates to a two-dimensional map obtained by transforming a corner portion of a triangular pattern, and (c-2) relates to the bit representation. It is about. It is the schematic diagram shown in order to demonstrate the technique which correct | amends the corner part of the pattern in the design image based on the conventional patent document 3, (a) is related with the coefficient value of a filter, (b) is the coefficient sum of a filter. It is about the area. It is the schematic diagram shown in order to demonstrate the technique which correct | amends the corner part of the pattern in the design image based on the conventional patent document 4, (a) is called a serif in the corner part of the main pattern of an optical proximity correction mask. (B) relates to the state in which the corner portion is surrounded by a broken line with respect to the detection of the line, (c) relates to the state in which the line detected by template matching is surrounded by the broken line, (D) relates to the post-correction state of the corner portion.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Processed pattern scanning real image input means 2 Design data input means 3 Design image correction processing filter coefficient calculation means 4 Image degradation reproduction filter coefficient calculation means 5, 5 'Design image correction processing execution means 6 Image degradation reproduction filter processing execution means 7 Image comparison means 8 Inspection result output means 11, 11 _{1 to} 11 _n Machining process characteristic filter processing execution means 12, 12 _{1 to} 12 _m Machining process response function calculation means 13 Machining process response function output value integration means 20 Filter integration means 21 Integration Filter processing execution means

Claims (1)

Considering the actual image obtained by beam scanning the processed pattern and converting it to an electrical signal, and the pattern processing process from the design image corresponding to the processed pattern and the image degradation process by the optical system at the time of acquisition of the actual image In the pattern appearance inspection method for outputting the determination result of the presence or absence of defects in the processed pattern by inputting the reference image obtained by comparing the actual image and the reference image,
The correction of the pattern in the design image is performed by controlling with a continuous amount of correction processing parameters,
The pattern correction processing is processing by one of the filters controlled by a continuous amount correction processing filter parameter, and a response function controlled by a continuous amount correction processing response function parameter by inputting a result of the filter processing. Performed by one of the
The correction filter coefficient of the filter is expressed by the sum of M Gaussian distributions, and the correction filter parameter is given by each height Ai and spread Bi (i = 1, 2,..., M) of the Gaussian distribution,
For the reference image Ref (θ) and the real image Real generated using the correction processing filter parameter θ = (Ai, Bi) (i = 1, 2,..., M), Err = (Real− R
ef (θ)) An image pattern appearance inspection method characterized by determining θ based on a steepest descent method so that ² is minimized.

Images