Inspection and Measuring Equipment

Abstract

To provide an inspection device and a measurement device that can improve the accuracy of evaluation and measurement, in evaluation and dimension measurement for a defect inspection or the like, by using a distortion correction technique suitable for a region excluding an abnormal portion such as a defect.SOLUTION: An inspection device 301 includes: an image distortion estimation unit 303 configured to estimate an amount of distortion between a reference image 101 and an inspection image 102; an image distortion correction unit 305 configured to use an estimated distortion amount 304 to correct the inspection image 102 and / or the reference image 101; and an inspection unit 307 configured to perform inspection using the corrected inspection image and the reference image, or the inspection image and the corrected reference image. The image distortion estimation unit 303 estimates, by adjusting correction conditions, a distortion amount that can correct only the distortion that occurs in the entire image.SELECTED DRAWING: Figure 3

Description

[Technical field]

[0001] The present invention relates to an image processing technique for processing image data, and more particularly to an inspection apparatus and a measurement apparatus suitable for inspection or measurement using image data. [Background technology]

[0002] A method is used to compare an inspection image with a reference image for evaluation of defects in semiconductor circuits, etc. The reference image can be the design data of the sample to be evaluated or the object to be measured, an image obtained by simulating a captured image using design data through machine learning, image statistics that represent the distribution of luminance values ​​for each pixel that can be taken by a captured image of a sample manufactured from the same design data, or another region of a captured image with the same pattern as the sample.

[0003] In defect inspection and dimensional measurement, image distortion due to image capture occurs in the captured image generated by the device, and local positional deviation may occur between the reference image and the inspection image. In this case, a part that is originally normal in an area with image distortion may be judged as defective, and measurement errors may increase. Patent Document 1 discloses a method of eliminating positional deviation and suppressing false detection of defects by estimating the shift amount that minimizes the error between two images using SSD (sum of squared difference) matching. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent No. 4239088 Summary of the Invention [Problem to be solved by the invention]

[0005] The misalignment that occurs between the reference image and the inspection image can be due to image distortion or due to anomalies such as defects. In order to apply an inspection image with image distortion to an evaluation such as defect inspection, it is necessary to remove only the misalignment due to image distortion. In Patent Document 1, the shift amount that minimizes the difference between the inspection image and the reference image, which are calculated for the entire image and for each partial region, is used as the distortion amount for position matching. However, since it does not take into account position shifts due to defects or other abnormalities, over-fitting may occur for inspection images that have defects or other abnormalities. As a result, there is a concern that areas that are actually abnormal may not be detected as normal.

[0006] Therefore, the present invention provides an inspection apparatus and a measurement apparatus that can improve the accuracy of evaluation and measurement in defect inspection and other evaluations and dimensional measurements by using a distortion correction technique that is adapted to areas excluding abnormal parts such as defects. [Means for solving the problem]

[0007] In order to solve the above problems, an inspection device according to the present invention is an inspection device that inspects an inspection image by comparing it with a reference image, and includes an image distortion estimation unit that estimates an amount of distortion between the reference image and the inspection image, an image distortion correction unit that corrects the inspection image and / or the reference image using the estimated amount of distortion, and an inspection unit that performs inspection using the corrected inspection image and reference image, or the inspection image and the corrected reference image, and the image distortion estimation unit estimates an amount of distortion by adjusting correction conditions. Except for local and high frequency image features Estimating the amount of distortion wherein the image distortion estimation unit performs pre-processing on an inspection image and a reference image, the processing content of which can be changed according to a correction condition, estimates a first distortion amount by inputting the inspection image and the reference image processed by the pre-processing, performs post-processing defined by a correction condition on the first distortion amount, and calculates a second distortion amount by comparing the estimated distortion amount with the It is characterized by: Further, a measurement apparatus according to the present invention is a measurement apparatus for measuring a length of a sample using an inspection image, and includes an image distortion estimation unit for estimating an amount of distortion between a reference image and the inspection image based on the reference image and the inspection image, an image distortion correction unit for correcting the inspection image using the estimated amount of distortion, and a measurement unit for measuring the length of the sample using the corrected inspection image, wherein the image distortion estimation unit estimates an amount of distortion between the reference image and the inspection image based on the reference image and the inspection image, Except for local and high frequency image features Estimating the amount of distortion The image distortion estimation unit performs pre-processing on an inspection image and a reference image, the processing content of which can be changed according to a correction condition, estimates a first distortion amount by inputting the inspection image and the reference image processed by the pre-processing, performs post-processing defined by the correction condition on the first distortion amount, and sets a second distortion amount as the estimated distortion amount. It is characterized by: Effect of the Invention

[0008] According to the present invention, it is possible to provide an inspection apparatus and a measurement apparatus that can improve the accuracy of evaluation and measurement in defect inspection and other evaluations and dimensional measurements by using a distortion correction technique that is adapted to areas excluding abnormal parts such as defects. Problems, configurations and effects other than those described above will become apparent from the following description of the embodiments. [Brief description of the drawings]

[0009] [Figure 1] FIG. 2 illustrates an example of a reference image and an inspection image having image distortion. [Figure 2A] 1A and 1B are diagrams showing a reference image and an inspection image having a position shift due to image distortion and a defect. [Figure 2B] 13A and 13B are diagrams showing the distortion correction amount that has caused overfitting to a defective portion and a post-correction inspection image. [Figure 2C] 4A and 4B are diagrams showing the amount of distortion correction based on the amount of distortion estimated by an image distortion estimation unit constituting an inspection device according to an embodiment of the present invention, and a corrected inspection image. [Diagram 3] 1 is a functional block diagram showing an example of the overall configuration of an inspection device according to a first embodiment of the present invention. [Figure 4] FIG. 4 is a diagram showing an example of a correction condition shown in FIG. [Diagram 5] 4 is a flowchart showing a processing operation of the inspection device according to the first embodiment. [Figure 6] 5 is a flowchart of an image distortion estimation unit included in the inspection device according to the first embodiment. [Figure 7] FIG. 11 is a functional block diagram showing an example of the overall configuration of an inspection device according to a second embodiment of the present invention. [Figure 8] 8 is a flowchart of the machine learning unit shown in FIG. 7. [Figure 9] 10 is a flowchart of an image distortion estimation unit included in the inspection device according to the second embodiment. [Figure 10A] FIG. 13 is a diagram showing an example of measuring an image region without image distortion. [Figure 10B] FIG. 13 is a diagram showing an example of measuring an image region having image distortion. [Figure 11] FIG. 11 is a functional block diagram showing an example of the overall configuration of a measurement device according to a third embodiment of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0010] The inspection and measurement devices exemplified in this specification relate to inspection and measurement devices that estimate only the distortion that occurs throughout the entire image between a reference image and an inspection image, and are capable of correcting for any misalignment caused by defects or other abnormalities, thereby improving inspection and measurement performance. In this specification, the sample to be inspected is a semiconductor circuit photographed by a scanning electron microscope (SEM), but is not limited to this. Needless to say, the present invention can also be applied to images photographed by other imaging devices. Hereinafter, an embodiment of the present invention will be described with reference to the drawings. EXAMPLES

[0011] FIG. 1 is a diagram showing an example of a reference image and an inspection image having image distortion. As shown in FIG. 1, the reference image 101 is an image that matches the inspection image 102 except for noise when there is no defect or image distortion. For example, the reference image 101 may be an image of the design data of a sample, a captured image generated by machine learning using the design data, an image of another area having a pattern similar to the inspection image, or an image statistic obtained for each pixel by assuming a Gaussian distribution or the like for the distribution of brightness values ​​that the captured image can have predicted from the design data. The inspection image 102 is a captured image of a sample manufactured using the same design data as the reference image 101. The presence or absence of a defect in the inspection image 102 is determined by evaluating the difference between the reference image 101 and the inspection image 102. In a difference image 103 between the reference image 101 and the inspection image 102, a region with a high brightness value is detected as an abnormal part such as a defect. When the inspection image 102 has image distortion, a difference occurs between the reference image 101 and the inspection image 102 due to a position shift caused by the image distortion occurring throughout the image, resulting in a false detection in which a location that is not actually a defect is mistakenly determined to be a defect.

[0012] Here, the reference image 101 in Fig. 1 shows an example without image distortion, but as described above, the reference image 101 may be a photographed image of a sample manufactured using the same design data as the inspection image 102, and may have image distortion. In that case, the inspection device of this embodiment estimates the position shift due to the image distortion between the reference image 101 and the inspection image 102. Also, the inspection image 102 in Fig. 1 shows an example of radial distortion as the image distortion, but is not limited to this. For example, any distortion may be used, such as translational deviation, rotational deviation, or wavy deviation in which wiring is wavy.

[0013] 3 is a functional block diagram showing an example of the overall configuration of an inspection device according to a first embodiment of the present invention. Note that the inspection device 301 according to this embodiment performs inspection by comparing a reference image 101 with an inspection image 102. 3, inspection device 301 is made up of image distortion estimation section 303, estimated distortion amount 304, image distortion correction section 305, corrected inspection image 306, and inspection section 307. Here, image distortion estimation section 303, image distortion correction section 305, and inspection section 307 are realized by, for example, a processor such as a CPU (not shown), a ROM for storing various programs, a RAM for temporarily storing data in the calculation process, and a storage device such as an external storage device, and the processor such as a CPU reads and executes the various programs stored in the ROM, and stores the execution results, which are the calculation results, in the RAM or the external storage device.

[0014] The image distortion estimation unit 303 receives the reference image 101, the inspection image 102, and the correction condition 302, estimates the positional shift due to image distortion between the reference image 101 and the inspection image 102, and outputs an estimated distortion amount 304. This estimated distortion amount 304 is calculated so as to increase the similarity between the reference image 101 and the inspection image 102 while satisfying the constraint condition defined by the correction condition 302 described later. The estimated distortion amount is expressed as a two-dimensional vector amount (dx, dy) corresponding to each pixel of the reference image 101 or the inspection image 102, and an image similar to the reference image 101 is obtained by moving each pixel of the inspection image 102 by the corresponding vector amount. Here, a vector amount is given as the format of the estimated distortion amount 304, but if it is assumed that the distortion amount is a formulatable amount such as a rotational shift or a translational shift, a parameter such as a rotation angle or a translational shift amount may be used. The similarity between reference image 101 and inspection image 102 is increased by reducing the error in brightness values, and by increasing the likelihood of the distribution of inspection image 102 and reference image 101 if reference image 101 is image statistics.

[0015] Image distortion correction section 305 receives estimated distortion amount 304 output by image distortion estimation section 303 and inspection image 102 as input, corrects inspection image 102 using estimated distortion amount 304 , and outputs corrected inspection image 306 . The inspection unit 307 inspects for defects or other abnormalities by comparing the corrected inspection image 306 with the reference image 101. When the reference image 101 has been corrected by the image distortion correction unit 305, the inspection unit 307 inspects by comparing the corrected reference image with the inspection image 102.

[0016] 4 is a diagram showing an example of the correction condition 302 shown in FIG. 3, and is a diagram explaining an example of variations of the correction condition 302 input to the image distortion estimation unit 303. The correction condition 302 is intended to enable the image distortion estimation unit 303 to estimate an amount of distortion capable of correcting only a position shift caused by a distortion occurring in the entire image, without considering local and high-frequency image features, by adjusting the processing contents and setting parameters. The correction condition 302 defines pre-processing for the inspection image 102 and the reference image 101 capable of removing local and high-frequency image features, and constraint conditions and post-processing that impose constraints on the estimated amount of distortion 304 so as not to correct a local and high-frequency position shift.

[0017] For example, as in correction condition 401, downsampling for reducing an image to, for example, one-fourth, or upsampling by four times for restoring the estimated distortion amount to the original image size can be defined as preprocessing. Also, as in correction condition 402, in addition to the downsampling magnification of preprocessing and the upsampling magnification for restoring the estimated distortion amount to the original image size, the filter size and standard deviation of a Gaussian filter for smoothing the estimated distortion amount, and the absolute error of the vector of the predicted distortion amount in the adjacent region can be defined to be within, for example, 0.1 as a constraint condition for the evaluation criterion in step S603 of Fig. 6 to be described later by the image distortion estimation unit. Furthermore, as in correction condition 403, it is possible to define without performing preprocessing and postprocessing.

[0018] The correction conditions 401, 402, and 403 are examples, and the process contents and the set parameter values ​​are determined according to the circuit size and defect size of the sample to be inspected, and are not limited to these. For example, a smoothing filter such as a Gaussian filter or a moving average filter may be applied to the inspection image as pre-processing, or a smoothing filter and downsampling or upsampling may be used in pre-processing and post-processing. In addition, a constraint on the distortion amount may be to reduce the variance value in the neighborhood of the estimated distortion amount.

[0019] 5 is a flowchart showing the processing operation of the inspection device 301 according to this embodiment. As shown in FIG. In step S 502 , the correction conditions 302 are input to the inspection device 301 . In step S503, the image distortion estimation unit 303 constituting the inspection device 301 estimates the amount of distortion between the reference image 101 and the inspection image 102 using the input reference image 101, the inspection image 102 and the correction condition 302, and outputs the estimated distortion amount 304. The output estimated distortion amount 304 is stored in the storage unit. Details of step S503 in this embodiment will be described later with reference to FIG. 6.

[0020] In step S504, image distortion correction unit 305 of inspection device 301 corrects inspection image 102 or reference image 101 using the estimated distortion amount output by image distortion estimation unit 303, and outputs corrected inspection image 306 or a corrected reference image (not shown). The output corrected inspection image 306 or corrected reference image (not shown) is stored in a memory unit.

[0021] In step S505, the inspection unit 307 constituting the inspection device 301 calculates the presence or absence of abnormalities such as defects and their coordinates by image comparison using the corrected inspection image 306 and the reference image 101, or the inspection image 102 and the corrected reference image (not shown). In step S506, inspection unit 307 constituting inspection device 301 outputs inspection result 308 calculated by inspection unit 307, and the inspection ends.

[0022] FIG. 6 is a flowchart of the image distortion estimation unit 303 constituting the inspection device 301 according to this embodiment. As shown in FIG. 6, in step S601, the image distortion estimation unit 303 constituting the inspection device 301 performs image pre-processing defined in the correction condition 302 on the inspection image 102 and the reference image 101 input in the above-mentioned step S501.

[0023] In step S602, the image distortion estimation unit 303 uses the inspection image and the reference image preprocessed in step S602 to estimate a shift amount that increases the similarity between the inspection image and the reference image, and outputs a positional shift (first distortion amount) due to image distortion between the reference image and the inspection image. As an estimation method, for example, a method of setting an arbitrary or random initial value of the distortion amount and updating the distortion amount according to an evaluation value in step S603 described later can be mentioned. More specifically, a method of solving a constrained optimization problem that minimizes an evaluation value calculated by an evaluation function d(R,f(I,D)) using a constraint condition L specified by the correction condition 302 for the reference image R, the inspection image I, the correction processing function f, the estimated distortion amount D, and the distance function d, by dynamic programming or the like can be mentioned. The distance function d evaluates the image similarity between the reference image and the inspection image, and can be an absolute error or a squared error. In addition, if the reference image is an image statistic such as a Gaussian distribution that the luminance value of the inspection image can take, a negative log likelihood can be mentioned.

[0024] In step S603, the image distortion estimation unit 303 evaluates the first distortion amount estimated in step S602, and determines whether the evaluation value satisfies an evaluation criterion. The evaluation value includes a value related to variation such as difference and variance of vector amount in a neighboring region of the estimated distortion amount, and image similarity between an image corrected using the estimated distortion amount and a reference image. The evaluation criterion includes whether constraint conditions specified by the correction condition 302 for variation such as difference and variance of the estimated distortion amount in a neighboring region are satisfied, and whether the image similarity is equal to or greater than a specified image similarity.

[0025] In step S604, the image distortion estimation unit 303 performs post-processing defined in the correction condition 302 on the first distortion amount, and outputs a second distortion amount. In step S605, the image distortion estimation unit 303 outputs the second amount of distortion as an estimated amount of distortion, and ends the image distortion estimation process.

[0026] The effects of this embodiment will be described with reference to Figures 2A, 2B, and 2C. Figure 2A is a diagram showing a reference image and a position shift due to image distortion, and an inspection image having a defect. Figure 2B is a diagram showing the amount of distortion correction resulting from overfitting to a defective portion, and the inspection image after correction. Figure 2C is a diagram showing the amount of distortion correction due to the amount of distortion estimated by this embodiment, and the inspection image after correction.

[0027] As shown in the inspection image of FIG. 2A, defects and other abnormalities appear on the inspection image as localized, high-frequency image features compared to image distortions. When using the method described in Patent Document 1, the image distortion amount is estimated so as to maximize the similarity between the inspection image and the reference image, and the distortion correction amount is estimated for each corresponding pixel of the inspection image as shown in Figure 2B, and the corrected inspection image matches the reference image. As a result, abnormal parts such as defects are judged to be normal and go undetected.

[0028] In contrast to this, in this embodiment, by adjusting the correction conditions such as the upsampling magnification and downsampling magnification, and the filter size and parameters of the smoothing filter that smoothes the inspection image and reference image, the image distortion estimation unit is prevented from estimating the amount of distortion based on local image variations due to defects in the inspection image, and it is possible to estimate only the image distortion, which is the variation that occurs across the entire image.

[0029] In addition, by adjusting the filter size and parameters of the smoothing filter that smoothes the estimated distortion amount, and by suppressing the difference in the predicted distortion amount in adjacent areas as a constraint condition for the evaluation criterion in step S603, it is possible to reduce the fluctuation in the distortion amount within the vicinity of the estimated distortion amount and suppress high-frequency correction amounts that cause defects to go undetected.

[0030] As a result, in this embodiment, as shown in Fig. 2C, it is possible to generate a corrected inspection image that leaves defective parts by not estimating local and high-frequency correction amounts, and therefore it is possible to ensure the accuracy of the inspection device for inspection images that have positional deviations due to image distortion.

[0031] In this embodiment, an example is shown in FIG. 3 in which inspection image 102 is input to image distortion correction unit 305 to remove image distortion from the inspection image. However, a configuration may also be used in which reference image 101 is input to image distortion correction unit 305, and the image with image distortion added is output as a corrected reference image.

[0032] As described above, according to this embodiment, it is possible to provide an inspection apparatus that can improve the accuracy of evaluation and measurement in defect inspection and other evaluations and dimensional measurements by using a distortion correction technique that is adapted to areas excluding abnormal parts such as defects. Specifically, by adjusting the upsampling magnification and downsampling magnification, which are positive conditions, and the filter size and parameters of the smoothing filter that smoothes the inspection image and reference image, the image distortion estimation unit is prevented from estimating the amount of distortion based on local image fluctuations caused by defects in the inspection image, and it is possible to estimate only the image distortion, which is the fluctuation that occurs across the entire image.

[0033] In addition, by not estimating the amount of local and high-frequency correction, it is possible to generate a corrected inspection image that retains defects, thereby ensuring the accuracy of the inspection device for inspection images that have misalignment due to image distortion. EXAMPLES

[0034] Fig. 7 is a functional block diagram showing an example of the overall configuration of an inspection device according to a second embodiment of the present invention. An inspection device 704 according to this embodiment corresponds to the inspection device 301 in the first embodiment described above, but differs from the first embodiment in that an image distortion estimation model generated by machine learning is used to estimate a position shift caused by image distortion between a reference image and an inspection image in step S503 shown in Fig. 5. The same components as those in the first embodiment are denoted by the same reference numerals, and the following description will be omitted if they overlap with the first embodiment.

[0035] 7, inspection device 704 according to this embodiment is made up of image distortion estimation unit 705, estimated distortion amount 706, image distortion correction unit 305, corrected inspection image 306, and inspection unit 307. Here, image distortion estimation unit 705, image distortion correction unit 305, and inspection unit 307 are realized by, for example, a processor such as a CPU (not shown), a ROM for storing various programs, a RAM for temporarily storing data in the calculation process, and a storage device such as an external storage device, and the processor such as a CPU reads and executes the various programs stored in the ROM, and stores the calculation results that are the execution results in the RAM or the external storage device.

[0036] Note that the image distortion correction unit 305 and the inspection unit 307 are the same as those in the first embodiment. In addition, a machine learning unit 707 is provided for creating an image distortion estimation model 708 used in the image distortion estimation unit 705. The processing flow of the inspection device 704 is the same as that in the first embodiment shown in Fig. 5, but differs in the content of step S503. Only the points that differ from the first embodiment will be described below.

[0037] An image distortion estimation unit 705 constituting an inspection device 704 receives as input the reference image 101, the inspection image 102, and the correction conditions 703, and uses an image distortion estimation model 708 created by a machine learning unit 707 described later to estimate a position shift caused by image distortion between the reference image 101 and the inspection image 102, and outputs an estimated distortion amount 706. The image distortion estimation model 708 is trained so that the estimated distortion amount 706 satisfies the constraint conditions defined by the correction conditions 703 described later, while increasing the similarity between the reference image 101 and the inspection image 102.

[0038] Machine learning unit 707 uses learning reference image 701, learning inspection image 702, and correction condition 703 to create and store image distortion estimation model 708 to be used in image distortion estimation unit 705. The learning process in machine learning unit 707 may be performed simultaneously with the process in inspection device 704 or may be performed separately. Furthermore, as long as the computer that executes image distortion estimation unit 705 is capable of acquiring image distortion estimation model 708 via a network connection or the like, machine learning unit 707 may be configured to be executed by a computer different from that for inspection process 704.

[0039] The image distortion estimation model 708 is trained to increase the similarity between the reference image 101 and the inspection image 102 while satisfying the constraint conditions defined by the correction conditions 703 described below when the reference image 101 or the inspection image 102 is corrected by the image distortion correction unit 305. For the image distortion estimation model 708, for example, an Encoder-Decoder type CNN (Convolution Neural Network) such as U-Net or a CNN having another structure is used, but is not limited to a CNN.

[0040] Fig. 8 is a flowchart for explaining the operation of the machine learning unit 707 according to this embodiment. As shown in Fig. 8, in step S801, a learning reference image 701, a learning test image 702, and correction conditions 703 are input to the machine learning unit 707. The learning reference image 701 and the learning test image 702 are a pair of a reference image and a test image having a position shift due to image distortion between the two images, and one or more pairs are input to the machine learning unit 707. The correction conditions 703 are similar to the correction conditions 302 in the above-mentioned first embodiment, and in addition, weighting parameters in the calculation of an error function in step S806 described later and the like are defined.

[0041] In step S802, the machine learning unit 707 acquires one or more pairs of the input training test image 702 and training reference image 101. In step S803, the machine learning unit 707 performs pre-processing similar to step S601 in the image distortion estimation unit 303 of the above-mentioned first embodiment on the training test image 702 and the training reference image 701 acquired in step S802, based on the processing content and setting parameters defined in the correction conditions 703.

[0042] In step S804, the machine learning unit 707 estimates the amount of distortion (first distortion amount) between the reference image and the test image using the learning reference image and the learning test image processed in step S803, using the image distortion estimation model 708. In this embodiment, the image distortion estimation model 708 estimates the first amount of distortion using the reference image and the test image preprocessed in step S803 as input, but for example, when the reference image is an image simulated by a machine learning model such as CNN, the feature amount of the intermediate layer of the reference image generation model can be input to the image distortion estimation model 708 simultaneously with the reference image or as a substitute for the reference image.

[0043] In step S805, the machine learning unit 707 performs post-processing similar to step S604 in the image distortion estimation unit 303 of the above-mentioned first embodiment, based on the processing content and setting parameters defined in the correction condition 703, and calculates an estimated distortion amount (second distortion amount). In step S806, the machine learning unit 707 corrects the inspection image or the reference image using the estimated distortion amount (second distortion amount) to generate a corrected inspection image or a corrected reference image.

[0044] In step S807, machine learning unit 707 evaluates an error function or loss function of image distortion estimation model 708 using the estimated distortion calculated in step S805, the test image for learning and reference image for learning 701 acquired in step S802, and a corrected test image or a corrected reference image created in step S806 from test image for learning 702 and reference image for learning 701. The error function or loss function of image distortion estimation model 708 is, for example, a weighted sum of an evaluation value indicating a variation such as a difference in magnitude or variance of a vector in a neighborhood area of ​​the estimated distortion amount defined by correction condition 703, and an absolute error, a squared error, or a negative log likelihood due to a Gaussian distribution or the like between the corrected test image and reference image for learning, and the weighting parameter is defined in correction condition 703. The regularization term for the estimated distortion amount is, for example, an average of the difference in magnitude of the vector of the estimated distortion amount between adjacent pixels or the variance of the estimated distortion amount in the neighborhood of a specific pixel when the estimated distortion amount is in an image format.

[0045] In step S808, the machine learning unit 707 updates the parameters of the image distortion estimation model 708 based on the evaluation result in step S807 so as to reduce the value of the error function of the image distortion estimation model 708. This updating is performed by, for example, the stochastic gradient descent method.

[0046] In step S809, machine learning unit 707 determines whether or not a learning end condition has been reached, and if it is determined that the learning end condition has been reached (YES), the process proceeds to step S810, where machine learning unit 707 saves image distortion estimation model 708 and ends the learning process. On the other hand, if it is determined that the learning end condition has not been reached (NO), the process returns to step S802 and executes the processes from step S802 onwards again. The learning end condition may be whether the processes from step S802 to step S808 have been repeated a predetermined number of times or more, or whether the value of the error function of the image distortion estimation model calculated in step S807 does not become smaller even after the processes from step S802 to step S808 have been repeated a predetermined number of times, and thus it is determined that the learning of image distortion estimation model 708 has converged.

[0047] 9 is a flowchart of the image distortion estimation unit 705 constituting the inspection device 704 according to this embodiment. Only the points different from the processing flow of the inspection device 301 of the first embodiment shown in FIG.

[0048] In step S901, the image distortion estimation unit 705 constituting the inspection device 704 acquires the image distortion estimation model 708 created by the machine learning unit 707. In step S903, image distortion estimation unit 705 applies image distortion estimation model 708 to the test image and reference image preprocessed in step S601, and estimates the amount of distortion (first amount of distortion) between reference image 101 and test image 102. Thereafter, through steps S604 and S605, the estimated distortion amount is output, and the image distortion estimation process ends.

[0049] In this way, when machine learning is used, there is an advantage that by changing the configuration of the image distortion estimation model, it is possible to reduce the memory usage and calculation time of the computer performing the inspection, although this comes at a trade-off with the estimation accuracy. Methods for reducing calculation time by changing the configuration include, for example, reducing the number of channels in the convolutional layer used in CNN, or reducing the number of layers.

[0050] As described above, according to the present embodiment, in addition to the effects of the first embodiment, by changing the configuration of the image distortion estimation model, it is possible to reduce the memory usage and calculation time of the computer performing the inspection, although this reduces the estimation accuracy. EXAMPLES

[0051] 11 is a functional block diagram showing an example of the overall configuration of a measurement device of Example 3 according to another embodiment of the present invention. This embodiment is different from Example 1 in that the measurement device has a measurement for measuring the length of a sample using a corrected inspection image. The same components as those in Example 1 are given the same reference numerals, and in the following, descriptions that overlap with those in Example 1 will be omitted.

[0052] The position shift caused by image distortion has a characteristic that it becomes larger toward the image edge, such as the radial distortion of the inspection image 102 shown in FIG. 1. In such a case, the position shift causes the sample to deform on the image, and the measurement error increases. FIGS. 10A and 10B show an example of the measurement process, in which FIG. 10A shows an example of measuring an image area without image distortion, and FIG. 10B shows an example of measuring an image area with image distortion. As an example of the measurement method using an image, as shown in FIG. 10A, a line profile is obtained in which the luminance value of the measurement point is the height, and the width of the point at the height specified in the measurement recipe is taken as the measurement value. As shown in FIG. 10B, when measuring an area with image distortion, an error occurs between the obtained measurement value 2 and the measurement value 1 in the area without image distortion due to the deformation on the inspection image caused by the image distortion.

[0053] 11, measuring device 111 according to this embodiment is made up of image distortion estimation unit 303, estimated distortion amount 304, image distortion correction unit 305, corrected inspection image 306, and measurement unit 117. Here, image distortion estimation unit 303, image distortion correction unit 305, and measurement unit 117 are realized by, for example, a processor such as a CPU (not shown), a ROM for storing various programs, a RAM for temporarily storing data in the calculation process, and a storage device such as an external storage device, and the processor such as a CPU reads and executes the various programs stored in the ROM, and stores the calculation results that are the execution results in the RAM or the external storage device.

[0054] As in the above-described first embodiment, a corrected inspection image is obtained by operating the image distortion estimation unit 303 and the image distortion correction unit 305, and the measurement unit 117 constituting the measuring device 111 measures the length of the sample using the obtained corrected inspection image.

[0055] It is also possible to add a configuration for estimating the position shift due to image distortion between the reference image and the inspection image by using the image distortion estimation model generated by the machine learning shown in the second embodiment described above.

[0056] When local positional misalignment is corrected, measurements will not be performed correctly at locations that have deformations due to factors other than image distortion, such as defects. However, this embodiment estimates only the image distortion, which is the variation that occurs across the entire image, and corrects the inspection image, making it possible to reduce only the measurement error due to image distortion.

[0057] As described above, according to this embodiment, it is possible to provide a measuring apparatus capable of improving the accuracy of dimensional measurement by using a distortion correction technique adapted to an area excluding abnormal parts such as defects.

[0058] Specifically, by adjusting the upsampling magnification and downsampling magnification, which are positive conditions, and the filter size and parameters of the smoothing filter that smoothes the inspection image and reference image, the image distortion estimation unit is prevented from estimating the amount of distortion based on local image fluctuations caused by defects in the inspection image, and it is possible to estimate only the image distortion, which is the fluctuation that occurs across the entire image.

[0059] In addition, by not estimating the amount of local and high-frequency correction, it is possible to generate a corrected inspection image that retains defects, thereby ensuring the accuracy of the measurement device for inspection images that have misalignment due to image distortion. Furthermore, by changing the configuration of the image distortion estimation model, it is possible to reduce the memory usage and calculation time of the computer performing the inspection, although this is a trade-off with the estimation accuracy.

[0060] The present invention is not limited to the above-mentioned embodiment, but includes various modified examples. For example, the above-mentioned embodiment has been described in detail to easily explain the present invention, and the present invention is not necessarily limited to those including all of the configurations described. In addition, it is possible to replace a part of the configuration of one embodiment with the configuration of another embodiment, and it is also possible to add the configuration of another embodiment to the configuration of one embodiment. [Explanation of symbols]

[0061] 101...Reference image 102...Test image 103...Difference image 301,704…Inspection equipment 302,703…Correction conditions 401, 402, 403…Correction conditions 303,705…Image distortion estimation unit 304,706…Estimated distortion 305…Image distortion correction unit 306...Corrected inspection image 307…Inspection Department 308...Test results 701... Reference image for learning 702...Learning test image 707…Machine Learning Department 708…Image distortion estimation model 111…Measuring equipment 117…Measuring section 118…Measurement results

Claims

1. An inspection apparatus for inspecting an inspection image by comparing it with a reference image, comprising: an image distortion estimation unit that estimates an amount of distortion between the reference image and the inspection image; an image distortion correction unit that corrects the inspection image and / or the reference image using the estimated distortion amount; an inspection unit that performs inspection using the corrected inspection image and a reference image or the inspection image and the corrected reference image; the image distortion estimation unit estimates a distortion amount by excluding local and high-frequency image features by adjusting a correction condition, The image distortion estimation unit is A pre-processing is performed on the inspection image and the reference image, the processing contents of which can be changed according to a correction condition, and a first distortion amount is estimated by inputting the inspection image and the reference image processed by the pre-processing; an inspection apparatus which performs post-processing defined in a correction condition on the first distortion amount, and sets a second distortion amount as the estimated distortion amount. An inspection device characterized by:

2. 2. The inspection device according to claim 1, At least, the pre-processing defined in the correction condition is a down-sampling process for reducing the image sizes of the inspection image and the reference image, and the post-processing is an up-sampling process for processing the estimated distortion amount into a format that can be input to an image distortion correction unit, An inspection device characterized in that the image distortion estimation unit estimates the amount of distortion by excluding local and high-frequency image features by adjusting the magnifications of the downsampling process and the upsampling process and / or the filter size of a smoothing filter that smoothes the inspection image and the reference image as a correction condition.

3. 2. The inspection device according to claim 1, The image distortion estimation unit has a process for evaluating the similarity between the inspection image and the reference image and the variation in the amount of distortion within a region adjacent to the first amount of distortion, and estimates the amount of distortion so as to reduce the variation in the amount of distortion within a region adjacent to the estimated amount of distortion while increasing the similarity between the inspection image and the reference image using correction conditions.

4. An inspection apparatus for inspecting an inspection image by comparison with a reference image, comprising: an image distortion estimation unit that estimates an amount of distortion between the reference image and the inspection image; an image distortion correction unit that corrects the inspection image and / or the reference image using the estimated distortion amount; an inspection unit that performs inspection using the corrected inspection image and a reference image or the inspection image and the corrected reference image; the image distortion estimation unit estimates a distortion amount by excluding local and high-frequency image features by adjusting a correction condition, a machine learning unit that creates a model for estimating an amount of distortion between a learning test image and the test image by using a learning reference image corresponding to the learning test image as a teacher; The image distortion estimation unit is performing pre-processing of the inspection image and the reference image, the processing contents of which can be changed according to a correction condition, and estimating a first distortion amount from the inspection image and the reference image processed by the pre-processing using a model created by the machine learning unit; an inspection apparatus which performs post-processing defined in a correction condition on the first distortion amount, and sets a second distortion amount as the estimated distortion amount.

5. 5. The inspection device according to claim 4, At least, the pre-processing defined in the correction condition is a down-sampling process for reducing the image sizes of the inspection image and the reference image, and the post-processing is an up-sampling process for processing the estimated distortion amount into a format that can be input to an image distortion correction unit, An inspection device characterized in that the image distortion estimation unit estimates the amount of distortion by excluding local and high-frequency image features by adjusting the magnifications of the downsampling process and the upsampling process and / or the filter size of a smoothing filter that smoothes the inspection image and the reference image as a correction condition.

6. A measurement device for measuring a length of a sample using an inspection image, comprising: an image distortion estimation unit that estimates an amount of distortion between the reference image and the inspection image based on the reference image and the inspection image; an image distortion correction unit that corrects the inspection image using the estimated distortion amount; a measurement unit that measures the length of the sample by using the corrected inspection image, the image distortion estimation unit estimates a distortion amount by excluding local and high-frequency image features by adjusting a correction condition, The image distortion estimation unit is A pre-processing is performed on the inspection image and the reference image, the processing contents of which can be changed according to a correction condition, and a first distortion amount is estimated by inputting the inspection image and the reference image processed by the pre-processing; A measurement apparatus comprising: a first distortion amount being subjected to post-processing defined in a correction condition; and a second distortion amount being set as the estimated distortion amount.

7. 7. The measurement apparatus according to claim 6, At least, the pre-processing defined in the correction condition is a down-sampling process for reducing the image sizes of the inspection image and the reference image, and the post-processing is an up-sampling process for processing the estimated distortion amount into a format that can be input to an image distortion correction unit, A measurement device characterized in that the image distortion estimation unit estimates the amount of distortion by excluding local and high-frequency image features by adjusting the magnifications of the downsampling process and the upsampling process and / or the filter size of a smoothing filter that smoothes the inspection image and the reference image as a correction condition.

8. The measuring device according to claim 6, The measurement device is characterized in that the image distortion estimation unit has a process for evaluating the similarity between the inspection image and the reference image and the variation in the distortion amount within a region adjacent to the first distortion amount, and estimates the distortion amount so as to reduce the variation in the distortion amount within a region adjacent to the estimated distortion amount while increasing the similarity between the inspection image and the reference image using correction conditions.

9. A measurement device for measuring a length of a sample using an inspection image, comprising: an image distortion estimation unit that estimates an amount of distortion between the reference image and the inspection image based on the reference image and the inspection image; an image distortion correction unit that corrects the inspection image using the estimated distortion amount; a measurement unit that measures the length of the sample by using the corrected inspection image, the image distortion estimation unit estimates a distortion amount by excluding local and high-frequency image features by adjusting a correction condition, a machine learning unit that creates a model for estimating an amount of distortion between a learning test image and the test image by using a learning reference image corresponding to the learning test image as a teacher; The image distortion estimation unit is performing pre-processing of the inspection image and the reference image, the processing contents of which can be changed according to a correction condition, and estimating a first distortion amount from the inspection image and the reference image processed by the pre-processing using a model created by the machine learning unit; A measurement apparatus comprising: a first distortion amount being subjected to post-processing defined in a correction condition; and a second distortion amount being set as the estimated distortion amount.

10. 10. The measurement apparatus according to claim 9, At least, the pre-processing defined in the correction condition is a down-sampling process for reducing the image sizes of the inspection image and the reference image, and the post-processing is an up-sampling process for processing the estimated distortion amount into a format that can be input to an image distortion correction unit, A measurement device characterized in that the image distortion estimation unit estimates the amount of distortion by excluding local and high-frequency image features by adjusting the magnifications of the downsampling process and the upsampling process and / or the filter size of a smoothing filter that smoothes the inspection image and the reference image as a correction condition.

Images