JP2007198968A - Image-classifying method and image-classifying apparatus - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To reduce the labor of an adjusting person for determining the optimum image processing parameter used in ADC. <P>SOLUTION: The adjusting person only selects and sets a plurality of image processing parameters desired in the classifying sequence of ADC and the adjusting value range thereof on the screen of an image processing parameter setting screen 100 to perform image processing, with respect to all the combination of the adjusting values of a plurality of selected and set image processing parameters 111-113 by a missing classification device 10. The optimum combination pattern is provided to the adjusting person as a recommendation parameter, on the basis of the classifying processing result. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a method and apparatus for automatically classifying industrial product images, and is particularly suitable for use in automatically classifying defect images obtained by capturing foreign matters, pattern defects, and the like generated during the manufacturing process of semiconductor products. The present invention relates to an image classification method and an image classification apparatus.

  For example, in the manufacturing process of a semiconductor product, there is a risk that the product yield may be reduced due to a short circuit of the semiconductor formation pattern or a lack of disconnection due to foreign matters generated in the manufacturing apparatus. Therefore, in the manufacture of semiconductor products, it is important to identify the cause of the occurrence of a defect in the product at an early stage and take measures to improve the product yield.

  For this reason, after inspection by the semiconductor wafer visual inspection device, the defect image obtained at the time of inspection is analyzed to automatically classify the defect, or after inspection, a more detailed image of the defect portion is re-imaged by the review device. ADC (Automatic Defect Classification) that automatically classifies the images is performed. Then, efforts are made to improve the product yield by ascertaining the cause of the defect at an early stage based on the classification result by the ADC and taking countermeasures.

  By the way, in order to automatically classify defects, as described in Japanese Patent Application Laid-Open No. 2004-295879 (Patent Document 1), a defect sample (defect image) whose classification class is known before ADC is performed. ), It is necessary to set defect classification criteria in advance.

  As the defect classification criteria, those based on the rule form, those based on the teaching form, and those obtained by combining the rule form and the teaching form are known. According to the rule form, the defect features defined by each of a plurality of image feature quantities such as the brightness and shape of the defect extracted from the image of the defect sample are classified according to the defect class based on a predetermined rule. It is something to keep. In the teaching form, each item of the defect feature is represented by a scalar value, and the plurality of scalar values are combined into a multidimensional vector, and the defect class in the multidimensional space formed by the multidimensional vector is represented. Classification is based on the distribution.

  In order to create a defect classification standard based on the rule form, it is generally necessary to set a determination threshold value for each of several image feature quantities that define the feature of the defect. In order to create a reference, it is necessary to obtain the distribution of defect classes in a multidimensional space.

  For example, there is a method described in Japanese Patent Application Laid-Open No. 2001-331784 (Patent Document 2) for creating a defect classification standard based on a rule form, and Japanese Patent Laid-Open No. 2001 for creating a defect classification standard based on a teaching form. No. -256480 (Patent Document 3) describes a classification method and a teaching method. Furthermore, the creation of a defect classification standard combining these rule form and teaching form is described in Japanese Patent Application Laid-Open No. 2001-135692 (Patent Document 4).

  Further, in image processing, as described in Japanese Patent Application Laid-Open No. 2001-188906 (Patent Document 5), the image processing parameter used for the processing usually differs depending on the value of the same imaging target. An image processing result is obtained. That is, it is known that the value of the feature quantity such as the size and shape of the defect obtained by image processing changes depending on the value of the image processing parameter. For example, as described in Japanese Patent Laid-Open No. 8-21803 (Patent Document 6), a plurality of feature amounts such as a defect size (area) are extracted by image processing, and classification processing is performed using the feature amounts. In the image processing, the binarization processing of the image defines that the area of the image where the gradation value is larger than the threshold, that is, the area brighter than the threshold represents the area of the defect. When the image processing is performed, the calculated defect area value changes depending on how the binarization threshold is given in the binarization processing.

JP 2004-295879 A JP 2001-331784 A JP 2001-256480 A JP 2001-135692 A JP 2001-188906 A JP-A-8-21803

  Therefore, as described above, in the case where a plurality of feature amounts such as an area are extracted by image processing, and the ADC that automatically classifies the defect generated in the product or the defect image using the feature amount is performed, the classification is performed. In order to increase the accuracy, it has been necessary to adjust image processing parameters that are optimal manually in advance. In particular, the types of defects that occur in the semiconductor manufacturing process are different for each manufacturing process, such as an exposure process and an etching process. In addition, the number of manufacturing processes varies from product to product, but from several tens. There are up to hundreds.

  Therefore, when ADC is performed, it is necessary to manually adjust and set optimal image processing parameters corresponding to defects detected in each process of each semiconductor process to be classified. However, the adjustment setting work involves manually adjusting the image processing parameters for each manufacturing process even when defect inspection is performed only for a few processes selected in advance from among a number of semiconductor manufacturing processes as described above. Requires a great deal of labor for the adjuster, which has been a major issue in automatic classification of defective images, that is, ADC.

  Further, even for semiconductor wafers of the same type and process, the optimum image processing parameters generally differ depending on the imaging conditions (for example, magnification) of the inspection apparatus. For this reason, when the defect data that is determined to be appropriate as the teaching data in the current automatic defect classification work is created from the defect image group whose defect classification has already been acquired, the mutual imaging of the defect images is performed. There is also a problem that it is difficult to set an optimal image processing parameter due to a difference in conditions.

  The present invention has been made in view of the above-described problems, and it is possible to reduce the labor of an adjuster required to determine an optimum image processing parameter when performing ADC, and to perform troublesome image processing parameter adjustment. It is an object of the present invention to provide an image classification method and an image classification device that allow an adjuster to create teaching data to be used for an ADC without being conscious of it, and to classify classified parameters.

  In addition, the present invention provides an image classification method and image capable of setting optimal image processing parameters regardless of a difference in imaging conditions between defect images from among defect image groups with already acquired defect classifications. An object is to provide a classification apparatus.

  In order to solve the above-described problem, the image classification method of the present invention is a classification class registration step of registering a classification class representing a defect attribute in a defect group image detected by imaging a sample, and automatically detecting a defect group image. Set in the image processing parameter setting step for the image of the defect group for which the classification class is registered in the image processing parameter setting step and the classification class registration step for setting all types of image processing parameters and adjustment value ranges to be used for the classification processing. Using the image processing parameter setting value group obtained based on the type of image processing parameter and the adjustment value range, the image processing and the defect classification processing are executed, and the image processing result and the defect classification processing result are obtained. The processing execution step to be accumulated, and the classification processing results of defects acquired in the processing execution step are classified into image processing parameter setting value groups by classification class. An optimum parameter determination step for collating with the classification class of the corresponding image of the defect group registered in the registration step, and determining an image processing parameter setting value of one image processing parameter setting value group as an optimum parameter based on the matching result It is characterized by having.

  The image classification method of the present invention further includes a classified data setting step for setting a target image group to classify a defect as classified data, and an image group of classified data set in the classified data setting step. Automatic defect classification that executes image processing and defect classification processing using one image processing parameter setting value group determined as the optimal parameter in the optimal parameter determination step, and accumulates the image processing result and defect classification processing result A processing step.

The image classification method of the present invention includes a teaching data specifying step for specifying a plurality of teaching data as image processing parameters, a classified data setting step for setting a group of images to be classified as classified data, and a classified data setting step. The image data and the defect classification process are executed for each teaching data specified in the teaching data specifying step, and the classification processing result for each teaching data is integrated into the classified data for the group of classified data images set in step 1. And an automatic classification processing step for calculating a final classification processing result for.
The image classification device of the present invention is an image classification device to which these image classification methods are applied.

  According to the present invention, it is possible to greatly reduce the labor of an adjuster required to determine an optimal image processing parameter when performing ADC, and the adjuster is not aware of troublesome adjustment of the image processing parameter. The teaching data used for the ADC can be created.

  In addition, it is possible to solve the performance degradation due to the small number of teaching images and the difficulty in adjusting the image processing parameters due to the gathering of teaching data, and to realize more detailed classification stably.

  Embodiments of an image classification method and an image classification apparatus according to the present invention will be described with reference to the drawings, taking as an example a defect classification apparatus that classifies defects generated in a semiconductor wafer.

<First Embodiment>
FIG. 1 is a configuration diagram of a defect classification apparatus as a first embodiment of an image classification method and an image classification apparatus according to the present invention.

  The defect classification device 10 includes an ADC control unit 11, an input unit 12, a display unit 13, and a data storage unit 14. In the present embodiment, the defect classification device 10 is configured to be data-connected to an electron microscope device 30 as a review device.

  The ADC control unit 11 takes in the image data of the defect group detected by imaging the sample from the electron microscope apparatus 30 and stores it in the data storage unit 14. Further, the ADC control unit 11 executes various processes such as an image process and a classification process for each of the defect group image data stored in the data storage unit 14. At that time, the ADC control unit 11 also performs operation control of each unit including the data storage unit 14 of the defect classification apparatus 10.

  The input unit 12 is for inputting necessary information and the like when the ADC control unit 11 executes various processes including the ADC according to the guidance displayed on the screen of the display unit 13, for example. .

  The display unit 13 is for displaying and outputting a defect image based on the image data of the defect group captured by the ADC control unit 11 from the electron microscope apparatus 30 and stored in the data storage unit 14, various data, operation guidance, and the like.

  The data storage unit 14 stores various data such as defect group image information, image processing results, classification processing results, and the like, and stores and holds optimum image processing parameters used by the ADC set by the adjuster.

  On the other hand, an electron microscope apparatus 30 as a review apparatus that is connected to the defect classification apparatus 10 having the above-described configuration is configured to include a sample chamber 31 and an electron optical system 32.

  The sample chamber 31 includes a stage (sample stage) 33 on which the sample 34, that is, the semiconductor wafer 34 to be reviewed in this case is mounted for observation, inside the casing, and the casing chamber is maintained at a low vacuum during observation. It is configured to be.

  The electron optical system 32 controls the irradiation of the processing surface of the semiconductor wafer 34 to be reviewed mounted on the stage 33 in the sample chamber 31 with the electron beam generated by the electron beam source. The electron optical system 32 detects secondary electrons and reflected electrons generated by electron beam irradiation on the semiconductor wafer, converts the detection output into digital data, and supplies the digital data to the image generation unit. Two-dimensional image data (defect image group) of the defect group generated on the processed surface of the semiconductor wafer to be reviewed is generated. The generated two-dimensional image data of the defect group generated on the processed surface of the semiconductor wafer to be reviewed is supplied as defect group image data to the defect classification apparatus 10 that is data-connected to the electron microscope apparatus 30.

  Next, a method for determining image processing parameters applied to the defect classification apparatus 10 described above will be described.

  FIG. 2 is a flowchart of an ADC classification sequence performed by the defect classification apparatus according to the present embodiment.

  First, in step S10, in order to set a classification standard used for ADC by the defect classification apparatus 10, a visual classification operation of a defect image group is performed manually. Here, the ADC control unit 11 sequentially displays each defect image of the defect image group collected in the data storage unit 14 by the preliminary inspection on the display unit 13, and allows the adjuster to visually evaluate the input image. The classification destination of each defect image (or the defect represented by the defect image) is registered by the operation 12.

  Here, as an example of the classification destination of each defect image, a case of a semiconductor wafer pattern will be described as an example. For example, “foreign matter” in which foreign matter adheres to the pattern surface, and the pattern is missing in the middle of “ There are defect attributes such as “disconnection”, “scratch” in which the surface of the pattern has been scratched, and “short circuit” in which adjacent mutually independent patterns are connected.

  The adjuster visually evaluates which defect attribute each defect image of the defect image group displayed on the display unit 13 corresponds to, and registers the classification class information of the corresponding defect attribute as the visual classification class information. .

  At that time, the visual classification class information of the defect image registered by the adjuster is associated with the defect ID assigned to the defect image or the defect location in the defect image by the ADC control unit 11 for identification. And stored in the data storage unit 14.

  In step S20, an image processing parameter desired to be used for the ADC classification sequence is set.

  Setting the image processing parameters is an important operation in the ADC classification sequence. This is because, in order to accurately classify defect images, it is necessary to accurately grasp defect information, and in order to accurately calculate defect information, for example, unevenness threshold, detection sensitivity, noise removal threshold This is because an image processing parameter setting value group composed of a plurality of image processing parameter setting values for calculating a feature amount of a defect such as the above needs to be appropriately adjusted in advance.

  However, even if the image processing parameters are wafers of the same type and process, the optimum image processing parameters used in the ADC are generally different depending on the imaging conditions (for example, magnification) of the inspection apparatus. For this reason, from the group of known defect images for which the defect classification class has already been acquired, the defect images judged to be suitable as the teaching data in the current automatic defect classification work are gathered together, and the teaching data used for the current automatic defect classification work However, it is difficult to set the optimum image processing parameters for extracting the suitability of the teaching data because of the difference in the imaging conditions between the defect images. In addition, image processing parameters are easily affected by image capturing conditions (for example, magnification), and adjustment may be required even for samples of the same type and manufactured by the same process. The work load on the coordinator was enormous.

  Considering this, the defect classification apparatus 10 of the present embodiment is configured to be able to adjust an image processing parameter setting value group composed of a plurality of image processing parameter setting values as shown in FIG. In the description, for the sake of easy understanding, the adjustment of the image processing parameters will be described by taking an example of an image processing parameter setting value group having three unevenness threshold values, detection sensitivity, and noise removal threshold values as individual image processing parameter setting values. explain.

  FIG. 3 is a configuration diagram of an example of an image processing parameter setting screen in the defect classification apparatus according to the present embodiment.

  The image processing parameter setting screen 100 is displayed on the display unit 13 by the ADC control unit 11 manually based on a predetermined operation of the input unit 12 or automatically based on completion of registration of the classification standard used for the ADC in step S10. .

  In the illustrated example, the image processing parameter setting screen 100 includes an OSD (On Screen) having a parameter name column 110, a reference value column 120, a setting value column 130, an all selection column 140, an image processing start unit 150, and an end unit 160. Display) screen.

  The parameter name column 110 displays items of image processing parameters that need to be adjusted (that is, names of individual image processing parameter setting values). In this embodiment, the unevenness threshold value 111, the detection sensitivity 112, and the noise removal threshold value are displayed. An item 113 (image processing parameter setting value) is provided.

  The reference value column 120 is used to display adjustment values (reference values) that are set in advance as default values corresponding to the individual image processing parameter setting values 111 to 113. In the image processing parameter setting values 111 to 113, the intermediate level “2” is set in advance and displayed.

  This is for setting and inputting adjustment values for the setting value field 130 and the image processing parameter setting values 111 to 113. In this embodiment, the setting range of each adjustment value is set for both the image processing parameter setting values 111 to 113. The image processing parameter set values 111 to 113 are normalized by the integer values of 1 to 3, and at least one integer value corresponding to a desired adjustment value is selected from the integer values of 1 to 3. Each adjustment value can be set and input. Therefore, the setting value column 130 is provided with three selected / unselected setting boxes 1301 to 1303 corresponding to the image processing parameter setting values 111 to 113, respectively. Here, if the image processing parameter setting value 111 of the unevenness threshold is described as an example, the setting value = 2 corresponds to an adjustment value (unevenness threshold) that makes the unevenness identification sensitivity better than the setting value = 1, and This corresponds to an adjustment value (concave / convex threshold) that makes the discrimination sensitivity of the concavity and convexity dull compared to the set value = 3. Of course, the adjustment values of the image processing parameter setting values 111 to 113 need not be limited to those normalized to integer values of 1 to 3, but can be set from a number of further subdivided integer values. Alternatively, the adjustment value may be directly manually input based on a predetermined operation of the input unit 12.

  The all selection column 140 is a selection operation for selecting all the selected / unselected setting boxes 1301 to 1303 when selecting the adjustment values of the image processing parameter setting values 111 to 113 over the entire setting range of the adjustment values. Is provided so that the entire adjustment value range of the corresponding image processing parameter setting values 111 to 113 can be set by selecting the selected / unselected setting boxes 141 to 143. ing.

  The image processing start unit 150 is for instructing the start of image processing for all combinations of adjustment values of the image processing parameter setting values 111 to 113 set in the setting value column 130.

  Here, with respect to all combinations of the adjustment values of the image processing parameter setting values 111 to 113, the selection state of the adjustment values of the image processing parameter setting values 111 to 113 shown in FIG. 3 will be described as an example. 1. A combination (identification code “121”) having a detection sensitivity of 2 and a noise removal value of 1 (identification code “121”), and a combination of unevenness threshold of 3, detection sensitivity of 3 and noise removal value of 2 (identification code “332”) Thus, a total of 12 combinations of image processing parameter setting values are obtained.

  The ending unit 160 is a means for ending the selection operation of the image processing parameter setting values 111 to 113 in response to the adjustment processing of the image processing parameter.

  According to the present embodiment, the adjuster confirms the image processing parameter setting values 111 to 113 on the image processing parameter setting screen 100 displayed on the display unit 13, and selects the desired selection / non-selection setting boxes 1301 to 1301. By simply selecting 1303 and 141 to 143 by the operation of the input unit 12, all combinations of adjustment values set for the image processing parameter setting values 111 to 113 such as the unevenness threshold 111, the detection sensitivity 112, and the noise removal threshold 113 are selected. An image processing parameter set value group can be easily set as an image processing parameter.

  Also in this case, a desired setting value is selected for each image processing parameter setting value such as the unevenness threshold value 111, the detection sensitivity 112, and the noise removal threshold value 113 for the image processing parameter to be set, that is, the image processing parameter setting value group. As a result, these can be automatically combined and set as image processing parameters on the defect classification device 10 side, so that it is possible to reduce the load on the adjustment work of the image processing parameters of the adjuster.

  The ADC control unit 11 receives the operation input from the input unit 12 corresponding to the image processing start unit 150 on the image processing parameter setting screen 100 by the adjuster, and then displays the image processing parameter setting screen 100 on the screen. Based on the setting values selected for each of the image processing parameter setting values 111 to 113 such as the displayed unevenness threshold value 111, detection sensitivity 112, and noise removal threshold value 113, an image processing parameter setting value group as described above is calculated. The generated image processing parameter group is temporarily stored, for example, in its own internal memory.

  When the setting of the image processing parameter shown in step S20 of FIG. 2 is completed as described above, the ADC control unit 11 sets the set image processing parameter (image processing parameter) as shown in step S30 of FIG. Image processing is executed for each set value group), and classification processing is executed based on the image processing result of the defect group obtained by execution of this image processing. Thereby, the image processing result and the classification processing result based on each of the plurality of image processing parameters can be obtained simultaneously.

  Here, an example of the internal sequence of the image processing and classification processing shown in step S30 will be described in detail based on FIG. 4 and FIG.

FIG. 4 is a flowchart for one embodiment of image processing and classification processing.
In FIG. 4, in the image processing in step S31, the ADC control unit 11 performs step S20 for each defect image to which the defect ID of the defect image group in which the defect class classification class information is registered in the visual classification operation in step S10. The image processing parameter setting value group is selected from among a plurality of (for example, 12 patterns in the case of FIG. 3) image processing parameter setting value groups that are set in step S3 and temporarily stored in the internal memory. The detection sensitivity, the noise removal value, and the unevenness threshold value in the processing parameter set value group are applied to calculate each of the image processing results.

  In the classification process of step S32, the ADC control unit 11 calculates a classification result (ADC class) for each of the image processing results corresponding to all the defect images calculated by the image processing of step S31.

  As a method of calculating the classification result, an image feature is calculated using the image processing parameter setting value of the image processing parameter setting value group used for the image processing, and the calculated image feature is used. For example, a rule type, a feature amount teaching unit, or a combination thereof can be applied.

  Then, the ADC control unit 11 stores the image processing result calculated in step S31 and the classification result calculated in step S32 in correspondence with the defect ID of each defect image and the selected image processing parameter setting value group. Stored in the storage unit 14.

  When the ADC control unit 11 completes the image processing shown in step S31 and the classification processing shown in step S32 for the selected image processing parameter setting value group, the image processing and classification are performed in step S33. It is determined whether or not an image processing parameter set value group that is not used for processing remains, that is, whether or not another set image processing parameter still remains.

  As a result of this determination, the ADC controller 11 determines that the image processing parameter setting value group that is not used in the image processing and classification processing shown in steps S31 and S32 is included in addition to the one image processing parameter setting value group selected this time. If it still remains, one image processing parameter setting value group is selected from the unused image processing parameter setting value groups, and the image processing and classification processing shown in steps S31 and S32 are repeated. .

  On the other hand, the ADC control unit 11 includes, in addition to the one image processing parameter setting value group selected this time, an image processing parameter setting value group that is not used in the image processing and classification processing shown in steps S31 and S32. If it does not already exist, the image processing and classification processing shown in step S30 of FIG. 2 is terminated, and the process proceeds to result browsing and performance evaluation processing shown in step S40 of FIG.

  Thus, in the image processing and classification processing shown in FIG. 4 corresponding to step S30 in FIG. 2, the ADC control unit 11 sets the image processing parameter setting value for each defect image stored as teaching data. The image processing result and defect classification result for each group can be acquired.

  FIG. 5 is a flowchart of a variation of the image processing and classification processing detailed in FIG.

  In the image processing in step S31, as in the image processing in step S31 of FIG. 3, the ADC control unit 11 is set in step S20 for each defect image to which the defect ID of the defect image group as teaching data is attached. One image processing parameter setting value group is selected from a plurality of (for example, twelve) image processing parameter setting value groups, and detection sensitivity, noise removal value, and unevenness threshold value in the selected image processing parameter setting value group are selected. Apply and calculate each image processing result.

  Next, in step S35, as in the discrimination process in step S33 of FIG. 3, the ADC control unit 11 determines whether or not there remains an image processing parameter set value group that is not used for image processing, that is, another image. It is determined whether or not processing parameters are still set.

  As a result of this determination, the ADC control unit 11 determines that, in addition to the one image processing parameter setting value group selected this time, an image processing parameter setting value group not used in the image processing shown in step S31 still remains. Selects one image processing parameter setting value group from the unused image processing parameter setting value group, and repeats the image processing shown in step S31.

  On the other hand, the ADC control unit 11 does not already have an image processing parameter setting value group that is not used for the image processing shown in step S31 other than the one image processing parameter setting value group selected this time. After image processing is executed for all image processing parameter setting value groups, and after confirming that all the image processing results for each image processing parameter setting value group have been acquired for each defect image group stored as teaching data In step S36, a classification result (ADC class) is calculated for each of the acquired image processing results.

  In the case of this embodiment as well, as in the case of the embodiment shown in FIG. 4, the image processing result and defect classification for each image processing parameter set value group for each defect image stored as teaching data. Results and can be obtained.

  As a result, by the execution of the image processing and the classification process by the ADC control unit 11 shown in step S30, the identification given to each image processing parameter set value group in order to identify each image processing parameter set value group The image processing result and the defect classification result corresponding to each defect ID of the defect image group of the teaching data acquired using the code and the image processing parameter setting value group corresponding to each of these identification codes are stored in the data storage It is stored in the unit 14.

  As described above, after completing the image processing and the classification process shown in step S30 of FIG. 2, the ADC control unit 11 performs a result browsing process of the image processing result and the classification process result, and an automatic process as shown in step S40. Perform the performance evaluation process of classification.

  In this step S40, the ADC control unit 11 performs classification processing for each defect image or defect corresponding to each defect ID of the defect image group of the teaching data acquired using the image processing parameter setting value group corresponding to each identification code. The result is compared with the defect image corresponding to each defect ID of the defect image group of the teaching data obtained in the visual classification process shown in step S10 or the classification class information for each defect, and the classification success / failure determination process, The correct classification rate calculation process and the optimum parameter determination process are performed.

  The classification success / failure determination process is performed by classifying the corresponding image processing parameter setting value group acquired in step S30 according to the identification code of the image processing parameter setting value group and the defect ID of each defect location in the defect image group of the teaching data. This is a process for determining whether the classification process is successful or unsuccessful by determining whether the result matches the classification class information of the corresponding defective portion obtained in the visual classification process shown in step S10.

  The classification accuracy rate calculation process is performed on the entire number of defect images in the defect image group collected in the data storage unit 14 by the preliminary inspection that is visually classified in step S10 for each image processing parameter set value group, that is, for each identification code. This is a process of calculating the percentage of the number of matches between the classification process result and the classification class information in the classification success / failure determination process.

  The optimum parameter determination process compares the classification accuracy rates calculated in the classification accuracy rate calculation process for each image processing parameter set value group, that is, for each identification code, and determines a predetermined determination condition, for example, for each identification code. This is a process of determining an image processing parameter setting value group of an identification code having the highest classification accuracy rate, that is, the highest classification accuracy rate according to the level of the classification accuracy rate.

  In the present embodiment, the following (Condition 1) to (Condition 3) are adopted in selecting the image processing parameter setting value group having the highest classification accuracy rate for each identification code as the optimum parameter group determination conditions. Yes.

(Condition 1) When there is only one image processing parameter setting value group that maximizes the classification accuracy rate, the values of the image processing parameter setting values 111 to 113 that form this image processing parameter setting value group are the optimum parameters. And

(Condition 2) When there are a plurality of image processing parameter setting value groups with the maximum classification accuracy rate, the default value is set in the reference value column 110 shown in FIG. 3 between these image processing parameter setting value groups. The values of the image processing parameter setting values 111 to 113 that form the image processing parameter setting value group having the closest distance to the adjustment value (reference value) being set are set as the optimum parameters.

(Condition 3) There are a plurality of image processing parameter setting value groups with the maximum classification accuracy rate, and the default value is set in the reference value column 120 of FIG. 3 between the plurality of image processing parameter setting value groups. When the distance to the adjustment value (reference value) is the same, the values of the image processing parameter setting values 111 to 113 forming the image processing parameter setting value group having the smallest identification code are set as the optimum parameters.

  Then, based on the completion of the image processing and classification processing in step S30, the ADC control unit 11 automatically displays the processing results on the display unit 13 as shown in FIG. The condition setting of (Condition 3) is various. In the above embodiment, for example, by changing the step size (setting value range) of each setting value, it is possible to devise so that the distances are not equal. is there.

FIG. 6 shows an example of a result browsing screen for image processing results and classification processing results.
The result browsing screen 200 is automatically displayed on the display unit 13 by the ADC control unit 11 based on the completion of the image processing and the classification processing shown in step S30.

  In the case of the present embodiment, the result browsing screen 200 of the image processing result and the classification processing result is configured by an OSD screen having a recommendation column 210, an identification code column 220, a classification correct rate column 230, and an adoption parameter selection column 240. Yes.

  The recommendation column 210 guides an image processing parameter set value group as an optimum parameter determined by the ADC control unit 11 by performing the optimum parameter determination process, that is, its identification code.

  The identification code column 220 includes image processing parameter setting value groups each consisting of all combinations of adjustment values of the image processing parameter setting values 111 to 113 set by the adjuster in the image processing parameter setting processing in step S20. Displays the identification code.

  In this embodiment, a three-digit numerical value obtained by combining three image processing parameter setting values is used as the identification code. For example, the identification code display “121” indicates that the image processing parameter setting value of the unevenness threshold 111 shown in FIG. 3 is “1”, the image processing parameter setting value of the detection sensitivity 112 is “2”, and the image processing of the noise removal threshold 113 is performed. This represents an image processing parameter setting value group consisting of combinations of parameter setting values “3”.

  For each identification code, that is, for each image processing parameter set value group, the classification correct answer rate column 230 classifies all the defect images in the defect image group of the teaching data using the image processing parameter set value group. The classification accuracy rate in the entire defect image of the defect image group of the teaching data is displayed by collating with the classification class information of the defect image corresponding to the defect image group registered in the visual classification process of step S10. In the case of the present embodiment, for example, when one image processing parameter set value group whose identification code can be represented by “121” is used, the classification accuracy rate as a result of classifying the entire defect image of the defect image group of the teaching data is 33. It represents 3%.

  The adopted parameter selection column 240 is a column for uniquely selecting an image processing parameter setting value group actually used by the adjuster for the ADC.

  According to the present embodiment, the adjuster recommends between the image processing parameter setting values in the identification code column 220 on the image processing result and classification processing result browsing screen 200 displayed on the display unit 13. While confirming the presence / absence of “*” indicating the optimum parameter group in the column 210 and the value in the classification accuracy rate column 230, the image processing parameter setting value group in the desired identification code column 220 is displayed on the adopted parameter selection column 240. Just by selecting, the optimum adjustment of the image processing parameter setting values 111 to 113 is selected from all combinations of adjustment values for the image processing parameter setting values 111 to 113 such as the unevenness threshold 111, the detection sensitivity 112, and the noise removal threshold 113. A combination of values can be easily set as an image processing parameter. As a result, the load on the adjuster in the image processing parameter adjustment operation can be reduced, and the image processing parameter setting value group as the optimum parameter can be set uniquely.

  Further, in the present embodiment, there is a special classification case in which the optimum parameter determination method itself that determines an image processing parameter setting value group having a high classification accuracy rate for all defect images in the defect image group as an optimum parameter is not necessarily applied. Therefore, in the case of the embodiment shown in FIG. 5, an adopted parameter selection field 240 is provided, and even if an image processing parameter set value group not marked with “*” is determined as the optimum parameter by the coordinator's judgment. The configuration can be changed.

  In this case, for example, the adjuster needs to confirm the collation result between the classification processing result and the classification class information for the specific defect image in the defect image group instead of all the defect images. According to the form, the operator operates the input unit 12 on the result browsing screen 200 further configured with the OSD screen, and the desired image processing parameter setting value group displayed in the identification code column 220 is selected. The ADC control unit 11 is configured to display the detailed image processing result and the classification processing result result browsing screen 300 for the identification code on the display unit 13 by specifying one identification code. .

  FIG. 7 shows an example of a detailed image processing result and classification processing result result browsing screen for a desired identification code.

  In the present embodiment, the result browsing screen 300 includes a classification processing result display area 310 and an image processing result display area 350. The classification processing result display area 310 includes an identification code setting / display box 320 and a detailed result display area 330. The detailed result display area 330 includes a defect ID column 331, a target classification class column 332, and an ADC class column. 333, configured in a table format having a success / failure determination column 334.

  The image processing result display area 350 includes a defect ID setting / display box 360 and a defect image / reference image display area 370. The defect image / reference image display area 370 includes a defect ID setting / display box 360. The defect image 371 corresponding to the defect ID set / displayed in the above and the reference image 372 are displayed in comparison.

  Next, the classification process result display area 310 of the result browsing screen 300 is displayed in the identification code setting / display box 320 at the beginning of the display of the result browsing screen 300, for example, on the result browsing screen 200 shown in FIG. One operation-designated identification code “132” is displayed, and the detailed result display area 330 is associated with the defect ID column 331, the target classification class column 332, the ADC class column 333, and the success / failure determination column 334, respectively. A classification process result obtained by classifying all the defect images in the defect image group of the teaching data using the image processing parameter set value group corresponding to the one identification code “132” is displayed. That is, the defect ID column 331 contains the defect IDs of the defect images constituting the defect image group or the defect locations in the defect images, respectively, and the purpose classification class column 332, the ADC class column 333, and the success / failure determination column 334, respectively. Is the classification class information registered by the visual classification process, the classification process result automatically classified by the ADC, and the matching result of both, for the defect represented in the defect image corresponding to the defect ID in the defect ID column 331. The determination result of success / failure of the ADC based on each is displayed.

  The identification code “132” initially displayed in the identification code setting / display box 320 can be changed by an identification code input by a predetermined operation of the input unit 12 thereafter. When the setting of the identification code displayed in the identification code setting / display box 320 is changed, the display contents of the detailed result display area 330 and the defect image / reference image display area 370 described later are changed according to the setting change. The configuration corresponding to the identified identification code is changed and displayed. In addition, the display contents of each column 331 to 334 of the detailed result display area 330 can be sorted by the defect ID input in the identification code setting / display box 320, and further, a predetermined or predetermined range of defect ID, disconnection, foreign matter, It can be sorted according to the contents of each of the columns 331 to 334 such as a defect attribute (classification class) such as a scratch and a determination result of success / failure, and the readability is improved. Yes.

  Thereby, in the detailed result display area 330, the adjuster compares the contents of the visual classification class column 332 and the ADC class column 333 corresponding to each defect ID column 331, thereby determining which classification class the ADC has made a mistake. It is possible to obtain information as to whether it is easy to make mistakes. Further, by checking the contents of the success / failure determination column 334, it is possible to check the defect ID that has failed in classification.

  Also, in the image processing result display area 350, the defect ID specified from the display of the classification processing result display area 310 of the result browsing screen 300 or the defect image / reference image display area 370 according to a predetermined operation of the input unit 12. The defect image 371 and the reference image 372 corresponding to the defect ID set in the defect ID setting / display box 360 are configured to be displayed in comparison. Thereby, the adjuster can estimate the cause of the classification error visually from the image by browsing the defect image 371 and the reference image 372 in comparison.

  Then, the image processing parameter setting value group as the optimum parameter uniquely set using the adopted parameter selection field 240 on the result browsing screen 200 of FIG. It is registered and stored in the data storage unit 14 by the ADC control unit 11 as image processing parameters used when the ADC is executed by the apparatus 10.

  As described above, according to the defect classification apparatus 10 of the present embodiment, the adjuster can use the image processing parameter setting screen 100 shown in FIG. Only by selecting and setting a range, the defect classification apparatus 10 performs image processing on all combinations of adjustment values of the plurality of image processing parameters 111 to 113 that have been selected and set, and based on the classification processing result, FIG. 6 shows an optimum combination pattern (image processing parameter setting value group) as a recommended parameter among all the combination patterns (that is, image processing parameter setting value group) of the adjustment values of the image processing parameters 111 to 113 selected and set. The user can make a proposal on the result browsing screen 200 and at the same time select and set it on the detailed result browsing screen 300 shown in FIG. For every combination pattern (that is, image processing parameter setting value group) of adjustment values of the image processing parameters 111 to 113, an opportunity to examine and compare the results of the image processing and the classification processing for each defect image or defect in the defect image group is obtained. The determination material for determining the image processing parameter can be obtained without any effort.

<Second Embodiment>
Next, a defect classification apparatus 10 ′ as a second embodiment of the image classification method and the image classification apparatus of the present invention will be described with reference to the drawings.

  Since the overall configuration of the defect classification apparatus 10 ′ of the present embodiment is the same as the configuration of the defect classification apparatus 10 of the first embodiment shown in FIG. 1, the detailed description thereof is omitted, and the same or similar configuration The parts will be described using the same reference numerals.

  FIG. 8 is a flowchart of an ADC classification sequence performed by the defect classification apparatus according to the present embodiment.

  In step S110, classification target data to be classified is set using the ADC of the defect classification apparatus 10 '. For example, from the image information of the defect group which is supplied in advance from the electron microscope apparatus 30 as the review apparatus and stored in the data storage section 14 by a predetermined input from the input section 12 based on a predetermined operation by the adjuster. Data to be classified for performing defect classification work is set.

  Steps S120, S130, and S140 correspond to the visual classification process in step S10, the image processing parameter setting process in S20, and the image process and classification process in S30 described in FIG.

  In step S120, the defect classification device 10 ′ causes the adjuster to visually evaluate the defect images in the defect image group collected in the data storage unit 14 by the preliminary inspection, and classifies each defect image, that is, Register classification class information. In step S <b> 130, the defect classification device 10 ′ causes the adjuster to set an image processing parameter desired to be used for the ADC classification sequence and its adjustment value range. In step S140, image processing is performed on the defect image of the previous defect image group by each of the image processing parameters (image processing parameter setting value group) set in step S130, and the classified object obtained by executing this image processing is executed. Based on the image processing result of the data, the classification process is executed.

  Here, the image processing and classification processing shown in step S130 will be described in detail with reference to FIG.

  FIG. 9 is a configuration diagram of an example of an image processing parameter setting screen in the defect classification apparatus according to the present embodiment.

  The image processing parameter setting screen 100 ′ in the defect classification device 10 ′ of the present embodiment further includes a mode area 170 compared to the image processing parameter setting screen 100 shown in FIG. 3 by the defect classification device 10 of the above-described embodiment. Is composed of an OSD screen to which is added. The added mode area 170 has a result browsing setting box 171, an automatic classification setting box 172, and a classified data setting unit 173.

  Here, the result browsing setting box 171 and the automatic classification setting box 172 in the mode area 170 are set by the adjuster selecting and setting one of the result browsing setting box 171 and the automatic classification setting box 172 by a predetermined operation of the input unit 12. This is to instruct the ADC control unit 11 of the defect classification apparatus 10 ′ to execute the result browsing process or the automatic classification process.

  Further, the classified data setting unit 173 in the mode area 170 allows the adjuster to set the classified data by a predetermined search of the input unit 12 in the same manner as in step S110, and the ADC control unit 11 of the defect classification device 10 ′. It is for instructing. The classified data setting unit 173 is operated when changing the contents set in step S110.

  According to the present embodiment, the adjuster confirms the image processing parameter setting values 111 to 113 on the image processing parameter setting screen 100 ′ displayed on the display unit 13, and selects a desired selection / non-selection setting box 1301. ... To 1303 by simply operating the input unit 12, image processing parameters including all combinations of adjustment values set for the image processing parameter setting values 111 to 113 such as the unevenness threshold value 111, the detection sensitivity 112, and the noise removal threshold value 113. A set value group can be easily set as an image processing parameter.

  Further, according to the present embodiment, the adjuster selects and sets the result browsing setting box 171 and the automatic classification setting box 172 on the image processing parameter setting screen 100 ′, thereby browsing the results of the classified data and evaluating the performance. And whether to perform automatic classification of the classified data can be selected and set.

  Here, when the result browsing or automatic classification is selected, the implementation of the selection setting content is applied in step S150 and step S170 of FIG. 8 described later.

  Then, when the operation input from the input unit 12 corresponding to the image processing start unit 150 on the image processing parameter setting screen 100 ′ by the adjuster is supplied, the ADC control unit 11 displays the screen of the image processing parameter setting screen 100 ′. Based on the setting values selected for each of the image processing parameter setting values 111 to 113 such as the unevenness threshold 111, the detection sensitivity 112, and the noise removal threshold 113 displayed above, in the case shown in FIG. A parameter set value group is calculated, and the generated image processing parameter group is temporarily stored, for example, in its own internal memory.

  When the setting of the image processing parameter set value group shown in step S130 of FIG. 8 is completed in this way, the ADC control unit 11 performs the image parameter processing of step S30 of FIG. 2 as shown in step S140 of FIG. In the same manner as in, image processing is performed for each of the set image processing parameters (image processing parameter setting value group), and classification processing is performed on the image processing result of the defect group obtained by execution of the image processing. Execute.

  When the image processing and classification processing shown in step S140 of FIG. 8 is completed, the ADC control unit 11 performs the setting operation of the mode area 170 of the image processing parameter setting processing described in the previous step S130 as shown in step S150. Then, it is determined whether or not the result browsing and performance evaluation process is selected and set.

  When the result view setting box 171 in the mode area 170 on the image processing parameter setting screen 100 ′ shown in FIG. 9 is selected and set, the ADC control unit 11 performs the process as shown in step S160 of FIG. As in the case of the image parameter processing in step S40 of 2, the result browsing and performance evaluation processing are performed.

  In this case, the adjuster can uniquely set the image processing parameter setting value group as the optimum parameter on the result browsing screen 200 of the image processing result and the classification processing result shown in FIG.

  Then, the ADC control unit 11 is a case where the result browsing and the performance evaluation process are not selected or set in the confirmation process of the selection setting of the result browsing process shown in step S150, or the case where the result browsing and the performance evaluation process are selected and set, and step S160. When the result browsing and performance evaluation process shown in FIG. 5 is completed, as shown in step S170, the automatic classification process is selected and set by the setting operation in the mode area 170 of the image processing parameter setting process described in the previous step S130. It is determined whether or not.

  When the automatic classification setting box 172 in the mode area 170 on the image processing parameter setting screen 100 ′ shown in FIG. 9 is selected and set, the ADC control unit 11 performs the classified data setting process in the previous step S110. For the set classified data or the classified data set by operating the classified data setting unit 173 of the mode area 170 on the image processing parameter setting screen 100 ′ in the image parameter setting processing in step S130, The automatic classification process is executed using the adopted parameters set in step S160.

  As described above, according to this embodiment, the optimum parameter automatic setting is used to perform the process from setting the classified data to the automatic classification of the classified data without being aware of the image processing parameters. Defect classification can be realized.

<Third Embodiment>
Next, a defect classification apparatus 10 ″ as a third embodiment of the image classification method and the image classification apparatus of the present invention will be described with reference to the drawings.

  The overall configuration of the defect classification apparatus 10 ″ of this embodiment is the same as that of the defect classification apparatus 10 ′ of the first embodiment, and the configuration of the defect classification apparatus 10 of the first embodiment shown in FIG. Therefore, detailed description thereof is omitted, and the same or similar components are described using the same reference numerals.

  FIG. 10 is a flowchart of an ADC classification sequence performed by the defect classification apparatus according to the present embodiment.

  Step S210 is a step of selecting teaching data to be used for automatic classification processing of classified data.

  In the case of the present embodiment, unlike the case of the above-described embodiment, a plurality of image processing parameters (image processing parameter setting) are set for automatic classification processing for image processing parameters (image processing parameter setting value group) as teaching data. The feature is that the selection of (value group) is allowed.

  Here, for each selected image processing parameter (image processing parameter setting value group), the image processing parameter is optimal for a predetermined defect image (defect location), for example, by the image processing parameter determination method of the above-described embodiment. Assuming that If the individual image processing parameters are not optimized with respect to a predetermined defect image (defect location), the image processing parameters can be optimized by the above-described image processing parameter determination method.

  When selecting teaching data to be used for the current automatic classification process from among a plurality of teaching data to be selected at the time of selecting teaching data, it is necessary to select and set teaching data displayed on the display unit 13. On the OSD screen, for each teaching data to be selected, it is configured to be able to confirm what classification class is set in the teaching data by comments etc., and the suitability for use in this automatic classification processing The coordinator can make a judgment by referring to the comments. In addition, when the teaching data is selected and set, it is included separately from the plurality of teaching data to be selected or included in the plurality of teaching data to be selected regardless of the actual selection by the adjuster in advance. You may employ | adopt the structure which prepares the teaching data selected without fail.

  That is, the greatest feature of the image classification method according to the present embodiment is the classification method using a plurality of teaching data. In step S210, the ADC control unit 11 selects the plurality of teaching data, for example, the display unit 13. This is done by displaying a teaching data selection setting screen (not shown) and allowing the adjuster to operate the input unit 12 in a predetermined manner to select a desired plurality of teaching data.

  In step S220, the ADC control unit 11 is a step of setting a defect image group as the classified data based on a predetermined operation of the input unit 12 of the adjuster. In the second embodiment, the step of FIG. This corresponds to the processing shown in S110.

  In step S230, the ADC control unit 11 performs image processing of the classified data using image processing parameters included in each of the selected teaching data, calculates defect information based on the image information, and calculates the calculation result as data. Store in the storage unit 14. Furthermore, based on each teaching data, a classification process is executed based on the calculation result of defect information corresponding to the teaching data, and the classification result corresponding to each teaching data is stored in the data storage unit 14. Step S230 corresponds to the process shown in step S110 of FIG. 8 in the second embodiment.

  In step S240, the ADC control unit 11 performs each of the teaching data on the classified data in step S230 and integrates the classification processing stored in the data storage unit 14 to generate a classification result group. From this, the final classification result is determined.

  Next, the operation of the image classification method according to this embodiment for classifying using a plurality of such teaching data will be described.

In general, the classification problem becomes simpler in the following cases.
(Case 1) There are few classification classes.
(Case 2) There are many images included in the teaching data (= reliable distribution of classification)
If these cases are explained concretely, for example, in case 1, the classification with only two classification classes is more difficult for the processing and the procedure than the classification with five classes having more classification classes. Means low.

  Case 2 is a case where there are three classification classes. The teaching data in which 100 images exist in each classification class is more preferable than the teaching data in which each classification class has only one image. The classification problem tends to be stable because the characteristics of the classification class are reliable.

  In other words, as in the image classification method according to the present embodiment, the meaning of using a plurality of teaching data for image classification is to classify difficult teaching problems to individual teaching data, thereby classifying each teaching data. This is to reduce the number of classes. On the other hand, to increase the number of image groups that can be used as teaching data as a whole.

  In order to realize these effects, the step of integrating the taxon as shown in step S240 in FIG. 10 is an important key. Therefore, based on a specific example, the classification process result group integration process shown in step S240 will be described below.

  For example, when it is desired to classify a pattern defect, artifact defect, and scratch defect, the number of classification classes is changed to three classes with three as it is as the first teaching data. Specialized in the classification of non-foreign defects as the second teaching data, specialized in the classification of foreign / non-foreign defects, and specialized in the classification of scratch / non-scratch defects as the third teaching data Prepare things.

Specifically, a defect (defect image) determined to be suitable as teaching data in the current automatic defect classification work is prepared from a known defect group (defect image group) for which a defect classification class has already been acquired. It is assumed that each prepared defect (defect image) is given an ID as an identification code, and the content of each defect is as follows.
ID1: Pattern defect ID2: Pattern defect ID3: Foreign object defect ID4: Foreign object defect ID5: Scratch defect ID6: Scratch defect

  In the first teaching data, it is assumed that the defect of ID1 and the defect of ID2 are classified as pattern defects, and the other defects of ID3, ID4, ID5, and ID6 are classified as non-pattern defects. Similarly, in the second teaching data, it is assumed that the defect of ID3 and the defect of ID4 are classified as foreign matter defects, and the other defects of ID1, ID2, ID5, and ID6 are classified as non-foreign matter defects. Further, in the third teaching data, it is assumed that the defect of ID5 and the defect of ID6 are classified as scratch defects, and the other defects of ID1, ID2, ID3, and ID4 are classified as non-scratch defects.

  When attention is paid to the defect of ID1, the defect of ID1 is classified as a pattern defect, a non-foreign object defect, and a non-scratch defect according to the first to third teaching data. As described above, when the first to third teaching data are applied to each ID, the defects of ID1 to ID6 are “pattern defects, non-foreign matter defects and non-scratch defects”, “non-pattern defects and foreign matter defects”. And “non-scratch defect” and “non-pattern defect and non-foreign defect and scratch defect”.

  In this example, by using the first to third teaching data, it was reduced to 3 class classification, but depending on the result, it was expanded to 23, that is, 8 class classification, enabling classification of boundary cases between pattern defects and foreign object defects. It is also possible to accurately classify unknown defects that do not belong to pattern defects, foreign object defects, or scratch defects.

  That is, the image classification method according to the present embodiment, which selects a plurality of teaching data specialized for detecting different specific defects and integrates the classification results of defects based on the teaching data, simplifies the classification problem. In addition to improving the classification performance for specific defects, it is also possible to realize detailed classification of defects.

  The realization of the detailed classification of defects will be described by taking a semiconductor wafer defect classification apparatus as an example.

  In defect classification of semiconductor wafers, the optimum image processing parameters are generally different depending on the imaging conditions (for example, magnification) of the inspection apparatus even for wafers of the same type and process. For this reason, it is difficult to set optimal image processing parameters due to differences in imaging conditions when creating teaching data in which defective images determined to be appropriate for teaching data are created from already acquired teaching data groups. Met.

  However, even for such a problem, the problem can be solved by employing the image classification method according to the present embodiment and preparing a large number of gathered teaching data having different image processing parameters. This is because, for example, there are a plurality of teaching data for the same defect group according to the difference in imaging conditions (for example, magnification), the teaching data of the optimum parameters for the first imaging condition (for example, magnification) group, and the second If the teaching data of the optimum parameter for other imaging condition (for example, magnification) groups is different, conventionally, it is assumed that defect classification is performed using the optimum parameter for the first imaging condition group. The appropriate image information of the defect group of the second imaging condition group could not be calculated. In addition, new separate teaching data obtained by synthesizing the optimum parameter for the first imaging condition (for example, magnification) group and the optimum parameter for the second other imaging condition (for example, magnification) group is adopted. Even so, the creation of the new teaching data requires the efforts of the coordinator, and the creation of the new teaching data itself is difficult.

  Therefore, in the image classification method according to the present embodiment, the first teaching data adopting the optimum parameter for the first imaging condition group and the optimum parameter for the second imaging condition group are adopted. By separately preparing the second teaching data, it is possible to calculate appropriate image information for all images regardless of the difference in imaging conditions.

  As a result, it is possible to effectively use a group of defects having low utility value because it is difficult to adjust image processing parameters. Further, the increase in the number of defects used in the teaching data also leads to stable classification performance.

  As described above, according to the present embodiment, the classification data is classified using a plurality of optimum parameters specialized for individual defects, which are obtained by using the automatic setting of optimum parameters, and the processing results are obtained. By integrating, it is possible to reduce the effort required to adjust the image processing parameters without making the adjuster aware of the adjustment of one image processing parameter that can be dealt with for all the defects of the classified data, and to classify them. Defect classification with stable performance can be realized.

  The image classification method and the image classification device of the present invention are as described above, but the specific procedure and configuration are not limited to the specific configuration of the above-described embodiment. For example, the OSD screens 100 to 300 appropriately displayed on the display unit 13 are not limited to those illustrated.

1 is an overall configuration diagram of a defect classification apparatus as a first embodiment of the present invention. It is a flowchart of the ADC classification sequence by the defect classification apparatus of the present embodiment. It is a block diagram of an Example of the image processing parameter setting screen by the defect classification apparatus of this Embodiment. It is a flowchart about one Example of the image processing and classification processing by the defect classification device of this Embodiment. 6 is a flowchart of a modification of the image processing and classification processing detailed in FIG. 4. It is a block diagram of one Example of the result browsing screen of an image processing result and a classification process result. It is a block diagram of one Example of the detailed image processing result about a desired identification code, and the result browsing screen of a classification processing result. It is a flowchart of the classification sequence of ADC by the defect classification device as the 2nd Embodiment of this invention. It is a block diagram of an example of an image processing parameter setting screen in the defect classification apparatus of the present embodiment. It is a flowchart of the classification sequence of ADC by the defect classification device as the 3rd Embodiment of this invention.

Explanation of symbols

10, 10 ', 10 "defect classification device (image classification device)
DESCRIPTION OF SYMBOLS 11 ADC control part 12 Input part 13 Display part 14 Data storage part 30 Electron microscope apparatus 31 Sample chamber 32 Electron optical system 33 Stage (sample stand)
34 Sample 100 Image processing parameter setting screen 200 Result viewing screen 300 Result viewing screen

Claims (9)

A classification class registration step of registering a classification class representing a defect attribute in an image of a defect group detected by imaging a sample;
An image processing parameter setting step for setting all types of image processing parameters and adjustment value ranges to be used when automatically classifying images of defect groups;
For the image of the defect group in which the classification class is registered in the classification class registration step, the image processing parameter setting value group obtained based on the image processing parameter of the type set in the image processing parameter setting step and the adjustment value range is obtained. A process execution step for performing image processing and defect classification processing and storing the image processing results and defect classification processing results;
The defect classification processing result acquired in the processing execution step is collated with the classification class of the corresponding image of the defect group registered in the classification class registration step for each image processing parameter setting value group, and based on the collation result, the optimum An optimum parameter determining step for determining an image processing parameter setting value of one image processing parameter setting value group as a parameter;
An image classification method characterized by comprising: In the optimum parameter determination step, the defect classification processing result acquired in the processing execution step is collated with the classification class of the corresponding image of the defect group registered in the classification class registration step for each image processing parameter setting value group. The image classification method according to claim 1, wherein an image processing parameter setting value of one image processing parameter setting value group having high matching result is determined. In the optimum parameter determination step, the defect classification processing result acquired in the processing execution step is collated with the classification class of the corresponding image of the defect group registered in the classification class registration step for each image processing parameter setting value group, 2. The image classification method according to claim 1, further comprising a collation result notifying step for notifying a collation result for each processing parameter set value group. The optimum parameter determining step includes:
The defect classification processing result acquired in the processing execution step is collated with the classification class of the corresponding image of the defect group registered in the classification class registration step for each image processing parameter setting value group, and for each image processing parameter setting value group A verification result notification step for notifying the verification result;
A parameter selection step of selecting an image processing parameter setting value of one image processing parameter setting value group from among the image processing parameter setting value groups notified of the matching result by the matching result notification step;
The image classification method according to claim 1, comprising: A classification class registration step of registering a classification class representing a defect attribute in an image of a defect group detected by imaging a sample;
An image processing parameter setting step for setting all types of image processing parameters and adjustment value ranges to be used when automatically classifying images of defect groups;
For the image of the defect group in which the classification class is registered in the classification class registration step, the image processing parameter setting value group obtained based on the image processing parameter of the type set in the image processing parameter setting step and the adjustment value range is obtained. A process execution step for performing image processing and defect classification processing and storing the image processing results and defect classification processing results;
The defect classification processing result acquired in the processing execution step is collated with the classification class of the corresponding image of the defect group registered in the classification class registration step for each image processing parameter setting value group, and based on the collation result, the optimum An optimum parameter determining step for determining an image processing parameter setting value of one image processing parameter setting value group as a parameter;
A classified data setting step for setting a target image group for classifying defects as classified data;
For the image group of the classified data set in the classified data setting step, image processing and defect classification processing are performed using the one image processing parameter setting value group determined as the optimal parameter in the optimal parameter determining step. Automatic defect classification processing step for executing and storing the image processing result and the defect classification processing result,
An image classification method characterized by comprising: A classification class registration step of registering a classification class representing a defect attribute in an image of a defect group detected by imaging a sample;
An image processing parameter setting step for setting all types of image processing parameters and adjustment value ranges to be used when automatically classifying images of defect groups;
For the image of the defect group in which the classification class is registered in the classification class registration step, the image processing parameter setting value group obtained based on the image processing parameter of the type set in the image processing parameter setting step and the adjustment value range is obtained. A process execution step for performing image processing and defect classification processing and storing the image processing results and defect classification processing results;
The defect classification processing result acquired in the processing execution step is collated with the classification class of the corresponding image of the defect group registered in the classification class registration step for each image processing parameter setting value group, and based on the collation result, the optimum An optimum parameter determining step for determining an image processing parameter setting value of one image processing parameter setting value group as a parameter;
The image processing parameter setting step, the processing execution step, and the optimum parameter determination step are executed according to predetermined conditions, and each of the optimum parameters for each condition is obtained as each teaching data.
A classified data setting step for setting a target image group for classifying defects as classified data;
For the group of classified data images set in the classified data setting step, image processing and defect classification processing are executed for each optimum parameter acquired as teaching data in the teaching data setting step, Automatic defect classification processing step for calculating the final defect classification processing result for the classified data by integrating the defect classification processing results of
An image classification method characterized by comprising: The predetermined condition in the teaching data acquisition determination step is any one of a predetermined defect and an imaging condition of a defect group image in the classification class registration step. Item 7. The image classification method according to Item 6. Teaching data specifying step for specifying a plurality of teaching data as image processing parameters;
A classified data setting step for setting an image group to be classified as classified data;
For the classified data image group set in the classified data setting step, image processing and defect classification processing are executed for each teaching data specified in the teaching data specifying step, and the classification processing result for each teaching data Automatic classification processing step to calculate the final classification processing result for the classified data by integrating
An image classification method characterized by comprising: Classification class registration means for registering a classification class representing a defect attribute in an image of a defect group detected by imaging a sample,
Image processing parameter setting means for setting all types of image processing parameters and the adjustment value range to be used when automatically classifying images of defect groups,
The image processing parameter setting value group obtained based on the image processing parameter of the type set by the image processing parameter setting unit and the adjustment value range for the image of the defect group in which the classification class is registered by the classification class registration unit. Image processing and classification processing means for executing image processing and defect classification processing, and storing the image processing results and defect classification processing results;
The defect classification processing result acquired by the image processing and classification processing means is collated with the classification class of the corresponding image of the defect group registered in the classification class registration step for each image processing parameter setting value group, and based on the collation result An optimum parameter determining means for determining an image processing parameter setting value of one image processing parameter setting value group as an optimum parameter;
An image classification apparatus comprising:

Images