WO2011155123A1 - 観察画像の分類基準の最適化方法、および画像分類装置 - Google Patents
観察画像の分類基準の最適化方法、および画像分類装置 Download PDFInfo
- Publication number
- WO2011155123A1 WO2011155123A1 PCT/JP2011/002661 JP2011002661W WO2011155123A1 WO 2011155123 A1 WO2011155123 A1 WO 2011155123A1 JP 2011002661 W JP2011002661 W JP 2011002661W WO 2011155123 A1 WO2011155123 A1 WO 2011155123A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- defect
- adc
- category
- classification
- image
- Prior art date
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2200/00—Indexing scheme for image data processing or generation, in general
- G06T2200/24—Indexing scheme for image data processing or generation, in general involving graphical user interfaces [GUIs]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10056—Microscopic image
- G06T2207/10061—Microscopic image from scanning electron microscope
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
- G06T2207/30148—Semiconductor; IC; Wafer
Definitions
- the present invention is based on a recipe in which observation conditions are registered, classifies observation images for the purpose of identifying the cause of defects by observing samples under various observation conditions and classifying the observation images by type.
- the present invention relates to a standard optimization method and an observation apparatus.
- this technique is effective when the optimum classification standard changes according to the sample characteristics, manufacturing process, and DOI (Defect Of Interest) during user use, such as a scanning electron microscope (SEM) type observation device.
- SEM scanning electron microscope
- the SEM observation apparatus is an apparatus for observing such various defects, and generally observes defects based on defect position information detected by a host inspection apparatus. Furthermore, in order to identify a problem process, the operation
- Patent Document 1 discloses a method of quantifying the features of a defective portion and automatically classifying them using the feature amounts. Yes.
- Patent Document 2 discloses an invention for displaying the ADC result in the form of a confusion matrix.
- JP 2001-135692 A (US Pat. No. 6,922,482) Japanese Patent Laid-Open No. 2008-082821 (US Publication No. 2008/0075352)
- the main purpose of the present invention is to provide an ADC recipe optimization method for reducing the burden of MDC work and an observation apparatus having an ADC function.
- the ADC function mounted on the conventional defect observation apparatus or inspection apparatus is set to select either ADC or MDC.
- ADC or MDC there are only two options: whether all defect classification is left to the device or manual manual classification, and the device executes part of the defect classification work and manually classifies some of the defects.
- a function of performing defect classification by arbitrarily combining ADC and MDC has not been installed.
- the present invention solves the problem of the present invention by displaying on the screen judgment information for judging which category should be classified by the MDC.
- This determination information is more convenient to display on the screen display means for displaying the ADC result together with the ADC result, but can be displayed on another screen.
- the present invention it is possible to reduce the amount of manual work in the work of classifying defect images detected as a result of defect review. As a result, the defect observation result can be quickly fed back to the manufacturing process, and the manufacturing yield can be improved.
- the block diagram of a SEM type defect observation apparatus The block diagram of an observation image classification device.
- the flowchart which shows the optimization procedure of a classification parameter Explanatory drawing of a confusion matrix.
- Parameter optimization setting and evaluation screen for MDC work reduction The flowchart which shows the procedure of the parameter optimization for MDC work reduction.
- the conceptual diagram for demonstrating the idea of parameter optimization in the classification algorithm which uses the boundary setting in a feature-value space as a parameter.
- the conceptual diagram for demonstrating the concept of parameter optimization in the classification algorithm which uses the weight setting of a feature-value as a parameter.
- category algorithm The flowchart which shows the procedure of the parameter optimization for DOI oversight reduction. Parameter optimization setting and evaluation screen for reducing missed DOI. Parameter optimization setting and evaluation screen to reduce MDC work and missed DOI. 6 is a configuration example of a defect selection GUI screen used for MDC.
- Example 1 In the first embodiment, a configuration example of an SEM type defect observation apparatus having a function of displaying determination information for dividing ADC and MDC on a confusion matrix will be described.
- the SEM type defect observation apparatus acquires a high definition SEM image of a defect position on a sample detected by an external defect detection apparatus such as an optical type or SEM type appearance inspection apparatus, and the obtained high definition SEM image is determined in advance. It is a device for classifying by criteria.
- an external defect detection apparatus such as an optical type or SEM type appearance inspection apparatus
- the performance of the observation apparatus is remarkably improved, and the number of defect images that can be acquired by defect observation per hour reaches several hundred to several thousand. It is virtually impossible to manually classify such a large amount of defect images manually, and it is essential to introduce ADC in order to classify defects for defect images that are sufficient for feedback to the manufacturing process. It is.
- the ADC classification performance is not sufficient, and it is necessary to perform manual MDC for defects in some categories.
- the determination result indicating which category the ADC result is reliable and which category the ADC result is unreliable is used as the result of ADC and MDC for a part of the defect image acquired from the observation target sample. This is calculated and displayed on the display unit 206, so that the judgment information of the ADC / MDC separation is presented to the apparatus user.
- FIG. 1 is a schematic diagram showing the overall configuration of the SEM type defect observation apparatus of this example.
- the SEM type defect review apparatus shown in FIG. 1 includes an electron optical system composed of optical elements such as an electron gun 101, a lens 102, a scanning deflector 103, an objective lens 104, a sample 105, a secondary particle detector 109, an observation object, and the like.
- the image forming apparatus includes a storage device 116 having a recording medium for storing image data used for the optical microscope 119, an optical microscope 119, and the like.
- the electron optical system, the electron optical system control unit 110, the A / D conversion unit 111, the stage 106, and the stage control unit 112 described above constitute a scanning electron microscope that is an imaging unit for SEM images.
- the primary electron beam 107 emitted from the electron gun 101 is focused by the lens 102, deflected by the scanning deflector 103, focused by the objective lens 104, and irradiated on the sample 105.
- Secondary particles 108 such as secondary electrons and reflected electrons are generated from the sample 105 irradiated with the primary electron beam 107 in accordance with the shape and material of the sample.
- the generated secondary particles 108 are detected by the secondary particle detector 109 and then converted into a digital signal by the A / D converter 111.
- the output signal of the secondary particle detector converted into a digital signal may be referred to as an image signal.
- the output signal of the A / D conversion unit 111 is input to the image processing unit 114 to form an SEM image.
- the image processing unit 114 includes an ADR processing unit 117 that executes image processing such as defect detection using the generated SEM image, and executes various types of image processing.
- the electron optical system control unit 110 Control of optical elements inside the electron optical system such as the lens 102, the scanning deflector 103, and the objective lens 104 is performed by the electron optical system control unit 110.
- the sample position control is executed by the stage 106 controlled by the stage control unit 112.
- the overall control unit 113 is a control unit that comprehensively controls the entire SEM type defect observation apparatus.
- the overall control unit 113 interprets inputs from the keyboard 116, the mouse 118, and the storage device 116, and the electron optical system control unit 110 and the stage control unit 112.
- the image processing unit 114 and the like are controlled, and the processing result is output to the display unit 206 and the storage device 116 included in the operation unit 115 as necessary.
- the automatic defect classification process is executed by the overall control unit 113, and the overall control unit 113 includes an ADC processing unit 118 for performing ADC.
- the ADC processing unit executes ADC using the defect image extracted by the ADR processing unit or the defect image stored in the storage device 116.
- the ADR processing unit 117 or the ADC processing unit 118 described above can be realized by either hardware or software.
- the ADR processing unit 117 or the ADC processing unit 118 is configured by hardware, a plurality of arithmetic units that perform processing necessary for ADR or ADC are integrated on a wiring board or in one semiconductor chip or package.
- the ADR processing unit 117 or the ADC processing unit 118 is configured by software, a high-speed general-purpose CPU is installed in the ADR processing unit 117 or the ADC processing unit 118, and a program for performing ADR or ADC processing is provided. This can be realized by executing.
- FIG. 2 shows a more detailed view of the ADC processing unit 118 shown in FIG.
- the ADC processing unit 118 shown in FIG. 2 includes a plurality of functional blocks that are realized by a CPU provided in the overall control unit 113 in FIG. 1 executing a predetermined program, and controls the entire ADC process.
- An ADC control unit 202 that performs an image processing unit 203 that performs preprocessing necessary for defect classification, a classification processing unit 204 that performs actual defect classification processing using image data preprocessed by the image processing unit 203, and the like. It is configured.
- the above functional blocks can also be realized by hardware. In that case, a semiconductor device in which an arithmetic unit that realizes the ADC control unit 202, the image processing unit 203, and the classification processing unit 204 is integrated is an overall control unit. 113 will be provided.
- an image information storage unit 201 that stores image data preprocessed by the image processing unit 203 and a classification information storage unit 205 that stores a classification result executed by the classification processing unit 204. Has been.
- the classification information storage unit 205 also stores MDC result information as verification data for verifying the ADC result. Physically, these storage units correspond to partitions, logical volumes, file systems, or the like provided in the storage device 116.
- an operation unit 115 constituted by a keyboard 207 and a mouse 208, and a display unit 206 for displaying a processing result of the ADC and a GUI (Graphical User Interface) for giving an instruction to the apparatus.
- GUI Graphic User Interface
- the image information of the defect image acquired by the scanning electron microscope that is the imaging means is stored in the image information storage unit 201.
- the ADC control unit 202 reads image information from the image information storage unit 201 and transfers it to the image processing unit 203.
- the image processing unit 203 performs various feature amounts of the observation image necessary for the classification process from the transferred image information, for example, the size, shape, luminance distribution, texture, etc. of the defective portion, or the size, shape, luminance distribution of the background pattern.
- Corresponding data such as the positional relationship between the defective portion and the background pattern is calculated and stored in the image information storage unit 201.
- the ADC control unit 202 reads the feature amount data of the observation image stored in the image information storage unit 201 and transfers it to the classification processing unit 204.
- the classification processing unit 204 performs defect classification processing based on a predetermined classification model, and stores the processing result in the classification information storage unit 205.
- the teaching type automatically configures a classifier by teaching feature amount data associated with a correct classification result. For example, in the feature amount space based on the teaching data, the feature amount space is divided by defining the boundary to correspond to each category, and it is classified into the category by determining which feature amount space the classification target belongs to be able to.
- a method of defining the boundary of the feature amount space a method of classifying into the category of the taught defect whose distance in the feature amount space is the closest, or a feature amount distribution of each defect category is estimated based on the teaching data.
- the rule-based type is a method of classifying according to the rules described in if-then-else, for example, and repeatedly classifying the set of defects to be classified into two, and finally classifying each category
- a binary tree structure classification model is typical. It has been put into practical use as an expert system or BRMS (Business Rules Management System).
- BRMS Business Rules Management System
- a program for performing defect classification based on the classification model is stored in the storage device 116, and the CPU in the overall control unit 113 executes this program, thereby realizing the classification processing function of the classification processing unit 204.
- the Various processing execution instructions, selection of processing target data, and the like can be instructed from the operation unit 115 including the keyboard 207 and the mouse 208. Execution processes such as the above instruction contents, classification processing, and storage are displayed on the display unit 206, and the apparatus user can confirm the processing contents of the ADC by confirming the display contents.
- the ADC processing unit 118 is not necessarily provided inside the overall control unit 113, and may be realized by an information processing unit provided independently of the overall control unit. Furthermore, the same ADC processing can be performed in parallel by a plurality of information processing means connected to the overall control unit 113.
- a classification table called a confusion matrix is often used when evaluating a classification model that performs some sort of classification.
- the confusion matrix is a table in which the number when correctly classified by the model and the number when mistakenly classified are displayed in a matrix. Basically, as shown in FIG. 4 (A), True (Positive ( (True positive), False ⁇ ⁇ Positive (false positive), False Negative (false negative), and True Negative (true negative).
- the confusion matrix shown in FIG. 4B is a classification table in which the ADC results are displayed in the horizontal direction and the true classification results are displayed in the vertical direction in the matrix. In the example shown in FIG.
- the number of defect points (True Positive) that was classified and the true classification result was A was a
- the number of defect points (False Positive) that were classified into Category A by the ADC but the true classification result was B was a
- the result of MDC is used as the true classification result.
- Purity is an index indicating the classification performance, and indicates the purity of the ADC result. That is, it is a value calculated using the total number of certain categories classified by the ADC as the denominator and the number of correct classifications within that category as the numerator. It can be determined that the higher the Purity, the higher the reliability of the ADC result.
- Accuracy refers to a category in the confusion matrix.
- the MDC result of that category that is, the total number of defects to be classified into that category
- the ADC succeeds in classification within that category. It is a value calculated using the number as a numerator. It can be determined that the higher this Accuracy is, the fewer defects that should be classified into that category are.
- the above-described Purity and Accuracy information is used as judgment information indicating which category the ADC result is reliable and which category the ADC result is unreliable.
- the classification accuracy rate in MDC work is considered to be about 80%. Therefore, if the Purity of the ADC is 80% or more, it can be determined that the category does not require the MDC work, that is, the Purity can be used as a criterion for determining whether visual confirmation by the MDC work is necessary.
- FIG. 4C shows an example of a confusion matrix in which the number of classification categories is increased to three.
- FIG. 4C is a schematic diagram showing an actual GUI displayed on the display unit 206. In addition to the confusion matrix 401, ADC / MDC separation determination information is displayed.
- the ADC results corresponding to the category of interest or the MDC results corresponding to the classification result of interest are those other than true positive. What is necessary is just to calculate considering that all the classification results correspond to false negative or false positive.
- Purity for other ADC results “B” and “C”, or Accuracy for other categories B and C can be calculated in the same manner.
- the confusion matrix shown in FIG. 4C is generated when the ADC control unit 202 reads out the classification information stored in the classification information storage unit 205.
- the classification information storage unit 205 stores the MDC result data together with the ADC result data in association with the common defect ID, and the ADC control unit 202 refers to the ADC with reference to the defect ID.
- the result data of MDC and the result data of MDC are read, and the number of defects in the same category is counted to generate a confusion matrix.
- the ADC control unit 202 similarly classifies True ⁇ ⁇ Positive and False Negative or True Positive and False Positive for each category in which the numerical value in the confusion matrix is focused. Based on the above-described formula, True Positive number / Total of True Positive number and False Positive number, or True Positive number / True Positive number and total of False Negative number.
- a message 402 indicating a category that requires MDC, a Purity highlight 403 that does not require MDC, a category highlight 404 that does not require MDC, or the like can be used.
- a highlighting method an appropriate object such as a circle as shown in FIG. 4C is superimposed on the confusion matrix, and numbers and backgrounds corresponding to Purity and categories are highlighted. Alternatively, a method such as changing the color can be used.
- the message 402 gives a category that requires MDC
- the highlights 403 and 404 give a numeric value or category that does not require MDC.
- a message may be displayed, or highlighting may be performed on a numerical value or category that requires MDC.
- the highlighting 403 or 404 is applied to a category or numerical value that satisfies the criterion that Purity is 80% or more, that is, a category or numerical value that does not require MDC.
- This numerical value is an initial value that the ADC processing unit 118 has, but it is also possible to change the initial value. Therefore, the ADC control unit 202 displays a determination criterion setting box 405 on the GUI so that the device user can input an arbitrary numerical value.
- the ADC control unit 202 reads a numerical value input to the GUI, and changes a message displayed on the GUI and a target item (category or numerical value) to be highlighted on the confusion matrix based on the numerical value.
- a method for optimizing a threshold value used for defect classification processing using a confusion matrix will be described. For example, when classifying defects using a classification model of a binary tree structure, classification of each tree constituting the binary tree is performed based on whether or not the defect feature amount exceeds a certain threshold value. Therefore, in order for the apparatus user to classify defects with a desired accuracy, the threshold value needs to be optimized.
- the threshold optimization method will be described. Since the threshold is a parameter-set amount, the threshold is referred to as “parameter” in the following description.
- FIG. 3 is a flowchart showing the procedure of the classification process parameter optimization.
- a target Purity or Accuracy selection request is displayed on the display unit 206 (step 301).
- the device user inputs the Purity or Accuracy to which attention is paid according to the selection request on the GUI.
- a classification process is executed with a combination of a plurality of classification parameters (step 302). This processing is executed by the ADC processing unit 118 as described with reference to FIG.
- the classification result is displayed on the display unit 206 (step 303). In order to compare and display the classification results for a plurality of classification parameters, if the confusion matrix is displayed in parallel for each of the plurality of classification parameters, the evaluation results can be displayed easily.
- a classification parameter selection button is displayed on the GUI on which the display result is displayed. Therefore, the user selects the best result from the classification results executed with the plurality of classification parameters (step 304).
- the ADC control unit 202 determines a parameter to be adopted based on the input result (step 305). With the above procedure, it is possible to determine an optimum parameter for improving the classification performance of the focused category by paying attention to the accuracy rate of the specific category.
- FIG. 5 is an example of a GUI for confirming the evaluation results with a plurality of parameters when optimizing parameters for the purpose of reducing MDC work.
- a confusion matrix generated for a plurality of parameters is shown in FIG. Displayed for each parameter.
- a determination criterion setting box 501 for setting a determination criterion that does not require MDC work and a sort criterion setting box 502 for displaying the components of the confusion matrix in the order desired by the user.
- Parameter input boxes 503 and 504 a manual switching button 505 for parameter setting, and an auto switching button 506 are displayed.
- the “MDC work unnecessary rate” displayed in the figure is obtained by dividing the number of defects (or the number of defect images) classified into the category determined to be MDC unnecessary by the total number of defects (or the total number of defect images). It is a numerical value and is calculated by the ADC control unit 202. It is also possible to calculate and display the “MDC work necessary rate” instead of the “MDC work unnecessary rate”, and the definition in that case is the number of defects classified into the categories determined as necessary by the MDC (or the defects). (Number of images) divided by the total number of defects (or the total number of defective images).
- the criterion for not requiring MDC work is set as Purity ⁇ 80%, and the evaluation results are sorted and displayed in descending order of the number of defects that are judged to require no MDC work, that is, in descending order of Purity. ing.
- FIG. 6 is a flowchart of classification parameter optimization for the purpose of reducing MDC work.
- the ADC control unit 202 first displays on the GUI a determination criterion setting request (for example, a determination criterion setting box 501) that does not require MDC work and a setting request for whether the optimum parameter is automatically set on the apparatus side or manually. Step 601).
- a determination criterion setting request for example, a determination criterion setting box 501
- a setting request for example, an input box or setting button for setting these items is displayed on the GUI.
- the device user inputs a criterion according to the setting request. In the case of FIG. 5, Purity ⁇ 80% is set.
- a classification process is executed with a plurality of classification parameters (step 602).
- the processing executed by the apparatus in step 602 is as described above with reference to FIG. 4 or FIG.
- the process branches between when the optimum parameter is automatically set and when the optimum parameter is selected and set manually.
- the ADC control unit 202 calculates the MDC work unnecessary rate (step 604), and employs a classification parameter that maximizes the number of defects belonging to the MDC unnecessary category.
- the finally adopted parameters are displayed on the GUI (step 605).
- the optimum parameter is selected and set manually, as shown in FIG. 5, the result of classification execution by a plurality of parameters is compared and displayed on the GUI (step 606), and the apparatus enters a state waiting for input of the optimum parameter (step 606).
- Step 607 The user inputs the best result on the GUI as shown in FIG. 5, and the ADC control unit 202 adopts the input result as the optimum parameter (step 608).
- Manual selection setting is required when the classification parameter that maximizes the MDC work unnecessary rate is not necessarily adopted as the optimum parameter.
- the selection setting is performed in consideration of other judgment criteria. There is a need to. For example, the importance of the classification category may be taken into account, or the determination may be made together with the classification parameter setting for the purpose of reducing oversight described later.
- the best results can be easily selected by displaying in parallel the confusion matrix, the criteria for determining that the MDC work is not required, the MDC work unnecessary or required category name, the MDC work unnecessary rate, and the like. According to the above procedure, the optimum classification parameter for reducing the MDC work can be set automatically or manually.
- the confusion matrix is also displayed on the optimal parameter display screen at the same time, but this is only displayed for ease of use by the device user, and is essentially important. It is a screen display of discrimination information indicating which category requires (or does not require) MDC. Therefore, as shown in FIG. 5, the confusion matrix is not displayed, and the calculated Purity and information on the category that requires or does not require MDC or only the message that indicates the category that requires (or does not need) MDC is displayed. Good.
- Example 2 In this embodiment, a method for optimizing a boundary value (parameter) when an algorithm that sets a boundary line between categories in a feature amount space and performs classification based on the boundary line will be described.
- the description is simplified to a two-dimensional space and uses a classification model of a binary tree structure, the algorithm is not limited to the binary tree structure, and can be applied to boundary setting in a multidimensional space.
- an SEM type defect observation apparatus As an apparatus on which the optimization method of the present embodiment is mounted, an SEM type defect observation apparatus is assumed. However, the hardware configuration of the apparatus is the same as that of the first embodiment, and thus the description thereof is omitted.
- FIG. 7 shows a feature distribution diagram for explaining a parameter optimization method in the algorithm of this embodiment.
- the boundary line is a parameter.
- the feature quantity is 1 on the horizontal axis and the feature is on the vertical axis.
- the quantity 2 is adopted and expressed by two feature quantities.
- the feature amount distribution of category A is represented by ⁇
- the feature amount distribution of category B is represented by ⁇
- three types of boundary lines are represented by parameters 1, 2, and 3.
- FIG. 8 compares the classification results when parameter 1 and parameter 2 are adopted in the example of FIG. In general, it is often set to about 80%, but since the number of categories is small, the determination criterion indicating that the MDC work is not required shown in the determination criterion setting box 801 is strictly set as Purity ⁇ 95%.
- the parameter 2 aiming at improving the Purity of the category D as expected does not require the category D to require the MDC work.
- the MDC work unnecessary rate is greatly improved to 60%. This indicates that introduction of ADC can reduce 60% of manual MDC work.
- the overall accuracy rate (calculated by (a + d) / (a + b + c + d) in FIG. 4) is 92% to 90% compared to the case where parameter 2 is adopted.
- the MDC work unnecessary rate drastically increases from 0% to 60%. It has improved.
- Example 3 In this embodiment, a weight optimization method when an algorithm that improves classification performance by changing the weight of each feature amount will be described.
- the weight corresponds to the parameter.
- the number of features is limited to two.
- the algorithm can be applied to a multidimensional space, and the distance between the feature amount distribution of teaching data and the classification target data in the multidimensional space.
- the present invention can also be applied to a method of quantifying the amount corresponding to the distance in the feature amount space, such as calculation or phage voting, by another calculation method.
- an SEM type defect observation apparatus is assumed as an apparatus on which the optimization method of the present embodiment is mounted.
- the hardware configuration of the apparatus is the same as that in the first embodiment, Description is omitted.
- FIGS. 9A and 9B show feature histograms for explaining the concept of parameter optimization in the algorithm of this embodiment.
- 9A and 9B three types of categories, categories F, G, and H, and feature amounts of feature amounts 3 and 4 are extracted and shown in schematic diagrams for explanation.
- This feature amount histogram represents the feature amount distribution of the teaching data.
- the feature amount histogram is created by taking the value of feature amount 3 on the horizontal axis and the occurrence frequency on the vertical axis.
- the data taught as the category F tends to have a smaller feature amount than the categories G and H.
- the feature amount of the classification target data is small, the data is likely to be classified into the category F.
- This feature quantity distribution is created by the ADC processing unit 118 in FIG. 1 and the ADC control unit in FIG.
- the image processing unit 203 uses the image stored in the image information storage unit, the image processing unit 203 performs image processing calculation to calculate a feature amount, and the classification processing unit 204 converts it into a format suitable for classification processing. It is displayed on the display unit 206.
- the result is preferably stored in the classification information storage unit 205 in order to facilitate reuse.
- the feature amount histogram of the category F increases the weight of the feature amount having a feature amount distribution that does not overlap with the other categories G and H, and conversely, the feature amount histogram of the category F
- the weight of the feature amount having the feature amount distribution overlapping with the other categories G and H may be reduced.
- the weight of the feature amount 3 shown in FIG. 9A is increased, and the weight of the feature amount 4 shown in FIG. 9B is reduced.
- the feature weight parameter can be set automatically.
- the initial value is a value equally divided by the number of feature amounts
- the weight parameter is changed to a plurality of values with respect to the initial value, and the parameter having the highest Purity of the category of interest is adopted as the optimum value.
- the purpose is to prevent an overlooked category from being overlooked
- manual setting is possible instead of automatic setting. For example, when there is knowledge based on experience, an optimum solution can be obtained efficiently in a short time by setting the initial value, step size, and upper and lower limits of the parameter.
- the initial value, the step size, and the upper and lower limits of the swing width are set for the GUI displayed on the display unit 206 of FIG.
- the image processing unit 203 calculates the feature amount using the data stored in the image information storage unit 201 or reads the feature amount stored in the classification information storage unit, and the classification processing unit 204 reads the feature amount.
- the weight parameter is optimized and the result is displayed on the display unit 206.
- This system can improve the Purity, Accuracy, or overall accuracy rate of the category of interest in a system that improves classification performance by optimizing feature weights. This is particularly effective in a system that employs a learning type classification algorithm.
- Example 4 an embodiment of a method for optimizing the connection at each stage in a system that classifies classification targets in stages will be described.
- the connection at each stage is a parameter.
- the classification algorithm at each stage is not limited to a combination of different ones, but can be applied to a model that narrows down the classification target in multiple stages within a single classification algorithm, for example.
- the present embodiment also assumes mounting on an SEM type defect observation apparatus, but the hardware configuration of the apparatus is the same as that in the first embodiment, and thus description thereof is omitted.
- FIG. 10A is a schematic diagram of category links for explaining a parameter optimization method when an algorithm that improves classification performance by classifying into multiple stages is adopted.
- the categories I and J are respectively obtained at the n + 1th stage.
- Categories J and K, and categories K and I so that the difficulty of classification can be reduced.
- the classification parameters correspond to the nth and n + 1th category links.
- parameter 1 indicates that the category link is not actually disconnected
- parameter 1 indicates that the category link 1001 is disconnected
- parameter 2 indicates that the category link 1002 is disconnected
- parameter 3 indicates that both the category links 1001 and 1002 are disconnected.
- Example 5 In the present embodiment, a parameter optimization method for reducing detection omissions / missing as much as possible for a defect type of interest, that is, a specific defect category that is particularly interesting to an apparatus user will be described.
- the above-mentioned interest defect is referred to as DOI. Since the overall configuration of the apparatus is the same as that of the first embodiment, description thereof is omitted.
- Accuracy is a value calculated in the confusion matrix using the number of ADC correct defects for the target category as the numerator and the MDC result for the category as the denominator. Therefore, since the higher the Accuracy, the fewer the defects that should be classified into the category are less likely to be missed, so that it can be used as an evaluation index for the DOI miss rate.
- FIG. 11 is a flowchart showing a parameter optimization procedure of the present embodiment
- FIG. 12 is a schematic diagram of a GUI screen on which the parameter optimization result of the present embodiment is displayed.
- the GUI screen in FIG. 12 is displayed on the display unit 206 in FIG.
- a defect request (hereinafter referred to as DOI) for which a miss rate is to be reduced, a determination criterion, and an optimum parameter are set on the GUI, and a setting request for whether to automatically set or manually set the defect category.
- DOI defect request
- the setting request for example, the determination criterion setting box 1201, the sorting criterion setting box 1202, the parameter input boxes 1205 and 1206 for setting the evaluation criterion threshold shown in FIG.
- the device user inputs a desired setting value via these boxes and buttons.
- the determination criteria there are the accuracy of a specific DOI, the number of defects belonging to a category exceeding the accuracy that is the reference for the DOI, and the like.
- the category C is DOI
- the DOI miss rate of category C is adopted as the DOI miss criteria 1201.
- Accuracy ⁇ 90% is set.
- a plurality of categories can be set as the DOI.
- the setting value of the determination criterion can be changed for each category. By changing the judgment criterion for each category, weighting can be performed according to the importance of the defect type to be classified, and the DOI can be set more flexibly.
- the ADC processing unit 118 executes a classification process using the input plurality of classification parameters (step 1102).
- step 1103 the process branches between when the optimum parameter is automatically set and when the optimum parameter is selected and set manually. If the Manual switching button 1207 is pressed at the start of the flow, Step 1103 branches to No. If the Auto switch button 1208 is pressed, the process branches to Yes.
- the ADC control unit 202 calculates a DOI miss rate based on each input parameter (step 1104), determines a parameter that minimizes the DOI miss rate, and further calculates the calculation result. Display on GUI.
- the ADC control unit 202 displays the result of the classification execution with a plurality of parameters (step 1105) and waits for parameter input (step 1106).
- the user selects the best result by clicking a parameter selection button 1205, 1206 as shown in FIG. 12, for example. Instead of the parameter selection button, an input box for inputting the text of the parameter itself may be displayed.
- the ADC control unit 202 adopts the selected parameter (step 1107).
- the GUI shown in FIG. 12 corresponds to a display screen after execution of step 1105 or step 1106.
- Calculation results for a plurality of parameters are displayed in parallel as a confusion matrix, and messages 1203 and 1204 indicating the DOI miss rate are displayed for each category.
- the confusion matrix for each parameter can be sorted and displayed for any parameter.
- the matrix is arranged in the ascending order of the DOI miss rate as the sort criterion.
- the DOI miss rate is 10% for parameter 1 and 30% for parameter 2, and it can be seen that it is better to use parameter 1 when it is desired to reduce the miss of category C.
- the parameter adjustment method may be different from that for the purpose of reducing MDC work.
- the Purity of category A which is frequently generated, is improved. This is effective in reducing the MDC work, and can be expected to be optimized in the order of parameter 3, parameter 1, and parameter 2.
- category A is DOI and the purpose is to reduce the miss of DOI
- improving the accuracy of category A is effective for reducing the miss of DOI, so parameter 2, parameter 1, and parameter 3 are optimal in this order.
- the Purity of category I having a high occurrence frequency can be improved for the purpose of reducing MDC work. Since it is effective in reducing the MDC work, consider cutting the category link of category I as shown in FIG. However, for the purpose of reducing the oversight of the DOI, if the category I is the DOI, improving the accuracy of the category I is effective in reducing the oversight of the DOI. Therefore, as shown in FIG. Consider not only leaving the I category links 1003 and 1005 without deleting them, but also adding an intermediate classification ⁇ and category I category links 1004.
- the number of missed images due to misclassification of observation images that should be classified into the DOI category can be minimized, so that it is possible to prevent omission of countermeasures to problematic processes due to missed DOIs. it can.
- FIG. 13 is an example of a GUI showing a result of optimization of classification parameters based on two determination criteria of reduction of MDC work and reduction of missed DOI.
- the category C is selected as the DOI
- the miss rate of the category C is set as one of the determination criteria 1301.
- Purity ⁇ 80% is set as the determination standard 1301 that does not require the MDC work.
- the display order of the confusion matrix is such that the sorting criterion 1302 is in the order of low DOI miss rate, so that the parameter displayed at the top can be determined as the optimum parameter.
- the number of defects that should be classified into the category set in the DOI that is, the number of defects classified as category C in the MDC is used as the denominator, and the category that does not perform the confirmation work by the MDC work is mistaken.
- Example 7 a configuration of an SEM type defect observation apparatus having a function of selecting a defect to be used for MDC performed in advance to generate a confusion matrix on a defect map will be described.
- FIG. 14 shows an example of a GUI screen for selecting a defect for performing MDC.
- the defect map 1401 is displayed on the left side of the screen, and the apparatus user operates a pointer 1402 shown on the GUI using a pointing device such as a mouse to select an arbitrary defect on the defect map.
- the selected defect is activated and displayed on the GUI as indicated by 1403, and a high-magnification image of the selected defect is displayed as a thumbnail on the left side of the screen.
- a defect ID is assigned to each defect, and a defect ID display field 1405 is also displayed below the thumbnail image 1404.
- the personal computer When selecting a defect to perform MDC after ADR execution, the personal computer provided in the display unit 206 searches and displays the storage device 116 based on the defect ID.
- defect selection is performed before ADR execution, an SEM image is captured via the overall control unit.
- thumbnail images patch images acquired by an appearance inspection apparatus can be used.
- the device user inputs defect type information in the defect type input box 1406 while referring to the displayed thumbnail image 1404.
- the active display 1403 is changed to an inactive display 1407 as indicated by a black circle in the drawing.
- the input defect ID and defect type information of the defect ID are stored in the classification information storage unit 205 and used when generating a confusion matrix.
- the MDC execution screen of the present embodiment may be displayed not only on the GUI displayed on the operation unit 115 but also on an offline computer (with the SEM type defect observation apparatus).
- Electron gun 102 Lens 103 Deflector 104 Objective lens 105
- Sample 106 Stage 107
- Electron beam 108 Secondary particle 109
- Secondary particle detector 110
- Electron optical system control unit 111
- a / D conversion unit 112
- Stage control unit 113
- Image processing unit 115
- Operation unit 116
- Storage device 117
- ADR processing unit 118
- ADC processing unit 119
- Optical microscope 201 Image information storage unit 202
- ADC control unit 204
- Classification processing unit 205
- Classification information storage unit 206
- Display unit 207 Keyboard 208
- Mouse 401 Confusion matrix 402 Message 403, 404 indicating category that requires MDC 405, 501, 801, 1201 Judgment standard setting box 502, 1202 Sort standard setting box 503, 504, 1205, 1206 Meter input box 505, 1207 Manual switching button 506, 1208 Auto switching button 802, 803, 1304, 1305 Result 1001, 1002, 1003, 1004, 1005 Category link 1203, 1204 Message
Landscapes
- Engineering & Computer Science (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Analysing Materials By The Use Of Radiation (AREA)
- Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)
- Testing Or Measuring Of Semiconductors Or The Like (AREA)
- Image Processing (AREA)
Abstract
Description
実施例1では、コンフュージョンマトリックス上にADCとMDC切り分けの判断情報を表示する機能を備えたSEM式欠陥観察装置の構成例について説明する。
・True Positive:モデルはPositive(陽性)と分類し、その結果も正しかった(True)、すなわち実際もPositiveであった場合の数
・False Positive:モデルはPositive(陽性)と分類したが、その判定は誤り(False)、すなわち実際はNegativeであった場合の数
・False Negative:モデルはNegative(陰性)と分類したが、その判定は誤り(False)、すなわち実際はPositiveであった場合の数
・True Negative:モデルはNegative(陰性)と分類し、その判定も正しかった(True)、すなわち実際もNegativeであった場合の数
図4(A)に示すコンフュージョンマトリックスをADCの正しさ評価に応用する場合、図4(B)に示される様式のコンフュージョンマトリックスが使用される場合が多い。図4(B)に示すコンフュージョンマトリックスは、横方向にADC結果を、縦方向に真の分類結果をマトリックス表示した分類表であり、図4(B)に示す例では、ADCによりカテゴリAに分類され、真の分類結果もAであった欠陥点数(True Positive)がa個、ADCによりカテゴリAに分類されたが真の分類結果はBであった欠陥点数(False Positive)がc個、ADCによりカテゴリBに分類されたが真の分類結果はAであった欠陥点数(False Negative)がb個、ADCによりカテゴリBに分類され、真の分類結果もBであった欠陥点数(True Negative)がd個となっている。なお、実際にコンフュージョンマトリックスを作成する場合には、真の分類結果としてはMDCの結果を使用する。
本実施例では、特徴量空間における各カテゴリ間の境界線を設定しこの境界線に基づき分類を行うアルゴリズムを採用した場合の境界値(パラメータ)を最適化する手法について説明する。説明では2次元空間に簡略化しており、二分木構造の分類モデルを使用しているが、アルゴリズムは二分木構造に限定するものではなく、多次元空間における境界設定にも適用できる。
本実施例では、各特徴量の重みを変化させることで、分類性能を向上させるアルゴリズムを採用している場合の重み最適化の手法について説明する。本実施例の場合、パラメータとしては重みが相当することになる。図9では、説明簡略化のために2つの特徴量に限定しているが、アルゴリズムは多次元空間においても適用可能であり、多次元空間における教示データの特徴量分布と分類対象データとの距離算出、あるいはファージーボーティングなど特徴量空間における距離に相当する量を別の算出方法で定量化する方式にも適用可能である。
本実施例では、段階的に分類対象を分類するシステムにおいて、各段階の接続を最適化する手法の実施例について説明する。ここでは、各段階の接続がパラメータとなる。各段階における分類アルゴリズムは、異質なものの組み合わせに限定するものではなく、一つの分類アルゴリズム内で多段階に、例えば分類対象を限定して絞り込んでいくモデルにも適用できる。また、実施例2,3と同様、本実施例でもSEM式欠陥観察装置への実装を想定しているが、装置のハードウェア構成は実施例1と同様であるので、説明は省略する。
本実施例では、着目する欠陥種、つまり装置ユーザーが特に興味のある特定の欠陥カテゴリについて、検出漏れ・見逃しをなるべく低減するためのパラメータ最適化手法について説明する。以下では、上記の興味欠陥のことをDOIと称する。装置の全体構成は実施例1と同様であるので説明は省略する。
図13は、MDC作業の低減と、DOIの見逃し低減という二つの判定基準に基づいて、分類パラメータの最適化を行った結果を示す、GUIの一例である。ここでは、DOIとしてカテゴリCを選択し、カテゴリCの見逃し率を、判定基準1301の一つとして設定している。さらに、MDC作業の低減を目的とし、Purity≧80%をMDC作業不要の判定基準1301として設定している。コンフュージョンマトリックスの表示順は、ソート基準1302をDOIの見逃し率が低い順にしているので、上位に表示されるパラメータが、最適なパラメータと判断できる。ここでは、DOIの見逃し率の定義として、DOIに設定したカテゴリに分類されるべき欠陥数、つまりMDCでカテゴリCと分類された欠陥数を分母とし、MDC作業による確認作業を実施しないカテゴリに誤分類された欠陥画像、つまり、ADCによって、カテゴリC以外のカテゴリで、かつPurityが設定閾値以上のカテゴリに分類された欠陥数を分子としている。したがって、パラメータ2を用いた場合のDOIの見逃し率は、真値がカテゴリCにも関わらずADCでカテゴリAと分類された欠陥数/カテゴリCのMDC結果=0/7=0%(MDC作業による確認作業を実施するカテゴリBに誤分類した欠陥画像の数1303は、DOIの見逃しとしてはカウントしない)となり、パラメータ1を用いた場合のDOI見逃し率10%を上回る結果となる。本実施例によれば、このような複雑な判定基準による分類パラメータの良否判定を、簡単に実行することが可能になる。
本実施例では、コンフュージョンマトリックスを生成するために事前に行うMDCに使用する欠陥を欠陥マップ上で選択する機能を備えたSEM式欠陥観察装置の構成について説明する。
102 レンズ
103 偏向器
104 対物レンズ
105 試料
106 ステージ
107 電子ビーム
108 二次粒子
109 二次粒子検出器
110 電子光学系制御部
111 A/D変換部
112 ステージ制御部
113 全体制御部
114,203 画像処理部
115 操作部
116 記憶装置
117 ADR処理部
118 ADC処理部
119 光学顕微鏡
201 画像情報記憶部
202 ADC制御部
204 分類処理部
205 分類情報記憶部
206 表示部
207 キーボード
208 マウス
401 コンフュージョンマトリックス
402 MDCが必要なカテゴリを示すメッセージ
403,404 強調表示
405,501,801,1201 判定基準設定ボックス
502,1202 ソート基準設定ボックス
503,504,1205,1206 パラメータ入力ボックス
505,1207 Manual切り替えボタン
506,1208 Auto切り替えボタン
802,803,1304,1305 結果
1001,1002,1003,1004,1005 カテゴリリンク
1203,1204 DOI見逃し率を示すメッセージ
1301 判定基準
1302 ソート基準
1303 誤分類した欠陥画像の数
Claims (11)
- 被観察試料の画像を取得し、当該画像から欠陥に該当する領域を欠陥画像として抽出する機能を備えた欠陥レビュー装置において、
前記被観察試料を撮像し、撮像した画像を画像信号として出力する電子光学系と、
前記欠陥画像の抽出処理を行うADR処理部と、
前記欠陥を複数のカテゴリに自動分類するADC処理部と、
当該ADC処理部による自動分類結果のうち、人手によるマニュアル分類を行うべきカテゴリを判断するための判断情報が表示される画面表示手段とを備えたことを特徴とする欠陥レビュー装置。 - 請求項1に記載の欠陥レビュー装置において、
前記欠陥に対する前記マニュアル分類結果と前記ADCによる分類結果とを、前記複数のカテゴリ毎にマトリックス表示するコンフュージョンマトリックスが前記画面表示手段上に表示されることを特徴とする欠陥レビュー装置。 - 請求項2に記載の欠陥レビュー装置において、
前記ADC処理部の分類結果と当該ADCの対象となった欠陥画像に対する人手によるマニュアル分類結果とが格納される記憶手段を備え、
前記ADC処理部は、前記ADCによる分類結果と、前記マニュアル分類結果とを用いて、各カテゴリに対する前記ADCによる欠陥数および前記マニュアル分類による欠陥数をカテゴリ毎に配列したコンフュージョンマトリックスおよび前記判断情報とを生成し、
前記画面表示手段に表示することを特徴とする欠陥レビュー装置。 - 請求項2に記載の欠陥レビュー装置において、
前記判断情報が、前記各カテゴリに対する欠陥数の純度あるいは正解率であることを特徴とする欠陥レビュー装置。 - 請求項1に記載の欠陥レビュー装置において、
前記ADC処理部は、前記欠陥分類を行うための閾値情報を複数備え、
当該複数の閾値に対して得られる前記ADCの結果に基づき、前記複数の閾値に対する前記判断情報を算出することを特徴とする欠陥レビュー装置。 - 請求項1に記載の欠陥レビュー装置において、
前記判断情報が、人手によるマニュアル分類が必要なカテゴリ,当該人手によるマニュアル分類の不要率,着目欠陥に対する見逃し率のいずれかであることを特徴とする欠陥レビュー装置。 - 請求項4に記載の欠陥レビュー装置において、
前記純度がコンフュージョンマトリックス上で強調表示されることを特徴とする欠陥レビュー装置。 - 被観察試料の画像を取得し、当該画像から欠陥に該当する領域を欠陥画像として抽出する機能を備えた欠陥レビュー装置において、
前記被観察試料を撮像し、撮像した画像を画像信号として出力する操作電子顕微鏡と、
前記欠陥画像の抽出処理を行うADR処理部と、
前記欠陥を複数のカテゴリに自動分類するADC処理部と、
前記ADR処理部により得られた欠陥画像の一部を選択するための選択画面が表示される画面表示手段と、
当該選択画面により選択された欠陥画像に対する前記ADC処理部でのADC結果と、当該ADC結果の対象となった欠陥画像に対する人手によるマニュアル分類結果とが格納される記憶手段と、
前記ADC結果に対して、前記マニュアル分類を行うべきカテゴリを判断するための判断情報を計算する演算手段とを備えたことを特徴とする欠陥レビュー装置。 - 請求項8に記載の欠陥レビュー装置において、
前記判断情報が表示される画面表示手段を備えたことを特徴とする欠陥レビュー装置。 - 被観察試料の走査電子顕微鏡画像から得られる欠陥画像を所定の分類モデルに基づき複数のカテゴリに自動分類する演算装置と、当該演算処理に必要な設定情報の読み込み手段とを備えた欠陥レビュー装置で使用されるプログラムであって、
所定数の欠陥画像に対する前記自動分類結果と、前記所定数の欠陥画像に対する人手によるマニュアル分類結果を読み込むステップと、
前記自動分類結果と前記マニュアル分類結果を用いて、前記自動分類結果のうち前記マニュアル分類を行うべきカテゴリを判断するための判断情報を計算するステップと、
を前記演算装置に実行させるプログラムを格納した記憶装置。 - 請求項10に記載のプログラムを格納した記憶装置であって、
前記プログラムは、
前記自動分類結果と前記マニュアル分類結果を用いて、前記自動分類による欠陥数および前記マニュアル分類による欠陥数とを前記複数のカテゴリ毎に配列したコンフュージョンマトリックスを生成するステップと、
前記各カテゴリに対する欠陥数の正答率と純度とを算出するステップと、
を前記演算装置に実行させることを特徴とする記憶装置。
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US13/702,674 US9020237B2 (en) | 2010-06-07 | 2011-05-13 | Method for optimizing observed image classification criterion and image classification apparatus |
JP2012519217A JPWO2011155123A1 (ja) | 2010-06-07 | 2011-05-13 | 観察画像の分類基準の最適化方法、および画像分類装置 |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2010-129521 | 2010-06-07 | ||
JP2010129521 | 2010-06-07 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2011155123A1 true WO2011155123A1 (ja) | 2011-12-15 |
Family
ID=45097745
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2011/002661 WO2011155123A1 (ja) | 2010-06-07 | 2011-05-13 | 観察画像の分類基準の最適化方法、および画像分類装置 |
Country Status (3)
Country | Link |
---|---|
US (1) | US9020237B2 (ja) |
JP (1) | JPWO2011155123A1 (ja) |
WO (1) | WO2011155123A1 (ja) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2013225313A (ja) * | 2012-04-20 | 2013-10-31 | Applied Materials Israel Ltd | 自動欠陥分類における分類器の準備及び保守 |
JP2013224943A (ja) * | 2012-04-19 | 2013-10-31 | Applied Materials Israel Ltd | 自動及び手動欠陥分類の統合 |
WO2016092614A1 (ja) * | 2014-12-08 | 2016-06-16 | 株式会社日立ハイテクノロジーズ | 欠陥検査装置、表示装置、及び欠陥分類装置 |
US10114368B2 (en) | 2013-07-22 | 2018-10-30 | Applied Materials Israel Ltd. | Closed-loop automatic defect inspection and classification |
JP2020101958A (ja) * | 2018-12-21 | 2020-07-02 | カシオ計算機株式会社 | 識別装置、識別方法及びプログラム |
WO2021075152A1 (ja) * | 2019-10-18 | 2021-04-22 | 株式会社日立ハイテク | 欠陥分類装置、欠陥分類プログラム |
Families Citing this family (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP5608575B2 (ja) * | 2011-01-19 | 2014-10-15 | 株式会社日立ハイテクノロジーズ | 画像分類方法および画像分類装置 |
US9305219B2 (en) * | 2014-01-23 | 2016-04-05 | Mitsubishi Electric Research Laboratories, Inc. | Method for estimating free space using a camera system |
US11182280B2 (en) * | 2014-10-13 | 2021-11-23 | Microsoft Technology Licensing, Llc | Application testing |
US9652688B2 (en) * | 2014-11-26 | 2017-05-16 | Captricity, Inc. | Analyzing content of digital images |
KR102392057B1 (ko) * | 2014-12-22 | 2022-04-28 | 삼성전자주식회사 | 자동 결함 분류 방법 |
US9659359B1 (en) * | 2015-05-05 | 2017-05-23 | Flextronics Ap, Llc | Method of and device for quality control process optimization |
WO2018216629A1 (ja) * | 2017-05-22 | 2018-11-29 | キヤノン株式会社 | 情報処理装置、情報処理方法、及びプログラム |
US11138507B2 (en) * | 2017-09-28 | 2021-10-05 | Applied Materials Israel Ltd. | System, method and computer program product for classifying a multiplicity of items |
CN111712769B (zh) * | 2018-03-06 | 2023-08-01 | 欧姆龙株式会社 | 用于设定照明条件的方法、装置、系统以及存储介质 |
JP7015001B2 (ja) * | 2018-03-14 | 2022-02-02 | オムロン株式会社 | 欠陥検査装置、欠陥検査方法、及びそのプログラム |
RU2698402C1 (ru) * | 2018-08-30 | 2019-08-26 | Самсунг Электроникс Ко., Лтд. | Способ обучения сверточной нейронной сети для восстановления изображения и система для формирования карты глубины изображения (варианты) |
JP6869490B2 (ja) * | 2018-12-28 | 2021-05-12 | オムロン株式会社 | 欠陥検査装置、欠陥検査方法、及びそのプログラム |
EP3836019A1 (de) | 2019-12-09 | 2021-06-16 | Siemens Aktiengesellschaft | Verfahren und assistenzsystem zur überprüfung von mustern auf fehlerhaftigkeit |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2001189362A (ja) * | 2000-01-04 | 2001-07-10 | Toshiba Corp | 半導体欠陥分類装置及び半導体欠陥分類方法 |
JP2003515942A (ja) * | 1999-11-29 | 2003-05-07 | ケイエルエイ−テンコー コーポレイション | 半導体欠陥用のパワー・アシスト自動監視分類器作製ツール |
JP2006300517A (ja) * | 2005-04-15 | 2006-11-02 | Hitachi High-Technologies Corp | 検査装置 |
JP2009103508A (ja) * | 2007-10-22 | 2009-05-14 | Hitachi Ltd | 欠陥分類方法及びその装置 |
JP2010087322A (ja) * | 2008-10-01 | 2010-04-15 | Hitachi High-Technologies Corp | 欠陥観察装置及び欠陥観察方法 |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6922482B1 (en) | 1999-06-15 | 2005-07-26 | Applied Materials, Inc. | Hybrid invariant adaptive automatic defect classification |
JP4908995B2 (ja) | 2006-09-27 | 2012-04-04 | 株式会社日立ハイテクノロジーズ | 欠陥分類方法及びその装置並びに欠陥検査装置 |
-
2011
- 2011-05-13 JP JP2012519217A patent/JPWO2011155123A1/ja active Pending
- 2011-05-13 US US13/702,674 patent/US9020237B2/en active Active
- 2011-05-13 WO PCT/JP2011/002661 patent/WO2011155123A1/ja active Application Filing
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2003515942A (ja) * | 1999-11-29 | 2003-05-07 | ケイエルエイ−テンコー コーポレイション | 半導体欠陥用のパワー・アシスト自動監視分類器作製ツール |
JP2001189362A (ja) * | 2000-01-04 | 2001-07-10 | Toshiba Corp | 半導体欠陥分類装置及び半導体欠陥分類方法 |
JP2006300517A (ja) * | 2005-04-15 | 2006-11-02 | Hitachi High-Technologies Corp | 検査装置 |
JP2009103508A (ja) * | 2007-10-22 | 2009-05-14 | Hitachi Ltd | 欠陥分類方法及びその装置 |
JP2010087322A (ja) * | 2008-10-01 | 2010-04-15 | Hitachi High-Technologies Corp | 欠陥観察装置及び欠陥観察方法 |
Cited By (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
TWI639824B (zh) * | 2012-04-19 | 2018-11-01 | 應用材料以色列公司 | 用於自動及手動缺陷分類之整合的方法、設備及非暫態電腦可讀取儲存媒介 |
JP2013224943A (ja) * | 2012-04-19 | 2013-10-31 | Applied Materials Israel Ltd | 自動及び手動欠陥分類の統合 |
US10043264B2 (en) | 2012-04-19 | 2018-08-07 | Applied Materials Israel Ltd. | Integration of automatic and manual defect classification |
JP2013225313A (ja) * | 2012-04-20 | 2013-10-31 | Applied Materials Israel Ltd | 自動欠陥分類における分類器の準備及び保守 |
US10901402B2 (en) | 2013-07-22 | 2021-01-26 | Applied Materials Israel, Ltd. | Closed-loop automatic defect inspection and classification |
US10114368B2 (en) | 2013-07-22 | 2018-10-30 | Applied Materials Israel Ltd. | Closed-loop automatic defect inspection and classification |
US9964500B2 (en) | 2014-12-08 | 2018-05-08 | Hitachi High-Technologies Corporation | Defect inspection device, display device, and defect classification device |
WO2016092614A1 (ja) * | 2014-12-08 | 2016-06-16 | 株式会社日立ハイテクノロジーズ | 欠陥検査装置、表示装置、及び欠陥分類装置 |
JP2020101958A (ja) * | 2018-12-21 | 2020-07-02 | カシオ計算機株式会社 | 識別装置、識別方法及びプログラム |
JP7268347B2 (ja) | 2018-12-21 | 2023-05-08 | カシオ計算機株式会社 | 識別装置、識別方法及びプログラム |
WO2021075152A1 (ja) * | 2019-10-18 | 2021-04-22 | 株式会社日立ハイテク | 欠陥分類装置、欠陥分類プログラム |
JP2021068745A (ja) * | 2019-10-18 | 2021-04-30 | 株式会社日立ハイテク | 欠陥分類装置、欠陥分類プログラム |
KR20220062568A (ko) | 2019-10-18 | 2022-05-17 | 주식회사 히타치하이테크 | 결함 분류 장치, 결함 분류 프로그램 |
JP7390851B2 (ja) | 2019-10-18 | 2023-12-04 | 株式会社日立ハイテク | 欠陥分類装置、欠陥分類プログラム |
Also Published As
Publication number | Publication date |
---|---|
US9020237B2 (en) | 2015-04-28 |
JPWO2011155123A1 (ja) | 2013-08-01 |
US20130077850A1 (en) | 2013-03-28 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2011155123A1 (ja) | 観察画像の分類基準の最適化方法、および画像分類装置 | |
US11526979B2 (en) | Method of defect classification and system thereof | |
US7873205B2 (en) | Apparatus and method for classifying defects using multiple classification modules | |
JP6391083B2 (ja) | 自動欠陥分類における分類器の準備及び保守 | |
US9280814B2 (en) | Charged particle beam apparatus that performs image classification assistance | |
US8625906B2 (en) | Image classification standard update method, program, and image classification device | |
JP5452392B2 (ja) | 欠陥観察方法及び欠陥観察装置 | |
JP6285640B2 (ja) | 自動及び手動欠陥分類の統合 | |
US20080298670A1 (en) | Method and its apparatus for reviewing defects | |
KR101955268B1 (ko) | 결함 관찰 장치 및 결함 관찰 방법 | |
WO2018061067A1 (ja) | 欠陥検査装置及び欠陥検査方法 | |
JP2012225768A (ja) | 欠陥分類方法及び欠陥分類システム | |
KR102530950B1 (ko) | 반도체 시편에서의 결함들의 분류 | |
JP2012181209A (ja) | 欠陥分類方法及びその装置 | |
KR20230119625A (ko) | 결함 검사 장치, 결함 검사 방법 및 제조 방법 | |
JP2021002354A (ja) | 表示制御装置、表示制御方法及び表示制御プログラム | |
WO2016092614A1 (ja) | 欠陥検査装置、表示装置、及び欠陥分類装置 | |
WO2021075152A1 (ja) | 欠陥分類装置、欠陥分類プログラム | |
JP2017003305A (ja) | 欠陥画像分類装置 | |
WO2013008531A1 (ja) | 走査電子顕微鏡、欠陥検査システム、および欠陥検査装置 | |
JP2021162527A (ja) | 半導体の欠陥検出装置及びプログラム |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 11792086 Country of ref document: EP Kind code of ref document: A1 |
|
WWE | Wipo information: entry into national phase |
Ref document number: 2012519217 Country of ref document: JP |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
WWE | Wipo information: entry into national phase |
Ref document number: 13702674 Country of ref document: US |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 11792086 Country of ref document: EP Kind code of ref document: A1 |