Pattern matching device

Abstract

The present disclosure proposes a pattern matching technique using machine learning, wherein the technique greatly reduces the workload related to data collection and true value creation work and eliminates manual work in a training process (creation of a training data set, training of a machine learning model, performance evaluation, and the like) of a machine learning model. A pattern matching device according to the present disclosure performs pattern matching using either a first algorithm that does not use a learning device or a second algorithm that uses the learning device, performs the pattern matching using the second algorithm when the pattern matching using the first algorithm fails, and performs re-training of the learning device when the pattern matching using the second algorithm fails (see FIG. 9).

Description

Pattern Matching Device

[0001] The present disclosure relates to techniques for performing pattern matching between images.

[0002] Devices that measure and inspect patterns formed on semiconductor wafers often use template matching technology to align the field of view of the device with the desired measurement or inspection position. Template matching is a process of finding an area in a search image that most closely matches a pre-registered template image. Patent Document 1 describes an example of such template matching.

[0003] Patent Document 2 describes a template matching device with a learning function that estimates a correlation image having pixel values ​​that represent the correlation between a template image and a searched image. This document describes a template matching device that can absorb the visual discrepancy between the template image and the searched image through learning, enabling highly accurate matching.

[0004] Patent Literature 3 describes a method for selecting the minimum amount of training data necessary to achieve the target performance of a machine learning model based on semiconductor design data when applying machine learning to a semiconductor inspection device that images patterns formed on a semiconductor wafer. When using machine learning for image inspection, it is necessary to collect a large amount of image data and generate accurate inspection results corresponding to each image data, which places a heavy burden on the device user. Patent Literature 3 reduces the burden on the device user by selecting the minimum amount of data necessary when training the machine learning model.

[0005] Japanese Patent No. 4218171 (corresponding US Patent No. 6,627,888) WO2022 / 009357 JP 2020-35282 A

[0006] Conventionally, the template matching technology used in semiconductor inspection and measurement equipment has been a rule-based pattern matching method that performs matching based on a predetermined algorithm. Generally, rule-based methods are considered highly versatile because they allow algorithms to be designed based on a variety of known cases. However, when matching fails due to a discrepancy in appearance between the template image and the searched image, it is difficult to improve the algorithm, and it takes time to address the issue.

[0007] Therefore, a pattern matching method using machine learning with a learning function has been devised, as disclosed in Patent Document 2. This method absorbs the visual discrepancy between a template image and a searched image through learning, enabling highly accurate matching. However, in order to train a machine learning model with sufficient performance, it is necessary to collect a large amount of training data, such as images, and create correct pattern matching results (hereinafter referred to as true values) for the collected training data. Creating true values ​​for a large amount of data requires manual work and time, placing a heavy burden on the worker. Furthermore, if the types of data included in the training data (e.g., image shape, image quality, etc.) are insufficient, the learning will be compatible with only specific data, reducing the versatility of the machine learning model.

[0008] Patent Literature 3 describes a method for analyzing semiconductor design data and captured images to select and collect the minimum amount of training data necessary to achieve the target performance of a machine learning model. This method ensures a variety of training data while reducing the amount of data, thereby improving the versatility of the machine learning model and reducing the effort required for creating true values. The same document also describes the creation of estimated true value images that simplify the task of creating true values ​​for each pixel in segmentation, a machine learning task. However, this method does not eliminate the need for manual work, and there is still room for automation.

[0009] The present disclosure has been made in consideration of the above-mentioned problems, and proposes a method for significantly reducing the workload associated with data collection and true value creation in a pattern matching method using machine learning, and for reducing the amount of manual work required in the learning process of a machine learning model (creating a learning dataset, learning a machine learning model, performance evaluation, etc.).

[0010] The pattern matching device according to the present disclosure performs pattern matching using either a first algorithm that does not use a learning device or a second algorithm that uses the learning device, and if the pattern matching using the first algorithm fails, performs the pattern matching using the second algorithm, and if the pattern matching using the second algorithm fails, re-learns the learning device.

[0011] The pattern matching device according to the present disclosure can reduce the workload associated with collecting training data used in training a machine learning model and creating true values, thereby reducing the amount of manual work required in the training process of the machine learning model (creating a training dataset, training the machine learning model, performance evaluation, etc.), thereby significantly reducing the burden on users of devices equipped with machine learning-based pattern matching.

[0012] 1 is a configuration diagram of a pattern dimension measurement system according to a first embodiment. FIG. 2 is a flowchart explaining the operation of the pattern dimension measurement system. FIG. 3 is a flowchart explaining the specific operations of S206 to S207. FIG. 4 is a flowchart explaining the specific operation of S214. FIG. 5 is a flowchart explaining the specific operation of S213. FIG. 6 shows an example of data held by the temporary storage data 406 and the learning data set 407. FIG. 7 shows an example of data in a learning log. FIG. 8 shows an example of the configuration of a measurement apparatus 300 including an SEM 600 and a GUI 307. FIG. 9 is a flowchart explaining the operation of a pattern dimension measurement system according to a third embodiment. FIG. 10 is a block diagram of a recipe generation apparatus 200.

[0013] 1 is a configuration diagram of a pattern dimension measurement system according to a first embodiment of the present disclosure. The pattern dimension measurement system is a system that measures the size of a shape pattern formed on a semiconductor wafer. The pattern dimension measurement system includes a design data storage unit 100, a recipe generation device 200, a measurement device 300 (pattern matching device), and a machine learning model training device 400 (pattern matching device). Data created and output by each device constituting the system can be transmitted and received between the devices.

[0014] The matching result of the pattern matching is output as a matching shift amount. The matching shift amount represents the position shift amount or position difference between the design data 101 and the SEM (Scanning Electron Microscope) image 310, and can be represented by a two-dimensional scalar value, for example, (shift amount in the X direction, shift amount in the Y direction). Note that the template image is not limited to the SEM image, and other types of images, such as the design data 101, may be used. Also, the searched image is not limited to the design data 101, and other types of images, such as the SEM image 310, may be used.

[0015] The design data storage unit 100 holds design data 101 that indicates the pattern structure of a semiconductor device created by an apparatus user.

[0016] The recipe generation device 200 has an interface through which the device user inputs measurement points on a device pattern referenced from the design data 101, and outputs a measurement recipe 201 based on the measurement point information input from the interface and the design data 101, the measurement recipe 201 including imaging conditions for the measurement device 300, an algorithm used in rule-based pattern matching, measurement point information, and the like.

[0017] The measurement apparatus 300 measures the dimensions of a device pattern based on the measurement recipe 201 and outputs the measurement values. The measurement apparatus 300 includes an imaging unit 301 that images a device pattern formed on a semiconductor wafer, a rule-based pattern matching unit 302, a machine learning pattern matching unit 303, a matching success / failure determination unit 304, a measurement unit 305, a measurement success / failure determination unit 306, a GUI (Graphical User Interface) 307, and a measurement information storage unit 308.

[0018] The imaging unit 301 captures an image of a device pattern based on the imaging conditions of the measurement recipe 201 and outputs an SEM image 310. The rule-based pattern matching unit 302 performs pattern matching between the SEM image 310 and the design data 101 using an algorithm described in the measurement recipe 201 and outputs a matching shift amount 314 between the two images. The machine learning pattern matching unit 303 performs pattern matching between the SEM image 310 and the design data 101 using a machine learning model 312 and outputs a matching shift amount 314 between the two images. The data used for pattern matching is not limited to the above combination, and any data that can identify a matching point in the field of view between the template image and the searched image may be used. For example, a design drawing obtained from the SEM image and the design data 101 may be used, or point cloud information of the SEM image and the design data 101 may be used. The matching success / failure determination unit 304 performs matching success / failure determination based on the design data 101, the SEM image 310, and the matching shift amount 314 and outputs a matching success / failure determination result. The measurement unit 305 measures the dimensions of the device pattern based on the measurement point information described in the measurement recipe 201, the design data 101, and the SEM image 310, and outputs the measurement values. The measurement success / failure determination unit 306 determines the success / failure of the measurement based on the measurement values, and outputs a measurement success / failure determination result. If the measurement success / failure determination result is failure, the device user operates the GUI 307 to input a correct matching shift amount that matches the pattern shapes between the design data 101 and the SEM image 310, and outputs a matching shift amount 314. The measurement information storage unit 308 stores the measurement recipe 201, the SEM image 310, the design data 101, and the machine learning model 312. In an initial state, the machine learning model 312 holds a base model learned by the machine learning model training device 400.

[0019] After completing the dimension measurement of the device pattern, the measurement device 300 outputs the measurement recipe 201, design data 101, SEM image 310, and matching shift amount 314 to the machine learning model training device 400. The design data 101 and SEM image 310 output at this time can be used as training data for training the machine learning model, and the matching shift amount 314 is a value obtained when the dimension measurement of the device pattern is normally performed, and can be used as a true value for training the machine learning model. Therefore, manual work in collecting training data and creating true values ​​can be almost completely eliminated.

[0020] The machine learning model learning device 400 includes a learning data analysis unit 401, a machine learning model learning unit 402, a GUI 403, a learning data storage unit 404, and a machine learning model storage unit 405. The learning data storage unit 404 stores temporary storage data 406 and a learning dataset 407. The temporary storage data 406 includes a measurement recipe storage folder 4042, a design data storage folder 4043, and an SEM image storage folder 4044. The temporary storage data 406 further includes a temporary data table 4047 and a collected data table 4049. The learning dataset 407 includes a base model learning data table 4045, a test data table 4046, and an additional learning data table 4048. Specific examples of these data configurations will be described later.

[0021] The machine learning model learning device 400 stores the measurement recipe 201, design data 101, and SEM image 310 received from the measuring device 300 in a measurement recipe save folder 4042, a design data save folder 4043, and an SEM image save folder 4044 of the temporary storage data 406, respectively. At this time, the machine learning model learning device 400 creates a collected data table 4049 in which an arbitrary data ID, the file names of the measurement recipe 201, the design data 101, and the SEM image 310, and the numerical value of the matching shift amount 314 are compiled into one row of a record, and stores the created data table in the temporary storage data 406. The collected data table 4049 is created to manage the collected data, but the method of managing the collected data is not limited to this, and any method may be used as long as it can manage combinations of the measurement recipe 201, the design data 101, the SEM image 310, the matching shift amount 314, etc., and the data may be managed by a folder structure or a file name, for example.

[0022] The training data analysis unit 401 reads each data (measurement recipe 201, design data 101, SEM image 310) based on each record in the collected data table 4049 and an existing data table (e.g., base model training data table 4045) of the training data set 407, and extracts features such as pattern shape features, SEM image features, and imaging conditions (the extraction method will be described later with reference to FIG. 5 ). Furthermore, the training data analysis unit 401 selects records from each data table so that the extracted features are not biased (a method for selecting to avoid bias will be described later with reference to FIG. 5 ), and stores each data (measurement recipe 201, design data 101, SEM image 310) included in the record selected from the collected data table 4049 in the measurement recipe storage folder 4042, the design data storage folder 4043, and the SEM image storage folder 4044. Furthermore, the records selected from each data table are saved as new data tables with unique names. This ensures a sufficient variety of data included in the training data set 407, thereby improving the performance of the machine learning model while ensuring versatility. Furthermore, since a new training data set can be created while retaining a training data set that was previously used, management of multiple training data sets can be facilitated. When analyzing data in the training data analysis unit 401, if there is an error in the matching shift amount 314, the error can be corrected manually using a GUI 403, which will be described later.

[0023] The machine learning model learning unit 402 reads data necessary for learning (such as the design data 101, the SEM image 310, and the matching shift amount 314) based on a data table (e.g., the additional learning data table 4048) in the learning dataset 407, and learns a machine learning model based on default learning conditions (details of the machine learning model learning method will be described later using FIGS. 3 and 4). To evaluate the performance of the machine learning model after learning, the machine learning model learning unit 402 reads data necessary for inference (such as the design data 101, the SEM image 310, and the matching shift amount 314) from a test data table 4046 stored in the learning dataset 407, and evaluates the performance of the learned model. If the target performance is not achieved, the learning conditions can be changed and re-learning can be performed. This allows a machine learning model with the required performance to be obtained. If the target performance is achieved, the learning log and the data file of the learned machine learning model 408 are given unique names and stored in the machine learning model 312. At this time, previously learned machine learning models are retained.

[0024] If there is an error in the matching shift amount 314 and it is necessary to instruct the correct matching shift amount, the device user operates the GUI 403 to instruct and output the correct matching shift amount that will match the pattern shapes of the design data 101 and the SEM image 310.

[0025] 2 is a flowchart illustrating the operation of the pattern dimension measurement system. Each step in FIG. 2 will be described below.

[0026] (FIG. 2: S200) Based on measurement point information of a device pattern input by an apparatus user from the interface of the recipe generation apparatus 200 and design data 101 obtained from the design data storage unit 100, the recipe generation apparatus 200 generates a measurement recipe 201 including imaging conditions (which may include imaging position information, etc.) for the measurement apparatus 300, an algorithm used in rule-based pattern matching, measurement point information, etc.

[0027] (FIG. 2: S201) The imaging unit 301 of the measurement device 300 images a device pattern based on the imaging conditions of the measurement recipe 201 and outputs an SEM image 310.

[0028] (FIG. 2: S202) The rule-based pattern matching unit 302 calculates a matching shift amount 314 between the design data 101 and the SEM image 310. The data used for pattern matching is not limited to this combination, and any data capable of identifying a matching point in the field of view between the template image and the searched image may be used. For example, a design drawing obtained from the SEM image 310 and the design data 101 may be used, or point cloud information of the SEM image 310 and the design data 101 may be used. The matching shift amount 314 is calculated using a matching algorithm determined based on the measurement recipe 201. The matching algorithm may be, for example, an algorithm that calculates the similarity at each candidate position between the design data 101 and the SEM image 310 using a mean square error and outputs the matching shift amount for the position with the highest similarity. The matching algorithm is not limited to this method, and may be an algorithm that extracts each image feature based on a predetermined process and calculates the matching shift amount based on the extracted image feature.

[0029] (FIG. 2: S203) The matching success / failure determination unit 304 determines whether the matching is successful or not based on the design data 101, the SEM image 310, and the matching shift amount 314. One example of a method for determining whether the matching is successful or not is to calculate a mean square error representing the degree of similarity between the design data 101 and the SEM image 310 based on the matching shift amount 314, and determine whether the matching is successful if the value is equal to or greater than a threshold. The method for determining whether the matching is successful or not is not limited to this method, and any method can be used as long as it can determine whether a position on the design data 101 based on the matching shift amount 314 corresponds to the field of view of the SEM image 310. If the result of the matching success / failure determination is successful, the process skips to S208. If the result of the matching success / failure determination is unsuccessful, the process proceeds to S204.

[0030] (FIG. 2: S204) The machine learning pattern matching unit 303 calculates the matching shift amount 314 using a machine learning model that has been trained to estimate the matching shift amount 314 based on the design data 101 and the SEM image 310. The data used for pattern matching is not limited to the above combination, and any data that can identify the matching point in the field of view between the template image and the searched image may be used. For example, a design drawing obtained from the SEM image and the design data 101 may be used, or point cloud information of the SEM image and the design data 101 may be used. As the machine learning model, a specified machine learning model (such as the machine learning model with the highest performance or the latest machine learning model) is read from the machine learning model 312 and used. As an example, a pattern matching method using machine learning can be implemented by estimating a correlation image between images as described in Patent Document 2 and calculating the matching shift amount from the correlation image. The pattern matching method using machine learning is not limited to this, and any method that can identify the matching point in the field of view between the template image and the searched image may be used. For example, a method in which the matching shift amount between images is directly estimated using a machine learning model may be used.

[0031] (FIG. 2: S205) The matching success / failure determination unit 304 determines whether the matching is successful or not based on the matching shift amount 314. If the result of the matching success / failure determination is success, S208 is executed. If the result of the matching success / failure determination is failure, the design data 101 and the SEM image 310 are stored in the temporary storage data 406 of the machine learning model learning device 400, and S206 is executed to perform learning of the machine learning model. At this time, the machine learning model learning device 400 creates a temporary data table 4047 having records with the file names of the received design data 101 and SEM image 310, and stores them in the temporary storage data 406 with unique names.

[0032] (FIG. 2: S206) The device user uses the GUI 403 to read the design data 101 and the SEM image 310 based on the temporary data table 4047. The device user operates the GUI 403 to input the correct matching shift amount that will cause the pattern shapes of the two images to match. The method of inputting the correct matching shift amount is not limited to using the GUI 403, and any method that can input the correct matching shift amount may be used. For example, pattern matching may be performed between the design data 101 and the SEM image 310 using a pattern matching algorithm based on multiple rules, and the most similar matching shift amount may be set as the correct matching shift amount. The machine learning model learning device 400 creates a failure pattern learning data table 4041 that includes the generated correct matching shift amount, the design data 101, and the SEM image 310. Details will be described later using FIG. 3.

[0033] (FIG. 2: S207) The machine learning model learning unit 402 performs learning of the machine learning model using the failure pattern learning data table 4041. The machine learning model 408 after learning and the learning log are stored in the machine learning model storage unit 405. The machine learning model 408 after learning is also stored in the machine learning model 312 of the measuring device 300. Details will be described later using FIG. 3. S204 is performed again using this machine learning model 408 after learning.

[0034] (FIG. 2: S208) The measurement unit 305 measures the dimensions of the device pattern based on the measurement recipe 201, the design data 101, the SEM image 310, and the matching shift amount 314, and outputs the measurement values.

[0035] (FIG. 2: S209) The measurement success / failure determination unit 306 determines whether the measurement is successful or not based on the measurement recipe 201, the design data 101, the SEM image 310, and the measurement value, and outputs the success / failure determination result. If the measurement success / failure determination result is success, skip to S212. If the measurement success / failure determination result is failure, S210 is executed to perform manual assistance.

[0036] (FIG. 2: S210) The device user uses GUI 307 to input a matching shift amount 314 at which the pattern shapes of the design data 101 and the SEM image 310 match, based on the two. The matching shift amount may be input by directly inputting a numerical value for the matching shift amount, or by calculating a matching shift amount that results in a correct matching position while checking an image in which the design data 101 and the SEM image 310 are superimposed.

[0037] (FIG. 2: S211) The measurement unit 305 measures the dimensions of the device pattern based on the measurement recipe 201, the design data 101, the SEM image 310, and the matching shift amount 314, and outputs the measurement values.

[0038] (FIG. 2: S212) After completing measurement of the dimensions, etc. of the device pattern, the measurement device 300 outputs the measurement recipe 201, design data 101, SEM image 310, and matching shift amount 314 to the machine learning model learning device 400. The machine learning model learning device 400 stores the measurement recipe 201, design data 101, and SEM image 310 received from the measurement device 300 in a measurement recipe storage folder 4042, a design data storage folder 4043, and an SEM image storage folder 4044 of the temporary storage data 406, respectively. At this time, the machine learning model learning device 400 adds the file names of the received measurement recipe 201, design data 101, and SEM image 310, and the matching shift amount 314, as records to the collected data table 4049.

[0039] (FIG. 2: S213) The learning data analysis unit 401 analyzes the data stored in the temporary storage data 406 and creates a new learning data set. Details will be described later with reference to FIG. 5. S213 may be performed each time a pattern is measured, or may be performed after a certain amount of data has been collected.

[0040] (FIG. 2: S214) The machine learning model learning unit 402 learns the machine learning model based on the new learning dataset created in S213, and saves the learning log and the trained machine learning model 408 in the machine learning model storage unit 405. The trained machine learning model 408 is also stored in the machine learning model 312 of the measurement device 300. Details will be described later using FIG. 4. S214 may be performed every time a pattern is measured, or may be performed after a certain amount of data has been collected.

[0041] 3 is a flowchart illustrating the specific operations of S206 to S207. Each step in FIG. 3 will be described below. S206 to S207 may be executed immediately when pattern matching by machine learning fails, or may be executed at a later date after the semiconductor wafer is unloaded from the measuring device 300.

[0042] (FIG. 3: S300) The device user uses the GUI 403 to read the design data 101 and the SEM image 310 based on the temporary data table 4047, and inputs the correct matching shift amount at which the pattern shapes of the two images match. Then, a failure pattern training data table 4041 is created that includes the input correct matching shift amount, the design data 101, and the SEM image 310. At this time, all or part of the records in the base model training data table 4045 or the additional training data table 4048 may be added to the failure pattern training data table 4041. This allows the machine learning model to adapt to failure patterns while ensuring its general-purpose performance.

[0043] (FIG. 3: S301) The machine learning model training unit 402 trains the machine learning model based on the failure pattern training data table 4041 and the default training conditions, and outputs the trained machine learning model 408. Generally, the training conditions for the machine learning model include the number of training cycles per record in the data table (hereinafter referred to as the number of epochs), values ​​that can be set arbitrarily for the machine learning model, such as the learning rate (hereinafter referred to as hyperparameters), and the presence or absence of a model to be used as the basis for training. All or only some of the internal parameters of the machine learning model may be updated. The default training conditions are preferably optimal values ​​empirically obtained through prior experiments, but are not limited thereto. For example, the number of epochs may be set until a certain level of performance is achieved while monitoring the progress of the training. The model to be used as the basis for training is preferably a machine learning model trained using the base model training data table 4045 or the latest machine learning model stored in the machine learning model storage unit 405, but are not limited thereto. For example, a new machine learning model may be trained. The machine learning model 408 after training can also be used as a model to serve as the basis for future training.

[0044] (FIG. 3: S302) The machine learning model learning unit 402 performs inference on the learned machine learning model 408 using as input the design data 101 and the SEM image 310 for which the correct matching shift amount was taught in S300, and evaluates whether matching is successful. If matching is successful, S303 is performed. If matching is unsuccessful, S301 is performed again. At this time, records selected from the base model training data table 4045 or the additional training data table 4048 in the failure pattern training data table 4041 may be deleted. This reduces the versatility of the machine learning model, but allows it to be specialized for failure cases, which may result in successful matching. Because the learned machine learning model 408 can handle failure patterns, a prompt response to matching failures is possible.

[0045] (FIG. 3: S303) The machine learning model learning unit 402 stores the machine learning model 408 after learning and the learning log in the machine learning model storage unit 405.

[0046] (FIG. 3: S304) The machine learning model 408 after learning is stored in the machine learning model 312.

[0047] 4 is a flowchart illustrating the specific operation of S214. Each step in FIG. 4 will be described below.

[0048] (FIG. 4: S400) The machine learning model training unit 402 uses the additional training data table 4048 to train the machine learning model based on default training conditions. The default training conditions are preferably optimal values ​​empirically obtained through prior experiments, but are not limited to these. For example, the number of epochs may be set until a certain level of performance is achieved while monitoring the progress of the training. Furthermore, the model used as the basis for training is preferably a machine learning model that has been trained using the base model training data table 4045, or the latest machine learning model stored in the machine learning model storage unit 405 and obtained by referencing the learning log. However, this is not limited to these. For example, a new machine learning model may be trained. The method for selecting the model used as the basis for training is not limited to referencing the learning log; it is sufficient to select an optimal model used as the basis for training. For example, the latest machine learning model can be identified from the save date and time of the machine learning model, or from the file name of the machine learning model.

[0049] (FIG. 4: S401) The machine learning model learning unit 402 performs inference on the trained machine learning model 408 using data from the test data table 4046, for which true values ​​are known, as input, and evaluates its performance. The test data table 4046 may store data with various shape features in records in order to evaluate the general-purpose performance of the machine learning model, or may store other data, such as data selected by the device user that the user wants to achieve successful matching. Performance may be evaluated, for example, by having an accuracy rate equal to or greater than a threshold, or by other methods, such as having a sufficiently small difference between the inferred matching shift amount and the true matching shift amount, or by a combination of these methods. If sufficient performance is obtained, S403 is performed. If performance is insufficient, S402 is performed.

[0050] (Figure 4: S402) The machine learning model learning unit 402 may check the data contained in the additional learning data table 4048 and delete data that may have a negative impact on learning (e.g., data in which the template image does not exist in the field of view of the searched image due to field of view misalignment, data in which the device pattern shape formed on the wafer is significantly different from the design data due to deformation, data in which the image quality or other characteristics are significantly different from other data in the same data table, etc.) (adversely influencing data type). The learning conditions may also be adjusted. Methods for adjusting the learning conditions include changing the model used as the basis for learning, or increasing the number of epochs if learning is insufficient. Then, S400 is performed again. This step is optional and does not necessarily have to be performed.

[0051] (FIG. 4: S403) The machine learning model learning unit 402 stores the machine learning model 408 after learning and the learning log in the machine learning model storage unit 405.

[0052] (FIG. 4: S404) The machine learning model 408 after learning is stored in the machine learning model 312.

[0053] 5 is a flowchart illustrating the specific operation of S213. Each step in FIG. 5 will be described below.

[0054] (FIG. 5: S500) The learning data analysis unit 401 reads the design data, SEM images, and measurement recipes contained in the records from the collected data table 4049, the failure pattern learning data table 4041, the latest additional learning data table 4048, and the base model learning data table 4045. All four data tables may be used, or only some of them may be used. The learning data analysis unit 401 analyzes the read data and extracts features such as pattern shape features, SEM image features, and imaging conditions. Specific examples of each feature are as follows: Pattern shape features include the pattern type (line-space pattern, hole array pattern, etc.), pattern density, and constituent material; SEM image features include the presence or absence of transparency of the underlying pattern, the amount of deformation, and image quality; and imaging conditions include the acceleration voltage, detector type, imaging magnification, number of image quality integration frames, and image quality improvement processing. Specific methods for extracting pattern shape features include, for example, estimation based on information obtained from image edge and corner detection, image frequency analysis, etc. Alternatively, the recipe creator may include such information in the measurement recipe when creating the measurement recipe. As a specific method for extracting SEM image features, for example, the contours of the pattern portions in the SEM image and the design data are extracted, and by comparing the extracted contours, it is possible to estimate the presence or absence of transparency in the underlying pattern, the amount of pattern deformation, etc. Furthermore, the recipe creator may include such information in the measurement recipe when creating the measurement recipe. The imaging conditions may use the information described in the measurement recipe as is, or may be subjected to some preprocessing. The features to be extracted are not limited to these, and may also include the model of the measurement tool 300, the identification number of the same type of tool, the process name (photolithography process, etching process, etc.), etc., described in the measurement recipe by the recipe creator when creating the measurement recipe.

[0055] (FIG. 5: S501) The training data analysis unit 401 selects records from each data table so that there is no bias in the features extracted in S500, and creates a new additional training data table. One method of selecting records is, for example, clustering based on the features of the records, and selecting records from each class so that there is no bias in each class. A specific clustering method that can be applied is the k-means method, which places each record in a space with the same number of dimensions as the number of extracted features, and performs clustering on this space. However, the method is not limited to this, and any method that can cluster similar records into one class based on the extracted features may be used.

[0056] (FIG. 5: S502) The device user checks the matching shift amount in the new additional learning data table using the GUI 403, and if an error is found, manually corrects the incorrect matching shift amount. The method for correcting the matching shift amount is not limited to the method using the GUI 403, but may be any method that can correct matching errors. For example, pattern matching may be performed between the design data 101 and the SEM image using a pattern matching algorithm based on multiple rules, and the most similar matching shift amount may be determined as the correct matching shift amount. Alternatively, as target data for which the matching shift amount is to be confirmed, only data that is likely to have an incorrect matching shift amount may be selected in advance, and the matching shift amount may be confirmed for that data. One selection method is, for example, to perform pattern matching between the design data 101 and the SEM image using a pattern matching algorithm based on multiple rules, and then select data for which no similar matching shift amount was obtained using each algorithm. S502 is not necessarily required. This is because the collected data is data for which measurements were performed correctly, and therefore the matching is likely to be correct. This step is intended to ensure that correct data is provided.

[0057] (FIG. 5: S503) The learning data analysis unit 401 stores each piece of data (measurement recipe 201, design data 101, SEM image 310) in the temporary storage data 406 included in the new additional learning data table 4048 in the measurement recipe storage folder 4042, design data storage folder 4043, and SEM image storage folder 4044. The new additional learning data table 4048 is given a unique name and saved in the learning data set 407.

[0058] 6 shows an example of data held in the temporary storage data 406 and the learning data set 407. As shown in Fig. 6, the temporary storage data 406 and the learning data set 407 include a plurality of data tables for managing each piece of data, a measurement recipe storage folder 4042, a design data storage folder 4043, and an SEM image storage folder 4044. The temporary storage data 406 and the learning data set 407 hold three folders, each of which separately stores the measurement recipe, design data, and SEM images, but all data files may be stored at the same level as the data tables, all data files may be stored in one folder, or the data files may be stored separately in multiple folders.

[0059] The training dataset 407 holds a base model training data table 4045 for training a base machine learning model in the initial state, a test data table 4046 for evaluating the performance of the machine learning model, and each data file described in these two data tables. The base model refers to a machine learning model trained using a frequently measured or general pattern.

[0060] As shown in FIG. 6 , each data table holds a data ID, measurement recipe, design data, SEM image, and matching shift amount in one row of a record. By holding multiple such data tables, multiple data sets can be managed as data tables. The method for managing data sets is not limited to management using data tables, and any method can be used as long as it can manage combinations of measurement recipes, design data, SEM images, matching shift amounts, etc. For example, management may be performed using folder structures, file names, etc. Furthermore, each data table holds information on the data ID, measurement recipe, design data, SEM image, and matching shift amount in its records, but the information held may be more or less than this, as long as it holds information necessary for data management. For example, it is not necessary to hold a data ID, and other information such as the registration date and time may be held.

[0061] FIG. 7 shows an example of learning log data. As shown in FIG. 7, the learning log can be recorded in the form of a data table, with each record containing the update date, machine learning model name, data table used, and learning conditions. This facilitates management of trained machine learning models and also makes it possible to retrieve past machine learning models if necessary. In FIG. 7, the learning log stores information such as the update date, machine learning model name, data table used, and learning conditions in the record, but the stored information may be more or less than this, as long as it is possible to manage the machine learning model after training. For example, other information such as a machine learning model ID may also be stored.

[0062] FIG. 8 shows an example configuration of the measurement apparatus 300, including the SEM 600 and the GUI 307. The SEM 600 can be operated as the imaging unit 301. The control unit 614 can be operated as the rule-based pattern matching unit 302, the machine learning pattern matching unit 303, the matching success / failure determination unit 304, the measurement unit 305, and the measurement success / failure determination unit 306. The image memory 615 can be operated as the measurement information storage unit 308. The SEM 600 measures, for example, the pattern dimensions of a semiconductor device formed on a semiconductor wafer. The arithmetic processing unit or computer system in the measurement apparatus 300 can be configured as, for example, the control unit 614. The control unit 614 includes a calculation unit (for example, a CPU / GPU 616) and a storage device (for example, a memory including the image memory 615). The storage device can store information, such as a program related to the pattern matching process. The CPU / GPU 616 executes this program to perform the pattern matching process shown in FIG. 2. In other words, the control unit 614 functions as each pattern matching unit. In other words, this program causes the computer system to function as a processing unit included in the pattern matching unit and to execute the pattern matching process shown in FIG.

[0063] The SEM 600 generates an electron beam from an electron gun 601. A deflector 604 and an objective lens 605 are controlled so that the electron beam is focused and irradiated at a desired position on a semiconductor wafer 603, which is a sample placed on a stage 602. Secondary electrons are emitted from the semiconductor wafer 603 irradiated with the electron beam and detected by a secondary electron detector 606. The detected secondary electrons are converted into a digital signal by an A / D converter 607. The image represented by the digital signal is stored in an image memory 615 in a control unit 614. This image is used, for example, as an SEM image 310, and pattern matching processing is performed based on this image by the control unit 614 or a CPU / GPU 616. The settings required for these processing and the display of processing results can be performed on a GUI 307. For alignment using an optical camera with a lower magnification than that of an SEM, an optical camera 611 may also be used. The signal obtained by imaging the semiconductor wafer 603 with the optical camera 611 is converted into a digital signal by the A / D converter 612 (if the signal from the optical camera 611 is a digital signal, the A / D converter 612 is not necessary), and the image represented by the digital signal is stored in an image memory 615 in the control unit 614, and the CPU / GPU 616 performs image processing according to the purpose.

[0064] The SEM 600 may be equipped with a backscattered electron detector 608. If the backscattered electron detector 608 is equipped, backscattered electrons emitted from the semiconductor wafer 603 are detected by the backscattered electron detector 608, and the detected backscattered electrons are converted into digital signals by an A / D converter 609 or 610. An image represented by the digital signal is stored in an image memory 615 in a control unit 614, and a CPU / GPU 616 performs image processing according to the purpose. The control unit 614 may control the stage 602 via a stage controller 630, and may control the objective lens 605 and the like via a deflection control unit 631.

[0065] Summary of First Embodiment The pattern dimension measurement system according to the first embodiment uses the successful results of rule-based matching or machine learning matching to train the learner (S214). These successful results can be obtained each time pattern matching and measurement based on the matching are performed, and are the results of successful measurements, so they are suitable as learning data. Therefore, the workload for collecting learning data and providing correct answers can be reduced.

[0066] Second Embodiment A second embodiment of the present disclosure describes a configuration example for selectively using a learning device suitable for a matching target having specific characteristics. Similar to the first embodiment, the second embodiment is configured using the pattern dimension measurement system of Fig. 1. The basic operation is the same as the first embodiment, and the following mainly describes the differences in operation from the first embodiment.

[0067] As in the first embodiment, the design data storage unit 100 holds design data 101 indicating the pattern structure of a semiconductor device created by an apparatus user.

[0068] In addition to the interface, the recipe generation device 200 also includes a pattern feature analysis unit. The pattern feature analysis unit analyzes the design data 101 and measurement recipe 201 obtained from the design data storage unit 100 to extract pattern shape features and imaging conditions. Specific examples of each feature are as follows: Pattern shape features include the pattern type (e.g., line-space pattern, hole array pattern, etc.), pattern density, and constituent material. Imaging conditions include the acceleration voltage, detector type, imaging magnification, number of image quality integration frames, and image quality improvement processing. Specific methods for extracting pattern shape features include, for example, estimation based on information obtained from image edge and corner detection, image frequency analysis, etc. Alternatively, the device user may operate the interface to input pattern type information, etc. The imaging conditions may use the information written in the measurement recipe as is, or may be subjected to some preprocessing. The extracted features are not limited to these, and may also include the model of the measurement device, the identification number of the same type of device, and process name (e.g., photolithography process, etching process, etc.) written in the measurement recipe by the recipe creator when creating the measurement recipe.

[0069] The recipe creation device 200 uses a unique feature group ID for each feature group (feature groups are described in detail below) created in advance to select a machine learning model trained using data from the feature group into which the extracted features are classified. The method for selecting a machine learning model is not limited to this method, and any machine learning model trained using data similar to the extracted features can be selected. For example, the device user may operate an interface to select the machine learning model optimal for the design data.

[0070] Based on the measurement point information, design data 101, and selected machine learning model input from the interface, the recipe creation device 200 outputs a measurement recipe 201 including the imaging conditions for the measurement device 300, an algorithm used in rule-based pattern matching, a machine learning model used in machine learning pattern matching, measurement point information, etc. The selected machine learning model has previously undergone specialized learning for specific features and has high matching performance for similar data, so it is possible to reduce cases of matching failure in the machine learning pattern matching unit 303.

[0071] The measurement apparatus 300 includes an imaging unit 301 that images a device pattern formed on a semiconductor wafer, a rule-based pattern matching unit 302, a machine learning pattern matching unit 303, a matching success / failure determination unit 304, a measurement unit 305, a measurement success / failure determination unit 306, a GUI 307, and a measurement information storage unit 308. The machine learning pattern matching unit 303 extracts a machine learning model to be used for pattern matching from a machine learning model 312 based on the measurement recipe 201. Alternatively, the machine learning pattern matching unit 303 receives a machine learning model to be used for pattern matching from a machine learning model storage unit 405 of the machine learning model learning device 400 based on the measurement recipe 201. The machine learning pattern matching unit 303 performs pattern matching using the acquired machine learning model.

[0072] The machine learning model learning device 400 includes a learning data analysis unit 401, a machine learning model learning unit 402, a GUI 403, a learning data storage unit 404, and a machine learning model storage unit 405. A learning dataset 407 in the learning data storage unit 404 has additional learning datasets created in advance for each feature group. A method for creating a dataset for each feature group will be described later. The machine learning model storage unit 405 holds a machine learning model for each feature group. A method for learning and managing the machine learning model for each feature group will be described later.

[0073] The machine learning model learning device 400 stores the measurement recipe 201, design data 101, and SEM image 310 received from the measuring device 300 in a measurement recipe save folder 4042, a design data save folder 4043, and an SEM image save folder 4044 of the temporary storage data 406, respectively. At this time, the machine learning model learning device 400 creates a collected data table 4049 in which an arbitrary data ID, the file names of the measurement recipe 201, the design data 101, and the SEM image 310, and the numerical values ​​of the matching shift amount 314 and the feature group ID are compiled into one row of a record, and stores the created data table in the temporary storage data 406. The collected data table 4049 is created to manage the collected data, but the management method for the collected data is not limited to the management method using the collected data table 4049. Any method can be used as long as it can manage combinations of measurement recipes, design data, SEM images, matching shift amounts, feature group IDs, etc., and the data may be managed by folder structure or file name, for example.

[0074] As with the first embodiment, the operation of the second embodiment can be explained using the flowchart shown in Fig. 2. Here, we will mainly describe the steps in Fig. 2 in the second embodiment that are different from those in the first embodiment. Steps that are not described are the same as those in the first embodiment.

[0075] (FIG. 2: S200) Based on measurement point information of a device pattern input by the device user from the interface, design data 101 obtained from the design data storage unit 100, and feature groups obtained in the pattern feature analysis unit, the recipe generation device 200 generates a measurement recipe 201 that includes imaging conditions for the measurement device 300 (which may include imaging position information, etc.), an algorithm used in rule-based pattern matching, a machine learning model used in machine learning pattern matching, measurement point information, etc.

[0076] (FIG. 2: S204) The machine learning pattern matching unit 303 extracts the machine learning model to be used for pattern matching from the machine learning model 312 based on the measurement recipe 201. Alternatively, it receives the model from the machine learning model storage unit 405 of the machine learning model learning device 400. Thereafter, the machine learning pattern matching unit 303 performs pattern matching using the acquired machine learning model.

[0077] (FIG. 2: S206) In the same manner as in embodiment 1, the device user operates the GUI 403 to input the correct matching shift amount. Thereafter, the machine learning model training device 400 creates a failure pattern training data table 4041 that includes the input correct matching shift amount, the design data 101, and the SEM image 310. When creating the failure pattern training data table 4041, all or part of the data set used to train the machine learning model acquired in S204 may be added to the failure pattern training data table 4041. This allows the machine learning model to respond to failure patterns while ensuring its performance for specific patterns.

[0078] (FIG. 2: S207) The machine learning model learning unit 402 learns the machine learning model in the same manner as in embodiment 1 and outputs the learned machine learning model 408. The model used for learning is preferably the machine learning model acquired in S204, but is not limited to this. For example, a new machine learning model may be learned. The learned machine learning model 408 is stored in the machine learning model 312 of the measurement device 300.

[0079] (FIG. 2: S212) After completing measurement of the dimensions, etc. of the device pattern, the measuring device 300 outputs the measurement recipe 201, design data 101, SEM image 310, and matching shift amount 314 to the machine learning model learning device 400. The machine learning model learning device 400 stores the measurement recipe 201, design data 101, and SEM image 310 received from the measuring device 300 in a measurement recipe storage folder 4042, a design data storage folder 4043, and an SEM image storage folder 4044 of the temporary storage data 406, respectively. At this time, the machine learning model learning device 400 adds, as records to the collected data table 4049, the file names of the received measurement recipe 201, design data 101, and SEM image 310, as well as the numerical values ​​of the matching shift amount 314 and pattern feature group ID.

[0080] (FIG. 2: S213) The training data analysis unit 401 extracts feature group ID information from the failure pattern training data table 4041 and the collected data table 4049. Records from each data table are added to the additional training data table 4048 for each feature group so that the feature group ID information matches. This makes it possible to collect and accumulate training data for each feature group. Therefore, it is possible to train a machine learning model with higher performance for specific features.

[0081] (Figure 2: S214) The machine learning model learning unit 402 uses the additional learning data table 4048 for each feature group created in S213 to learn a machine learning model adapted to each feature group using the same learning method as in embodiment 1, and saves each learning log and each trained machine learning model, which is given a unique name, in the machine learning model storage unit 405. In addition to the information shown in embodiment 1, the learning log may be assigned a feature group ID. This makes it possible to link each feature group with each machine learning model. The trained machine learning model is stored in the machine learning model 312 of the measurement device 300.

[0082] A clustering method, for example, can be considered as a method for creating feature groups in the second embodiment. A method for creating feature groups using a clustering method will be described below.

[0083] The design data and measurement recipes contained in the records are read from the collected data table 4049 and the base model training data table 4045, which have been obtained in advance. Both data tables may be used, or only one may be used. The read data is analyzed, and features such as pattern shape features and imaging conditions are extracted using a method similar to that of embodiment 1 ( FIG. 5 ). Each record is placed in a space with a number of dimensions equal to the number of extracted features, and clustering is performed on this space. A specific clustering method that can be used is the k-means method, but is not limited to this method. Any method that can cluster similar records into one class based on the extracted features is sufficient. An arbitrary feature group ID is assigned to each class obtained as described above, and the arbitrary feature group ID assigned to each record is added. Then, data tables are created, separated by feature group ID.

[0084] Although the method for creating feature groups by clustering has been described above, the method for creating feature groups is not limited to this. For example, feature groups can also be created by a classification task based on feature group IDs assigned by the device user. Specifically, the device user assigns feature group IDs to some or all of the records obtained from the previously obtained collected data table 4049 and the base model training data table 4045 based on pattern shape features, SEM image features, imaging conditions, and the device user's requirements for the machine learning model. The device user's requirements for the machine learning model may include, for example, "I want to prepare one machine learning model for each pattern type," "I want to separate machine learning models based on whether or not the lower layer in the SEM image is transparent," or "I want to prepare one machine learning model for each measurement device." Features such as pattern shape features and imaging conditions are then extracted from each record. Each record is placed in a space with a number of dimensions equal to the number of extracted features, and classification is performed in this space using the feature group IDs assigned by the device user as true values. A specific classification method may be a support vector machine, which is a common method for classification tasks, but the method is not limited to this and any method may be used as long as it can estimate a feature group ID from features such as pattern shape features and imaging conditions. The feature group classification model obtained in this manner can estimate a feature group ID that reflects the requirements of the equipment user based on the input design data and measurement recipe information. The feature group ID estimated by the feature group classification model is added to each record. After that, a data table is created, separated by feature group ID.

[0085] The records in the data table for each feature group obtained by the above two methods have similar features, so by using them in additional learning, it is possible to obtain a machine learning model specialized for a specific feature. It is preferable that the machine learning model for each feature group is trained using the machine learning model created by training the base model training data table 4045 as the base model for learning, but this is not limited to this. For example, other methods may be used, such as training a new machine learning model without using the base model for learning. The machine learning model trained for each feature group is given a unique name and saved together with the learning log.

[0086] For machine learning models specialized for each feature group, the corresponding feature group, model version, etc. can be determined from the learning log, etc., and this information can be used to select a machine learning model to use for pattern matching by machine learning, or to select a model to use as the basis for learning. The method of determining the machine learning model is not limited to this method, and it is sufficient if the optimal machine learning model can be selected. For example, it is also possible to select the optimal machine learning model based on the file name or save date and time of the machine learning model.

[0087] The pattern feature analysis unit of the recipe creation device 200 selects the optimal machine learning model for the input data using the class information of each feature group or the feature group classification model obtained by the above method. The specific procedure is as follows.

[0088] (1) Pattern shape features, imaging conditions, etc. are extracted from the design data 101 and the measurement recipe 201. (2) A feature group ID is estimated for the features extracted in (1) using class information or a feature group classification model for each feature group. (3) A machine learning model trained using the feature group data estimated in (2) is written in the measurement recipe 201 as a machine learning model to be used in pattern matching.

[0089] Third Embodiment Fig. 9 is a flowchart illustrating the operation of a pattern dimension measurement system according to a third embodiment of the present disclosure. In the first embodiment, rule-based pattern matching is performed first, and if that fails, machine learning pattern matching is then performed. Alternatively, the algorithm performed first may be either rule-based or machine learning, as appropriate. Specifically, steps S901 and S902, which will be described below, are performed between steps S201 and S202. The rest of the process is the same as in the first and second embodiments.

[0090] (FIG. 9: S901) The recipe generation device 200 analyzes the measurement recipe 201, and from the information on past measurement recipes 201 and the corresponding matching success / failure results, refers to the matching success / failure results with information on similar measurement recipes 201, and determines whether rule-based pattern matching or machine learning pattern matching is more suitable for performing the pattern matching process. Past matching that has been successful (or has a higher success rate) is considered more suitable. If there is no measurement recipe 201 with similar information or if there are no past matching success / failure results, rule-based pattern matching is selected. In other words, at the time the measurement recipe 201 is generated, the matching algorithm corresponding to that measurement recipe 201 is also determined. The specific operation of the recipe generation device 200 will be described later using FIG. 10.

[0091] (FIG. 9: S902) The measurement device 300 switches the pattern matching algorithm based on the result of S901. If rule-based pattern matching is specified, S202 is performed. If a machine learning pattern matching device is specified, skip to S204. This step allows the optimal pattern matching method to be selected before executing the pattern matching process, thereby reducing unnecessary processing and improving the matching success rate.

[0092] FIG. 10 is a block diagram of the recipe generation device 200. The recipe generation device 200 analyzes a measurement recipe 201 and selects an optimal pattern matching method. The measurement recipe 201 includes a pattern type, imaging conditions, an apparatus number, etc. The pattern type includes pattern shape information, pattern density information, the constituent material of the semiconductor pattern, whether or not the underlying semiconductor pattern is transparent, the amount of deformation that has occurred, etc. The imaging conditions include the acceleration voltage of the electron microscope used for imaging, the type of detector, the imaging magnification, the number of image frames for image accumulation performed to reduce noise, and whether or not prior image quality improvement processing using AI technology or image processing technology has been performed. The apparatus number includes the model of the apparatus, the identification number of apparatuses of the same model, etc.

[0093] The recipe generation unit 210 creates a measurement recipe 201. The recipe analysis unit 220 analyzes all or part of the information in the measurement recipe 201 and extracts its features. The recipe analysis unit 220 references, from the matching result storage unit 230, the matching success / failure determination result linked to the measurement recipe 201 having the feature most similar to the extracted feature, and selects a method that is likely to be successful in the pattern matching process. The matching result storage unit 230 is a database that has information such as the pattern type, imaging conditions, and apparatus number of the measurement recipe 201, as well as keys for the matching success / failure result at that time, and stores the information of each key in the form of characters, numbers, etc.

[0094] An example of a recipe analysis method is a classification model such as a random forest based on machine learning. In this learning, the input is each piece of data corresponding to the measurement recipe 201 in the matching result storage unit 230, and the teacher label is the matching success / failure determination result corresponding to the measurement recipe 201. This enables the machine learning model to estimate the optimal matching method corresponding to the measurement recipe. When performing estimation using the machine learning model, the input is each piece of information in the measurement recipe 201, and the output is the optimal pattern matching method. Another method is to refer to the column with the most items that most match the measurement recipe 201 from the database in the matching result storage unit 230 and refer to the matching success / failure result at that time. These methods make it possible to select the optimal pattern matching method based on the results of pattern matching processes performed in the past. The recipe analysis method is not limited to these methods, and any method that allows the optimal pattern matching method to be selected is sufficient.

[0095] <Summary of Third Embodiment> The pattern dimension measurement system according to the third embodiment performs either pattern matching using a rule-based algorithm or pattern matching using machine learning, and if the rule-based matching fails, performs machine learning matching, and if the machine learning matching fails, performs re-learning. Re-learning of the learner needs to be performed only when the machine learning matching fails, and can also be performed automatically (see S206). Therefore, in the third embodiment as well, it is possible to reduce the workload for learning.

[0096] <Regarding Modifications of the Present Disclosure> The present disclosure is not limited to the above-described embodiments and includes various modifications. For example, the above-described embodiments have been described in detail to clearly explain the present disclosure, and are not necessarily limited to those including all of the described configurations. Furthermore, it is possible to replace part of the configuration of one embodiment with the configuration of another embodiment, or to add the configuration of another embodiment to the configuration of one embodiment. Furthermore, it is possible to add, delete, or replace part of the configuration of each embodiment with other configurations.

[0097] In the above embodiment, a scanning electron microscope has been described as an example of the measuring device 300, but the present disclosure is not limited to this, and can be applied to any measuring device or inspection device that acquires images and performs pattern matching.

[0098] In the above embodiments, the design data storage unit 100, the recipe creation device 200, the measuring device 300, and the machine learning model learning device 400 are shown as separate devices, but they may be implemented as one or more devices (pattern matching devices configured by one or more computer systems) that can perform similar processing. For example, a system may be configured by the recipe creation device 200 and one device that has the functions of the design data storage unit 100, the measuring device 300, and the machine learning model learning device 400.

[0099] In the second embodiment, the following features can be used to perform clustering: (a) image features: brightness distribution of an image, frequency components of an image, (b) shape pattern features: type of shape pattern, plane density of shape pattern, (c) imaging condition features: optical conditions, electrical conditions of an imaging device that acquires an image.

[0100] In the third embodiment, after steps S200 to S207 are performed, the success or failure of the measurement is determined (S209) as in the first embodiment, and learning is performed using the successful results (S210 to S214). By performing steps S209 and onward, learning data can be automatically collected as in the first embodiment, but these steps do not necessarily have to be performed. That is, since the third embodiment implements either machine learning or rule-based, whichever is more appropriate, the objective can be achieved as long as this can be realized. Of course, performing steps S209 and onward is more effective.

[0101] In the above embodiments, the functional units of each device (e.g., recipe generation unit 210, recipe analysis unit 220, rule-based pattern matching unit 302, machine learning pattern matching unit 303, matching success / failure determination unit 304, measurement unit 305, measurement success / failure determination unit 306, GUI 307, learning data analysis unit 401, machine learning model learning unit 402, GUI 403, etc.) can be configured by hardware such as a circuit device that implements these functions, or by software that implements these functions being executed by a computing device such as a CPU (Central Processing Unit).

[0102] In the above embodiments, the rule-based matching algorithm can be said to be configured to identify the correspondence between image features and the positional shift amounts between two images corresponding to the image features using the internal logic of the algorithm. On the other hand, it should be noted that the machine learning matching algorithm differs from the rule-based algorithm in that, in addition to the internal logic that implements the matching process, there is a learning model that describes the results of learning the correspondence between image features and positional shift amounts.

[0103] In the above embodiment, S201 is performed after S200, but the order in which the steps are performed is not limited to this and may be changed within a range that does not contradict the method according to the present disclosure. For example, S902 may be performed after S200, or S201 may be performed as shown in FIG.

[0104] 100: Design data storage unit 200: Recipe generation device 300: Measuring device 400: Machine learning model learning device

Claims

1. A pattern matching device that performs pattern matching between images, comprising: a computer system that performs the pattern matching; the computer system having a learning device that outputs the results of the pattern matching based on machine learning; the computer system performs the pattern matching using either a first algorithm that does not use the learning device or a second algorithm that uses the learning device; the computer system performs the pattern matching using the second algorithm if the pattern matching using the first algorithm fails; and the computer system re-trains the learning device if the pattern matching using the second algorithm fails.

2. The pattern matching device according to claim 1, characterized in that, if the pattern matching using the second algorithm fails, the computer system generates training data for the learning device to perform machine learning and causes the learning device to perform re-learning using the training data, and the computer system re-performs the pattern matching using the learning device that has performed the re-learning.

3. The pattern matching device according to claim 2, characterized in that the computer system is configured to supply the learning device with information including the amount of positional shift between the two images on which the pattern matching is performed as the learning data, and the computer system receives a designated input specifying the amount of positional shift via an interface provided in the computer system, and generates the learning data based on the designated input.

4. The pattern matching device according to claim 2, characterized in that the computer system is configured to supply the learning device with information including the amount of positional shift between the two images on which the pattern matching is performed as the learning data, the computer system obtains the amount of positional shift from each of the first algorithms by implementing a plurality of the first algorithms, each of which implements a different matching procedure, and the computer system generates the learning data using the amount of positional shift obtained from each of the first algorithms that provides the best match between the two images.

5. The pattern matching device according to claim 1, characterized in that the first algorithm implements a rule-based algorithm configured to identify, by the internal logic of the first algorithm, a correspondence between image features and positional shift amounts between two images corresponding to the image features.

6. The pattern matching device according to claim 1, characterized in that, if the pattern matching by the first algorithm or the learning device is successful, the computer system supplies the result to the learning device as learning data for the learning device to perform re-learning.

7. The pattern matching device according to claim 6, characterized in that the computer system evaluates the inference performance of the learning device that has undergone the re-learning, and if the inference performance does not meet a reference value, the computer system deletes from the learning data used for the re-learning any portion that belongs to an adverse data type that has been predefined as a type that may degrade the inference performance, and the computer system performs the re-learning again using the learning data from which the adverse data type has been deleted.

8. The pattern matching device according to claim 6, characterized in that the computer system clusters the training data used to perform the re-learning according to its features, the computer system corrects the arrangement of the data belonging to each class so as to suppress bias in the data for each class obtained by the clustering, and the computer system performs the re-learning using the training data with the bias suppressed.

9. The pattern matching device according to claim 1, characterized in that the image is an observation image of a sample, and the computer system receives the results of measurement or inspection performed on the sample if the pattern matching is successful, and if the results indicate that the measurement or inspection has failed, the computer system receives a specified input via an interface provided in the computer system that specifies the amount of positional shift between the images, and causes the learning device to perform re-learning based on the specified input.

10. The pattern matching device according to claim 1, characterized in that the computer system comprises a plurality of learning devices that have performed machine learning using mutually different learning data, the computer system groups the learning data used by each learning device when performing machine learning according to features of the learning data, and the computer system performs the pattern matching using the learning devices that have performed machine learning using the learning data that corresponds to at least one of features of the image, features of a shape pattern contained in the image, or features of the image capturing conditions.

11. A pattern matching device as described in claim 10, characterized in that the image feature includes at least one of the luminance distribution of the image and the frequency components of the image, the shape pattern feature includes at least one of the type of the shape pattern and the plane density of the shape pattern, and the imaging condition feature includes at least one of the optical conditions and electrical conditions of an imaging device that acquires the image.

12. A pattern matching device that performs pattern matching between images, comprising: a computer system that performs the pattern matching; the computer system having a learning device that outputs the results of the pattern matching based on machine learning; the computer system is configured to perform the pattern matching by executing one of two or more algorithms that implement the pattern matching; the computer system selects one of the algorithms based on the matching success rate when each of the algorithms was previously executed; and the computer system performs the pattern matching by executing the selected algorithm.

13. The pattern matching device according to claim 12, characterized in that the computer system performs the pattern matching using either a first algorithm that does not use the learning device or a second algorithm that uses the learning device as the two or more algorithms, the computer system performs the pattern matching using the second algorithm if the pattern matching using the first algorithm fails, and the computer system re-learns the learning device if the pattern matching using the second algorithm fails.

14. A pattern matching device that performs pattern matching between images, comprising: a computer system that performs the pattern matching; the computer system having a learning device that outputs the result of the pattern matching based on machine learning; the computer system is configured to perform the pattern matching by executing one of two or more algorithms that implement the pattern matching; the computer system is configured to perform the pattern matching in accordance with conditions described in a measurement recipe; when a new measurement recipe is acquired, the computer system extracts features of the new measurement recipe; the computer system selects one of the algorithms by identifying the measurement recipe that is most similar to the extracted features from among measurement recipes that have been used when each of the algorithms has been successfully executed in the past; and the computer system performs the pattern matching by executing the selected algorithm.

15. The pattern matching device according to claim 14, characterized in that the computer system performs the pattern matching using either a first algorithm that does not use the learning device or a second algorithm that uses the learning device as the two or more algorithms, the computer system performs the pattern matching using the second algorithm if the pattern matching using the first algorithm fails, and the computer system re-learns the learning device if the pattern matching using the second algorithm fails.

Images