Parameter optimization system, parameter optimization method, and storage medium

Abstract

In the present invention, if a converted image generated by using a learning model that generates a converted image is utilized for measurement, a parameter of the learning model is optimized so that dimensional metrics can be improved. For this purpose, the present disclosure proposes a parameter optimization system for optimizing a parameter for generating converted images on the basis of input images, said system comprising a storage device for retaining a parameter optimization program, and a processor for reading in the parameter optimization program from the storage device and executing same. The processor executes: processing for calculating a conversion error between the converted images and training images inputted during learning; processing for calculating dimensional metrics of a pattern included in the training images on the basis of the converted images and the training images; processing for calculating a composite error on the basis of the conversion error and the dimensional metrics; training of a machine learning model for generating converted images so as to optimize the composite error; and processing for optimizing a parameter of the machine learning model (see fig.

Description

[Technical Field]

[0001] This disclosure relates to a parameter optimization system, a parameter optimization method, and a memory medium carrying the parameter optimization program for optimizing the parameters of a learning model that performs image transformation. [Previous Technology]

[0002] In recent years, attempts have been made to effectively perform image measurement using AI (Artificial Intelligence) technology. For example, Patent Document 1 discloses a technique for generating high-quality images using a learning model that generates transformed images based on input images generated by an image generation tool. In particular, this learning model applies different learning methods to specific areas, such as the edges of objects being measured or inspected, within the transformed image to improve the generation quality of those areas. [Prior Art Documents] [Patent Documents]

[0003] [Patent Document 1] International Publication No. 2021 / 140620 [Summary of the Invention]

[0004] [The problem the invention aims to solve]

[0005] However, in Patent Document 1, when learning a model to output a transformed image that becomes the measurement object, only the transformation error between the transformed image and the teacher image input during learning is used. Therefore, the size index is not utilized. Here, the size index refers to an index value that includes the measurement accuracy derived from the size value and the yield calculated from the measurement accuracy of the transformed image generated from the input image of the image generation tool. Therefore, even if the technology disclosed in Patent Document 1 is used, there are cases where it is impossible to generate a transformed image with sufficiently improved size index.

[0006] On the other hand, in order to perform deep learning using neural networks and the like, it is necessary to differentiate the error used for learning. Therefore, it is impossible to use size indicators, such as size values, which are not differentiable, as errors. As a result, it is impossible to achieve the target value of the size indicator expected by the user, and it is difficult to generate a learning model that can perform image transformations suitable for practical applications.

[0007] In view of this situation, this disclosure proposes a technique for optimizing the parameters of the learning model in a way that improves dimensional indices when using transformed images generated by a learning model that generates transformed images for measurement. [Means of the Problem]

[0008] To address the aforementioned issues, this disclosure proposes a parameter optimization system that optimizes the parameters required to generate a transformed image based on an input image. The system comprises: a memory device for storing the parameter optimization program; a processor for reading and executing the parameter optimization program from the memory device; the processor performing: processing to calculate the transformation error between the transformed image and the teacher image input during learning; processing to calculate the size index of a pattern contained in the teacher image based on the transformed image and the teacher image; processing to calculate a composite error based on the transformation error and the aforementioned size index; and processing to optimize the learning of the machine learning model for generating the transformed image and the parameter optimization of the machine learning model in a manner that optimizes the composite error.

[0009] Other features related to this disclosure are understood from the description and drawings in this specification. Furthermore, the form of this disclosure is achieved by the combination of elements and various elements, as well as the detailed description thereafter and the appended claims. The description in this specification is merely typical illustration, and the claims or applicable examples disclosed herein are not intended to be limiting in any sense. [Effects of the Invention]

[0010] According to the technique disclosed herein, it is possible to generate a learning model that best suits the size parameters of the transformed image. This makes it easy to generate a transformed image that meets the size parameters desired by the user.

Implementation Method

[0012] Hereinafter, embodiments of the present disclosure will be described with reference to the drawings. In the drawings, functionally identical elements may be represented by the same number. Furthermore, although the drawings show specific embodiments in accordance with the principles of the present disclosure, they are for the purpose of understanding the present disclosure and are not intended to limit or interpret the present disclosure.

[0013] In this embodiment, it is necessary to understand that a full and detailed description is provided to enable those skilled in the art to implement this disclosure, but other embodiments / forms are also possible, with changes in configuration / structure or substitution of various elements without departing from the technical concept and spirit of this disclosure. Therefore, the following description should not be limited to this interpretation.

[0014] Furthermore, in this embodiment, the measurement of the circuit pattern size is described with regard to the measurement of the circuit pattern of a semiconductor image as the object. However, the technology disclosed herein is not limited to the measurement of semiconductor circuit patterns, but can be widely and generally applied to image measurement.

[0015] (1) First embodiment The first embodiment is a technology that improves productivity by performing image transformation on an input image to generate an image with higher image quality than the input image.

[0016] For example, in order to perform size measurement using a scanning electron microscope, signals obtained from multiple scans of the same field of view are averaged to generate a high-quality image with a high signal-to-noise ratio (S / N ratio) for measurement. However, performing multiple scans requires a long imaging time, resulting in reduced productivity. In this embodiment, a low-quality image generated from fewer scans than that used to generate a high-quality image (an image with quality sufficient for size measurement with predetermined accuracy) is transformed to generate an image of comparable high quality. This eliminates the need for generating high-quality images using low-speed imaging, thereby increasing productivity.

[0017] <Example of the construction of the parameter optimization system> Figure 1 is a diagram showing a schematic construction example of the parameter optimization system 10 in the first embodiment. The parameter optimization system 10 includes, for example, a calculation device 100 with calculation parameters, an input device 101 used by the user (operator) to input the target value of the size index described later, an output device 102 that outputs (displays) the optimization result, and a storage device 201 that holds a high-quality image (e.g., a SEM image) with a predetermined accuracy for size measurement and a low-quality image (e.g., a SEM image: which may be measurable) with a lower quality than the high-quality image.

[0018] As a means of generating input images (high-quality images and low-quality images stored in storage device 201) for optimizing the parameters of the learning model, a scanning electron microscope or the like can be used as an image generation tool. Furthermore, the input device 101 can be configured as a keyboard or mouse, etc. Also, the output device 102 can be configured as a display, etc. The input device 101 receives target values ​​of various size parameters input by the user (refer to FIG. 4), and outputs the optimization result to the output device 102.

[0019] Although the computing device 100 is not shown in FIG1, it can be a computer system consisting of one or more CPUs (Central Processing Units) or GPUs (Graphics Processing Units), RAMs (Random Access Memory) or ROMs (Read Only Memory), which hold various programs or parameters and other memory devices.

[0020] The computing device 100 uses a processor (CPU, etc.) to read the parameter optimization program stored in memory, expand the program in the processor's internal memory, and realize the parameter optimization function.

[0021] As shown in FIG1, the calculation device 100, as the above-mentioned parameter optimization function, includes, for example, an image transformation unit 1001, a transformation error calculation unit 1002, a size index calculation unit 1003, a composite error calculation unit 1004, a transformation parameter update unit 1005, and a learning parameter update unit 1006.

[0022] The image transformation unit 1001 takes an image generated using a scanning electron microscope or an image (low-resolution image) stored in a storage device (memory medium) 201 as input image, and generates a transformed image based on the input image. The image transformation unit 1001 includes a learning model that adjusts transformation parameters (weights and biases) and hyperparameters, and performs image transformation using the learning model. The learning model can be, for example, a neural network, a convolutional neural network, or a neural network with other structures, which are also machine learning models.

[0023] The transformation parameter update unit 1005 adjusts the transformation parameters using hyperparameters obtained from the learning parameter update unit 1006, and supplies the adjusted transformation parameters to the image transformation unit 1001. Hyperparameters are parameters that determine the learning performance of the learning model. Hyperparameters are various parameters used to determine the computation (learning computation) required for each coefficient of the learning model, including, for example, the learning rate, batch size, structure of the learning model, and number of layers, which determine the update amount of the transformation parameters.

[0024] The learning parameter update unit 1006 adjusts the hyperparameters based on the composite error described later and supplies the adjusted hyperparameters to the transformation parameter update unit 1005. Furthermore, although described later, the adjustment of the transformation parameters and hyperparameters is performed by repeatedly updating each parameter in a way that minimizes the index value of the error (transformation error and composite error) so that the learning model outputs the desired result.

[0025] The transformation error calculation unit 1002 calculates the transformation error between the transformed image generated from the low-quality image by the image transformation unit 1001 and the high-quality image used as the teacher image. More specifically, the transformation error calculation unit 1002 calculates the transformation error using the pixel values ​​of the transformed image output by the learning model and the mean absolute error and mean square error calculated from the pixel values ​​of the corresponding high-quality image. The mean absolute error can be calculated using the following formula (1). The pixel value of the coordinates (i,j) of the transformed image is represented by x_i,j, and the pixel value of the coordinates (i,j) of the high-quality image is represented by y_i,j. N represents the number of pixels in the transformed image and the high-quality image. The mean square error can be calculated by converting the absolute value of formula (1) into a square.

[0026]

[0027] The size index calculation unit 1003 calculates the size index of the transformed image generated from the low-resolution image by the image transformation unit 1001. The size index includes, for example, the measurement accuracy obtained by applying arbitrary calculations to size values ​​obtained from the dimensions of the circuit pattern of the measuring semiconductor, or the yield from inputting an input image generated by a scanning electron microscope to calculating the measurement accuracy of the transformed image. The size values ​​are obtained by calculating the distance between edges and roughness values ​​using two or more edge coordinates (x_edge, y_edge) based on the brightness gradient on the image using a predetermined measurement algorithm. The measurement algorithm includes, for example, measurement parameters for measurement methods such as controlling smoothing processing and edge detection thresholds. On the other hand, the yield is affected because the computation processing time of the neural network changes depending on the construction and number of layers of the learning model adjusted by hyperparameters.

[0028] Measurement accuracy can be calculated using a measurement index that statistically represents the similarity between the size values ​​obtained from the transformed image and the high-resolution image. For example, this includes the average error, sensitivity, coefficient of determination, and reproducibility between the size values ​​obtained from the transformed image and the high-resolution image. Productivity, for example, can be expressed as the number of image slices processed per unit time from the input of an image (or the generation of an input image by an image generation tool such as a scanning electron microscope) to the calculation of the size value. Size indices calculated by statistical processing generally do not participate in the output of the neural network, just like size values, making it impossible to differentiate them using the gradient descent method described later with respect to the transformation parameters. Therefore, deep learning methods using neural networks are not applicable. On the other hand, algorithms for hyperparameter optimization are not limited by whether differentiation is possible. In this disclosure, the size index is included only in the error (composite error) during hyperparameter optimization.

[0029] Furthermore, the aforementioned error is generally used as the transformation error of the brightness value of the transformed image output by the learning model. However, in cases where the generated transformed image is used for measurement, as represented in Patent Document 1, no size index was previously used. Therefore, it is impossible to generate a transformed image with sufficiently improved size index. Moreover, in order to perform deep learning using neural networks, etc., since there is a prerequisite that the error used for learning is differentiable, it is difficult to use the corresponding non-differentiable calculated size index as the error, like the size value. Therefore, it is impossible to achieve the target value of the size index expected by the user, and it is difficult to generate a learning model for image transformation suitable for practical applications. Therefore, in this disclosure, by using the composite error described later in the adjustment (optimization) of hyperparameters, it is possible to generate a learning model that generates a transformed image with optimal size index.

[0030] The composite error calculation unit 1004 combines the transformation error L_image calculated by the transformation error calculation unit 1002 with the size index L_measure calculated by the size index calculation unit 1003 with additional weights to calculate the composite error. For example, the composite error can be calculated based on the formula (2).

[0031]

[0032] Where α and β are weighting coefficients expressed as arbitrary real numbers. For example, setting the weighting coefficient β of the size index L_measure to a value larger than the weighting coefficient α of the transformation error L_image can cause the size index to become an error larger than the transformation error, and update it to a hyperparameter that places greater emphasis on minimizing the size index.

[0033] As an example, the measurement indices included in the size index include the average error, sensitivity, coefficient of determination, and reproducibility between the size values ​​obtained from the transformed image and the size values ​​obtained from the high-definition image. Furthermore, the size index may also be composed of multiple optimization indices. Hereinafter, the calculation method for each optimization index is shown for n sets of transformed images and high-definition images. The average error can be obtained by averaging the absolute error between the size value X1 obtained from the transformed image and the size value X2 obtained from the high-definition image across n sets. Sensitivity can be represented by plotting the n sets of X1 and X2 on the x-axis and y-axis respectively, using the slope of an approximate straight line. The coefficient of determination represents the degree of misalignment of the n sets of X1 and X2 used to calculate the sensitivity, and can be obtained from the sum of squared deviations of the n sets of X1 and X2. Reproducibility can be determined by taking multiple images of the same field of view with a scanning electron microscope, and calculating the standard deviation (3σ) of the size value when generating the image by averaging the images multiple times (images within the range of 3σ are considered reproducible).

[0034] In the case where the size index includes a complex optimization index, for example, as in equation (3), each optimization index can be weighted and added together to form the size index.

[0035] γ×(average error)+δ×|1-sensitivity| +ε×(1-determination coefficient) -ζ×yield・・・(3)

[0036] Where the coefficients γ, δ, ε, and ζ are weighting coefficients expressed as arbitrary real numbers. For example, by setting the coefficient δ for sensitivity to a value larger than that of the coefficients γ, ε, and ζ, the error of the sensitivity-related terms can be strongly reflected in the optimization index, and the hyperparameter can be updated to place greater emphasis on sensitivity improvement.

[0037] Furthermore, in the above formula (3), a specific optimization index that causes the target value input by the input device 101 to meet the target value can also be configured as a size index. For example, as in formula (4), if the specific optimization index is lower than the target value, it can be configured by adding the residual from the target value to the size index.

[0038] γ×f(mean error, target value)+δ×g(|1-sensitivity|, target value) +ε×h(determination coefficient, target value) +ζ×i(yield, target value) ・・・(4)

[0039] Where f, g, h, and i represent arbitrary functions. Functions f and g, for example, when the value of the first parameter (mean error, |1-sensitivity|) is lower than the target value, output 0; when it is higher than the target value, output the residual calculated as (first parameter) - (target value). Functions h and i, for example, when the value of the first parameter (coefficient of determination, yield) is higher than the target value, output 0; when it is lower than the target value, output the residual calculated as (target value) - (first parameter). In addition, learning models and hyperparameters can also be used that take the power or root of (first parameter) - (target value) or (target value) - (first parameter), determine whether it is higher or lower than the target value by taking a certain boundary, or use additional constraints for optimization, discarding (not adopting) size indicators that are above or below the target value.

[0040] The weighting coefficients γ, δ, ε, and ζ are parameters that the user can adjust (set) according to the desired priority of the size indicators. By inputting each weighting coefficient in the priority coefficient input area 3013 included in GUI301 (refer to Figure 4), the user can reflect the priority of the desired size indicator improvement in the composite error. For example, if the user wants the determination coefficient to improve more strongly than other size indicators, by inputting a value larger than the text box for the determination coefficient in the priority coefficient input area 3013 for other size indicators, the user can optimize the parameters so that the value of the determination coefficient term in equations (3) and (4) is relatively larger than the term for other size indicators, thereby more strongly improving the determination coefficient. In this way, the user can control the improvement of each size indicator according to the desired priority.

[0041] The predetermined values ​​of each weighting coefficient can, for example, all be set to 1, which can be adjusted by the user. Alternatively, the predetermined values ​​can be determined in advance through preliminary experiments using predetermined sample images in a way that does not deviate from the user's perception (in a way that does not become an empirically uncommon value). Furthermore, the size index is not limited to the above-mentioned optimization index, but may also include other optimization index values.

[0042] The learning parameter update unit 1006 adjusts (optimizes) the hyperparameters based on the composite error calculated by the composite error calculation unit 1004 to control the composite error, and supplies them to the transformation parameter update unit 1005. For example, the learning parameter update unit 1006 optimizes the hyperparameters by using hyperparameters arbitrarily selected by the user from the set of available hyperparameter values. Alternatively, the learning parameter update unit 1006 may optimize the hyperparameters by successively selecting hyperparameters using parameter exploration algorithms such as grid search, random search, or Bayesian optimization, exploring the combination of hyperparameters that minimizes or maximizes the predetermined optimization index. The hyperparameters are set based on the above methods. When learning ends and the end condition for parameter optimization is not met, the hyperparameters are updated successively as parameter settings that determine the learning performance when learning the model again. By adjusting (optimizing) the hyperparameters, the image transformation performance and computation processing time of the image transformation model change. Therefore, by setting appropriate parameters for learning, size indicators can be improved. For example, when the number of layers in the learning model is adjusted as a hyperparameter, fewer layers result in shorter processing time but lower image quality. On the other hand, more layers can generate faithful images of teachers, but the processing time increases. The transformation parameter update unit 1005 adjusts (optimizes) the transformation parameters based on the transformation error calculated by the transformation error calculation unit 1003, and supplies them to the image transformation unit 1001. The transformation parameters are optimized, for example, by performing partial differentiation of the applied error with the transformation parameters using a method called gradient descent, and then using a successive update method. In the above calculations, the transformation parameter update unit 1005 uses the hyperparameters obtained by the learning parameter update unit 1006.

[0043] <Details of Parameter Optimization Processing> Figure 2 is a detailed flowchart illustrating the parameter optimization processing (optimization learning completes model generation and size index generation). Although the main body of the action in each step is the respective processing unit (image transformation unit 1001, transformation parameter update unit 1005, learning parameter update unit 1006, etc.), the functions of these processing units are realized by expanding the program into the processor's internal memory, and the actions of each step can be replaced by actions with the calculation device 100 as the main body.

[0044] (i) Step S201 The image conversion unit 1001 obtains the pairing of low-quality images and high-quality images stored in the storage device (memory medium) 201. Alternatively, the image conversion unit 1001 may obtain the pairing of low-quality images and high-quality images captured by a scanning electron microscope.

[0045] (ii) Step S202 The learning parameter update unit 1006 uses the GUI screen shown in FIG4 to obtain the target value of the size index input by the user from the input device 101.

[0046] (iii) Step S203 The learning parameter update unit 1006 sets the hyperparameters of the learning model in a predetermined method. As described above, the method for setting hyperparameters can be, for example, a method in which the user arbitrarily selects from the set of available hyperparameter values, or a method determined by a parameter exploration algorithm such as grid search, random search, or Bayesian optimization, or other exploration algorithms.

[0047] (iv) Step S204: The transformation parameter update unit 1005 performs learning calculations using the hyperparameters set in step S203 to optimize the transformation parameters of the learning model (generating a completed learning model). Details of step S204 will be described later with reference to FIG3.

[0048] (v) Step S205 The image transformation unit 1001 applies the learning completed model learned in step S204 to low-quality images and generates a transformed image.

[0049] (vi) Step S206 The transformation error calculation unit 1002 calculates the transformation error using the high-quality image paired with the low-quality image and the transformation image generated in step S205. Also, the size index calculation unit 1003 calculates the size index. The size index can be calculated, for example, by applying image processing (e.g., edge extraction processing) to the transformation image generated in step S205, using the measurement value of a predetermined pattern (e.g., the width of a circuit pattern) in the image (a complex measurement value obtained from the complex transformation image) or the yield through statistical processing. Next, the composite error calculation unit 1004 calculates the composite error from the transformation error and the size index (refer to Equation (2)).

[0050] (vii) Step S207 The learning parameter update unit 1006 determines whether the optimization termination condition is met. The optimization termination condition can be determined, for example, by whether the size index (each index value obtained by statistical processing) calculated in step S206 meets the target value set by the user in step S202, or by whether a predetermined number of optimization times has been reached.

[0051] In step S207, if the optimization condition is not met (the case of NO in step S207: for example, if the dimensional index does not meet the target value), the process proceeds to step S208. That is, the hyperparameter value and the dimensional index at this time are used as the selection criteria for the hyperparameter. On the other hand, if the optimization termination condition is met (the case of YES in step S208), the process proceeds to step S209.

[0052] (viii) Step S208: The parameter update unit 1006 updates the hyperparameters based on the composite error. After the hyperparameters are updated, the process returns to step S203.

[0053] The learning parameter update unit 1006, in the same manner as described in step S203, may use a hyperparameter arbitrarily selected by the user from the set of available hyperparameter values ​​as the update parameter, or may use a parameter exploration algorithm such as grid search, random search, or Bayesian optimization, or other exploration algorithms to determine the update parameter.

[0054] (ix) Step S209 The calculation device 100 outputs the optimized learning completed model and size index to the output device 102 using the GUI shown in Figure 5.

[0055] Through the above-described processing, the parameters of the learning model can be optimized. On the other hand, even if the optimization processing caused by the parameter optimization system 10 is performed, there may be cases where the target value of the size index input by the user through the input device 101 is not met. In this case, for example, if the user changes the generation conditions of the image generation tool (filtering, etc.), the above-described parameter optimization processing can be performed again using an input image with better image quality (as a result, the productivity may be sacrificed to some extent). For example, if the image generation tool is a scanning electron microscope, the generation conditions such as the number of average slices, filtering, and autofocus processing of low-quality images can be changed. Therefore, by generating an input image with better image quality, a transformed image close to a higher-quality image can be generated, increasing the possibility of improving the size index. Furthermore, in the case of changing the image quality of the input image (changing it to high quality), when actually using the learned model to generate a transformed image for measurement processing, a low-quality image with the changed image quality level is input to the image transformation unit 1001 to generate a transformed image of the measurement object.

[0056] Furthermore, in the first embodiment, the size index is obtained from the size value obtained by measuring the size of the circuit pattern of the semiconductor. However, as the size value, the overlap measurement value of the centroid deviation between layers of a multilayer semiconductor or the size value of any shape within other images can also be used.

[0057] <Details of the transformation parameter optimization process (step S204)> Figure 3 is a detailed flowchart illustrating the transformation parameter optimization process (step S204) in the parameter optimization process (Figure 2).

[0058] (i) Step S2041 The transformation parameter update unit 1005 obtains the transformation parameters (initial values) set by the user. When setting the initial values ​​of the transformation parameters of the learning model, the user can, for example, set the initial values ​​obtained by He, Xavier, or other initialization methods.

[0059] (ii) Step S2042 The image transformation unit 1001 inputs the low-quality image for learning into the learning model and generates a transformed image.

[0060] (iii) Step S2043 The transformation error calculation unit 1002 calculates the transformation error using the high-quality image paired with the low-quality image used for learning and the transformed image generated in step S2042. The transformation error can be, for example, mean absolute error (refer to Equation (1)) and mean square error.

[0061] (iv) Step S2044: The transformation parameter update unit 1005 updates (adjusts) the transformation parameters in a way that suppresses the transformation error, based on the transformation error calculated in step S2043 and the hyperparameters set in step S203. The transformation parameters are updated as described above, for example, using the gradient descent method.

[0062] (v) Step S2045 The transformation parameter update unit 1005 determines whether the learning termination condition is met. The learning termination condition may be set by whether the transformation error calculated in step S2043 has reached a predetermined number of learning times, or it may be determined by whether the change of transformation error (learning curve) during learning updates to the minimum value within a predetermined number of times.

[0063] If the learning termination condition is met (if YES is met in step S2045), the learning process in S204 ends. On the other hand, if the learning termination condition is not met (if NO is met in step S2045), the process returns to step S2042.

[0064] <Example of Optimization Condition Setting GUI> Figure 4 is a diagram showing an example of the structure of an optimization condition setting GUI (Graphical User Interface) 301 displayed on the output device 102 for inputting optimization conditions via the input device 101. Through this GUI screen, optimization conditions that serve as the optimization reference in the calculation device 100 of the parameter optimization system 10 can be set.

[0065] The optimization condition setting GUI 301 consists of an optimization index selection area 3011, an optimization index input area 3012, and a priority coefficient input area 3013. The user can select the desired optimization index by inputting the check box of the size index to be used as the optimization index in the optimization index selection area 3011.

[0066] Furthermore, the user can set the target value of the optimization index by inputting the desired target value in the optimization index input area 3012. Next, the user can adjust the intensity of the improvement of the dimensional index by inputting a priority coefficient (e.g., the larger the value, the higher the priority) in the priority coefficient input area 3013.

[0067] The selected optimization index, the target value of each optimization index, and the priority coefficient are sent to the calculation device 100, which can then calculate the dimensional index. Furthermore, the dimensional index shown in Figure 4 is a simple example and is not limited to the index shown in Figure 4 (average error, sensitivity, coefficient of determination). Additional dimensional indices other than those shown in Figure 4 can be added, along with corresponding check boxes and text boxes.

[0068] <Example of Optimization Result Output GUI> Figure 5 is a diagram showing an example of the structure of an optimization result output GUI 401 displayed on the output device 102. Through this GUI screen, the optimization results resulting from the parameter optimization processing of the calculation device 100 of the parameter optimization system 10 can be displayed to the user.

[0069] The optimization result output GUI 401 consists of a target reach display area 4011, an optimal parameter display area 4012, and a parameter contribution rate display area 4013. The target reach display area 4011 displays the target reach of the optimal size index output by the parameter optimization system 10, based on the optimal index input in the optimal index input area 3012 and its target value. For example, a size index that reaches the target value shows a target reach rate of 100%. A size index that does not reach the target value shows the target reach rate at that time. In the case of not reaching the target value, it means that even with parameter optimization, it is difficult to achieve the image quality improvement required to reach the target value in the given input image. In this case, the user can adjust the imaging parameters of the scanning electron microscope to obtain a lower-quality image with better image quality, and then perform parameter optimization again to try to achieve the target value.

[0070] The optimal parameter display area 4012 displays the hyperparameters optimized by the parameter optimization system 10. Furthermore, the parameter contribution rate display area 4013 displays the extent to which the parameters of the optimization index object input in the optimization index input area 3012 contribute to the improvement of the composite error. This allows the user to know the parameters contributing to the improvement of the dimensional index, making it easier to determine whether to add or exclude parameters from the parameter optimization object.

[0071] <Modification Example> (Modification Example 1) The parameter optimization system 10 of Modification Example 1 of the first embodiment, in addition to the components shown in FIG1, also has a loop for optimizing measurement parameters. According to Modification Example 1, the user can perform parameter optimization of the learning model instead of manually adjusting the measurement parameters in a way that does not hinder the improvement of dimensional indicators.

[0072] (i) A modified example of the optimization condition setting GUI. Figure 6 is a diagram showing a construction example of the optimization condition setting GUI 301' of modified example 1. This optimization condition GUI 301' can select whether to include measurement parameters. Regarding the same components as in Figure 4, since the same reference numerals are added in Figure 6, repeated descriptions are omitted below.

[0073] The optimization condition setting GUI 301', as an additional component, has a measurement parameter selection area 5001. In the measurement parameter selection area 5001, by inputting a checkbox for the measurement parameter to be optimized, the user can select the measurement parameters included in the update object (index). In Figure 6, after inputting the "Smoothing" checkbox, the parameters for smoothing the transformed image during pattern measurement are included in the optimization object. Similarly, after inputting the "Edge Threshold" checkbox, the threshold used for edge determination in the transformed image during pattern measurement is included in the optimization object. Furthermore, the measurement parameters shown in Figure 6 are just one example and are not limited to the measurement parameters shown in Figure 6 (smoothing, edge threshold). Measurement parameters other than those shown in Figure 6 can also be added, and corresponding checkboxes can be added accordingly.

[0074] (ii) Construction Example of Parameter Optimization System 10 Figure 7 is a diagram showing a construction example of the parameter optimization system 10 of Modification Example 1. The parameter optimization system 10 of this modification example has a channel for feeding back measurement parameters from the learning parameter update unit 1006 to the dimension index calculation unit 1003. The learning parameter update unit 1006 includes the measurement parameters used to calculate the dimension value as a part of the hyperparameters in the update object (optimization object).

[0075] In situations where the measurement algorithm cannot calculate the dimensional value (or the dimensional value is inappropriate) even when the measurement parameters are not included in the optimized object, the measurement parameters are included in the optimized object (by adding a check to the measurement parameter check box). Because parts that are difficult to see become clear (e.g., edges are emphasized), the dimensional value can be measured. Including the measurement parameters in the optimized object in this way is effective when the measurement algorithm becomes a factor hindering the improvement of dimensional indicators. For example, in order to perform measurements on images containing noise and where edges are difficult to see, there are cases where the image is smoothed during measurement. Because the value of the measurement parameter that determines the degree of smoothing is small, the edges obtained from the image are uneven, and there are cases where an appropriate dimensional value cannot be calculated. Due to uneven edges, the dimensional value becomes larger or cannot be measured, and the measurement accuracy deteriorates. Therefore, by optimizing the parameters to improve measurement accuracy, the measurement accuracy can be improved, and the dimensional indicators can be improved.

[0076] (Modification 2) According to Modification 2, the transformation parameter update unit 1005 (refer to FIG. 1) can also perform data expansion on either low-resolution or high-resolution images. This data expansion is an image processing that adds Gaussian noise, smooths the image, or enlarges or reduces it with random parameters. It is a process that improves the robustness of learning low-resolution and high-resolution images by increasing the variation between them. In this case, the learning parameter update unit 1006 can add at least one of the parameters related to the data expansion, namely, the type of image processing performed, the number of image processing operations, or the maximum intensity of the random parameters of the image processing (e.g., in the case of Gaussian noise addition, the maximum standard deviation of the added noise), as part of the hyperparameters and adjust them.

[0077] According to Variation Example 2, the dimensional indicators shown in Equation (3) or Equation (4) above can be easily improved. Therefore, users can easily obtain a learning model to achieve the target value of the desired dimensional indicator.

[0078] (Modification 3) According to Modification 3, when the dimensional index is differentiable, the transformation parameter can be adjusted (optimized) using a composite error. Therefore, the transformation parameter update unit 1005 can also adjust the transformation parameter using the composite error output by the composite error calculation unit 1004. In this case, in FIG1, the parameter optimization system 10 of this modification example replaces the channel from the transformation error calculation unit 1002 to the transformation parameter update unit 1005 with a channel from the composite error calculation unit 1004 to the transformation parameter update unit 1005. Next, in FIG3, the transformation parameter optimization process (step S204) is configured to include steps S2046 and S2047 between steps S2043 and S2044 of the parameter optimization process in the first embodiment (refer to FIG2).

[0079] (i) Step S2046: The size index calculation unit 1003 calculates the size index using the high-quality image paired with the low-quality image used for learning and the transformed image generated in step S2042. The size index calculation method can be the method described in step S206.

[0080] (ii) Step S2047 The composite error calculation unit 1004 calculates the composite error using the transformation error calculated in step S2043 and the dimensional index calculated in step S2046. According to Modification 3, the transformation parameters can be adjusted to minimize the dimensional index without relying on the adjustment (optimization) of hyperparameters. Therefore, by improving the dimensional index shown in Equation (3) or Equation (4) above, the user can obtain a learning model that achieves the target value of the desired dimensional index. In Modification 3, although the adjustment (optimization) of hyperparameters can be omitted, because of the adjustment (optimization) of hyperparameters, the transformation parameters can be optimized using excellent hyperparameters. Compared with the case of only optimizing the transformation parameters, it is easier to achieve the target value of the desired dimensional index.

[0081] <Technical Effects of the First Embodiment> According to Patent Document 1, as described above, a method for generating a transformed image or direct size value that aims to improve image quality in a measurement area by predicting the transformation image through a learning model is shown. On the other hand, according to the first embodiment, a transformed image can be generated based on a composite error including the image's transformation error and the user's desired size index (through statistical processing). Furthermore, by constructing the size index with the elements desired by the user, the user's desired size index can be improved, and a transformed image with improved size index can be generated.

[0082] (2) Second embodiment The second embodiment is a process in which, in addition to the features of the first embodiment, the image generation parameters are included in the optimization object and the optimized image generation parameters are fed back to the image generation unit (e.g., SEM device).

[0083] <Example of the construction of the parameter optimization system> Figure 8 is a diagram showing a schematic example of the construction of the parameter optimization system 10 of the learning model in the second embodiment. Since the same reference numerals are used in Figure 8 for the same constituent elements as in Figure 1, repeated descriptions are omitted below.

[0084] This parameter optimization system 10, replacing the storage device 201 (refer to FIG. 1) that holds SEM images, includes an image generation unit 7001 that generates images of the input learning model using a scanning electron microscope. Next, the parameter optimization system 10 includes the image generation parameters of the scanning electron microscope (image generation unit 7001) as the object of parameter optimization, and feeds back the image generation parameters from the learning parameter update unit 1006 to the image generation unit 7001. Furthermore, the image generation unit 7001 may be a storage device that pre-memorizes images generated using the scanning electron microscope with each image generation parameter and those image generation parameters, and outputs the memorized images.

[0085] Furthermore, the parameter optimization system 10 in the second embodiment is a system designed to improve composite errors by controlling the image generation parameters of the input image to control the image quality and image generation time of the low-quality image input to the learning model. This improves the image quality of the transformed image and expands the control width of the size index. For example, when the image quality of the low-quality image input to the learning model is improved, it becomes easier to generate a transformed image close to a high-quality image, and improvements in image quality and measurement indexes can be expected. Also, in order to control the image generation parameters of the input image, the generation time of the input image also affects the yield. Therefore, shortening the generation time of the input image can be expected to increase the yield.

[0086] <Examples of Image Generation Parameters That Can Be Optimized> Examples of image generation parameters that can be optimized in the image generation unit 7001 include the average number of images for low-quality images, scanning speed (high-speed scanning or low-speed scanning), filtering processing (smoothing filtering processing or edge filtering processing), and autofocus processing (image processing caused by a predetermined value when there is no autofocus adjustment). For example, if the average number of images is increased, the scanning speed is set to low speed, or filtering processing and autofocus processing are applied, a low-quality image with better image quality can be generated. However, on the other hand, the productivity decreases because the image generation time becomes longer. On the other hand, if the average number of images is not reduced, the scanning speed is set to high speed, or filtering processing and autofocus processing are not applied, a low-quality image can be generated with a higher productivity. However, on the other hand, because the image quality of the low-quality image deteriorates, the image quality and size indicators of the transformed image also deteriorate.

[0087] Thus, the relationship between measurement indicators (measurement indicators of indicators: such as average error, sensitivity, coefficient of determination, reproducibility, etc.) or image quality and productivity is considered. When the target values ​​of both measurement indicators and productivity are input into the input device 101, the calculation device 100 in the parameter optimization system 10 performs optimization processing by outputting appropriate balanced image generation parameters, transformation parameters, hyperparameters, and measurement parameters that satisfy the target values ​​for each size indicator (both measurement indicators and productivity).

[0088] Generally, the number of combinations of image generation parameters, transformation parameters, hyperparameters, and measurement parameters required to achieve measurement indicators and yields is enormous. Therefore, it is very difficult for users to predetermine their desired combinations (appropriate combinations). However, by performing parameter optimization processing using the parameter optimization system 10 of this embodiment, appropriate combinations of various parameters can be derived.

[0089] <Processing after image generation parameter update> In the case where the image generation unit 7001 is configured as a scanning electron microscope, the scanning electron microscope obtains the image generation parameters fed back from the learning parameter update unit 1006, changes the imaging conditions according to the image generation parameters, and then takes an image. However, it is not limited to this. The computer (not shown) can also save the image settings for multiple imaging conditions in a storage device (not shown) in advance, and input the image settings that are applicable to the fed-back image generation parameters into the learning model. Alternatively, the computer can also re-encode the scanning electron microscope indicator image when there are no image settings that are applicable to the fed-back image generation parameters.

[0090] <Technical Effects of the Second Embodiment> According to the second embodiment, the user can optimize the parameters of the learning model without manually adjusting the image generation parameters that improve the size index. Furthermore, in the second embodiment, since the image generation parameters are also included in the parameter optimization object, the image generation parameters can also be displayed in the parameter contribution rate display area 4013. Therefore, as an additional effect of the parameter contribution rate display area 4013, the user can see the image generation parameters that contribute to the improvement of the size index, making it easier to adjust the image generation parameters that improve the size index.

[0091] Although the second embodiment describes using a scanning electron microscope as an image generation tool, it is not limited to this. For example, an image generation tool whose image quality can be controlled by image generation parameters can also be used.

[0092] (3) Third embodiment The third embodiment is about the optimization of preprocessing / postprocessing parameters of a learning model for generating transformed images from low-quality images by using composite errors instead of hyperparameters.

[0093] <Example of the Construction of the Parameter Optimization System> Figure 9 is a diagram showing a schematic construction example of the parameter optimization system 10 in the third embodiment. This parameter optimization system 10 is a system that optimizes the preprocessing or postprocessing parameters when inferring transformed images using a learned model. In Figure 9, since the same reference numerals are added in Figure 8, repeated descriptions of the same components as in Figure 8 are omitted below. As shown in Figure 9, the parameter optimization system 10 in the third embodiment replaces the transformation parameter update unit 1005 and the learning parameter update unit 1006 shown in Figure 8, and includes an inference parameter update unit 9001 that updates the parameters of the postprocessing of low-quality images or transformed images applicable to the inference input to the learning model during inference, and updates the parameters of the preprocessing.

[0094] <Contents of Post-processing and Pre-processing> (i) Post-processing represents image processing applied to the transformed image (processing performed after image transformation). For example, as one example of such image processing, there are computational processes such as Gaussian noise addition processing, smoothing processing, and brightness normalization processing, or compositing processing with a second image different from the transformed image, but it is not limited to these. Furthermore, as a second image different from the transformed image, examples include high-quality images caused by analog (e.g., images created by predetermined analog simulation from semiconductor circuit pattern designs), designs with overlapping noise, or low-quality images, but it is not limited to these. In addition, the second image can be input from the input device 101 to the image transformation unit 1001, or the image transformation unit 1001 can obtain it from a storage device not shown in the figure.

[0095] It is deduced that the parameter update unit 9001 calculates a parameter for the ratio of the two images (image blending ratio) by combining parameters (e.g., the two images) of the composite of the transformed image and a second image different from the transformed image. This post-processing has the effect of changing the image quality (noise level in the transformed image, or smoothness of the transformed image, etc.). Because it changes the size value obtained from the transformed image by the post-processing, the size index obtained from the size value can be improved by optimizing the parameters of the post-processing.

[0096] (i) Preprocessing refers to image processing applied to a low-quality image (processing performed before image transformation). Specifically, as image processing applied to a low-quality image, similar to post-processing, it can include computational processing such as Gaussian noise reduction processing, smoothing processing, and brightness normalization processing, or compositing processing with a second image different from the low-quality image. This preprocessing has the effect of changing the image quality of the low-quality image. Because it changes the image quality and size values ​​of the transformed image obtained from the preprocessed low-quality image through this preprocessing, the size index obtained from the size values ​​can be improved by optimizing the parameters of the preprocessing.

[0097] <Details of Parameter Optimization Processing> Figure 10 is a flowchart illustrating the detailed parameter optimization processing of the third embodiment. In the third embodiment, the learned model is used to infer the transformation image. Furthermore, regarding the same components as in Figure 2, since the same reference numerals are added in Figure 10, repeated descriptions are omitted below.

[0098] In the third embodiment, the computing device 100 uses parameters related to the above-mentioned image processing, namely the type or number of image processing operations and the parameters required for image processing (in the case of combining with a noisy image, the synthesis rate of the image containing noise), as hyperparameters to optimize the composite error expressed by the above-mentioned formula (3) or formula (4). This parameter optimization process consists of the same steps as step S204 of the parameter optimization process in the first embodiment (refer to FIG2). However, step S203 "setting hyperparameters" is replaced by step S1001 "setting pre-processing / post-processing parameters". Also, step S208 "updating hyperparameters" is replaced by step S1002 "updating pre-processing / post-processing parameters". The hyperparameters are parameters that determine the learning performance as described above, but the pre-processing / post-processing parameters are parameters used when inferring the transformed image using the learning model that performs image transformation.

[0099] (i) Step S1001: The inference parameter update unit 9001 sets the preprocessing / postprocessing parameters (e.g., parameters of image blending ratio) of the learning model using a predetermined method. Methods for setting the preprocessing / postprocessing parameters include, for example, as described above, methods where the user arbitrarily selects from a set of available preprocessing / postprocessing parameter values, methods using parameter exploration algorithms such as grid search, random search, or Bayesian optimization, or other exploration algorithms.

[0100] (ii) Step S1002 The parameter update unit 9001 updates the preprocessing / postprocessing parameters (the parameters of the image mixing ratio) based on the composite error (used as an optimization index). After the preprocessing / postprocessing parameters are updated, the processing returns to step S1001.

[0101] The parameter update unit 9001, in the same manner as described in step S1001, may use a parameter arbitrarily selected by the user from the set of available preprocessing / postprocessing parameter values ​​as the update parameter, or may use a parameter exploration algorithm such as grid search, random search, or Bayesian optimization, or other exploration algorithms to determine the update parameter.

[0102] <Technical Effects of the Third Embodiment> According to the third embodiment, the preprocessing / postprocessing parameters used for transform image inference can be optimized. Furthermore, a transformed image with improved size parameters derived from the learning model can be obtained. Therefore, users can easily achieve the target value of the desired size parameters.

[0103] (4) In this embodiment (the first to the third embodiments), the computing device (processor or computer) 100 executes a parameter optimization program, which performs the following operations: processing the transformation error of the transformed image (low-resolution image) applied to the machine learning model and the high-resolution image (teacher image) input during the learning of the learning model; processing the size index of the pattern contained in the teacher image based on the transformed image and the teacher image; processing the composite error based on the transformation error and the size index; learning the machine learning model that generates the transformed image in a way that optimizes the composite error; and processing the parameter optimization of the machine learning model. In the first embodiment, the hyperparameters of the machine learning performance and the transformation parameters of the machine learning model that performs image transformation are defined as the objects of parameter optimization processing. In the second embodiment, in addition to defining the hyperparameters of the machine learning performance and the transformation parameters of the machine learning model that performs image transformation, the image generation parameters in the image generation unit (image generation tool) 7001 are defined as the objects of parameter optimization processing. Furthermore, in the third embodiment, instead of the transformation parameters and hyperparameters of the first embodiment, the parameters of the preprocessing and postprocessing performed during the transformation of the image become the objects of parameter optimization processing. This allows for the improvement of the user-desired size parameters and the generation of a transformed image with improved size parameters.

[0104] (ii) In this embodiment, the size index (the size index of the transformed image) includes the measurement accuracy calculated using a measurement index that statistically represents the similarity between the size values ​​obtained from the transformed image and the teacher image, and the productivity from the input of the input image (low-resolution image) to the calculation of the size value using the transformed image. The size index is expressed as a function of measurement accuracy and productivity as variables.

[0105] (iii) In the above parameter optimization process, the calculation device 100 determines that the parameter is optimized when the optimization termination condition of the size index (expressed by equation (3) or equation (4)) based on the set target value (the user-initiated set value) is met (e.g., the size index reaches the target value, or the predetermined number of optimization processes is reached) is satisfied (see step S207). By repeatedly performing optimization processing, the size index can be optimized with a high probability. In addition, even if the predetermined number of optimization processes is repeated and all the set indexes fail to reach the target value, the optimization process can be performed again by inputting an image with a better image quality than the original image (low-quality image). Furthermore, the calculation device 100 can also output the result of the optimization process and the target attainment degree of whether the size index has reached the set target value to the output device 102.

[0106] Furthermore, the computing device 100 repeatedly performs machine learning model learning and the aforementioned parameter optimization (hyperparameter and transform parameter optimization, hyperparameter, transform parameter, and image generation parameter optimization, or pre-processing / post-processing parameter optimization) until the composite error (refer to Equation (2)) is optimized. In addition, the computing device 100 optimizes the composite error by adjusting the hyperparameters using a predetermined parameter exploration algorithm. The composite error, as shown in Equation (2), is represented by multiplying the transform error and the size index by weighting coefficients (α and β) respectively. By setting the weighting coefficient (β) of the size index to be larger than the weighting coefficient (α) of the transform error, the transform parameters or hyperparameters, image generation parameters, and pre-processing / post-processing parameters that prioritize minimizing the size index can be updated.

[0107] (iv) In this embodiment, the hyperparameters defining the performance of the aforementioned machine learning include the type, number, or maximum intensity of image processing performed on at least one of the input image (low-resolution image) or the teacher image (high-resolution image). This allows for optimization of these metrics, further improving the size metrics. Therefore, users can easily obtain a machine learning model that achieves the desired target value for the size metrics.

[0108] (v) According to the second embodiment, the parameter optimization system 10 can replace the storage device 201 that stores the image (SEM image), and may also include an image generation tool (e.g., a scanning electron microscope) that generates the input image (low-resolution image). In this case, the calculation device 100 optimizes the image generation parameters when the image generation tool generates the input image by optimizing the aforementioned composite error. In this way, low-resolution images can be generated in real time using the image generation tool, while simultaneously constructing a machine learning model for generating appropriate transformed images.

[0109] (vi) According to the third embodiment, as described above, the computing device 100 optimizes the parameters of the inference about the transformed image using the machine learning model. Specifically, the computing device 100 optimizes at least one parameter of the preprocessing before generating the transformed image or the postprocessing after generating the transformed image as the parameters of the inference about the transformed image. Specifically, the preprocessing is image processing applied to the input image (low-quality image), including any one of Gaussian noise addition processing, smoothing processing, brightness normalization processing, or processing of synthesizing the input image and an image different from the input image (simulated image based on the design drawing). Furthermore, the postprocessing is image processing applied to the transformed image, including any one of Gaussian noise addition processing, smoothing processing, brightness normalization processing, or processing of synthesizing the transformed image and an image different from the transformed image (simulated image based on the design drawing). By improving the size metrics of the transformed image inferred through machine learning models, users can easily achieve the target values ​​of the desired size metrics.

[0110] (vii) The functions of each embodiment can also be implemented by software code. In this case, a memory medium storing the code is provided to a system or device, and the computer (or CPU or MPU) of the system or device reads the code stored in the memory medium. At this time, the code itself read from the memory medium can implement the functions of the aforementioned embodiments, and the code itself and the memory medium storing it constitute the present invention. As the memory medium providing such code, flexible magnetic disks, CD-ROMs, DVD-ROMs, hard disks, optical disks, optical discs, CD-Rs, magnetic disks, non-volatile memory cards, ROMs, etc. are used.

[0111] Furthermore, based on the instructions in the program code, the OS (operating system) or the like running on the computer may perform part or all of the actual processing, thereby achieving the functions of the aforementioned embodiment. Alternatively, after the program code read from the memory medium is written into the computer's memory, the computer's CPU (processor) may perform part or all of the actual processing based on the instructions in the program code, thereby achieving the functions of the aforementioned embodiment.

[0112] Furthermore, the program code of the software that implements the functions of each embodiment is distributed via the network and stored in the system or device's hard disk and memory or memory media such as CD-RW and CD-R. When in use, the computer (or CPU and MPU) of the system or device can read out and execute the program code stored in the memory or memory media.

[0113] The programs and technologies described herein are essentially unrelated to any specific device and can be implemented through the combination of their components. Furthermore, various types of devices for general purposes can be added. Dedicated devices can also be constructed to perform the functions of each embodiment. Moreover, various functions can be formed by appropriately combining the multiple constituent elements disclosed in each embodiment. For example, it is possible to delete several constituent elements from all the constituent elements shown in each embodiment, or to appropriately combine the constituent elements of different embodiments.

[0114] Although specific implementations are described in this disclosure, they are not intended to limit the scope of the invention, but rather to illustrate (the understanding of the technology disclosed herein). Those skilled in the art should be able to understand the various combinations of hardware, software, and firmware corresponding to the implementation of the technology disclosed herein. For example, the described software can be implemented in a wide range of programming or description languages ​​such as assembler, C / C++, Perl, Shell, PHP, and Java (registered trademark).

[0115] Furthermore, in the above embodiments, only those control lines and information lines are shown as necessary for the description; not all control lines and information lines may necessarily be shown on the finished product. All the structures may be interconnected.

[0116] Furthermore, those skilled in the art will understand other embodiments of this disclosure from an examination of the various implementations. The specification and specific examples are merely typical, and the scope and spirit of this disclosure will be further explained in the claims. [Simplified Explanation of the Diagram]

[0011] [Fig. 1] is a diagram showing a schematic construction example of the parameter optimization system 10 in the first embodiment. [Fig. 2] is a detailed flowchart illustrating the parameter optimization process (optimization learning completes model generation and size index generation). [Fig. 3] is a detailed flowchart illustrating the transformation parameter optimization process (step S204) in the parameter optimization process (Fig. 2). [Fig. 4] is a diagram showing a construction example of the optimization condition setting GUI (Graphical User Interface) 301 displayed on the display screen of the output device 102 for inputting optimization conditions via the input device 101. [Fig. 5] is a diagram showing a construction example of the optimization result output GUI 401 displayed on the output device 102. [Fig. 6] is a diagram showing a construction example of the optimization condition setting GUI 301' of Modified Example 1. [Fig. 7] is a diagram showing a construction example of the parameter optimization system 10 of Modified Example 1. [Figure 8] is a diagram illustrating a schematic construction example of the parameter optimization system 10 of the learning model in the second embodiment. [Figure 9] is a diagram illustrating a schematic construction example of the parameter optimization system 10 in the third embodiment. [Figure 10] is a detailed flowchart illustrating the parameter optimization process in the third embodiment.

Claims

1. A parameter optimization system for optimizing parameters required to generate a transformed image based on an input image, comprising: a memory device for storing a parameter optimization program; a processor for reading and executing the parameter optimization program from the memory device; the processor performing: processing to calculate a transformation error between the transformed image and a teacher image input during learning; processing to calculate a size index of a pattern contained in the teacher image based on the transformed image and the teacher image; processing to calculate a composite error based on the transformation error and the size index; and processing to optimize the learning of a machine learning model for generating the transformed image and the parameter optimization of the machine learning model in a manner that optimizes the composite error.

2. As in the parameter optimization system of request item 1, where, The aforementioned processor includes, in the aforementioned size parameters, a measurement accuracy calculated using a measurement index that statistically represents the similarity between the size values ​​obtained from the aforementioned transformed image and the aforementioned teacher image, and a productivity from the input of the aforementioned input image to the calculation of the aforementioned size value using the aforementioned transformed image.

3. As in the parameter optimization system of request item 1, where, The aforementioned processor determines that the aforementioned parameters have been optimized if the optimization termination condition based on the set target value of the aforementioned size index is met.

4. As in the parameter optimization system of request item 1, where, The aforementioned processor, until the aforementioned composite error is optimized, interactively repeats the learning of the aforementioned machine learning model and the optimization of the aforementioned parameters.

5. As in request item 4, the parameter optimization system, where, The aforementioned processor optimizes the aforementioned composite error by using a predetermined parameter exploration algorithm to adjust the hyperparameters that define the learning performance of the aforementioned machine learning model.

6. As in request item 1, the parameter optimization system, where, The aforementioned processor uses gradient descent to learn the aforementioned machine learning model.

7. As in the parameter optimization system of request item 1, where, The aforementioned machine learning models include at least those that use neural networks.

8. As in the parameter optimization system of request item 1, where, The aforementioned processor will optimize the hyperparameters that define the learning performance of the aforementioned machine learning model and the parameters that define the aforementioned size indicators.

9. As in request item 8, the parameter optimization system, where, The aforementioned processor optimizes the aforementioned hyperparameters by including the type, number, or maximum intensity of image processing to be performed on at least one of the aforementioned input image or the aforementioned teacher image.

10. As in the parameter optimization system of request item 1, where, The aforementioned composite error is expressed by the following formula: Composite error = α × L_image + β × L_measure; where L_image represents the transformation error, L_measure represents the size index, and α and β represent the weighting coefficients.

11. As in request item 2, the parameter optimization system, where, The aforementioned size index is expressed by the following formula: Size index = γ × L_bias + δ × |1-S| + ε × (1-R) ​​- ζ × T; where L_bias represents the average error of the aforementioned size value, S represents sensitivity, R represents the coefficient of determination, T represents yield, and γ, δ, ε, and ζ represent weighting coefficients.

12. As in request item 3, the parameter optimization system, where, The aforementioned processor outputs the result of the aforementioned optimization process and whether the aforementioned size index has reached the target value set above.

13. As in the parameter optimization system of request item 1, where, The aforementioned processor will optimize the parameters of the aforementioned size specifications.

14. As in the parameter optimization system of request item 1, where, It also includes: an image generation tool for generating the aforementioned input image; and a processor for optimizing the image generation parameters when the aforementioned image generation tool generates the aforementioned input image.

15. As in the parameter optimization system of request item 1, where, It further includes: an image generation tool for generating the aforementioned input image; and a processor, wherein the aforementioned size specification includes the productivity from generating the aforementioned input image from the aforementioned image generation tool to calculating the measurement accuracy of the aforementioned transformed image.

16. As in the parameter optimization system of request item 1, where, The aforementioned processor optimizes the parameters of the inferences about the aforementioned transformed image using the aforementioned machine learning model.

17. As in request item 16, a parameter optimization system, wherein, The aforementioned processor, as a parameter for inferences about the aforementioned transformed image, takes at least one parameter of the preprocessing before generating the aforementioned transformed image or the postprocessing after generating the aforementioned transformed image as the object of optimization.

18. As in request item 17, a parameter optimization system, wherein, The aforementioned preprocessing is image processing applied to the aforementioned input image, including any one of the following: adding Gaussian noise processing to the aforementioned input image, smoothing processing, brightness normalization processing, or compositing the aforementioned input image and an image different from the input image; the aforementioned postprocessing is image processing applied to the aforementioned transformed image, including any one of the following: adding Gaussian noise processing to the aforementioned transformed image, smoothing processing, brightness normalization processing, or compositing the aforementioned transformed image and an image different from the transformed image; 19. A parameter optimization method for optimizing parameters required to generate a transformed image based on an input image, comprising: a processor that reads and executes a parameter optimization program from a memory device and calculates a transformation error between the transformed image and a teacher image input during learning; the processor that calculates a size index of a pattern contained in the teacher image based on the transformed image and the teacher image; the processor that calculates a composite error based on the transformation error and the size index; and the processor that optimizes the learning of a machine learning model for generating the transformed image and the parameters of the machine learning model in a manner that optimizes the composite error.

20. A memory medium carrying a parameter optimization process for causing a computer to execute parameter optimization processing to optimize the parameters required to generate a transformed image based on an input image, wherein, The aforementioned program causes the aforementioned computer to perform the following: processing to calculate the transformation error between the aforementioned transformed image and the teacher image input during learning; processing to calculate the size index of the pattern contained in the aforementioned teacher image based on the aforementioned transformed image and the aforementioned teacher image; processing to calculate the composite error based on the aforementioned transformation error and the aforementioned size index; and processing to optimize the learning of the machine learning model that generates the aforementioned transformed image and the parameters of the aforementioned machine learning model in a manner that optimizes the aforementioned composite error.

Images