Device and method for training contour extraction model that detects contour of semiconductor lithography pattern

Abstract

A device for training a contour extraction model that detects the contour of a semiconductor lithography pattern, according to one embodiment of the present invention, comprises: a memory having a contour extraction learning program stored therein; and a processor for executing the program stored in the memory, wherein the contour extraction learning program inputs an SEM image of a new pattern to a contour extraction unit so as to extract a first contour image, inputs the first contour image into a style transfer model so as to generate a virtual SEM image, and trains the contour extraction model on the basis of a learning data set in which the first contour image and the virtual SEM image are matched.

Description

Learning device and method for a contour extraction model for detecting the contour of a semiconductor lithography pattern

[0001] The present invention relates to a learning device and method for an outline extraction model that detects the outline of a semiconductor lithography pattern.

[0002] Lithography is a crucial step in the semiconductor manufacturing process, using photoresist to create semiconductors. The wavelength of light determines the transistor integration, and recent efforts are underway to create precise patterns using ultrafine processing and extreme ultraviolet (EUV) lithography.

[0003] However, defects still occur in lithography patterns, and semiconductor companies are grappling with the problem of pattern defects. A single wafer contains hundreds to thousands of semiconductor chips, and minimizing defects among them and increasing yield is crucial. Therefore, analyzing repetitive patterns is essential to identify the process causing the problem. This requires accurately extracting the outline of the lithography pattern.

[0004] Semiconductor inspection systems provide outlines based on scanning electron microscope (SEM) images and measure critical dimension (CD) values ​​based on these outlines. Current outline extraction algorithms read pixel values ​​from SEM images and extract outlines if they exceed a threshold. However, because they rely on pixel values ​​from SEM images, they are susceptible to noise, and the outline extraction threshold can vary depending on manufacturing process conditions, resulting in inconsistent results.

[0005] Accordingly, recent research has proposed models utilizing deep learning. Deep learning can flexibly handle the conditions of SEM images because it focuses on features, not just pixels, by taking into account the surrounding context. However, most deep learning models are built on supervised learning, which means they require labeling a significant amount of ground-truth data.

[0006] A prior patent document related to this is Republic of Korea Patent No. 10-2588888 (Title of invention: Device and method for detecting pattern outline information of semiconductor layout).

[0007] The present invention is intended to solve the above-mentioned problem, and the purpose of the present invention is to provide a learning device and method for an outline extraction model that detects the outline of a semiconductor lithography pattern, which can extract the outline of a new lithography pattern without ground truth label data.

[0008] However, the technical task to be achieved by this embodiment is not limited to the technical task described above, and other technical tasks may exist.

[0009] As a technical means for solving the above-described technical problem, a learning device for an outline extraction model for detecting an outline of a semiconductor lithography pattern according to an embodiment of the present invention includes a memory in which an outline extraction learning program is stored; and a processor for executing the program stored in the memory, wherein the outline extraction learning program inputs an SEM image of a new pattern to an outline extraction unit to extract a first outline image, inputs the first outline image to a style transfer model to generate a virtual SEM image, and trains the outline extraction model based on a learning data set in which the first outline image and the virtual SEM image are matched.

[0010] A learning method for a contour extraction model for detecting the contour of a semiconductor lithography pattern performed by a learning device according to another embodiment of the present invention includes the steps of (a) inputting an SEM image of a new pattern into a contour extraction unit to extract a first contour image; (b) inputting the first contour image into a style transfer model to generate a virtual SEM image; and (c) training the contour extraction model based on a learning data set in which the first contour image and the virtual SEM image are matched.

[0011] According to any one of the problem solving means of the present invention described above, a learning data set composed of a first contour image and a virtual SEM image in pairs can be constructed using a new pattern SEM image and a corresponding layout image, and a contour extraction model can be trained based on the learning data set.

[0012] Additionally, by sharing the weights learned by the training data set with the pre-trained contour extraction model, the pre-trained contour extraction model can extract semiconductor lithography patterns even without ground truth label data.

[0013] Figure 1 is a configuration diagram of a learning device for an outline extraction model according to one embodiment of the present invention.

[0014] FIG. 2 is a drawing illustrating a detailed module of a learning device for an outline extraction model according to one embodiment of the present invention.

[0015] Figure 3 is a flowchart illustrating the learning process of a contour extraction model according to one embodiment of the present invention.

[0016] FIG. 4 is a diagram illustrating the learning process and inference process of an outline extraction model according to one embodiment of the present invention.

[0017] FIG. 5 is a drawing illustrating an outline extraction unit according to one embodiment of the present invention.

[0018] FIG. 6 is a diagram illustrating a style transfer model according to one embodiment of the present invention.

[0019] FIG. 7 is a drawing illustrating a contour extraction model according to one embodiment of the present invention.

[0020] Below, with reference to the attached drawings, embodiments of the present invention are described in detail so that those skilled in the art can easily implement them. However, the present invention may be implemented in various different forms and is not limited to the embodiments described herein. In the drawings, irrelevant parts have been omitted for clarity of description, and similar reference numerals have been used throughout the specification to indicate similar elements.

[0021] Throughout the specification, when a part is said to be "connected" to another part, this includes not only the cases where they are "directly connected" but also the cases where they are "electrically connected" with other elements intervening. Furthermore, when a part is said to "include" a component, this does not exclude other components, but rather includes the possibility of including other components, unless otherwise specifically stated.

[0022] In this specification, the term 'unit' includes a unit realized by hardware, a unit realized by software, and a unit realized using both. In addition, one unit may be realized by using two or more pieces of hardware, and two or more units may be realized by one piece of hardware. Meanwhile, the '~ unit' is not limited to software or hardware, and the '~ unit' may be configured to be in an addressable storage medium or may be configured to reproduce one or more processors. Therefore, as an example, the '~ unit' includes components such as software components, object-oriented software components, class components, and task components, as well as processes, functions, attributes, procedures, subroutines, segments of program code, drivers, firmware, microcode, circuits, data, databases, data structures, tables, arrays, and variables. The functionality provided within the components and '~sub-units' may be combined into a smaller number of components and '~sub-units' or further separated into additional components and '~sub-units'. Furthermore, the components and '~sub-units' may be implemented to recycle one or more CPUs within the device.

[0023] A network is a connection structure that enables information exchange between each node, such as terminals and servers, and includes a local area network (LAN), a wide area network (WAN), the Internet (WWW), wired and wireless data communication networks, telephone networks, and wired and wireless television communication networks. Examples of wireless data communication networks include, but are not limited to, 3G, 4G, 5G, 3GPP (3rd Generation Partnership Project), LTE (Long Term Evolution), WIMAX (World Interoperability for Microwave Access), Wi-Fi, Bluetooth communication, infrared communication, ultrasonic communication, visible light communication (VLC), LiFi, etc.

[0024] FIG. 1 is a configuration diagram of a learning device for a contour extraction model according to one embodiment of the present invention, FIG. 2 is a diagram explaining a detailed module of a learning device for a contour extraction model according to one embodiment of the present invention, and FIG. 3 is a flowchart explaining a learning process of a contour extraction model according to one embodiment of the present invention.

[0025] Referring to FIG. 1, the learning device (100) of the contour extraction model may include a communication module (110), a memory (120), a processor (130), and a database (140).

[0026] The learning device (100) of the contour extraction model may be implemented as a computer or portable terminal capable of connecting to a network. Here, the computer includes, for example, a notebook, a desktop, a laptop, etc., and the portable terminal may include, for example, a wireless communication device that guarantees portability and mobility, and may include all types of handheld-based wireless communication devices such as various smartphones, tablet PCs, smart watches, etc.

[0027] In addition, the learning device (100) of the contour extraction model can function as a server that provides the learning results of the contour extraction model using a learning data set including a first contour image and a virtual SEM image as a pair to an external computing device. At this time, the server can operate in a cloud computing service model such as SaaS (Software as a Service), PaaS (Platform as a Service), or IaaS (Infrastructure as a Service), or can be built in a form such as a private cloud, a public cloud, or a hybrid cloud.

[0028] The communication module (110) may be a device including hardware and software necessary to transmit and receive signals such as control signals or data signals through wired or wireless connections with other network devices.

[0029] The memory (120) may have a contour extraction learning program recorded therein. The contour extraction learning program includes a step (S110) of inputting a new pattern SEM image into the contour extraction unit (210) to extract a first contour image, a step (S120) of inputting the first contour image into the style transfer model (220) to generate a virtual SEM image, and a step (S130) of training the contour extraction model (230) based on a learning data set in which the first contour image and the virtual SEM image are matched. Here, the memory (120) may include a magnetic storage media or a flash storage media in addition to a volatile storage device requiring power to maintain stored information, but the scope of the present invention is not limited thereto.

[0030] The memory (120) may store a separate program, such as an operating system for processing and controlling the processor (130), or may perform a function for temporarily storing input or output data.

[0031] The processor (130) executes a contour extraction learning program (hereinafter, “program”) stored in the memory (120) and provides a function of controlling the hardware of the learning device (100) of the contour extraction model according to the execution of the program. That is, the processor (130) can perform hardware control functions such as a necessary file system, memory allocation, network, basic library, timer, device control (display, media, input device, 3D, etc.), and other utilities according to the execution of the program.

[0032] Referring to FIGS. 2 and 3, the processor (130) includes a step (S110) of inputting a SEM image of a new pattern into a contour extraction unit (210) to extract a first contour image, a step (S120) of inputting the first contour image into a style transfer model (220) to generate a virtual SEM image, and a step (S130) of training a contour extraction model (230) based on a learning data set in which the first contour image and the virtual SEM image are matched. In addition, each specific step of the learning process of the contour extraction model according to the execution of the program will be described later with reference to FIG. 4.

[0033] The processor (130) may include any type of device capable of processing data. For example, it may refer to a hardware-embedded data processing device having a physically structured circuit to perform a function expressed by a code or command included in a program. Examples of such hardware-embedded data processing devices include processing devices such as a microprocessor, a central processing unit (CPU), a processor core, a multiprocessor, an application-specific integrated circuit (ASIC), and a field programmable gate array (FPGA), but the scope of the present invention is not limited thereto.

[0034] The database (140) stores or provides data required for the contour extraction model learning device (100) under the control of the processor (130). For example, the database (140) may store results generated during the contour extraction model learning process. This database (140) may be included as a separate component from the memory (120), or may be constructed in a portion of the memory (120).

[0035] Referring to FIG. 2, the processor (130) may implement detailed modules that perform various functions according to the execution of the contour extraction learning program. For example, the contour extraction learning program may be executed by the processor (130) to implement a contour extraction unit (210), a style transfer model (220), and a contour extraction model (230).

[0036] FIG. 4 is a diagram illustrating the learning process and inference process of an outline extraction model according to one embodiment of the present invention.

[0037] Referring to FIG. 4, the learning device (100) of the contour extraction model according to the present invention can input a new pattern SEM image (10) into the contour extraction unit (210) during the learning process to extract a first contour image (201) (S110), and input the first contour image (201) into the style transfer model (220) to generate a virtual SEM image (202) (S120). Next, the contour extraction model (230) can be trained based on a learning data set in which the first contour image (201) and the virtual SEM image (202) are matched (S130). Next, the weights learned by the learning data set can be shared with the pre-trained contour extraction model (200).

[0038] For a pre-trained contour extraction model (200), the model must be retrained whenever a new lithography pattern appears. This requires new measured data combined with the new pattern, and the process of generating this data places a significant burden on semiconductor manufacturing facilities. However, the present invention eliminates the need for data labeling, automatically trains the contour extraction model (230) in response to changes in lithography patterns in semiconductor manufacturing facilities, and shares the learned weights with the pre-trained contour extraction model (200).

[0039] Therefore, the contour extraction model (200) learned in the inference process can output a contour image in which the contour of the pattern is accurately detected when an actual SEM image of a new pattern is input.

[0040] FIG. 5 is a drawing illustrating an outline extraction unit according to one embodiment of the present invention.

[0041] Referring to FIG. 5, the outline extraction unit (210) obtains a layout image (11) corresponding to an SEM image (10), distinguishes a pattern area and a non-pattern area in the layout image (11), extracts the center coordinates of each pattern corresponding to the pattern area, determines a coordinate range within a preset number of pixels based on the center coordinates of each pattern in the SEM image (10) matched with the layout image (11) as an outline extraction area (101, 102, ...), detects the outline of the pattern in the outline extraction area (101, 102, ...), but stops the outline detection if the outline of the detected pattern goes out of the outline extraction area (101, 102, ...) or exceeds the size of the pattern area, and merges the outlines (101-1, 102-1, ...) of the patterns whose outlines have been completed in the SEM image (10) to generate a first outline image (201).

[0042] For example, as illustrated in FIG. 5, the outline extraction unit (210) can determine the outline extraction area (101, 102, ...) based on the center coordinates of each pattern detected in the layout image (11) and the center coordinates of each pattern in the SEM image (10) that matches thereto as the coordinates that fall within 25 pixels. Next, the outline extraction unit (210) detects the outline of the pattern in the outline extraction area, and when it reaches the boundary (outline) of the pattern with a large pixel difference, the outline detection may be stopped, and when the boundary of the pattern is ambiguous and there is a risk of detection going beyond the pattern area and reaching the non-pattern area, or when the outline of the detected pattern exceeds the size of the pattern area distinguished in the layout image (11), the outline detection may be stopped in that specific instance.

[0043] Meanwhile, the SEM image (10) has patterned and non-patterned areas, but the two areas often have similar pixel value distributions due to noise. Therefore, it is difficult to distinguish between patterned and non-patterned areas by binarizing pixel values ​​based on a threshold value.

[0044] Therefore, the outline extraction unit (210) can more accurately detect the outline in the SEM image (10) by using the layout image (11) and the SEM image (10) that roughly distinguish between the pattern area and the non-pattern area.

[0045] FIG. 6 is a diagram illustrating a style transfer model according to one embodiment of the present invention.

[0046] Referring to FIG. 6, the style transfer model (220) may be a pre-trained model using a learning data set including a contour image and a SEM image matched thereto. Accordingly, the style transfer model (220) may identify a binarized pixel value in an input first contour image (201), detect a pattern area corresponding to the first pixel value and a non-pattern area corresponding to the second pixel value, and generate a virtual SEM image (202) in which the pattern area and the non-pattern area are converted.

[0047] For example, the style transfer model (220) may be comprised of one of the existing image transformation models. For example, the Pix2Pix model is an image transformation model based on Conditional Generative Adversarial Networks (CGANs) and can be trained using pairs of input images and corresponding output images.

[0048] That is, the style transfer model (220) can be formed as a fix-to-fix model, and can be trained using the SEM image corresponding to the contour image by utilizing the structure of this CGAN. For example, the style transfer model (220) can generate a virtual SEM image (202) by converting the part corresponding to 1 in the first contour image (201) into a pattern area and converting the part corresponding to 0 into a non-pattern area regardless of the input contour.

[0049] Meanwhile, CGANs are trained using two neural networks: a generator and a discriminator. The generator receives an input image and attempts to generate the desired output image, while the discriminator attempts to distinguish the generated image from the actual output image. In this process, the generator is trained to produce images that appear real, while the discriminator is trained to the point where the generator's output is indistinguishable from the actual image.

[0050] Referring again to FIG. 4 as an example, the program can generate a first contour image (201) for an SEM image (10) of a new pattern by the contour extraction unit (210), and generate a virtual SEM image (202) matching the first contour image (201) by the style transfer model (220). Accordingly, the program can build a learning data set in which the first contour image (201) and the virtual SEM image (202) for the new pattern are matched.

[0051] In this way, the present invention can construct a learning data set for a new lithography pattern without actual label data.

[0052] FIG. 7 is a drawing illustrating a contour extraction model according to one embodiment of the present invention.

[0053] The contour extraction model (230) is an autoencoder model built based on a learning data set in which a first contour image (201) and a virtual SEM image (202) are matched, and may include an encoder and a decoder. As an example, the contour extraction model (230) may include a semantic segmentation model that divides an image into meaningful categories at the pixel level.

[0054] For example, the encoder can extract a first feature of each pattern from an input virtual SEM image (202). The decoder can extract a second feature of each pattern from a layout image corresponding to the virtual SEM image (202), generate a third feature of each pattern by combining the first feature of the virtual SEM image (202) and the second feature of the layout image, and generate a second outline image (203) based on the third feature of each pattern.

[0055] For example, the encoder can extract features of each pattern from the virtual SEM image (202). For example, the encoder can extract features of each pattern outline as first features based on an object feature extraction algorithm including a Convolutional Block Attention Module (CBAM) and atrous spatial pyramid pooling (ASPP). The decoder can extract features of each pattern outline as second features from a layout image (CAD image) corresponding to the virtual SEM image (202). Subsequently, the decoder can generate third features of each pattern outline by combining the first features of each pattern outline extracted from the virtual SEM image (202) and the second features of each pattern outline extracted from the layout image. Thereafter, the decoder can generate the second outline image (203) by passing the third features of each pattern outline through a convolutional layer, a batch normalization layer, a ReLU (Recited Linear Unit) function, and upsampling.

[0056] For reference, the outline extraction model (230) was disclosed in a prior patent (Korean Patent No. 10-2588888 (Title of the invention: Device and method for detecting pattern outline information of semiconductor layout)), and for more detailed information, refer to the contents of the prior patent.

[0057] Referring again to FIG. 4 as an example, the program can train the contour extraction model (230) using a training data set constructed from a pair of a first contour image (201) and a virtual SEM image (202). At this time, the contour extraction model (230) has the same encoder and decoder structure as the pre-trained contour extraction model (200), and can learn appropriate weights for the training data set. In addition, the program can share the weights of the encoder and decoder of the contour extraction model (230) for which training has been completed with the pre-trained contour extraction model (200). At this time, the weights of the pre-trained contour extraction model (200) can be updated.

[0058] Accordingly, even if an actual SEM image of a new pattern is input to the pre-learned contour extraction model (200), an accurate contour image can be extracted. In other words, the present invention enables the pre-learned contour extraction model (200) to maintain high performance even in new semiconductor lithography patterns or under changed conditions while maintaining existing learned information.

[0059] Below, the description of the same configuration among the above-described configurations is omitted.

[0060] Referring again to FIG. 3, the learning method of the contour extraction model includes a step (S110) of inputting a new pattern SEM image into a contour extraction unit (210) to extract a first contour image, a step (S120) of inputting the first contour image into a style transfer model (220) to generate a virtual SEM image, and a step (S130) of training the contour extraction model (230) based on a learning data set in which the first contour image and the virtual SEM image are matched.

[0061] The step (S110) of extracting the first outline image may include a step of acquiring a layout image corresponding to an SEM image, a step of distinguishing a pattern area and a non-pattern area from the layout image and extracting the center coordinates of each pattern corresponding to the pattern area, a step of determining a coordinate range within a preset number of pixels based on the center coordinates of each pattern in the SEM image matched with the layout image as an outline extraction area, a step of detecting the outline of the pattern in the outline extraction area, but stopping the outline detection if the outline of the detected pattern leaves the outline extraction area or exceeds the size of the pattern area, and a step of merging the outlines of the patterns whose outlines have been completed in the SEM image to generate the first outline image.

[0062] The step (S120) of generating a virtual SEM image may include a step of identifying a binarized pixel value in an input first contour image, a step of detecting a pattern area corresponding to the first pixel value and a non-pattern area corresponding to the second pixel value, and a step of generating a virtual SEM image in which the pattern area and the non-pattern area are converted.

[0063] The step (S130) of training the contour extraction model (230) may include a step of extracting a first feature of each pattern from an input virtual SEM image, a step of extracting a second feature of each pattern from a layout image corresponding to the virtual SEM image, a step of generating a third feature of each pattern by combining the first feature of the virtual SEM image and the second feature of the layout image, and a step of generating a second contour image based on the third feature of each pattern.

[0064] Additionally, the learning method of the contour extraction model may further include a step of sharing the weights learned by the learning data set with the pre-learned contour extraction model (200).

[0065] The learning method according to one embodiment of the present invention may also be implemented in the form of a recording medium containing computer-executable instructions, such as program modules executed by a computer. Computer-readable media may be any available media that can be accessed by a computer, and include both volatile and nonvolatile media, removable and non-removable media. Furthermore, computer-readable media may include computer storage media. Computer storage media include both volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storing information, such as computer-readable instructions, data structures, program modules, or other data.

[0066] Although the devices and methods of the present invention have been described with respect to specific embodiments, some or all of their components or operations may be implemented using a computer system having a general-purpose hardware architecture.

[0067] The foregoing description of the present invention is for illustrative purposes only, and those skilled in the art will readily appreciate that the present invention can be readily modified into other specific forms without altering the technical spirit or essential characteristics of the present invention. Therefore, the embodiments described above should be understood as illustrative in all respects and not restrictive. For example, each component described as a single entity may be implemented in a distributed manner, and similarly, components described as distributed may be implemented in a combined manner.

[0068] The scope of the present invention is indicated by the claims described below rather than the detailed description above, and all changes or modifications derived from the meaning and scope of the claims and their equivalent concepts should be interpreted as being included in the scope of the present invention.

Claims

1. In a learning device of a contour extraction model that detects the contour of a semiconductor lithography pattern, Memory in which the contour extraction learning program is stored; and A processor that executes a program stored in the above memory, The above outline extraction learning program is a learning device for an outline extraction model, wherein the outline extraction learning program inputs a new pattern SEM image into an outline extraction unit to extract a first outline image, inputs the first outline image into a style transfer model to generate a virtual SEM image, and trains the outline extraction model based on a learning data set in which the first outline image and the virtual SEM image are matched.

2. In paragraph 1, The above outline extraction section Obtain a layout image corresponding to the above SEM image, In the above layout image, the pattern area and the non-pattern area are distinguished, and the center coordinates of each pattern corresponding to the pattern area are extracted, In the SEM image matched with the above layout image, the coordinate range within the preset number of pixels based on the center coordinate of each pattern is determined as the outline extraction area, Detect the outline of the pattern in the above outline extraction area, but if the outline of the detected pattern goes out of the outline extraction area or exceeds the size of the pattern area, outline detection is stopped. A learning device for an outline extraction model, which generates the first outline image by merging the outlines of patterns whose outlines have been detected in the SEM image.

3. In paragraph 1, The above style transfer model As a pre-trained model using a training data set including contour images and matching SEM images, A learning device for a contour extraction model, which identifies a binarized pixel value in the input first contour image, detects a pattern area corresponding to the first pixel value and a non-pattern area corresponding to the second pixel value, and generates a virtual SEM image in which the pattern area and the non-pattern area are converted.

4. In paragraph 1, The above outline extraction model is An autoencoder model built based on a learning data set in which the first contour image and the virtual SEM image are matched, An encoder that extracts the first feature of each pattern from the input virtual SEM image, and A learning device for an outline extraction model, comprising a decoder that extracts a second feature of each pattern from a layout image corresponding to the virtual SEM image, generates a third feature of each pattern by combining the first feature of the virtual SEM image and the second feature of the layout image, and generates a second outline image based on the third feature of each pattern.

5. In paragraph 1, The above outline extraction learning program is, A learning device for a contour extraction model, which shares weights learned by the above learning data set with a pre-learned contour extraction model.

6. A learning method of a contour extraction model for detecting the contour of a semiconductor lithography pattern performed by a learning device, (a) a step of inputting a SEM image of a new pattern into a contour extraction unit to extract a first contour image; (b) a step of inputting the first contour image into a style transfer model to generate a virtual SEM image; and (c) A learning method for a contour extraction model, comprising a step of learning a contour extraction model based on a learning data set in which the first contour image and the virtual SEM image are matched.

7. In paragraph 6, The step of extracting the above first outline image is A step of obtaining a layout image corresponding to the above SEM image; A step of distinguishing a pattern area and a non-pattern area in the above layout image and extracting the center coordinates of each pattern corresponding to the pattern area; A step of determining a coordinate range within a preset number of pixels based on the center coordinates of each pattern in the SEM image matched with the above layout image as an outline extraction area; A step of detecting the outline of a pattern in the outline extraction area, but stopping the outline detection if the outline of the detected pattern goes out of the outline extraction area or exceeds the size of the pattern area; and A learning method for an outline extraction model, comprising a step of generating the first outline image by merging the outlines of patterns whose outlines have been detected in the SEM image.

8. In paragraph 6, The steps for generating the above virtual SEM image are A step of identifying a binarized pixel value in the input first contour image; A step of detecting a pattern area corresponding to a first pixel value and a non-pattern area corresponding to a second pixel value; and A learning method for an outline extraction model, comprising the step of generating a virtual SEM image in which the pattern area and the non-pattern area are transformed.

9. In paragraph 6, The step of training the above outline extraction model is A step of extracting the first feature of each pattern from the input virtual SEM image; A step of extracting a second feature of each pattern from a layout image corresponding to the above virtual SEM image; A step of generating a third feature of each pattern by combining a first feature of the virtual SEM image and a second feature of the layout image; and A learning method for a contour extraction model, comprising a step of generating a second contour image based on a third feature of each of the above patterns.

10. In paragraph 6, A learning method for a contour extraction model, further comprising a step of sharing weights learned by the above learning data set with a pre-learned contour extraction model.

11. A computer-readable recording medium recording a computer program for executing a learning method according to any one of Articles 6 to 10.

Images