Sample observation device and method

Abstract

To provide a technique which allows a user to easily improve accuracy related to defect detection in a sample observation device.SOLUTION: A sample observation device acquires a low-quality image for learning under a first imaging condition for each defect position indicated by defect position information and determines the number of imaging and a plurality of imaging points of a plurality of high-quality images for learning associated with the low-quality image for learning for each defect position on the basis of the setting value of the imaging number in learning processing performed before sample observation processing (steps S705-S708), acquires the plurality of high-quality images for learning under a second imaging condition (step S702), learns a high-quality image estimation model by using the low-quality image for learning and the plurality of high-quality images for learning (step S703), and adjusts a parameter related to defect detection of sample observation processing by using the high-quality image estimation model (step S704).SELECTED DRAWING: Figure 7

Description

[Technical field]

[0001] The present invention relates to a sample observation technique, and to an apparatus having a function of observing defects, abnormalities, etc. (sometimes collectively referred to as defects) and circuit patterns, etc. in a sample such as a semiconductor wafer. [Background technology]

[0002] For example, in the manufacture of semiconductor wafers, it is important to quickly start up the manufacturing process and quickly transition to a mass production system with a high yield rate in order to ensure profits. For this purpose, various inspection devices, observation devices, measurement devices, etc. are introduced into the manufacturing line. A semiconductor wafer, which is a sample, is inspected for defects, for example, by an inspection device, and coordinate information indicating the position of the defect on the surface of the semiconductor wafer is output from the inspection device as defect position information. A sample observation device captures the position and part of the defect in the target semiconductor wafer with high resolution based on the defect position information. For example, an observation device using a scanning electron microscope (SEM) (also called a review SEM) is used as the sample observation device.

[0003] In mass production lines of semiconductor wafers, it is desirable to automate the observation work using a review SEM. To achieve this automation, systems including a review SEM are equipped with an Automatic Defect Review (ADR) function and an Automatic Defect Classification (ADC) function. The ADR function is a function that performs automatic defect image collection processing, which automatically collects images of defect positions in a sample. The ADC function is a function that performs automatic defect image classification processing, which automatically classifies defects from the collected defect images.

[0004] The defect coordinates in the defect position information provided by the inspection device contain errors. Therefore, in the ADR function, the review SEM first captures the defect coordinates in the defect position information at the center with a wide field of view and low magnification under the first imaging condition, and redetects the defect from the image obtained by this (in other words, a low-quality image). In the ADR function, the review SEM captures the defect area obtained by this redetection with a narrow field of view and high magnification under the second imaging condition, and outputs the image obtained by this (in other words, a high-quality image) as an image for observation. In this way, there is a method of observation with two-stage images, in which defects are searched for with a wide field of view image in the first stage, and the defects are observed in detail with a high-quality image in the second stage. Note that the high and low of the image quality, etc. mentioned above is a relative definition, and refers to the image quality according to the imaging conditions, for example, the magnification under the second imaging condition is higher than the magnification under the first imaging condition. The concept of image quality (image quality) includes magnification, resolution, signal-to-noise ratio, etc.

[0005] The following method is available for determining and detecting defects from an image (also referred to as an inspection image) captured by the above-mentioned sample observation device. This method uses an image captured of an area in which the same circuit pattern as the defective part is formed as a reference image, and determines and detects defects by comparing an inspection image captured of the defective part with a reference image that does not contain defects. In addition, Japanese Patent Application Laid-Open No. 2009-250645 (Patent Document 1) describes a method for omitting the capture of a reference image by synthesizing the reference image using an inspection image.

[0006] In the ADR function, it is necessary to adjust the processing parameters related to defect detection according to the appearance of the pattern formed on the semiconductor wafer. As a method for automatically adjusting the processing parameters, there is a method for searching for processing parameters capable of detecting a defective portion previously instructed by a user. There is also a method for searching for processing parameters capable of discriminating between defects and nuisances with high accuracy by using an inspection image capturing an image of the defective portion and multiple reference images.

[0007] In addition, in order to improve the throughput of the ADR function, there is a method of omitting imaging by estimating a high-resolution image. In this method, an inspection image captured with a wide field of view and a high-quality image captured at a high magnification are first obtained, and a model is created by learning the relationship between them. Then, in this method, when performing ADR, only the inspection image is captured, and the high-quality image for observation is estimated using the learned model. [Prior art documents] [Patent documents]

[0008] [Patent Document 1] JP 2009-250645 A Summary of the Invention [Problem to be solved by the invention]

[0009] For example, in semiconductor device manufacturing, attempts to increase integration density are being made continuously to improve device performance and reduce manufacturing costs, and the dimensions of circuit patterns formed on semiconductor wafers are being miniaturized. Accordingly, the dimensions of defects that are fatal to device operation are also tending to become smaller. Therefore, the ADR function of review SEM is also required to obtain high-quality images as observation images that allow minute defects to be visually recognized. For this purpose, a conventional technology for the ADR function of review SEM is a method that uses machine learning to estimate and obtain a high-quality observation image from a low-quality inspection image.

[0010] However, the conventional technology using this method has room for improvement in terms of accuracy. In this method, learning of a model for estimating a high-quality observation image from a low-quality inspection image is performed before adjustment of processing parameters related to defect detection. In this method, the inspection image and the high-quality image for learning are obtained using the initial setting (default) processing parameters before the adjustment. In this case, there is a possibility that defects present in the inspection image will not be captured in the high-quality image. As a result, the accuracy of the model decreases, and high-precision estimation is not possible.

[0011] An object of the present invention is to provide a technique for a sample observation device that allows a user to easily improve the accuracy of defect detection. [Means for solving the problem]

[0012] A representative embodiment of the present invention has the following configuration: A sample observation device according to the embodiment is a sample observation device including an imaging device and a processor that executes a learning process for learning a high-quality image estimation model and a sample observation process for performing defect detection, wherein (A) the learning process: (A1) acquires one or more learning defect positions related to a learning sample, (A2) acquires, for each of the learning defect positions, a learning low-quality image under a first imaging condition, (A3) acquires a first setting value related to the number of learning high-quality images to be captured, (A4) for each of the learning defect positions: (A4a) determines the number of the learning high-quality images to be captured based on the first setting value, (A4b) acquires the learning high-quality images based on the number of images determined in (A4a), (A4c) determine one or more imaging points which are positions for capturing a high-quality image, and for each of the one or more imaging points determined in (A4b), obtain the learning high-quality image under second imaging conditions, (A5) learn the high-quality image estimation model using the learning low-quality image and the learning high-quality image, (A6) adjust defect detection parameters using the high-quality image estimation model, and (B) the sample observation process is based on the adjusted defect detection parameters: (B1) obtain a first inspection image of a defect position of the sample to be observed under the first imaging condition, and (B2) detect defect candidates of the sample to be observed based on the first inspection image. Effect of the Invention

[0013] According to the representative embodiments of the present invention, the user can easily improve the accuracy of defect detection in relation to the technique of the sample observation device. Problems, configurations, effects, and the like other than those described above will be described in the description of the embodiment of the invention. [Brief description of the drawings]

[0014] [Figure 1] 1 is a diagram showing a configuration of a sample observation device according to a first embodiment of the present invention. [Diagram 2] FIG. 2 is a diagram showing a functional block configuration of a higher-level control device in the first embodiment. [Diagram 3] FIG. 2 is a diagram showing the arrangement of SEM detectors as viewed obliquely from above in the first embodiment. [Figure 4] FIG. 2 is a diagram showing the arrangement of SEM detectors as viewed from the z-axis direction in the first embodiment. [Diagram 5] FIG. 2 is a diagram showing the arrangement of SEM detectors as viewed from the y-axis direction in the first embodiment. [Figure 6] 1 is an explanatory diagram showing a schematic example of defect coordinates on a semiconductor wafer surface, which are indicated by defect position information of an inspection device; [Figure 7] 4 is a flow chart showing the overall processing of the sample observation device of the embodiment 1. [Figure 8] FIG. 11 is an explanatory diagram relating to step S707 of defect detection in the first embodiment. [Figure 9] FIG. 11 is a flow diagram of step S702 of acquiring an image in the first embodiment. [Figure 10] FIG. 4 is a diagram showing an example of a screen related to deciding the number of images to be captured in the first embodiment. [Figure 11] FIG. 11 is a flow diagram of step S905 of determining an imaging point in the first embodiment. [Figure 12] FIG. 2 is an explanatory diagram relating to an imaging point determination process in the first embodiment. [Figure 13] FIG. 2 is an explanatory diagram showing an example of the configuration of a high quality image estimation model in the first embodiment. [Figure 14] FIG. 11 is a flow diagram of step S704 of parameter adjustment in the first embodiment. [Figure 15] FIG. 2 is an explanatory diagram relating to the process of acquiring a high-quality inspection image in the first embodiment. [Figure 16] FIG. 4 is a schematic diagram showing an example of an estimation process for a high-quality inspection image in the first embodiment. [Figure 17] FIG. 2 is a schematic diagram showing an example of a high quality reference picture estimation process in the first embodiment. [Figure 18]FIG. 4 is a schematic diagram showing an example of a synthesis process of a high quality reference image in the first embodiment. [Figure 19] FIG. 4 is a diagram showing an example of a screen related to image acquisition in the first embodiment. [Figure 20] FIG. 4 is a diagram showing an example of a screen related to model learning and parameter adjustment in the first embodiment. [Figure 21] FIG. 11 is a diagram showing a functional block configuration of a higher-level control device in the second embodiment. [Figure 22] FIG. 11 is a flow chart showing the overall processing of the sample observation device in the second embodiment. [Diagram 23] FIG. 11 is a flow diagram of step S2102 of acquiring an image in the second embodiment. [Figure 24] FIG. 11 is a diagram showing a functional block configuration of a higher-level control device in the third embodiment. [Diagram 25] FIG. 11 is a flow chart showing the overall processing of the sample observation device in the third embodiment. [Figure 26] FIG. 11 is a flow diagram of step S2502 of acquiring an image in the third embodiment. [Figure 27] FIG. 13 is an explanatory diagram relating to an imaging point determination process in the third embodiment. [Figure 28] FIG. 11 is an explanatory diagram showing an example of a model configuration in the third embodiment. [Figure 29] FIG. 11 is a flow diagram of step S2504 of parameter adjustment in the third embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0015] Hereinafter, an embodiment of the present invention will be described in detail with reference to the drawings. In the drawings, the same parts are generally given the same reference numerals, and repeated explanations are omitted. In the drawings, the representation of each component may not show the actual position, size, shape, range, etc., in order to facilitate understanding of the invention. For the purpose of explanation, when describing processing by a program, the program, functions, processing units, etc. may be described as the main body, but the main body as hardware for them is a processor, or a controller, device, computer, system, etc., configured with the processor, etc. The computer executes processing according to the program read onto the memory while appropriately using resources such as memory and communication interfaces by the processor. This realizes a predetermined function, processing unit, etc. The processor is configured with a semiconductor device such as a CPU or GPU, for example. The processor is configured with a device or circuit capable of performing a predetermined calculation. The processing is not limited to software program processing, and can also be implemented by a dedicated circuit. The dedicated circuit can be an FPGA, an ASIC, a CPLD, etc. The program may be installed in the target computer as data in advance, or may be distributed as data from the program source to the target computer and installed. The program source may be a program distribution server on a communication network, or a non-transitory computer-readable storage medium (e.g., a memory card). The program may be composed of multiple modules. The computer system is not limited to one device, but may be composed of multiple devices. The computer system may be composed of a client-server system, a cloud computing system, an IoT system, etc. Various data and information are composed of structures such as, for example, tables and lists, but are not limited to these. Expressions such as identification information, identifiers, IDs, names, numbers, etc. are mutually interchangeable.

[0016] <Embodiment Mode> The sample observation device according to the embodiment is a device for observing circuit patterns and defects formed on a sample such as a semiconductor wafer. The sample observation device includes an inspection image acquisition unit that acquires an inspection image (in other words, a low-quality image) under a first imaging condition for each defect position in the defect position information created and output by the inspection device, a high-quality image acquisition unit that acquires a plurality of high-quality images under a second imaging condition for each defect position, a defect detection unit that calculates the position and feature amount of a defect candidate using the inspection image, a model learning unit that learns a model (high-quality image estimation model) that estimates an image acquired under the second imaging condition from an image acquired under the first imaging condition using the inspection image and the high-quality image, and a parameter adjustment unit that adjusts parameters related to the defect detection process of the defect detection unit using the model.

[0017] The sample observation method of the embodiment is a method having steps executed in the sample observation device of the embodiment, and has steps corresponding to each of the above-mentioned units. The processing in the sample observation device and the corresponding steps are roughly divided into sample observation processing (in other words, defect detection processing, etc.) and learning processing. The learning processing is processing for performing model learning by machine learning in correspondence with an image used in the sample observation processing, and adjusting parameters related to the sample observation processing.

[0018] In the following, an example of the sample observation device will be described in which a device that uses a semiconductor wafer as a sample and observes defects in the semiconductor wafer is used. This sample observation device includes an imaging device that images the sample based on the defect coordinates of the defect position information from the defect inspection device. In the following, an example will be described in which an SEM is used as the imaging device. The imaging device is not limited to an SEM, and may be a device other than an SEM, for example, an imaging device that uses charged particles such as ions.

[0019] <Embodiment 1> The specimen observation device and the like according to the first embodiment will be described with reference to Figs. 1 to 20. In the specimen observation device of the comparative example, high-quality images for learning a high-quality image estimation model are collected before adjusting processing parameters related to ADR processing for specimen defect observation. In this configuration, there are cases where defects are not captured in the high-quality images, and in such cases, the accuracy of the model is reduced. Therefore, the specimen observation device of the first embodiment (for example, Fig. 2) includes a high-quality image acquisition unit 206. The high-quality image acquisition unit 206 includes an image number determination unit 207 for determining the number of high-quality images to be captured, an imaging point determination unit 208 for determining multiple imaging points for a high-quality image based on the positions and feature amounts of defect candidates calculated by the defect detection unit 215, and the number of images to be captured, and a high-quality image imaging unit 209 for imaging a high-quality image under a second imaging condition for each of the determined multiple imaging points. This makes it possible to improve the accuracy of the model, and the accuracy of specimen defect observation can be improved by adjusting parameters using the model.

[0020] In the first embodiment, the defect detection process of the ADR function (step S707 in FIG. 7 described later) is realized using general image processing (i.e., determining the difference by comparing the inspection image with the reference image, etc.). In the first embodiment, machine learning (step S703) is used to adjust the parameters for the defect detection process (step S704). As described above, the high-low relationship of image quality, etc. in two-stage images is a relative definition. Furthermore, the imaging conditions (in other words, image quality conditions) that stipulate the image quality, etc. are conditions that are applied not only when actually capturing images with an imaging device, but also when creating an image by processing.

[0021] [1-1. Sample observation device] FIG. 1 shows the configuration of the sample observation device 1 of the first embodiment. The sample observation device 1 is broadly configured to have an imaging device 2 and a host control device 3. A specific example of the sample observation device 1 is a review SEM. A specific example of the imaging device 2 is an SEM 101. The host control device 3 is connected to the imaging device 2. The host control device 3 is a device that controls the imaging device 2, etc., in other words, a computer system. The sample observation device 1, etc., is equipped with necessary functional blocks and various devices, but the drawing shows only a part including essential elements. In other words, the whole including the sample observation device 1 in FIG. 1 is configured as a defect inspection system. A storage medium device 4 and an input / output terminal 6 are connected to the host control device 3, and a defect classification device 5, an inspection device 7, etc. are connected via a network.

[0022] The sample observation device 1 is a device or system having an automatic defect review (ADR) function. In this example, defect position information 8 is created in advance as a result of inspecting a sample in an external inspection device 7, and the defect position information 8 output / provided from the inspection device 7 is stored in advance in the storage medium device 4. The upper control device 3 reads out and refers to the defect position information 8 from the storage medium device 4 during ADR processing related to defect observation. The SEM 101, which is the imaging device 2, captures an image of a semiconductor wafer, which is the sample 9. The sample observation device 1 performs ADR processing based on the image captured by the imaging device 2 to obtain multiple high-quality inspection images.

[0023] The defect classification device 5 (in other words, a defect image classification device) is a device or system having an automatic defect classification (ADC) function, and performs ADC processing based on the information and data of the defect observation processing result using the ADR function of the sample observation device 1, to obtain a result of classifying defects (corresponding defect images). The defect classification device 5 supplies the information and data of the classification result to another device (not shown) connected to a network, for example. Note that the configuration is not limited to that of FIG. 1, and a configuration in which the defect classification device 5 is combined with the sample observation device 1 is also possible.

[0024] The upper control device 3 includes a control unit 102, a storage unit 103, a calculation unit 104, an external storage medium input / output unit 105 (in other words, an input / output interface unit), a user interface control unit 106, and a network interface unit 107. These components are connected to a bus 114, and are capable of mutual communication and input / output. Note that, although the example of Fig. 1 shows a case where the upper control device 3 is configured by one computer system, the upper control device 3 may be configured by multiple computer systems (e.g., multiple server devices), etc.

[0025] The control unit 102 corresponds to a controller that controls the entire sample observation device 1. The storage unit 103 stores various information and data including programs, and is composed of a storage medium device equipped with, for example, a magnetic disk or a semiconductor memory. The calculation unit 104 performs calculations according to the programs read from the storage unit 103. The control unit 102 and the calculation unit 104 are equipped with a processor and a memory. The external storage medium input / output unit (in other words, an input / output interface unit) 105 inputs and outputs data to and from the external storage medium device 4.

[0026] The user interface control unit 106 is a part that provides and controls a user interface including a graphical user interface (GUI) for inputting and outputting information and data with a user (in other words, an operator). An input / output terminal 6 is connected to the user interface control unit 106. Other input devices and output devices (e.g., a display device) may be connected to the user interface control unit 106. The defect classification device 5, the inspection device 7, etc. are connected to the network interface unit 107 via a network (e.g., a LAN). The network interface unit 107 is a part that has a communication interface that controls communication with external devices such as the defect classification device 5 via the network. Other examples of external devices include a DB server and an MES (manufacturing execution system).

[0027] A user uses the input / output terminal 6 to input information (e.g., instructions and settings) to the sample observing device 1 (particularly the upper control device 3) and checks the information output from the sample observing device 1. The input / output terminal 6 can be, for example, a PC equipped with a keyboard, mouse, display, etc. The input / output terminal 6 may be a client computer connected to a network. The user interface control unit 106 creates a GUI screen, which will be described later, and displays it on the display device of the input / output terminal 6.

[0028] The calculation unit 104 is configured with, for example, a CPU, a ROM, and a RAM, and operates according to a program read from the storage unit 103. The control unit 102 is configured with, for example, a hardware circuit or a CPU. When the control unit 102 is configured with a CPU, the control unit 102 also operates according to a program read from the storage unit 103. The control unit 102 realizes each function (each functional block described below) based on, for example, program processing. Data such as a program is supplied to and stored in the storage unit 103 from the storage medium device 4 via the external storage medium input / output unit 105. Alternatively, data such as a program may be supplied to and stored in the storage unit 103 from a network via the network interface unit 107.

[0029] The SEM 101 constituting the imaging device 2 includes a stage 109, an electron source 110, a detector 111, an electron lens (not shown), a deflector 112, etc. The stage 109 (in other words, a sample stage) is a stage on which a semiconductor wafer, which is a sample 9, is placed and which is movable at least in the horizontal direction. The electron source 110 is an electron source for irradiating an electron beam onto the sample 9. An electron lens (not shown) converges the electron beam onto the surface of the sample 9. The deflector 112 is a deflector for scanning the electron beam on the sample 9. The detector 111 detects electrons and particles such as secondary electrons and reflected electrons generated from the sample 9. In other words, the detector 111 detects the state of the surface of the sample 9 as an image. In this example, the detector 111 includes a plurality of detectors (described later) as shown in the figure.

[0030] Information detected by the detector 111 of the SEM 101 (in other words, an image signal) is supplied to a bus 114 of the host controller 3. The information is processed by the calculation unit 104 and the like. In this example, the host controller 3 controls the stage 109, deflector 112, detector 111 and the like of the SEM 101. Note that driving circuits for driving the stage 109 and the like are not shown. Defect observation processing for the sample 9 is realized by processing information from the SEM 101 (in other words, an image) by a computer system which is the host controller 3.

[0031] This system may be in the following form. The upper control device 3 is a server of a cloud computing system or the like, and the input / output terminal 6 operated by the user is a client computer. For example, when a large amount of computer resources is required for machine learning, the machine learning processing may be performed in a group of servers of a cloud computing system or the like. The processing function may be shared between the group of servers and the client computer. The user operates a client computer, and the client computer transmits a request to the server. The server receives the request and performs processing according to the request. For example, the server transmits data of the requested screen (e.g., a Web page) to the client computer as a response. The client computer receives the response data and displays the screen (e.g., a Web page) on the display screen of the display device.

[0032] [1-2. Functional blocks] 2 shows an example of a functional block configuration realized by the control unit 102, the storage unit 103, and the calculation unit 104 of the upper control device 3 in FIG. 1 in the first embodiment. In the control unit 102 and the calculation unit 104, each functional block as shown in FIG. 2 is configured based on program processing, and in the storage unit 103, each storage unit as shown in FIG. 2 is configured. The control unit 102 has a stage control unit 201, an electron beam scan control unit 202, a detector control unit 203, a reference image acquisition unit 204, an inspection image acquisition unit 205, and a high-quality image acquisition unit 206. The high-quality image acquisition unit 206 has an imaging number determination unit 207, an imaging point determination unit 208, and a high-quality image imaging unit 209. The calculation unit 104 has a defect detection unit 215, a model learning unit 216, a high-quality image estimation unit 217, and a parameter adjustment unit 218.

[0033] A stage control unit 201 controls the movement and stopping of the stage 109 in Fig. 1. An electron beam scan control unit 202 controls the polarizer 112 in Fig. 1 and the like so that the electron beam is irradiated within a predetermined field of view. A detector control unit 203 samples a signal from the detector 111 in Fig. 1 in synchronization with the scanning of the electron beam, adjusts the gain, offset, etc. of the sampled signal, and generates a digital image.

[0034] The reference image acquisition unit 204 operates the stage control unit 201, the electron beam scan control unit 202, and the detector control unit 203 to acquire a reference image not including defects for use in the defect observation process by capturing an image under a first imaging condition. The inspection image acquisition unit 205 operates the stage control unit 201, the electron beam scan control unit 202, and the detector control unit 203 to acquire an inspection image for use in the defect observation process by capturing an image under the first imaging condition.

[0035] The high-quality image acquisition unit 206 acquires a plurality of high-quality images corresponding to the inspection image acquired by the inspection image acquisition unit 205 by capturing images under the second imaging conditions. The high-quality image is an image for defect observation, and is required to have high visibility of defects and circuit patterns on the surface of the sample 9. Therefore, this high-quality image is made to have a higher quality than the inspection image, and generally, the second imaging conditions are made to be conditions that can capture an image of higher quality than the first imaging conditions. For example, the second imaging conditions have a slower electron beam scanning speed, a larger number of image addition frames, and a higher image resolution than the first imaging conditions. Note that the imaging conditions are not limited to these.

[0036] The storage unit 103 has an image storage unit 210, a defect detection parameter storage unit 211, a high-quality image estimation parameter storage unit 212, an imaging parameter storage unit 213, and an observation coordinate storage unit 214. The image storage unit 210 stores digital images generated by the detector control unit 203 (in other words, images captured by the SEM 101) together with incidental information (in other words, attribute information, metadata, management information, etc.), and stores images generated by the calculation unit 104.

[0037] The defect detection parameter storage unit 211 stores parameters (sometimes referred to as defect detection parameters) related to defect detection processing required for defect observation. The high-quality image estimation parameter storage unit 212 stores parameters (sometimes referred to as model parameters) related to a high-quality image estimation model. The imaging parameter storage unit 213 stores information and data such as conditions (sometimes referred to as imaging conditions) when imaging in the SEM 101 as imaging parameters. The imaging conditions include the above-mentioned first imaging conditions and second imaging conditions, and can be set in advance. Examples of the imaging parameters include an electron beam scanning speed, the number of image addition frames, and an image resolution. The observation coordinate storage unit 214 stores information and data such as defect coordinates (sometimes referred to as observation coordinates) of the defect position of the observation target input based on the defect position information 8.

[0038] The defect detection unit 215 calculates the position and feature amount of a defect candidate in the inspection image. The model learning unit 216 performs learning processing of a model (also described as a high-quality image estimation model) required for performing estimation processing for estimating an image acquired under a second imaging condition from an image acquired under a first imaging condition. Note that the meaning of estimation in the model is that a high-quality image that is not an image actually captured is created and output by estimation. Note that the image used when learning the model is different from the image used when observing an actual defect using the trained model.

[0039] The high quality image estimation unit 217 performs estimation processing from a first image acquired under a first imaging condition to a second image acquired under a second imaging condition using the model. The high quality image estimation unit 217 inputs the first image to the model and obtains the second image as an output of the estimation result.

[0040] The parameter adjustment unit 218 is a part that realizes a parameter adjustment function that automatically adjusts processing parameters (in other words, recipes, etc.) related to defect observation. The processing parameters include defect detection parameters, model parameters, and imaging parameters. The defect detection parameters include, for example, threshold values ​​for defect judgment. The model parameters are parameters that configure (in other words, set) a high-quality image estimation model, and are, for example, parameters that configure CNN, which will be described later.

[0041] [1-3. Detector] FIG. 3 is a projection view of the arrangement of detectors 111 relative to stage 109 in SEM 101 in the first embodiment, seen from diagonally above. In this example, five detectors 301-305 are provided as detectors 111, and these detectors 301-305 are mounted at predetermined positions in SEM 101 as shown in the figure. The number of detectors 111 is not limited to this. FIG. 4 is a plan view of the arrangement of detectors 111 as seen from the z-axis direction. The z-axis corresponds to the vertical direction. FIG. 5 is a cross-sectional view of the arrangement of detectors 111 as seen from the y-axis direction. The x-axis and y-axis correspond to the horizontal direction and are two orthogonal directions.

[0042] As shown in Figs. 3 to 5, detectors 301 and 302 are disposed at positions P1 and P2 along the y-axis, and detectors 303 and 304 are disposed at positions P3 and P4 along the x-axis. Although not particularly limited, these detectors 301 to 304 are disposed in the same plane on the z-axis. Detector 305 is disposed at position P5 along the z-axis, which is farther away from sample 9 on stage 109 than the positions on the z-axis plane on which detectors 301 to 304 are disposed. Note that detector 305 is omitted in Figs. 4 and 5.

[0043] The detectors 301 to 304 are arranged so as to selectively detect electrons having specific emission angles (elevation angle and azimuth angle). The detector 301 can detect electrons emitted from the sample 9 along the y-axis direction (positive direction indicated by an arrow), and the detector 302 can detect electrons emitted from the sample 9 along the y-axis direction (opposite direction to the positive direction). Similarly, the detector 304 can detect electrons emitted along the x-axis direction (positive direction), and the detector 303 can detect electrons emitted along the reverse direction of the x-axis. This makes it possible to obtain an image with contrast as if light was irradiated from the opposite directions to each detector. The detector 305 can mainly detect electrons emitted from the sample 9 along the z-axis direction.

[0044] As described above, in the first embodiment, a configuration in which multiple detectors are arranged at multiple positions along different axes allows for acquisition of contrasted images, enabling more detailed defect observation. The configuration of the detector 111 is not limited to this, and the detector may be arranged at a position or orientation different from that shown in FIG. 3.

[0045] [1-4. Defect location information] FIG. 6 is a schematic diagram showing an example of a defect position indicated by defect coordinates included in defect position information 8 from an external inspection device 7. In FIG. 6, defect coordinates are illustrated by points (marked with x) on the xy plane of the target sample 9. From the perspective of the sample observation device 1, these defect coordinates are the observation coordinates that are the object of observation. Wafer WW indicates a circular semiconductor wafer surface area. Die DI indicates the area of ​​multiple dies (in other words, semiconductor chips) formed on the wafer WW.

[0046] The sample observation device 1 of the first embodiment has an ADR function that automatically collects a high-definition image of a defective portion on the surface of the sample 9 as a high-quality image based on such defect coordinates. However, the defect coordinates in the defect position information 8 from the inspection device 7 contain errors. In other words, an error may occur between the defect coordinates in the coordinate system of the inspection device 7 and the defect coordinates in the coordinate system of the sample observation device 1. One cause of the error is imperfect alignment of the sample 9 on the stage 109, etc.

[0047] Therefore, the sample observation device 1 of the first embodiment first captures an image with a wide field of view and low magnification (in other words, an image with relatively low image quality) under first imaging conditions centered on the defect coordinates of the defect position information 8 as an inspection image, and redetects the defect portion based on that inspection image. Then, the sample observation device 1 estimates a high-quality image with a narrow field of view and high magnification under second imaging conditions for the redetected defect portion using a high-quality image estimation model learned in advance, and acquires that high-quality image as an image for observation.

[0048] A wafer WW includes a plurality of dies DI in a regular pattern. Therefore, when an image of another die DI adjacent to a die DI having a defective portion is captured, an image of a good die not including a defective portion can be acquired. In the defect detection process in the sample observation device 1, for example, such a good die image can be used as a reference image. In the defect detection process, for example, a comparison of shading (an example of a feature amount) is performed between the reference image and the inspection image to determine a defect, and a portion with a different shading can be detected as a defective portion. An example of the above-mentioned defect detection parameter is a threshold value that determines the difference in shading when determining a defective portion in this way.

[0049] [1-5. Defect inspection method] Next, a defect observation method in the sample observation device 1 will be described. First, the entire defect observation method and processing will be described, and each step performed in the defect observation method will be described in detail in order. Note that the steps are mainly executed by the upper control device 3 (particularly the processor), and the control unit 102, storage unit 103, and calculation unit 104 in FIG. 2 are used as appropriate. Information required in the steps is read from the storage unit 103, and information generated in the steps is stored in the storage unit 103. Note that the learning sample targeted during the learning process (steps S702 to S707 in FIG. 7 described later) and the observation target sample targeted during the sample observation process (steps S705 to S708 described later) may be different samples, and the target positions may be different positions on the same sample.

[0050] [1-6. Overall processing] FIG. 7 shows a defect observation processing flow as the overall processing and operation of the sample observation device 1 in the first embodiment. FIG. 7 has steps S701 to S708. At the start of defect observation, in step S701, the upper control device 3 reads information such as each parameter from the defect detection parameter storage unit 211, the high-quality image estimation parameter storage unit 212, the imaging parameter storage unit 213, etc. in FIG. 2, and judges whether a recipe related to the defect observation and defect detection processing has been set up. The recipe is a set of information required to perform the defect observation and defect detection processing, and includes processing parameters related to the ADR function, etc. If the recipe has not been set up (No), the process proceeds to step S702, and if the recipe has been set up (Yes), the process proceeds to step S705. Steps S702 to S704 are processing for setting up the recipe (the learning processing described above). Steps S705 to S708 are processing for the actual sample observation and defect detection.

[0051] Step S702 is an image acquisition step. In step S702, the host controller 3 acquires images (a low-quality image for learning and a high-quality image for learning) to be used in learning a high-quality image estimation model (step S703) and adjusting defect detection parameters (step S704) using the reference image acquisition unit 204, the inspection image acquisition unit 205, and the high-quality image acquisition unit 206 in FIG. 2. At this time, the host controller 3 reads defect coordinates of the wafer WW as shown in FIG. 6, which is the target for acquiring images, from the observation coordinate storage unit 214, acquires images corresponding to all the defect coordinates, and stores the acquired images in the image storage unit 210.

[0052] Next, a model learning step is performed in step S703. In step S703, the upper control device 3 reads the image acquired in step S702 from the image storage unit 210, and uses the model learning unit 216 to learn a high quality image estimation model related to the processing of the high quality image estimation unit 217. The learned model is stored in the high quality image estimation parameter storage unit 212.

[0053] Next, a parameter adjustment step of step S704 is performed. In step S704, the upper control device 3 adjusts the defect detection parameters using the parameter adjustment unit 218. The adjusted defect detection parameters are stored in the defect detection parameter storage unit 211.

[0054] If the recipe has already been set up before defect observation, or after the recipe has been set up in steps S702 to S704, a loop process of steps S705 to S708 is executed. In this loop process, the processes of steps S705 to S708 are repeated for each defect position indicated by the defect coordinates in the observation target sample. The defect position corresponding to the defect coordinates is designated as i, where i=1 to L.

[0055] Step S705 is a reference image acquisition step. In step S705, the upper control device 3 uses the reference image acquisition unit 204 to capture an image of a die DI adjacent to the die DI including the defect position (i) under the first imaging condition stored in the imaging parameter storage unit 213, thereby acquiring a reference image (first reference image) corresponding to the inspection image (first inspection image).

[0056] Step S706 is an inspection image acquisition step. In step S706, the upper control device 3 acquires an inspection image (first inspection image) at the defect position (i) by using the inspection image acquisition unit 205 to capture an image under the first imaging conditions stored in the imaging parameter storage unit 213. The reference image and the inspection image acquired in these steps are stored in the image storage unit 210 in association with each other.

[0057] Step S707 is a defect detection step. In step S707, the upper control device 3 performs defect detection from the inspection image in the defect detection section 215 by using the reference image obtained in step S705, the inspection image obtained in step S706, and the defect detection parameters that have been set up (in other words, adjusted) and stored in the defect detection parameter storage section 211. The defect detection parameters are, for example, an image mixing ratio in a mixing process described later.

[0058] Step S708 is a high-quality image estimation step. In step S708, the upper control device 3 performs high-quality image estimation using defect position information of the defect detection result in the inspection image obtained in step S707 and the high-quality image estimation unit 217. That is, the upper control device 3 estimates and obtains a high-quality image for defect observation based on a model from the inspection image. The estimated high-quality image is stored in the image storage unit 210 in association with the inspection image and the reference image. In the estimation in step S708, a high-quality image estimation model that has been set up or set up in step S703 may be used. In addition, in the sample observation process, highly accurate defect detection, observation, analysis, measurement, evaluation, etc. of defects may be performed using the high-quality image obtained in step S708. When the processes in steps S705 to S708 are performed for all defect positions (i), this defect observation flow ends.

[0059] [1-7. Step S707 Defect detection] 8 is an explanatory diagram regarding the defect detection process in step S707. In step S707, the defect detection unit 215 performs smoothing processing and image mixing processing of the images acquired by the detector 111 (301 to 305) on each of the inspection image 801 and the reference image 802 as preprocessing for step S805. Thereafter, in step S808, the defect detection unit 215 calculates the difference between the preprocessed inspection image 806 and the reference image 807, and binarizes the difference using a first threshold value to obtain a binarized image 803.

[0060] In step S809, the defect detection unit 215 calculates the degree of abnormality for each region 804 in the binarized image 803 that is determined to be greater than the first threshold. The defect detection unit 215 detects a portion in which the degree of abnormality is greater than the second threshold as a defect 811, and obtains a binarized image 810 (in other words, a defect detection result image). For example, the sum of the luminance values ​​of the difference images in the binarized region can be used as the degree of abnormality. In the process of FIG. 8, examples of the above-mentioned defect detection parameters include the smoothing degree in the smoothing process during preprocessing, the image mixing ratio in the mixing process, the first threshold, the second threshold, and the like. In the example of FIG. 8, a case where one defect 811 is detected as the binarized image 810 of the defect detection result is shown, but there are also cases where no defect is detected or multiple defects are detected.

[0061] [1-8. Step S702 Image Acquisition] 9 shows a detailed flow related to the image acquisition process in step S702. In the sample observation device 1 of the first embodiment, for defect observation, an inspection image including the defective portion, a reference image not including the defect, and multiple high-quality images are required for each defect position (i). Therefore, in this step S702, the upper control device 3 uses the inspection image acquisition unit 205, the reference image acquisition unit 204, and the high-quality image acquisition unit 206 to acquire those images (in other words, a low-quality image for learning, a reference image for learning, and a high-quality image for learning).

[0062] The high-quality image is acquired by performing defect detection on the inspection image and capturing an image of a defect position (corresponding defect candidate) detected under the second imaging condition stored in the imaging parameter storage unit 213. However, the defect detection parameters are default parameters that are not adjusted to the target process (semiconductor wafer manufacturing process). If only one (e.g., the one with the highest degree of abnormality) of the defect candidates detected by performing defect detection using the default parameters is captured, there is a possibility that the defect will not be captured in the high-quality image. Therefore, in the first embodiment, a high-quality image is acquired while changing the imaging area for each defect candidate detected from the inspection image. This increases the probability of acquiring a high-quality image in which the defect is captured. When multiple defect candidates are detected from one inspection image, the sample observation device 1 of the first embodiment captures a high-quality image corresponding to each defect candidate.

[0063] The flow in FIG. 9 has steps S901 to S906. First, in step S901, the upper control device 3 uses the number of images to be captured decision unit 207 to decide the number of images to be captured (referred to as N). This number of images to be captured (N) is the maximum number of high-quality images (high-quality images for learning) to be captured. In the first embodiment, the maximum number of high-quality images to be captured is the number of defect candidates to be captured. In step S901, the processor decides the number of images to be captured (N) to be associated with one inspection image, based on an input (in other words, a setting) by the user.

[0064] Fig. 10 shows an example of a GUI screen when the number of images to be captured is determined in step S901 in the first embodiment. On this screen, the user inputs the number of images to be captured (N) corresponding to the maximum number of high-quality images to be captured in area 1001. After inputting, the user can set the number of images to be captured (N) by pressing area 1002 which is a confirmation button. This number of images to be captured (N) is stored in the imaging parameter storage unit 213. In the example of Fig. 10, N=5.

[0065] In FIG. 9, after step S901, the upper control device 3 performs a loop process of steps S902 to S906 for each defect position (i). Step S902 is a reference image acquisition step. In step S902, the upper control device 3 acquires a reference image (learning reference image) of the defect position (i) under the first imaging condition stored in the imaging parameter storage unit 213 using the reference image acquisition unit 204. Step S903 is an inspection image acquisition step. In step S903, the upper control device 3 acquires an inspection image (learning low-quality image) of the defect position (i) under the first imaging condition stored in the imaging parameter storage unit 213 using the inspection image acquisition unit 205. Step S902 is the same process as the above-mentioned step S705, and step S903 is the same process as the above-mentioned step S706. The images acquired by these processes are stored in the image storage unit 210.

[0066] Step S904 is a defect detection step. In step S904, the host controller 32 performs a process of detecting defects in the inspection image using the inspection image and the reference image acquired in steps S902 and S903, and the defect detector 215. Step S904 is the same process as step S707 described above.

[0067] Next, in step S905, the upper control device 3 uses the imaging point determination unit 208 to determine one or more imaging points of multiple high quality images. The imaging point is j, where j = 1 to J. Thereafter, step S906 is a loop process of high quality image imaging. In this loop process, the upper control device 3 uses the high quality image imaging unit 209 to repeatedly obtain a high quality image for each imaging point (j). In step S906, the upper control device 3 obtains multiple high quality images for each imaging point (j) by imaging a high quality image at the imaging point (j) associated with the defect position (i) under the second imaging condition stored in the imaging parameter storage unit 213. The obtained high quality images are stored in the image storage unit 210. When the loop process for all defect positions (i) is completed, this image acquisition flow ends.

[0068] [1-9. Step S905: Deciding the imaging point] FIG. 11 shows a flow related to the high-quality image capturing point determination process in step S905. The flow in FIG. 11 has steps S1101 to 1103. First, in step S1101, the upper control device 3 determines the priority of the defect candidate in the inspection image (low-quality image for learning). For example, the degree of abnormality used in step S707 for defect detection can be used as this priority. It can be said that the greater the degree of abnormality, the greater the difference between the inspection image and the reference image, and the higher the probability of there being a defect. Therefore, in this example, the upper control device 3 calculates and sets so that the greater the degree of abnormality, the higher the priority of the corresponding defect candidate is.

[0069] Next, in step S1102, the upper control device 3 selects a defect candidate with a high priority. Specifically, the upper control device 3 reads the number of captured images (N) determined in step S901 from the imaging parameter storage unit 213, and selects up to N defect candidates corresponding to the number of captured images (N) in order of decreasing priority. For example, if the number of defect candidates in the defect image is 100 and the number of captured images (N) is 10, the top 10 defect candidates with the highest degree of abnormality are selected from the 100 defect candidates. Also, for example, if the number of defect candidates is 5 and N=10, 5 defect candidates are selected.

[0070] Next, in step S1103, the upper control device 3 determines one or more imaging points (j) capable of imaging up to N defect candidates selected in step S1102. The imaging points are, for example, imaging positions corresponding to the center point of the imaging area of ​​the learning high-quality image.

[0071] [1-10. Step S1103: Deciding the imaging point] FIG. 12 is an explanatory diagram of the process of determining the imaging point of the high-quality image in step S1103. First, the inspection image 1202 shown on the left side will be described. The inspection image 1202 is an inspection image under the first imaging condition corresponding to the defect position (i). In FIG. 12, the position of the defect candidate in the inspection image is represented by a white-filled x mark, like the defect candidate 1201. In this example, a plurality of defect candidates 1201, for example, five defect candidates 1201, are included in one inspection image 1202. Consider the case where a plurality of defect candidates 1201 are close to each other in the inspection image 1202, as in this example. The high-quality image imaging area 1203 is an area in which a high-quality image including the defect candidate 1201 is captured, and is illustrated as a rectangle.

[0072] The upper control device 3 determines a high-quality image capturing area 1203 centered on the position coordinates of each defect candidate 1201. In this example, since the multiple defect candidates 1201 are close to each other, the high-quality image capturing areas 1203 overlap significantly. For example, the two defect candidates 1201 indicated by a and b are relatively close to each other, and the two corresponding high-quality image capturing areas 1203 indicated by A and B have a relatively large overlapping area. In this case, some of the overlapping areas are imaged multiple times, resulting in poor imaging efficiency and a long setup time.

[0073] Therefore, in the first embodiment, the process of FIG. 11 is performed as in the inspection image 1204 shown on the right. When the plurality of defect candidates 1201 are close to each other in the inspection image 1204 and thus the plurality of defect candidates 1201 can be included in one high-quality image capturing area 1205, the upper control device 3 determines the center of gravity of the plurality of defect candidates 1201 included therein as the high-quality image capturing point 1206. The high-quality image capturing point 1206 is represented by a solid black x mark. The high-quality image capturing area 1205 is a rectangle with the high-quality image capturing point 1206 at its center. For example, for two defect candidates 1201 indicated by a and b, the corresponding high-quality image capturing points 1206 indicated by x are determined, and one high-quality image capturing area 1205 indicated by X is determined.

[0074] Through the above process, three high-quality image capturing areas 1205 are obtained in the right-side inspection image 1204 compared to five high-quality image capturing areas 1203 in the left-side inspection image 1202. In the right-side inspection image 1204, there is less overlap between the high-quality image capturing areas 1205, making it possible to capture multiple defect candidates 1201 (e.g., five) by capturing a smaller number of high-quality images (e.g., three). That is, the capturing efficiency is good, and as a result, the setup time can be shortened.

[0075] After the high-quality image capturing area 1205 and the high-quality image capturing point 1206 (=j) capable of capturing images of all the defect candidates selected in step S1102 are determined, information such as the coordinates and number of the capturing points (j) are stored in the capturing parameter memory unit 213.

[0076] [1-11. Step S703 Model Learning] 13 is an explanatory diagram of a configuration example of a high quality image estimation model related to the model learning in step S703. In the high quality image estimation process in the first embodiment, a correspondence relationship between an inspection image and a high quality image is machine-learned. Then, using the machine-learned model, an image captured under the second imaging condition is estimated from the inspection image or reference image captured under the first imaging condition.

[0077] The upper control device 3 performs a process of cutting out and enlarging the areas corresponding to the multiple high-quality images for each imaging point determined in step S905 for the inspection image captured in step S903, and creates a pair of the cut-out image and the high-quality image. Here, the image cut out from the inspection image is called a low-quality image. Although the low-quality image and the high-quality image each contain the same area, the low-quality image has been subjected to image processing such as enlarging the imaging field of view. Therefore, this low-quality image is, for example, a blurred image compared to the high-quality image.

[0078] In the first embodiment, a well-known deep learning is used as a method for implementing machine learning. Specifically, a convolutional neural network (CNN) is used as a model. Fig. 13 shows an example in which a neural network having a three-layer structure is used as a specific example of a CNN.

[0079] In the CNN in Fig. 13, Y indicates a low-quality image that is an input, and F(Y) indicates an estimated result that is an output. Furthermore, F1(Y) and F2(Y) indicate intermediate data between the input and the estimated result. The intermediate data F1(Y) and intermediate data F2(Y) and the estimated result F(Y) are calculated by the following formulas 1 to 3. Equation 1: F1(Y)=max(0,W1*Y+B1) Formula 2: F2(Y)=max(0,W2*F1(Y)+B2) Equation 3: F(Y)=W3*F2(Y)+B3

[0080] Here, the symbol * indicates a convolution operation, W1 indicates n1 filters of size c0×f1×f1, c0 indicates the number of channels of the input image, and f1 indicates the size of the spatial filter. In the CNN of FIG. 13, a n1-dimensional feature map is obtained by convolving a filter of size c0×f1×f1 with the input image Y n1 times. B1 is an n1-dimensional vector and is a bias component corresponding to the n1 filters. Similarly, W2 is n2 filters of size n1×f2×f2, B2 is an n2-dimensional vector, W3 is c3 filters of size n2×f3×f3, and B3 is a c3-dimensional vector. c0 and c3 are values ​​determined by the number of channels of the low-quality image and the high-quality image. In addition, f1, f2, n1, and n2 are determined by the user before the learning sequence. For example, f1=9, f2=5, n1=128, and n2=64 may be used. The above various parameters are examples of model parameters.

[0081] In step S703 of model learning, the model parameters to be adjusted are W1, W2, W3, B1, B2, and B3. The upper control device 3 sets a low-quality image, which is an image obtained by cutting out an image area from an inspection image captured under a first imaging condition and expanding the imaging field of view, as an input (Y) of the CNN model. The upper control device 3 adjusts the parameters of the CNN model using a high-quality image captured under a second imaging condition as an estimation result (F(Y)). The upper control device 3 stores, as an adjustment result, parameters in the high-quality image estimation parameter storage unit 212 such that the estimation result by the model is consistent with the high-quality image.

[0082] In the above model parameter adjustment, general error backpropagation may be used in neural network learning. In addition, when calculating the estimated error, all of the acquired learning image pairs (pairs of the above low-quality image and high-quality image) may be used, or a mini-batch method may be adopted. That is, several images may be randomly extracted from the learning image pairs, and the high-quality image estimation processing parameters may be repeatedly updated. Furthermore, a patch image may be randomly extracted from one learning image pair and used as an input image (Y) for the neural network. This allows efficient learning. Note that the configuration is not limited to the CNN configuration shown in this example, and other configurations may be used. For example, the number of layers may be changed, a network with four or more layers may be used, or a configuration with skip connections may be used.

[0083] [1-12. Step S704 Parameter Adjustment] FIG. 14 shows a detailed process flow for adjusting the defect detection parameters in step S704. The flow in FIG. 14 has steps S1401 to S1406. Here, an image estimated from an inspection image (a low-quality image for learning) using a high-quality image estimation model is called a high-quality inspection image (a high-quality image for learning). In this parameter adjustment, the upper control device 3 first performs a loop process of steps S1401 to S1403 on a defect candidate associated with the defect position (i) in order to determine a detailed defect position in the inspection image of the defect position (i). The defect candidate is k, where k=1 to K. At this time, by using the high-quality inspection image and a high-quality reference image corresponding to the high-quality inspection image, it is possible to determine the defect position with high accuracy.

[0084] [1-13. Step S1401 Acquiring high-quality inspection images] FIG. 15 is an explanatory diagram of the high-quality inspection image acquisition process in step S1401. One inspection image 2900 in FIG. 15 includes a plurality of defect candidates (k) 2901. In step S1401, the upper control device 3 acquires high-quality inspection images for a plurality of defect candidates (k) in the inspection image detected in step S904. Specifically, when the position of a defect candidate, such as the defect candidate 2901 indicated by k1, overlaps with (in other words, is included in) the imaging area 2902 of the high-quality inspection image determined in step S905, the high-quality inspection image corresponding to the defect candidate 2901 has been acquired in step S906, and therefore the high-quality inspection image is selected from the image storage unit 210. This process is referred to as a high-quality inspection image selection process.

[0085] If the number of defect candidates detected in step S904 is greater than the number of images N (the maximum number of high-quality images to be captured) determined in step S901, for example, defect candidate 2903 indicated by k3 in FIG. 15, a high-quality image including the defect candidate with a low priority has not been captured. Therefore, if the position of the defect candidate does not overlap (is not included in) the imaging area 2902 of the high-quality inspection image, the upper control device 3 estimates a high-quality inspection image using the high-quality image estimation unit 217. This process is referred to as a high-quality inspection image estimation process. The high-quality inspection image estimation process will be described in detail later.

[0086] After step S1401, in step S1402, the upper control device 3 acquires a high quality reference image corresponding to the high quality inspection image. The method of acquiring the high quality reference image can be a method of estimating a high quality reference image from a reference image or a method of synthesizing a high quality reference image from a high quality inspection image. In the first embodiment, either method may be applied. Details of the process of estimating a high quality reference image and the process of synthesizing a high quality reference image corresponding to these methods will be described later.

[0087] After step S1402, in step S1403, the host controller 3 performs defect discrimination processing using the high-quality inspection image and the high-quality reference image. In the defect discrimination processing, for example, the degree of abnormality obtained by comparing the high-quality inspection image with the high-quality reference image is used as an index value for discrimination. The greater the degree of abnormality, the greater the difference between the high-quality inspection image and the high-quality reference image. Therefore, the host controller 3 determines that the defect candidate in the inspection image corresponding to the high-quality inspection image with the maximum degree of abnormality is a defect. The index value for the defect discrimination processing is not limited to the degree of abnormality.

[0088] Next, in step S1404, the host controller 3 performs a process of determining a defect position in the inspection image based on the defect discrimination result obtained in step S1403. The host controller 3 determines the position on the inspection image of the defect candidate determined to be a defect by the defect discrimination process as the defect position.

[0089] After detailed defect positions are determined in all inspection images, in step S1405, the upper control device 3 evaluates whether or not defect positions in the inspection images can be detected for each defect detection parameter set (denoted as p). In the first embodiment, the number of detected defect positions determined in step S1404 is used as the evaluation index at this time.

[0090] Finally, in step S1406, the upper controller 3 stores and outputs the defect detection parameter resulting in the maximum number of detected defect positions in the defect detection parameter storage unit 211. This ends the parameter adjustment flow in FIG.

[0091] [1-14. Estimation processing of high-quality inspection images] The estimation process of the high quality inspection image will be described with reference to Fig. 16. First, the upper control device 3 determines an extraction region 1502 indicated by a dashed line, centered on a defect candidate in an inspection image 1501, and performs image extraction step S1503. In this example, the defect candidate is a portion that protrudes slightly horizontally from the vertical line as shown. The upper control device 3 then performs high quality inspection image estimation step S1505 on the extracted image 1504 using the high quality image estimation unit 217, thereby estimating and obtaining a high quality inspection image 1506.

[0092] [1-15. Method for estimating high-quality reference images] A method for estimating the high quality reference image will be described with reference to Fig. 17. First, the upper control device 3 performs step S1602 of determining a reference image cut-out region using an inspection image 1601. The inspection image 1601 includes a region 1603 from which a high quality inspection image was acquired in step S1401. Here, when the above-mentioned high quality inspection image selection process is performed, the imaging region of the selected high quality image is determined as the region 1603, and when the above-mentioned high quality inspection image estimation process is performed, the cut-out region in step S1503 of image cut-out in Fig. 16 is determined as the region 1603.

[0093] In step 1602, the upper control device 3 selects, from a reference image 1605 corresponding to the inspection image 1601, a cut-out area 1604 corresponding to an area 1603 from which a high-quality inspection image was acquired in step S1401, which is a defect candidate (k) in the inspection image 1601 at the defect position (i).

[0094] After determining the cropping region 1604, the upper control device 3 performs step S1606 of image cropping. Step S1606 is the same process as step S1503 in Fig. 16. Then, the upper control device 3 performs step S1608 of high quality reference image estimation on the cropped reference image 1607 using the high quality image estimation unit 217, thereby obtaining a high quality reference image 1609.

[0095] [1-16. How to synthesize high-quality reference images] A method for synthesizing the high-quality reference image will be described with reference to Fig. 18. In step S1701, the upper control device 3 performs high-quality reference image synthesis processing on the high-quality inspection image 1506 obtained in step S1401. This synthesis processing can be similar to the reference image synthesis processing described in Patent Document 1, for example. As a result, a high-quality reference image 1702 that does not include defects is synthesized. Note that "synthesis" here refers to processing in which a database is created in which reference images are collected in advance, and characteristic areas in the inspection image are replaced with similar areas in the database, but can be applied in other ways.

[0096] [1-17. User Interface] 1, a user operates the sample observation device 1 of the first embodiment using an input / output terminal 6. The user operates the sample observation device 1 by viewing a GUI screen provided by a user interface control unit 106 of the upper control device 3.

[0097] FIG. 19 shows an example of a GUI screen relating to image acquisition in step S702 in the first embodiment. In this screen, an area 1801 of an examination image list displays a list of IDs of examination images for which image acquisition has been completed. A user can select an ID from the list. An area 1802 of an examination image displays an examination image of an ID selected from the list. At this time, if a check box in an area 1803 of a check box for "display high-quality image capturing area" is selected, an area 1802 displays an imaging area of ​​a high-quality image, for example, as a dashed rectangle. Also, an area 1804 of a high-quality image displays a high-quality examination image (corresponding to the high-quality image capturing area in the area 1802) related to an examination image of an ID selected from the list. For example, a plurality of high-quality examination images are displayed in parallel. An area 1805 of a detector allows a detection target and an image type corresponding to the above-mentioned detector 111 to be selected.

[0098] FIG. 20 shows an example of a GUI screen relating to the high quality image estimation model learning in step S703 and the defect detection parameter adjustment in step S704. This screen is roughly divided into an area 1901 for high quality image estimation and an area 1910 for parameter adjustment. The area 1901 is used for learning the high quality image estimation model. The area 1902 is a button for instructing model learning, and is a button for manually calling the model learning unit 216 to execute the learning process of the high quality image estimation model. The area 1903 is a button for instructing high quality image estimation, and is a button for manually calling the high quality image estimation unit 217. The area 1904 displays an inspection image. The area 1905 displays a high quality image estimated using the high quality image estimation unit 217. The area 1906 displays a list of IDs of the inspection images. The area 1907 displays an acquired high quality image for comparison with the estimated high quality image in the area 1905. The upper controller 3 reads the acquired high quality image from the image storage unit 210 and displays it in this area 1907 .

[0099] An area 1910 is used for adjusting the defect detection parameters. In an area 1911 at the top of the area 1910, the defect detection parameters (multiple parameters) are displayed using parts such as sliders. In this example, the user can manually adjust each parameter value by changing the slider position of each parameter (#1 to #4). In addition, the default value for each parameter is displayed as a gray slider.

[0100] Area 1912 is a button for instructing automatic adjustment. When area 1912 is pressed, the upper control device 3 executes processing by the parameter adjustment unit 218 to automatically adjust the defect detection parameters. The upper control device 3 automatically changes the display of the slider values ​​(positions) of each parameter in the upper area 1911 according to the adjusted defect detection parameter values.

[0101] In area 1915, a list of inspection images is displayed as a list of IDs, and the success or failure (OK / NG) of detection of real defects in the inspection images is displayed for each execution item such as "Run 1" and "Run 2". In area 1916, defective areas (areas including defective parts) in the inspection images are displayed as areas surrounded by dashed lines. In area 1916, the user can also manually specify defective areas. Area 1913 is a button for instructing addition of a defective area, and is a button for calling a process for adding a defective area. In addition, area 1914 is a button for instructing deletion of a defective area, and is a button for calling a process for deleting a defective area on the inspection image. Area 1918 is an area for displaying the defect capture rate. Area 1919 is a button for instructing parameter output, and is a button for calling a process for storing the defect detection parameters set in area 1910 in defect detection parameter storage unit 211.

[0102] [1-18. Effects, etc.] As described above, in the sample observation device and method according to the first embodiment, the SEM 101, which is the imaging device 2, is used to acquire an inspection image, a reference image, and multiple high-quality images for each defect coordinate in the defect position information 8, the inspection image and the high-quality image are used to learn a high-quality image estimation model, and the defect detection parameters are adjusted using the high-quality inspection image and the high-quality reference image. According to the first embodiment, the user can easily improve the accuracy of defect detection. According to the first embodiment, the accuracy of the high-quality image estimation model can be improved, and the usability of creating a recipe for the ADR function can also be improved. According to the first embodiment, multiple high-quality images can be captured for each defect position indicated by the defect position information 8. Therefore, the possibility of capturing a defect in the inspection image in the high-quality image is improved, and the accuracy of the high-quality estimation model is improved. This improves the accuracy of the defect detection process parameter adjustment. Also, the accuracy of the high-quality image estimated in the defect observation is improved. Also, according to the first embodiment, the functions as described above are provided to the user, and the user can operate while viewing the above-mentioned GUI screen, thereby improving the usability during the defect observation work.

[0103] With regard to step S901 described above, the user can set the number of images (N) on the screen of FIG. 10. The effects of this include the following. In the ADR function using a machine learning model, there is generally a trade-off between accuracy and processing time regarding the number of images used. Increasing the number of images may improve accuracy, but the processing time will be longer. Therefore, in the first embodiment, the user can adjust the number of images (N), making it easy to adjust to the accuracy and processing time of the user's preference. The configuration is not limited to one in which the user can variably set the number of images (N), and as a modified example, a configuration in which the system including the sample observation device 1 automatically determines and sets the number of images (N) is also possible.

[0104] <Embodiment 2> The second embodiment will be described with reference to Figure 21 and subsequent figures. The basic configuration of the second embodiment etc. is the same as that of the first embodiment, and the following mainly describes the configuration of the second embodiment etc. that is different from the first embodiment.

[0105] In the second embodiment, when generating a reference image from an inspection image, both a low quality image and a high quality image are generated by synthesis. That is, in the second embodiment, a low quality reference image is generated by synthesis from a low quality inspection image, and a high quality reference image is generated by synthesis from a high quality inspection image. When acquiring images for defect observation, the fewer the number of images to be captured, the shorter the acquisition time and the higher the throughput. In the second embodiment, a reference image is generated from an inspection image, so that defect observation is performed without capturing a reference image. The GUI in the second embodiment can be similarly applied to GUIs such as those in FIGS. 10, 19, and 20 described in the first embodiment.

[0106] [2-1. Reference image acquisition part 2001] FIG. 21 shows a functional block configuration realized by the control unit 102 and the calculation unit 104 of the upper control device 3 in the second embodiment. The main component of the configuration of the second embodiment that differs from that of the first embodiment is the reference image acquisition unit. In the second embodiment, as shown in FIG. 21, the calculation unit 104 is provided with a reference image acquisition unit 2001. The reference image acquisition unit 2001 is provided in the calculation unit 104 in order to perform only calculation processing without controlling the SEM 101. The reference image acquisition unit 2001 reads the inspection image acquired under the first imaging condition in the inspection image acquisition unit 205 from the image storage unit 210 of the storage unit 103 (not shown; similar to FIG. 2), and executes a reference image synthesis process on the inspection image to acquire a reference image that does not include defects. The reference image synthesis process can be the same as the reference image synthesis process described in Patent Document 1, for example.

[0107] [2-2. Overall processing] A defect observation method executed in the sample observation device 1 of the embodiment 2 will be described. Fig. 22 shows the overall processing and operation of the sample observation device 1 of the embodiment 2. The flow in Fig. 21 has steps S2101 to S2108. At the start of defect observation, in step S2101 (corresponding to S701 in Fig. 7), it is determined whether a recipe related to the defect detection process has been set up. If it has not been set up, the process proceeds to step S2102, and if it has been set up, the process proceeds to step S2105.

[0108] In step S2102, the upper control device 3 acquires images to be used in high quality image estimation model learning and defect detection parameter adjustment using the reference image acquisition unit 2001, the inspection image acquisition unit 205, and the high quality image acquisition unit 206. In step S2103 (corresponding to S703 in FIG. 7), model learning is performed. In step S2104 (corresponding to S704 in FIG. 7), parameter adjustment is performed.

[0109] When the recipe has already been set up, or after the recipe has been set up by steps S2102 to S2104, a loop process of steps S2105 to S2108 is performed for each defect position (i) of the defect position information 8. In step S2105 (corresponding to S706 in FIG. 7), the upper control device 3 acquires an inspection image at the defect position (i) using the inspection image acquisition unit 205. In step S2106, the upper control device 3 performs a reference image synthesis process in the reference image acquisition unit 2001 using the inspection image read from the image storage unit 210 to obtain a reference image. The obtained reference image is stored in the image storage unit 210. In step S2107 (corresponding to S707 in FIG. 7), defect detection is performed. In step S2108 (corresponding to S708 in FIG. 7), high-quality image estimation is performed. When the processes of steps S2105 to S2108 are performed for all defect positions (i), this defect observation flow ends.

[0110] [2-3. Step S2102 Image Acquisition] 23 shows a flow of image acquisition processing in step S2102. The host controller 3 performs this image acquisition processing using the inspection image acquisition unit 205, the reference image acquisition unit 2001, and the high-quality image acquisition unit 206. First, in step S2201 (corresponding to S901 in FIG. 9), the host controller 3 determines the number of images (N) to be captured (the maximum number of high-quality images to be captured). After step S2201, a loop process of steps S2202 to S2206 is performed for each defect position (i) in the defect position information 8.

[0111] In step S2202 (corresponding to S903 in FIG. 9), the upper control device 3 acquires an inspection image of the defect position (i) under the first imaging condition stored in the imaging parameter storage unit 213. In step S2203, the upper control device 3 causes the reference image acquisition unit 2001 to synthesize a reference image of the defect position (i) using the inspection image read from the image storage unit 210. Step S2203 is the same process as step S2106 described above. The synthesized reference image is saved in the image storage unit 210.

[0112] Next, in step S2204 (corresponding to S904 in FIG. 9), the upper control device 3 performs defect detection, and in step S2205 (corresponding to S905 in FIG. 9), the imaging point (j) of a high quality image is determined. After the imaging point (j) is determined, in step S2206 (corresponding to S906 in FIG. 9), the upper control device 3 captures a high quality image at the imaging point (j) at the defect position (i) under the second imaging conditions stored in the imaging parameter storage unit 213. The captured high quality image is saved in the image storage unit 210. When the processes of steps S2202 to S2206 are completed for all defect positions, this image acquisition flow ends.

[0113] [2-4. Effects, etc.] As described above, according to the second embodiment, it is possible to synthesize a reference image by capturing only an inspection image. Therefore, according to the second embodiment, capturing a reference image can be omitted, and the throughput related to defect observation can be improved.

[0114] <Embodiment 3> The third embodiment will be described with reference to FIG. 24 and subsequent figures. In the first and second embodiments described above, a method of using a reference image to detect defects in an inspection image in high-quality image estimation model learning and defect detection parameter adjustment has been described. In the third embodiment, defect observation that does not require a reference image is realized by capturing the entire area of ​​the inspection image with a high-quality image. In the third embodiment, since the acquisition of a reference image is omitted, the throughput related to defect observation is improved. The ADR process in the third embodiment is realized using machine learning. The GUI in the third embodiment can be similarly applied to the GUI in the first embodiment described above.

[0115] [3-1. Functional blocks] 24 shows a functional block configuration of the control unit 102 and the calculation unit 104 of the higher-level control device 3 in the third embodiment. The main components that differ in the third embodiment are the configurations of the control unit 102 and the calculation unit 104, and first, the control unit 102 or the calculation unit 104 does not have a reference image acquisition unit. The configuration of the high-quality image acquisition unit 2301 is also different, and does not have an image capture number determination unit. In the third embodiment, an imaging point determination unit 2302 is provided in the high-quality image acquisition unit 2301 of the control unit 102 so as to eliminate the need for an image capture number determination unit. In addition, the processing and operation of the defect detection unit 2303 are also different from those in the first and second embodiments, and therefore the processing and operation of the parameter adjustment unit 2304 are also different.

[0116] [3-2. Defect Observation] A defect observation method executed by the specimen observation device 1 of the third embodiment will be described. FIG. 25 shows the overall processing and operation of the specimen observation device 1 of the third embodiment. At the start of defect observation, in step S2501 (corresponding to S701 in FIG. 7), it is determined whether a recipe related to the defect detection process has been set up. If it has not been set up, the process proceeds to step S2502, and if it has been set up, the process proceeds to step S2505. In step S2502, the upper control device 3 uses the inspection image acquisition unit 205 and the high-quality image acquisition unit 2301 to acquire images used in high-quality image estimation model learning and defect detection parameter adjustment. Then, in step S2503 (corresponding to S703 in FIG. 7), model learning is performed, and then, in step S2504 (corresponding to S704 in FIG. 7), parameter adjustment is performed. In the third embodiment, the model learning in step S2503 uses the same model as in the first embodiment (here, the first model). The parameter adjustment in step S2504 is the adjustment of parameters related to the defect detection process in step S2506.

[0117] When the recipe has already been set up, or after the recipe has been set up by steps S2502 to S2504, a loop process of steps S2505 to S2507 is performed for each defect position (i) of the defect position information 8. In step S2505 (corresponding to S706 in FIG. 7), the upper control device 3 uses the inspection image acquisition unit 205 to acquire an inspection image at the defect position (i) under the first imaging condition stored in the imaging parameter storage unit 213. In step S2506, the upper control device 3 uses the defect detection unit 2303 to detect defects in the inspection image. In the third embodiment, the defect detection process in step S2506 uses a model (here, a second model) different from the first model in step S2503. The defect detection process in step S2506 is a defect detection process by machine learning, unlike the defect detection process by image processing in the first embodiment, and therefore uses the second model by machine learning. In step S2507 (corresponding to S708 in FIG. 7), the upper control device 3 estimates a high-quality image. When the processes in steps S2505 to S2507 have been performed for all defect positions, this defect observation flow ends.

[0118] [3-3. Step S2502 Image Acquisition] 26 shows a flow of the image acquisition process in step S2502. The upper control device 3 performs this image acquisition process using the inspection image acquisition unit 205 and the high-quality image acquisition unit 2301. The sample observation device 1 of the third embodiment requires, for each defect position (i) in the defect position information 8, an inspection image including the defect site, and multiple high-quality images whose imaging ranges are determined so as to image the entire imaging range of the inspection image. Therefore, in step S2502, those images are acquired.

[0119] In the third embodiment, the number of images (N) to be captured is determined by the ratio of the size of the imaging range (in other words, the field of view range) of the inspection image and the high-quality image (described later). Therefore, in image acquisition step S2502, the user does not need to determine the number of images to be captured (N). In this step S2502, a loop process of steps S2601 to S2603 is executed for each defect position (i) in the defect position information 8.

[0120] In inspection image acquisition step S2601 (corresponding to S903 in FIG. 9), the upper controller 3 uses the inspection image acquisition unit 205 to acquire an inspection image of the defect position (i) under the first imaging condition stored in the imaging parameter storage unit 213. Step S2601 is the same as step S706 described above. The acquired image is stored in the image storage unit 210.

[0121] Thereafter, in step S2602, the host controller 3 determines imaging points (j) of a high-quality image using the imaging point determination unit 2302. In this step S2602, unlike the first and second embodiments, the host controller 3 determines the minimum number of imaging points (j) that can capture the entire imaging range of the inspection image. Details of step S2602 will be described later.

[0122] After the imaging point (j) is determined, in step S2603 (corresponding to S906 in FIG. 9), the host controller 3 uses the high quality image capturing unit 209 to capture a high quality image at the imaging point (j) corresponding to the defect position (i) under the second imaging conditions stored in the imaging parameter storage unit 213. The captured high quality image is saved in the image storage unit 210. When the processes of steps S2601 to S2603 are completed for all defect positions, this image acquisition flow ends.

[0123] [3-4. Step S2602: Deciding the imaging point] Fig. 27 is an explanatory diagram relating to the imaging point determination process of step S2602 in the third embodiment. Fig. 27 shows an example in which the size ratio of the imaging range (field of view range) in the inspection image and the high-quality image is 2. In this example, the inspection image 2701 is a square, and the horizontal size (expressed in number of pixels) is denoted by SX1. The high-quality image (shown by the dashed line) is a square, and the horizontal size is SX2, which is half of SX1. For example, the size ratio is SX1:SX2, and the size ratio is SX1 / SX2=2.

[0124] 27, when imaging points are determined using the inspection image 2701, the upper control device 3 configures regions 2702 by dividing the inspection image 2701 vertically and horizontally into the same number as the size ratio (=2). In this example, four regions 2702 are configured to cover the entire area of ​​the inspection image 2701. Then, the upper control device 3 determines multiple (four) imaging points (j) 2703 indicated by x marks so that these regions 2702 can be imaged as high-quality image regions.

[0125] [3-5. Step S2506 Defect detection] FIG. 28 is an explanatory diagram regarding the defect detection process in step S2506. In the defect detection process in the third embodiment, a machine learning model is used to detect defects in the inspection image. A known deep learning is used as a method for implementing machine learning for defect detection in the third embodiment. Specifically, a CNN as shown in FIG. 28 is used. In the defect detection in the third embodiment, the output is a defect detection result image (G(Z)). The defect detection result image is a binary image showing an area detected as a defect, and in the case of an 8-bit grayscale image, for example, the pixel value of the defect detection area is 255, and the pixel value of the other areas is 0. The output defect detection result image is stored in the image storage unit 210.

[0126] In Fig. 28, Z indicates an inspection image that is input, and G(Z) is a defect detection result image that is output as an estimation result. Furthermore, G1(Z) and G2(Z) indicate intermediate data between the input and the estimation result. The intermediate data G1(Z) and intermediate data G2(Z) and the defect detection result image G(Z), which is the estimation result, are calculated by the following formulas 4 to 6. Equation 4: G1(Z)=max(0,V1*Z+A1) Equation 5: G2(Z)=max(0,V2*G1(Z)+A2) Equation 6: G(Z)=V3*G2(Z)+A3

[0127] Here, the symbol * indicates a convolution operation, V1 indicates m1 filters of size d0×g1×g1, d0 indicates the number of channels of the input image, and g1 indicates the size of the spatial filter. By convolving the input image Z with the filters of size d0×g1×g1 m1 times, an m1-dimensional feature map is obtained. A1 is an m1-dimensional vector and is a bias component corresponding to the m1 filters. Similarly, V2 is m2 filters of size m1×g2×g2, A2 is an m2-dimensional vector, V3 is one filter of size m2×g3×g3, and A3 is a one-dimensional vector. g1, g2, m1, and m2 are values ​​determined by the user before defect detection, and may be, for example, g1=9, g2=5, m1=128, and m2=64.

[0128] In the parameter adjustment unit 2304 in the third embodiment, the model parameters to be adjusted are V1, V2, V3, A1, A2, and A3. The present invention is not limited to the above-mentioned CNN configuration.

[0129] [3-6. Step S2504 Parameter Adjustment] FIG. 29 shows a flow related to the defect detection parameter adjustment process in step S2504. FIG. 29 has steps S2801 to S2805. In the parameter adjustment in the third embodiment, first, in order to determine a detailed defect position in the inspection image of the defect position (i), step S2801 of selecting a high-quality inspection image and step S2802 of discriminating defects are performed for the imaging point (j) of the high-quality image. In step S2801, since the high-quality image corresponding to the entire region of the inspection image is acquired in step S2502 of acquiring an image, only the selection process of the high-quality image is performed. In step S2802, the upper control device 3 performs a defect discrimination process using the high-quality inspection image. In the defect discrimination process in the third embodiment, the defect detection unit 2303 uses the high-quality inspection image to acquire a defect detection result image corresponding to the high-quality inspection image. At this time, the defect detection parameters have not yet been adjusted for the target manufacturing process. Therefore, for example, the defect detection process is applied to the high-quality inspection image using the defect detection parameters previously adjusted in a different manufacturing process.

[0130] In step S2803 (corresponding to S1404 in FIG. 14), the host controller 3 performs a process of determining the defect position in the inspection image using the defect detection result image of the high-quality inspection image obtained in step S2802. In step S2804, the host controller 3 creates a defect detection result image corresponding to the inspection image using the detailed defect position on the inspection image determined in step S2803. The created defect detection result image is stored in the image storage unit 210.

[0131] After creating corresponding defect detection result images for all inspection images, finally, in step S2805 (corresponding to S1406 in FIG. 14), the host controller 3 adjusts the defect detection parameters using the above-mentioned model, with the inspection images as input and the defect detection result images as estimated results. The process using the CNN in FIG. 28 in step 2805 can apply various methods, similar to the process using the CNN in FIG. 13 described above, and can improve efficiency, etc. After the learning of the neural network is completed, the host controller 3 stores the defect detection parameters in the defect detection parameter storage unit 211 as the adjustment results, and this parameter adjustment flow ends.

[0132] [3-7. Effects, etc.] As described above, according to the third embodiment, by learning and adjusting the parameters of a high-quality image estimation model that does not acquire a reference image, it is possible to omit the acquisition of a reference image, and the throughput of the sample observation device can be improved.

[0133] Although the present invention has been specifically described based on the embodiment, the present invention is not limited to the above-described embodiment and can be modified in various ways without departing from the gist of the present invention. [Explanation of symbols]

[0134] 1...sample observation device (review SEM), 2...imaging device, 3...host control device (computer system), 4...storage media device, 5...defect classification device, 6...input / output terminal, 7...inspection device, 8...defect position information, 9...sample (semiconductor wafer), 101...SEM, 101...control unit.

Claims

1. An imaging device; A processor that executes a learning process for learning a high quality image estimation model and a sample observation process for detecting defects; A sample observation device comprising: (A) The learning process includes: (A1) obtaining one or more learning defect positions for a learning sample; (A2) acquiring a learning low-quality image under a first imaging condition for each of the learning defect positions; (A3) acquiring a first setting value related to the number of high-quality learning images to be captured; (A4) For each learning defect position: (A4a) determining the number of high-quality learning images to be captured based on the first set value; (A4b) determining one or more imaging points, which are positions for capturing the learning high-quality images, based on the number of images determined in (A4a); (A4c) acquiring the learning high-quality image under a second imaging condition for each of the one or more imaging points determined in (A4b); (A5) training the high quality image estimation model using the low quality training images and the high quality training images; (A6) adjusting defect detection parameters using the high quality image estimation model; (B) the sample observation process includes, based on the adjusted defect detection parameters: (B1) acquiring a first inspection image of a defect position of the observation target sample under the first imaging condition; (B2) detecting defect candidates in the observation target specimen based on the first inspection image; Sample observation device.

2. 2. The sample observation device according to claim 1, The processor sets the first setting value related to the number of images to be captured based on an input by a user. Sample observation device.

3. 2. The sample observation device according to claim 1, When determining the imaging point, the processor determining a priority for each of the defect candidates based on a feature amount of the defect candidates in the learning low-quality image; selecting a plurality of defect candidates to be imaged under the second imaging condition from the plurality of defect candidates in the learning low-quality image based on a priority of each of the defect candidates so that the number of images is equal to or less than the number of images; determining the imaging points for imaging the selected plurality of defect candidates; Sample observation device.

4. 4. The sample observation device according to claim 3, When determining the imaging point, the processor determining the imaging point such that, when several defect candidates are closely located in the learning low-quality image, a region centered on the imaging point becomes one region including the several defect candidates; Sample observation device.

5. 2. The sample observation device according to claim 1, The learning process includes: acquiring, for each of the learning defect positions, a learning reference image that does not include a defect and that is associated with the learning low-quality image under the first imaging condition; training the model using the training reference images; The sample observation process includes: acquiring a first reference image that does not include a defect and that is associated with the first inspection image under the first imaging condition for each defect position of the observation target sample; calculating positions and feature amounts of the defect candidate based on a comparison between the first inspection image and the first reference image; Sample observation device.

6. 6. The sample observation device according to claim 5, The processor, When acquiring the learning reference image, the learning reference image is acquired by capturing an image under the first capturing condition using the imaging device; When acquiring the first reference image, the first reference image is acquired by capturing an image under the first imaging condition using the imaging device. Sample observation device.

7. 6. The sample observation device according to claim 5, The processor, When obtaining the learning reference image, synthesizing the learning reference image from the learning low-quality image; When acquiring the first reference image, synthesizing the first reference image from the first inspection image; Sample observation device.

8. 6. The sample observation device according to claim 5, The processor, when adjusting the defect detection parameters, Extracting an image of the defect candidate from the learning low-quality image; Estimating the learning high-quality image from the cut-out image using the high-quality image estimation model, or selecting a learning high-quality image including the defect candidate from the acquired learning high-quality images; estimating the learning reference image from the learning reference image for a position associated with the defect candidate using the high quality image estimation model, or synthesizing the learning reference image from the learning high quality image; using the learning high-quality image and the learning reference image, discriminating whether the defect candidate is a defect or not, thereby determining a defect position in the learning high-quality image; selecting a parameter capable of detecting the determined defect location based on the evaluation; Sample observation device.

9. 9. The sample observation device according to claim 8, the processor displays on a screen the low-quality learning image, the high-quality learning image, the estimation result of a high-quality inspection image from the first inspection image by the high-quality image estimation model, the adjustment result of the defect detection parameters, and defect positions corresponding to defect candidates detected in the first inspection image after the adjustment of the defect detection parameters. Sample observation device.

10. A sample observation method in a sample observation device including an imaging device and a processor that executes a learning process for learning a high-quality image estimation model and a sample observation process for performing defect detection, comprising: As a step executed by the sample observation device, (A) The learning process includes: (A1) obtaining one or more learning defect positions for a learning sample; (A2) acquiring a learning low-quality image under a first imaging condition for each of the learning defect positions; (A3) acquiring a first setting value related to the number of high-quality learning images to be captured; (A4) For each learning defect position: (A4a) determining the number of high-quality learning images to be captured based on the first set value; (A4b) determining one or more imaging points, which are positions for capturing the learning high-quality images, based on the number of images determined in (A4a); (A4c) acquiring the learning high-quality image under a second imaging condition for each of the one or more imaging points determined in (A4b); (A5) training the high quality image estimation model using the low quality training images and the high quality training images; (A6) adjusting defect detection parameters using the high quality image estimation model; (B) the sample observation process includes, based on the adjusted defect detection parameters: (B1) acquiring a first inspection image of a defect position of the observation target sample under the first imaging condition; (B2) detecting defect candidates in the observation target specimen based on the first inspection image; Sample observation method.

Images