Sample observation device and method

Abstract

The present invention provides a sample observation device and method, providing a technique for enabling a user to easily improve the accuracy of defect detection related to the sample observation device. In a learning process performed prior to sample observation processing, the sample observation device acquires a low-quality learning image under a first shooting condition at a defect location indicated by defect location information. The device determines the number of shots and the number of shooting points for a plurality of high-quality learning images corresponding to the low-quality learning image at each defect location based on a set number of shots. The device acquires a plurality of high-quality learning images under a second shooting condition. A high-quality image estimation model is learned using the low-quality learning images and the plurality of high-quality learning images. The parameters related to defect detection in the sample observation processing are adjusted using the high-quality image estimation model.

Description

Technical Field

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

[0002] For example, in the manufacture of semiconductor wafers, in order to ensure profitability, it is important to quickly start the manufacturing process and transfer to a high-yield mass production system as soon as possible. For this purpose, various inspection devices, observation devices, and measuring devices are introduced into the manufacturing production line. For example, a semiconductor wafer as a sample is inspected for defects in 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. Based on the defect position information, the sample observation device captures the position and location of the defect in the semiconductor wafer as the object with high resolution. As the sample observation device, for example, an observation device using a scanning electron microscope (SEM) (also called a review SEM) is used.

[0003] Automating inspection work using a review SEM is desirable in mass production lines for semiconductor wafers. To achieve this automation, systems incorporating a review SEM feature automatic defect review (ADR) and automatic defect classification (ADC). The ADR function automatically collects images of defect locations within a sample. The ADC function automatically classifies defects based on the collected images.

[0004] The defect coordinates of the defect position information supplied from the inspection device contain errors. Therefore, in the ADR function, the review SEM first takes a picture with a wide field of view and low magnification under the first shooting condition, centering on the defect coordinates of the defect position information, and re-detects the defect from the image obtained thereby (in other words, a low-quality image). In the ADR function, the review SEM takes a picture of the defect portion obtained by the re-detection with a narrow field of view and high magnification under the second shooting condition, and outputs the image obtained thereby (in other words, a high-quality image) as an observation image. In this way, there is a method of observing through two-stage images, such as searching for defects with a wide-field image in the first stage and observing the defects in detail with a high-quality image in the second stage. In addition, the above-mentioned high and low image quality, etc. are relatively defined, and are image quality, etc. corresponding to the shooting conditions, for example, it means that the magnification under the second shooting condition is higher than the magnification under the first shooting condition. The concept of image quality (image quality) includes magnification, resolution, signal-to-noise ratio, etc.

[0005] The following method is a method for determining and detecting defects based on images captured by the sample observation device (also referred to as inspection images). This method uses an image captured of the same area where a circuit pattern is formed as the defective area as a reference image, and compares the inspection image captured of the defective area with a reference image that does not contain the defect, thereby determining and detecting the defect. Japanese Patent Application Laid-Open No. 2009-250645 (Patent Document 1) describes a method that uses inspection images to synthesize a reference image, thereby omitting the need to capture the reference image.

[0006] The ADR function requires adjusting the processing parameters related to defect detection based on the appearance of the pattern formed on the semiconductor wafer. One method for automatically adjusting these processing parameters involves searching for processing parameters that can detect defects pre-set by the user. Another method uses inspection images of the defective area and multiple reference images to search for processing parameters that can accurately distinguish between defects and obstructions.

[0007] To improve the throughput of the ADR function, there is a method that eliminates the need for image capture by estimating high-resolution images. This method first obtains an inspection image captured with a wide field of view and a high-quality image captured at a high magnification, and then creates a model by learning the relationship between them. Furthermore, during ADR execution, this method captures only the inspection image, and uses the learned model to estimate the observation image as a high-quality image.

[0008] For example, in semiconductor device manufacturing, efforts to increase integration density continue to improve device performance and reduce manufacturing costs, leading to a trend toward miniaturization of the circuit patterns formed on semiconductor wafers. This trend is accompanied by a trend toward miniaturization of the size of defects that can be critical to device operation. Consequently, the ADR function of an inspection SEM also requires high-quality observation images that can visually identify minute defects. Therefore, existing technologies related to the ADR function of an inspection SEM include methods that use machine learning to infer and generate high-quality observation images from low-quality inspection images.

[0009] However, existing technologies using this method leave room for improvement in terms of accuracy. In this method, a model for estimating high-quality observation images from low-quality inspection images is learned before adjusting processing parameters related to defect detection. This method uses the initial (default) processing parameters before these adjustments to acquire inspection images and high-quality images for learning. In this case, defects present in the inspection images may not be reflected in the high-quality images. Consequently, the accuracy of the model decreases, making it impossible to perform high-precision estimation.

[0010] Patent Document 1: Japanese Patent Application Laid-Open No. 2009-250645 Summary of the Invention

[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.

[0012] A representative embodiment of the present invention has the structure shown below. The sample observation device of the embodiment has: a photographing device; and a processor, which performs a learning process of learning a high-quality image estimation model and a sample observation process of performing defect detection. In the learning process (A), (A1) one or more learning defect positions related to the learning sample are obtained, (A2) according to the learning defect position, a learning low-quality image under a first shooting condition is obtained, (A3) a first setting value related to the number of shots of the learning high-quality image is obtained, (A4) according to the learning defect position, (A4a) the number of shots of the learning high-quality image is determined according to the first setting value, and (A4b) the number of shots of the learning high-quality image is determined according to the number of shots determined in (A4a). The position of the used high-quality image is one or more shooting points, (A4c) for each of the one or more shooting points determined in (A4b), the learning high-quality image under the second shooting condition is obtained, (A5) the learning low-quality image and the learning high-quality image are used to learn the high-quality image inference model, (A6) the defect detection parameters are adjusted using the high-quality image inference model, and in the sample observation process (B), according to the adjusted defect detection parameters, (B1) under the first shooting condition, a first inspection image of the defect position of the observation object sample is obtained, and (B2) based on the first inspection image, the defect candidate of the observation object sample is detected.

[0013] According to the representative embodiment of the present invention, the technology of the sample observation device allows the user to easily improve the accuracy of defect detection. Other issues, structures, and effects, etc., as described above, are shown in the embodiments for carrying out the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 This is a diagram showing the configuration of a sample observation device according to Embodiment 1 of the present invention.

[0015] Figure 2 This is a diagram showing a functional block configuration of a host control device in the first embodiment.

[0016] Figure 3 This is a diagram showing the arrangement of detectors of the SEM in the first embodiment as viewed from an oblique upper direction.

[0017] Figure 4 This is a diagram showing the arrangement of detectors in the SEM as viewed from the z-axis direction in the first embodiment.

[0018] Figure 5 This is a diagram showing the arrangement of detectors of the SEM as viewed from the y-axis direction in the first embodiment.

[0019] Figure 6 This is an explanatory diagram schematically showing an example of defect coordinates on the semiconductor wafer surface represented by defect position information of an inspection device.

[0020] Figure 7 This is a flowchart showing the overall processing of the sample observation device according to the first embodiment.

[0021] Figure 8 This is an explanatory diagram related to step S707 of defect detection in the first embodiment.

[0022] Figure 9 This is a flowchart of step S702 of image acquisition in the first embodiment.

[0023] Figure 10 This is a diagram showing an example of a screen related to determination of the number of shots in the first embodiment.

[0024] Figure 11 This is a flowchart of step S905 of photographing point determination in the first embodiment.

[0025] Figure 12 This is an explanatory diagram related to the imaging point determination process in the first embodiment.

[0026] Figure 13 This is an explanatory diagram showing a configuration example of a high-quality image estimation model in the first embodiment.

[0027] Figure 14 This is a flowchart of step S704 of parameter adjustment in the first embodiment.

[0028] Figure 15 This is an explanatory diagram related to the process of acquiring a high-quality inspection image in the first embodiment.

[0029] Figure 16 This is a schematic diagram showing an example of the estimation process of the high-quality inspection image in the first embodiment.

[0030] Figure 17 This is a schematic diagram showing an example of a high-quality reference image estimation process in the first embodiment.

[0031] Figure 18 This is a schematic diagram showing an example of the synthesis process of a high-quality reference image in the first embodiment.

[0032] Figure 19 This is a diagram showing an example of a screen related to image acquisition in the first embodiment.

[0033] Figure 20 This is a diagram showing an example of a screen related to model learning and parameter adjustment in the first embodiment.

[0034] Figure 21 This is a diagram showing a functional block configuration of a host control device in the second embodiment.

[0035] Figure 22 This is a flowchart showing the overall processing of the sample observation device in the second embodiment.

[0036] Figure 23 This is a flowchart of step S2102 of image acquisition in the second embodiment.

[0037] Figure 24 This is a diagram showing a functional block configuration of a host control device in the third embodiment.

[0038] Figure 25 This is a flowchart showing the overall processing of the sample observation device in the third embodiment.

[0039] Figure 26 This is a flowchart of step S2502 of image acquisition in the third embodiment.

[0040] Figure 27 This is an explanatory diagram related to the imaging point determination process in the third embodiment.

[0041] Figure 28 This is an explanatory diagram showing a structural example of a model in Implementation Method 3.

[0042] Figure 29 This is a flowchart of step S2504 of parameter adjustment in the third embodiment.

[0043] Explanation of symbols

[0044] 1…sample observation device (review SEM), 2…photographing device, 3…upper-level control device (computer system), 4…storage medium device, 5…defect classification device, 6…input / output terminal, 7…inspection device, 8…defect position information, 9…sample (semiconductor chip), 101…SEM, 101…control unit. DETAILED DESCRIPTION

[0045] Below, with reference to the attached Figure 1The embodiments of the present invention will be described in detail. In the drawings, identical parts are generally designated by identical reference numerals, and repeated descriptions will be omitted. In the drawings, to facilitate understanding of the invention, the representation of each component may not represent its actual position, size, shape, or range. In the description, when describing program-based processing, the program, function, processing unit, etc. may be used as the main subject, but the hardware underlying these components is the processor, or a controller, device, computer, system, etc. comprised of the processor, etc. The computer executes processing according to the program read from the memory while appropriately utilizing resources such as memory and communication interfaces through the processor. This implements the predetermined functions, processing units, etc. The processor may be comprised of, for example, semiconductor devices such as a CPU and a GPU. The processor may be comprised of devices and circuits capable of performing predetermined operations. Processing is not limited to software program processing and may also be performed by dedicated circuits. Dedicated circuits may employ FPGAs, ASICs, CPLDs, etc. The program may be pre-installed as data in the target computer, or it may be distributed and installed as data from a program source to the target computer. The program source may be a program distribution server on a communication network, or a non-transitory computer-readable storage medium (such as a memory card). The program may also be composed of multiple modules. The computer system is not limited to one device, but may also be composed of multiple devices. The computer system may also 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 tables and lists, but are not limited to these. Expressions such as identification information, identifiers, IDs, names, and numbers can be interchangeable.

[0046] <Implementation Method>

[0047] A sample observation device according to an embodiment is a device for observing circuit patterns and defects formed on a sample, such as a semiconductor wafer. The sample observation device comprises: an inspection image acquisition unit that acquires an inspection image (in other words, a low-quality image) under a first imaging condition, corresponding to the defect location in defect location information generated and output by the inspection device; a high-quality image acquisition unit that acquires multiple high-quality images under a second imaging condition, corresponding to the defect location; a defect detection unit that calculates the location and feature value of a defect candidate using the inspection image; a model learning unit that learns a model (a high-quality image estimation model) that uses the inspection image and the high-quality image to estimate an image acquired under the second imaging condition from the image acquired under the first imaging condition; and a parameter adjustment unit that uses the model to adjust parameters related to the defect detection process of the defect detection unit.

[0048] The sample observation method of the embodiment is a method comprising steps executed by the sample observation device of the embodiment, and includes steps corresponding to the aforementioned components. The processing and corresponding steps in the sample observation device generally include sample observation processing (in other words, defect detection processing, etc.) and learning processing. Learning processing involves performing machine learning-based model learning on the images used in the sample observation processing and adjusting parameters related to the sample observation processing.

[0049] The following describes an example of a sample observation device that uses a semiconductor wafer as a sample to observe defects and the like. This sample observation device includes an imaging device that images the sample based on defect coordinates obtained from defect location information obtained from a defect inspection device. The following describes an example using a SEM as the imaging device. The imaging device is not limited to an SEM and may also be a device other than an SEM, such as an imaging device that uses charged particles such as ions.

[0050] <Implementation Method 1>

[0051] use Figures 1 to 20 , the sample observation device of embodiment 1 is described. In the sample observation device of the comparative example, before adjusting the processing parameters related to the ADR processing for sample defect observation, high-quality images for learning the high-quality image estimation model are collected. In this structure, sometimes the defects are not reflected in the high-quality image, in which case the accuracy of the model is reduced. Therefore, the sample observation device of embodiment 1 (for example Figure 2 ) includes a high-quality image acquisition unit 206. The high-quality image acquisition unit 206 includes a number-of-shot determination unit 207 that determines the number of high-quality image shots; a shooting point determination unit 208 that determines multiple shooting points for the high-quality image based on the position and feature value of the defect candidate calculated by the defect detection unit 215 and the number of shots; and a high-quality image capturing unit 209 that captures a high-quality image under the second shooting condition at each of the multiple determined shooting points. This improves the accuracy of the model, and by adjusting the parameters of the model, the accuracy of sample defect observation can be improved.

[0052] In the first embodiment, the defect detection process of the ADR function (described later) Figure 7 Step S707) is implemented using general image processing (i.e., determination based on the difference between the inspection image and 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). Furthermore, as described above, the relationship between image quality and other qualities in the two stages is defined relatively. Furthermore, the shooting conditions that define image quality and other qualities (in other words, image quality conditions) are not limited to situations where actual shooting is performed by a camera, but are also applicable when images are produced through processing.

[0053] [1-1. Sample observation device]

[0054] Figure 1 The structure of the sample observation device 1 of embodiment 1 is shown. The sample observation device 1 is generally composed of a photographing device 2 and a host control device 3. As a specific example, the sample observation device 1 is a review SEM. As a specific example, the photographing device 2 is SEM101. The photographing device 2 is combined with the host control device 3. The host control device 3 is a device that controls the photographing device 2, etc., in other words, it is a computer system. The sample observation device 1, etc. has necessary functional blocks and various devices, but a part of the necessary elements is illustrated in the accompanying drawings. Including Figure 1 The entire sample observation device 1 is constituted as a defect inspection system. The host control device 3 is connected to the storage medium device 4 and the input / output terminal 6, and is connected to the defect classification device 5, the inspection device 7, etc. via a network.

[0055] The sample observation device 1 is a device or system with an automatic defect review (ADR) function. In this example, defect location information 8 is pre-produced as a result of inspecting a sample in an external inspection device 7, and this defect location information 8 output and provided by the inspection device 7 is pre-stored in the storage medium device 4. During the ADR processing related to defect observation, the upper control device 3 reads and refers to this defect location information 8 from the storage medium device 4. The SEM 101 serving as the imaging device 2 captures an image of a semiconductor wafer serving as 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.

[0056] The defect classification device 5 (in other words, the defect image classification device) is a device or system with an automatic defect classification (ADC) function. It performs ADC processing based on the information and data of the defect observation processing results using the ADR function of the sample observation device 1 to obtain the results of classifying the defects (corresponding defect images). The defect classification device 5 supplies the information and data of the classification results to other devices (not shown) connected to the network, for example. In addition, it is not limited to Figure 1 The structure may also be a structure in which the defect classification device 5 is integrated into the sample observation device 1.

[0057] The upper control device 3 includes: a control unit 102, a storage unit 103, a calculation unit 104, an external storage medium input and output unit 105 (in other words, an input and output interface unit), a user interface control unit 106, and a network interface unit 107. These components are connected to a bus 114 and can communicate with each other and perform input and output. Figure 1In the example shown in FIG, the host control device 3 is constituted by one computer system, but the host control device 3 may be constituted by a plurality of computer systems (eg, a plurality of server devices).

[0058] The control unit 102 functions as a controller for controlling the entire sample observation device 1. The storage unit 103 stores various information and data, including programs, and is comprised of a storage medium device such as a magnetic disk or 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 include a processor and memory. The external storage medium input / output unit (in other words, the input / output interface unit) 105 inputs and outputs data to and from the external storage medium device 4.

[0059] 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). The user interface control unit 106 is connected to the input / output terminal 6. The user interface control unit 106 can be connected to other input devices or output devices (for example, a display device). The network interface unit 107 is connected to the defect classification device 5, the inspection device 7, etc. via a network (for example, a LAN). The network interface unit 107 is a part having a communication interface for controlling communication with external devices such as the defect classification device 5 via a network. Other examples of external devices include a DB server, an MES (manufacturing execution system), etc.

[0060] The user uses the input / output terminal 6 to input information (e.g., instructions, settings) into the sample observation device 1 (particularly the host control device 3) and to confirm information output from the sample observation device 1. The input / output terminal 6 can be, for example, a PC equipped with a keyboard, mouse, and display. The input / output terminal 6 can also be a client computer connected to a network. The user interface control unit 106 creates a GUI screen, described below, and displays it on the display device of the input / output terminal 6.

[0061] The calculation unit 104 is composed of, for example, a CPU, ROM, and RAM, and operates according to the program read from the storage unit 103. The control unit 102 is composed of, for example, a hardware circuit or a CPU. In the case where the control unit 102 is composed of a CPU, the control unit 102 also operates according to the program read from the storage unit 103. The control unit 102 realizes each function (each functional block described later) according to program processing. Data such as programs are supplied from the storage medium device 4 to the storage unit 103 via the external storage medium input / output unit 105 and stored. Alternatively, data such as programs can also be supplied from the network to the storage unit 103 via the network interface unit 107 and stored.

[0062] The SEM 101 constituting the photographing device 2 includes: a workbench 109, an electron source 110, a detector 111, an electron lens (not shown), and a deflector 112. The workbench 109 (in other words, the sample stage) is a workbench on which a semiconductor wafer serving as the sample 9 is placed and which is movable at least in the horizontal direction. The electron source 110 is an electron source for irradiating the sample 9 with an electron beam. The electron lens (not shown) converges the electron beam on 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 or 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, as the detector 111, a plurality of detectors (described later) are provided as shown in the figure.

[0063] Information detected by the detector 111 of the SEM 101 (in other words, image signals) is supplied to the bus 114 of the host control device 3. This information is processed by the computing unit 104 and other components. In this example, the host control device 3 controls the stage 109, deflector 112, and detector 111 of the SEM 101. The drive circuits for driving the stage 109 and other components are not shown. The computer system serving as the host control device 3 processes the information (in other words, images) from the SEM 101 to implement defect observation on the sample 9.

[0064] This system can also be set up as follows. The upper control device 3 is set as a server of a cloud computing system, etc., and the input and output terminal 6 operated by the user is set as a client computer. For example, in a case where more computer resources are required in machine learning, machine learning processing can also be performed in a server group of a cloud computing system, etc. The processing function can also be shared between the server group and the client computer. The user operates the client computer, and the client computer sends a request to the server. The server receives the request and performs processing corresponding to the request. For example, the server sends the data of the requested screen (such as a web page) to the client computer as a response. The client computer receives the response data and displays the screen (such as a web page) on the display screen of the display device.

[0065] [1-2. Functional blocks]

[0066] Figure 2 In the embodiment 1, Figure 1 Example of a functional block structure implemented by the control unit 102, storage unit 103 and operation unit 104 of the upper control device 3. In the control unit 102 and the operation unit 104, the function blocks are configured according to program processing. Figure 2 Such functional blocks are constituted in the storage unit 103 Figure 2The control unit 102 includes a stage control unit 201, an electron beam scanning 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 includes a number of shots determination unit 207, an imaging point determination unit 208, and a high-quality image capturing unit 209. The calculation unit 104 includes a defect detection unit 215, a model learning unit 216, a high-quality image estimation unit 217, and a parameter adjustment unit 218.

[0067] The table control unit 201 controls Figure 1 The electron beam scanning control unit 202 controls the movement and stop of the workbench 109. Figure 1 The polarizer 112 is used to irradiate the electron beam within a predetermined field of view. The detector control unit 203 irradiates the electron beam from the detector in synchronization with the scanning of the electron beam. Figure 1 The signal of the detector 111 is sampled, and the gain, offset, etc. of the sampled signal are adjusted to generate a digital image.

[0068] The reference image acquisition unit 204 operates the stage control unit 201, the electron beam scanning control unit 202, and the detector control unit 203 to acquire a reference image that does not contain defects for use in defect observation processing by performing imaging under the first imaging conditions. The inspection image acquisition unit 205 operates the stage control unit 201, the electron beam scanning control unit 202, and the detector control unit 203 to acquire an inspection image for use in defect observation processing by performing imaging under the first imaging conditions.

[0069] The high-quality image acquisition unit 206 acquires a plurality of high-quality images corresponding to the inspection images acquired by the inspection image acquisition unit 205 by performing photography under the second photography condition. High-quality images are images used for defect observation, and require high visual recognition of defects on the surface of the sample 9 and circuit patterns. Therefore, the high-quality images are of higher quality than the inspection images. Generally, the second photography condition is a condition that enables the capture of higher-quality images compared to the first photography condition. For example, the second photography condition has a slower scanning speed of the electron beam than the first photography condition, a greater number of image frames are added, and an improved image resolution. In addition, the photography conditions are not limited to these.

[0070] The storage unit 103 includes an image storage unit 210, a defect detection parameter storage unit 211, a high-quality image estimation parameter storage unit 212, a shooting 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) along with accompanying information (in other words, attribute information, metadata, management information, etc.), and also stores images generated by the calculation unit 104.

[0071] The defect detection parameter storage unit 211 stores parameters related to the defect detection process required for defect observation (sometimes recorded as defect detection parameters). The high-quality image estimation parameter storage unit 212 stores parameters related to the high-quality image estimation model (sometimes recorded as model parameters). The shooting parameter storage unit 213 stores information and data such as the conditions when shooting in SEM101 (also recorded as shooting conditions) as shooting parameters. The shooting conditions include the first shooting conditions and the second shooting conditions, which can be set in advance. Examples of shooting parameters include electron beam scanning speed, image addition frame number, image resolution, etc. The observation coordinate storage unit 214 stores information and data such as the defect coordinates (also recorded as observation coordinates) of the defect position of the observation object input according to the defect position information 8.

[0072] The defect detection unit 215 calculates the position and feature value of defect candidates within the inspection image. The model learning unit 216 learns a model (also referred to as a high-quality image estimation model) required for estimating an image acquired under a second imaging condition based on an image acquired under a first imaging condition. The term "estimation" in the model means creating and outputting a high-quality image that is not an actual image captured. Furthermore, the images used during model learning differ from the images used during actual defect observation using the learned model.

[0073] The high-quality image estimation unit 217 uses the model to perform estimation processing from a first image acquired under a first shooting condition to a second image acquired under a second shooting condition. The high-quality image estimation unit 217 inputs the first image into the model and obtains the second image as an estimation result.

[0074] The parameter adjustment unit 218 automatically adjusts processing parameters (in other words, processes, etc.) related to defect observation. These processing parameters include defect detection parameters, model parameters, and imaging parameters. Examples of defect detection parameters include thresholds for defect determination. Model parameters are parameters that configure (in other words, set) a high-quality image estimation model, such as the CNN described below.

[0075] [1-3. Detector]

[0076] Figure 3 This is a projection diagram showing the arrangement of detectors 111 relative to stage 109 in SEM 101, viewed from an oblique upper direction, in Embodiment 1. In this example, five detectors 301 to 305 are provided as detectors 111, and these detectors 301 to 305 are mounted at predetermined positions within SEM 101 as shown. The number of detectors 111 is not limited to this. Figure 411 is a plan view of the arrangement of the detector 111 as viewed from the z-axis direction. The z-axis corresponds to the vertical direction. Figure 5 1 is a cross-sectional view of the arrangement of the detector 111 as viewed from the y-axis direction. The x-axis and the y-axis correspond to the horizontal direction and are two orthogonal directions.

[0077] like Figures 3 to 5 As shown, detectors 301 and 302 are arranged at positions P1 and P2 along the y-axis, and detectors 303 and 304 are arranged at positions P3 and P4 along the x-axis. These detectors 301 to 304 are not particularly limited, but are arranged in the same plane on the z-axis. Detector 305 is arranged at position P5 of the sample 9 along the z-axis, which is farther away from the workbench 109 than the position of the plane on which the detectors 301 to 304 are arranged. In addition, Figure 4 and Figure 5 , the detector 305 is omitted.

[0078] Detectors 301 to 304 are configured in a manner capable of selectively detecting electrons having specific emission angles (elevation angles and azimuth angles). Detector 301 is capable of detecting electrons emitted from sample 9 along the direction of the y-axis (positive direction indicated by the arrow), and detector 302 is capable of detecting electrons emitted from sample 9 along the direction of the y-axis (the reverse direction relative to the positive direction). Similarly, detector 304 is capable of detecting electrons emitted along the direction of the x-axis (positive direction), and detector 303 is capable of detecting electrons emitted along the reverse direction of the x-axis. Thus, an image with contrast can be obtained as if light were irradiated to each detector from opposite directions. Detector 305 is primarily capable of detecting electrons emitted from sample 9 in the direction of the z-axis.

[0079] As described above, in the first embodiment, a plurality of detectors are arranged at a plurality of positions along different axes, so that an image with contrast can be obtained, thereby enabling more detailed defect observation. The structure of the detector 111 is not limited to this, and may also be configured to be Figure 3 Structures with different positions and orientations.

[0080] [1-4. Defect location information]

[0081] Figure 6 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. Figure 6 In the diagram, defect coordinates are indicated by dots (× marks) on the xy plane of the target sample 9. These defect coordinates are the observation coordinates of the target when observed from the sample observation device 1. Wafer WW represents a circular semiconductor wafer surface area. Bare chip DI represents the area of multiple bare chips (in other words, semiconductor chips) formed on wafer WW.

[0082] The sample observation device 1 of Embodiment 1 has an ADR function that automatically collects high-definition images of defect locations on the surface of the sample 9 as high-quality images based on these defect coordinates. However, the defect coordinates within the defect position information 8 from the inspection device 7 contain errors. In other words, errors 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. Major causes of these errors include incomplete alignment of the sample 9 on the worktable 109.

[0083] Therefore, the specimen observation device 1 of Embodiment 1 first captures a wide-field-of-view, low-magnification image (in other words, a relatively low-quality image) under first imaging conditions, centered on the defect coordinates of the defect location information 8, as an inspection image. The defect location is then re-detected based on this inspection image. The specimen observation device 1 then uses a previously learned high-quality image estimation model to estimate a high-quality image with a narrow field of view and high magnification under second imaging conditions for the re-detected defect location, and acquires this high-quality image as an observation image.

[0084] The wafer WW regularly includes a plurality of bare chips DI. Therefore, when photographing a bare chip DI having a defective portion, for example, another adjacent bare chip DI, an image of a qualified bare chip that does not contain the defective portion can be obtained. In the defect detection process in the sample observation device 1, such a qualified bare chip image can be used as a reference image, for example. Furthermore, in the defect detection process, as a defect determination, for example, a comparison of lightness and darkness (an example of a feature quantity) is performed between the reference image and the inspection image, and a portion having a different lightness and darkness can be detected as a defective portion. As an example of the defect detection parameter, there is a threshold value for determining the lightness and darkness difference when determining a defective portion.

[0085] [1-5. Defect observation method]

[0086] Next, the defect observation method in the sample observation device 1 is described. First, the defect observation method and the overall process are described, and each step implemented by the defect observation method is described in detail. In addition, the main body of the execution of the steps is mainly the upper control device 3 (especially the processor), and the appropriate use Figure 2 The control unit 102, storage unit 103 and operation unit 104 are configured. The information required in the step is read from the storage unit 103, and the information generated in the step is stored in the storage unit 103. Figure 7The learning sample used as the object in steps S702 to S707) and the observation sample used as the object in the sample observation process (steps S705 to S708 described later) can be different samples, and the target positions can also be different positions of the same sample.

[0087] [1-6. Overall processing]

[0088] Figure 7 The defect observation process flow as the overall process and operation of the sample observation device 1 in the first embodiment is shown. Figure 7 There are steps S701 to S708. When defect observation starts, in step S701, the host control device 3 starts Figure 2 The defect detection parameter storage unit 211, the high-quality image estimation parameter storage unit 212, the shooting parameter storage unit 213, etc. read in information such as various parameters to determine whether the process related to defect observation and defect detection processing is set up. The process is a set of information required for defect observation and defect detection processing, including processing parameters related to the ADR function. If the process is not set (no), transfer to step S702; if it is set (yes), transfer to step S705. Steps S702 to S704 are the processing for setting the process (the learning processing mentioned above). Steps S705 to S708 are the actual processing of sample observation and defect detection.

[0089] Step S702 is an image acquisition step. In step S702, the host control device 3 uses Figure 2 The reference image acquisition unit 204, the inspection image acquisition unit 205, and the high-quality image acquisition unit 206 acquire images (low-quality learning images and high-quality learning images) used in learning the high-quality image estimation model (step S703) and adjusting the defect detection parameters (step S704). At this time, the upper control device 3 reads the target of the image acquisition from the observation coordinate storage unit 214. Figure 6 The defect coordinates of the wafer WW are determined, and images corresponding to all the defect coordinates are acquired, and the acquired images are stored in the image storage unit 210 .

[0090] Next, the model learning step of step S703 is performed. In step S703, the host 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.

[0091] Next, the parameter adjustment step of step S704 is performed. In step S704, the host 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.

[0092] After the process setup is complete before defect observation, or after the setup has been completed in steps S702 to S704, the loop processing of steps S705 to S708 is executed. In this loop processing, steps S705 to S708 are repeated for each defect position represented by the defect coordinates in the observation target sample. The defect position corresponding to the defect coordinates is set to i, and i is set to 1 to L.

[0093] Step S705 is a reference image acquisition step. In step S705, the host control device 3 uses the reference image acquisition unit 204 to capture the bare chip DI adjacent to the bare chip DI containing the defect location (i) under the first imaging conditions stored in the imaging parameter storage unit 213, thereby acquiring a reference image (first reference image) corresponding to the inspection image (first inspection image).

[0094] Step S706 is the inspection image acquisition step. In step S706, the host control device 3 uses the inspection image acquisition unit 205 to capture an inspection image (first inspection image) at the defect location (i) under the first imaging conditions stored in the imaging parameter storage unit 213. The reference image and inspection image acquired in these steps are associated and stored in the image storage unit 210.

[0095] Step S707 is a defect detection step. In step S707, the host control device 3 uses the reference image obtained in step S705, the inspection image obtained in step S706, and the set (in other words, adjusted) defect detection parameters stored in the defect detection parameter storage unit 211 to perform defect detection in the inspection image. The defect detection parameters include the image blending ratio during the blending process described later.

[0096] Step S708 is a high-quality image estimation step. In step S708, the upper control device 3 uses the defect position information of the defect detection result in the inspection image obtained in step S707 and the high-quality image estimation unit 217 to perform high-quality image estimation. That is, the upper control device 3 can estimate the high-quality image for defect observation from the inspection image based on the model. The estimated high-quality image is stored in the image storage unit 210 in the same manner as the inspection image and the reference image. In the estimation of this step S708, the high-quality image estimation model that has been set or set in step S703 can also be used. In addition, in the sample observation process, the high-quality image obtained in this step S708 can also be used to implement high-precision defect detection, defect observation, analysis, measurement, evaluation, etc. When the processing of steps S705 to S708 is performed at all defect positions (i), this defect observation process ends.

[0097] [1-7. Step S707 Defect Detection]

[0098] Figure 8 This diagram illustrates the defect detection process in step S707. In step S707, the defect detection unit 215 performs smoothing processing on each of the inspection image 801 and the reference image 802, and blending processing on the images acquired by the detectors 111 (301-305) as preprocessing for step S805. Subsequently, in step S808, the defect detection unit 215 calculates the difference between the preprocessed inspection image 806 and the reference image 807, and binarizes this difference using a first threshold value to obtain a binarized image 803.

[0099] In step S809, the defect detection unit 215 calculates the abnormality of the area 804 in the binarized image 803 that is determined to be greater than the first threshold. The defect detection unit 215 detects the area with an abnormality greater than the second threshold as a defect 811, and obtains a binarized image 810 (in other words, a defect detection result image). As the abnormality, for example, the sum of the brightness values of the differential image within the binarized area can be used. Figure 8 In the process of , examples of the defect detection parameters are the degree of smoothing in the smoothing process in the preprocessing, the image mixing ratio in the mixing process, the first threshold, the second threshold, etc. Figure 8 In the example of FIG, the binary image 810 as the defect detection result shows a case where one defect 811 is detected, but there are also cases where no defect is detected or where multiple defects are detected.

[0100] [1-8. Step S702 Image Acquisition]

[0101] Figure 9This section shows a detailed flow of image acquisition processing in step S702. In the sample observation device 1 of Embodiment 1, defect observation requires, for each defect location (i), an inspection image containing the defect, a reference image not containing the defect, and multiple high-quality images. Therefore, in step S702, the host 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 these images (in other words, low-quality learning images, reference learning images, and high-quality learning images).

[0102] For the defect positions (corresponding defect candidates) detected by defect detection on the inspection image, shooting is performed under the second shooting conditions stored in the shooting parameter storage unit 213, thereby obtaining a high-quality image. However, the defect detection parameters are default parameters that have not been adjusted in the target process (the manufacturing process of the semiconductor wafer). In the case where only one of the defect candidates detected by defect detection using the default parameters (for example, the defect candidate with the largest abnormality) is photographed, the defect may not be reflected in the high-quality image. Therefore, in embodiment 1, a high-quality image is obtained while changing the shooting area according to the defect candidates detected from the inspection image. Thereby, the probability of obtaining a high-quality image with the defect reflected is increased. In the case where a plurality of defect candidates are detected from one inspection image, the sample observation device 1 of embodiment 1 shoots the corresponding high-quality images according to the defect candidates.

[0103] Figure 9 The process includes steps S901 to S906. First, in step S901, the upper control device 3 uses the number of shots determination unit 207 to determine the number of shots (set to N). The number of shots (N) is the maximum number of high-quality images (high-quality images for learning) that are taken. In embodiment 1, the maximum number of high-quality images that are taken is the number of defect candidates that are desired to be photographed. In step S901, the processor determines the number of shots (N) corresponding to one inspection image through user input (in other words, setting).

[0104] Figure 10 This is an example of a GUI screen for determining the number of shots in step S901 in embodiment 1. In this screen, the user enters the number of shots (N) corresponding to the maximum number of shots for capturing high-quality images in area 1001. After entering the number, the user can set the number of shots (N) by pressing the decision button in area 1002. The number of shots (N) is stored in the shooting parameter storage unit 213. Figure 10 In the example, N=5.

[0105] exist Figure 9In the process, after step S901, the upper control device 3 performs a loop processing of steps S902 to S906 according to the defect position (i). Step S902 is a reference image acquisition step. In step S902, the upper control device 3 uses the reference image acquisition unit 204 to obtain a reference image (reference image for learning) of the defect position (i) under the first shooting condition stored in the shooting parameter storage unit 213. Step S903 is an inspection image acquisition step. In step S903, the upper control device 3 uses the inspection image acquisition unit 205 to obtain an inspection image (low-quality image for learning) of the defect position (i) under the first shooting condition stored in the shooting parameter storage unit 213. Step S902 is the same processing as the above-mentioned step S705, and step S903 is the same processing as the above-mentioned step S706. The image thus obtained is stored in the image storage unit 210.

[0106] Step S904 is a defect detection step. In step S904, the host control device 32 performs defect detection processing on the inspection image using the inspection image and reference image acquired in steps S902 and S903 and the defect detection unit 215. Step S904 is the same as step S707.

[0107] Next, in step S905, the upper control device 3 uses the shooting point determination unit 208 to determine one or more shooting points for a plurality of high-quality images. Let the shooting point be j, and let j=1 to J. Thereafter, step S906 is a loop process of high-quality image shooting. In this loop process, the upper control device 3 uses the high-quality image shooting unit 209 to repeatedly obtain a high-quality image for each shooting point (j). In step S906, the upper control device 3 shoots a high-quality image at the shooting point (j) corresponding to the defect position (i) under the second shooting condition stored in the shooting parameter storage unit 213, thereby obtaining a plurality of high-quality images for each shooting point (j). 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 process ends.

[0108] [1-9. Step S905: Shooting Point Determination]

[0109] Figure 11 The flow of the high-quality image capturing point determination process in step S905 is shown. Figure 11The process has steps S1101 to S1103. First, in step S1101, the upper control device 3 determines the priority of the defect candidates in the inspection image (low-quality image for learning). As the priority, for example, the abnormality used in step S707 of defect detection can be used. It can be said that the larger the abnormality, the greater the difference between the inspection image and the reference image, and the higher the probability of a defect. Therefore, in this example, the upper control device 3 calculates and sets it in such a way that the larger the abnormality, the higher the priority of the corresponding defect candidate.

[0110] Next, in step S1102, the upper control device 3 selects defect candidates with high priority. Specifically, the upper control device 3 reads the number of shots (N) determined in step S901 from the shooting parameter storage unit 213, and selects the maximum N defect candidates corresponding to the number of shots (N) in descending order of priority. For example, when the number of defect candidates in the defect image is 100 and the number of shots (N) is 10, the top 10 defect candidates with high abnormality are selected from the 100 defect candidates. In addition, for example, when the number of defect candidates is 5 and N=10, 5 defect candidates are selected.

[0111] Next, in step S1103, the host control device 3 determines one or more imaging points (j) that can capture images of the maximum 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 high-quality learning image.

[0112] [1-10. Step S1103: Shooting Point Determination]

[0113] Figure 12 This is an explanatory diagram regarding the process of determining the shooting point of the high-quality image in step S1103. First, the inspection image 1202 shown on the left is explained. The inspection image 1202 is an inspection image under the first shooting condition corresponding to the defect position (i). Figure 12 In the example, the positions of defect candidates within the inspection image are indicated by white-out x marks, as in defect candidate 1201. In this example, a single inspection image 1202 contains multiple, for example, five, defect candidates 1201. Consider a situation where multiple defect candidates 1201 are close together within inspection image 1202, as in this example. High-quality image capture area 1203 is the area where a high-quality image containing defect candidates 1201 is captured, and is illustrated as a rectangle.

[0114] The host control device 3 determines a high-quality image capture area 1203 centered around the position coordinates of each defect candidate 1201. In this example, multiple defect candidates 1201 are close together, resulting in a significant overlap between the high-quality image capture areas 1203. For example, the two defect candidates 1201 shown in a and b are relatively close together, resulting in a large overlap between the corresponding high-quality image capture areas 1203 shown in A and B. In this case, the overlapping area is captured multiple times, resulting in poor capture efficiency and increased setup time.

[0115] Therefore, in the first embodiment, the inspection image 1204 shown on the right is performed. Figure 11 When multiple defect candidates 1201 are close to each other in the inspection image 1204 and can be contained within a single high-quality image capturing region 1205, the host control device 3 determines the center of gravity of the multiple defect candidates 1201 contained therein as a high-quality image capturing point 1206. A blackened × mark indicates a high-quality image capturing point 1206. The high-quality image capturing region 1205 is a rectangle centered on the high-quality image capturing point 1206. For example, for the two defect candidates 1201 shown in a and b, a high-quality image capturing point 1206 shown in x is determined corresponding to them, and a single high-quality image capturing region 1205 shown in X is determined.

[0116] Through the above processing, the five high-quality image capture areas 1203 in the left-side inspection image 1202 yield three high-quality image capture areas 1205 in the right-side inspection image 1204. In the right-side inspection image 1204, the high-quality image capture areas 1205 have minimal overlap, allowing the capture of multiple defect candidates 1201 (e.g., five) using fewer (e.g., three) high-quality images. This improves image capture efficiency, resulting in reduced setup time.

[0117] After determining the high-quality image capturing area 1205 and high-quality image capturing points 1206 (=j) that can capture all defect candidates selected in step S1102, information such as the coordinates and number of the capturing points (j) is stored in the capturing parameter storage unit 213.

[0118] [1-11. Step S703 Model Learning]

[0119] Figure 13This diagram illustrates an example configuration of a high-quality image estimation model, related to model learning in step S703. In the high-quality image estimation process in Embodiment 1, machine learning is performed on the correspondence between inspection images and high-quality images. Using this machine-learned model, an image captured under a second imaging condition is estimated based on an inspection image or a reference image captured under a first imaging condition.

[0120] The host control device 3 extracts and enlarges the region corresponding to the multiple high-quality images at the imaging point determined in step S905 from the inspection image captured in step S903, creating a pair of the extracted image and the high-quality image. The image extracted from the inspection image is referred to herein as a low-quality image. The low-quality image and the high-quality image each contain the same region, but the low-quality image has been processed to enlarge the imaging field of view. Consequently, the low-quality image is, for example, blurred compared to the high-quality image.

[0121] In the first embodiment, well-known deep learning is used as a method for implementing machine learning. Specifically, a convolutional neural network (CNN) is used as a model. Figure 13 An example of using a neural network having a three-layer structure as a specific example of CNN is shown.

[0122] exist Figure 13 In the CNN, Y represents the low-quality image as input, and F(Y) represents the estimated output. Furthermore, F1(Y) and F2(Y) represent the intermediate data between the input and the estimated output. Intermediate data F1(Y) and F2(Y) and the estimated output F(Y) are calculated using the following equations 1 to 3.

[0123] Formula 1: F1(Y)=max(0, W1*Y+B1)

[0124] Formula 2: F2(Y)=max(0, W2*F1(Y)+B2)

[0125] Formula 3: F(Y)=W3*F2(Y)+B3

[0126] Here, the symbol * represents the convolution operation, W1 represents n1 filters of size c0×f1×f1, c0 represents the number of channels of the input image, and f1 represents the size of the spatial filter. Figure 13In the CNN, an n1-dimensional feature map is obtained by convolving the input image Y with n1 filters of size c0×f1×f1. B1 is an n1-dimensional vector and is the 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 determined by the number of channels in the low-quality and high-quality images. Furthermore, f1, f2, n1, and n2 are determined by the user before learning the sequence. For example, f1 = 9, f2 = 5, n1 = 128, and n2 = 64 can be used. The various parameters described above are examples of model parameters.

[0127] In step S703 of model learning, the model parameters that are adjusted are W1, W2, W3, B1, B2, and B3. The upper control device 3 sets the low-quality image as the input (Y) of the above-mentioned CNN model, wherein the low-quality image is an image obtained by cutting out an image area from the inspection image captured under the first imaging condition and enlarging the imaging field of view. The upper control device 3 adjusts the parameters of the above-mentioned CNN model using the high-quality image captured under the second imaging condition as the estimation result (F(Y)). The upper control device 3 stores the parameters such that the estimation result of the model matches the high-quality image as the adjustment result in the high-quality image estimation parameter storage unit 212.

[0128] In the above-mentioned model parameter adjustment, general error back propagation can be used in the learning of the neural network. In addition, when calculating the estimated error, all the obtained learning image pairs (the above-mentioned pairs of low-quality images and high-quality images) can be used, but a small batch method can also be used. That is, it is also possible to repeatedly extract several images randomly from the learning image pairs and update the high-quality image estimation processing parameters. In addition, patch images can also be randomly cut out from 1 learning image pair as the input image (Y) of the neural network. In this way, efficient learning can be performed. In addition, it is not limited to the structure of the CNN shown in this example, and other structures can also be used. For example, the number of layers can be changed, a network with more than 4 layers can be used, etc., and it can also be constructed with jump connections.

[0129] [1-12. Step S704 Parameter Adjustment]

[0130] Figure 14 The detailed processing flow related to the defect detection parameter adjustment in step S704 is shown. Figure 14The process has steps S1401 to S1406. Here, the image inferred from the inspection image (low-quality image for learning) using the high-quality image estimation model is called a high-quality inspection image (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 the defect candidates corresponding to the defect position (i) in order to determine the detailed defect position in the inspection image of the defect position (i). Let the defect candidate be k, and let k=1~K. At this time, by using the high-quality inspection image and the high-quality reference image corresponding to the high-quality inspection image, the defect position can be determined with high precision.

[0131] [1-13. Step S1401: Acquisition of High-Quality Inspection Image]

[0132] Figure 15 This is an explanatory diagram related to the acquisition process of the high-quality inspection image in step S1401. Figure 15 In one inspection image 2900, a plurality of defect candidates (k) 2901 are included. In step S1401, the host control device 3 obtains a high-quality inspection image for the plurality of defect candidates (k) in the inspection image detected in step S904. Specifically, as in the case of defect candidate 2901 indicated by k1, if the position of the defect candidate overlaps with (in other words, is included in) the imaging region 2902 of the high-quality inspection image determined in step S905, a high-quality inspection image corresponding to the defect candidate 2901 is obtained in step S906, and the high-quality inspection image is selected from the image storage unit 210. This processing is referred to as high-quality inspection image selection processing.

[0133] If the number of defect candidates detected in step S904 is greater than the number of shots N (the maximum number of shots of high-quality images) determined in step S901, for example, Figure 15 For example, in the case of defect candidate 2903 shown in k3 of FIG, a high-quality image containing the defect candidate is not captured for the defect candidate with a low priority. Therefore, when the position of the defect candidate does not overlap with (is not included in) the captured area 2902 of the high-quality inspection image, the host control device 3 uses the high-quality image estimation unit 217 to estimate the high-quality inspection image. This processing is referred to as high-quality inspection image estimation processing. Details of the high-quality inspection image estimation processing are described later.

[0134] Following step S1401, in step S1402, the host control device 3 obtains a high-quality reference image corresponding to the high-quality inspection image. Methods for obtaining the high-quality reference image include estimating the high-quality reference image from the reference image and synthesizing the high-quality reference image from the high-quality inspection image. In Embodiment 1, either method can be used. Details of the high-quality reference image estimation and synthesis processes corresponding to these methods will be described later.

[0135] Following step S1402, in step S1403, the host control device 3 performs defect identification processing using the high-quality inspection image and the high-quality reference image. In the defect identification process, for example, the degree of abnormality obtained by comparing the high-quality inspection image with the high-quality reference image is used as an indicator for identification. 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 control device 3 determines the defect candidate in the inspection image corresponding to the high-quality inspection image with the highest degree of abnormality as a defect. The indicator used for defect identification processing is not limited to the degree of abnormality.

[0136] Next, in step S1404, the host control device 3 performs a process for determining the defect position in the inspection image based on the defect identification result obtained in step S1403. The host control device 3 determines the position of the defect candidate determined as a defect by the defect identification process on the inspection image as the defect position.

[0137] After the detailed defect locations are determined in all inspection images, in step S1405, the host control device 3 evaluates whether the defect locations in the inspection images can be detected according to the defect detection parameter set (referred to as p). In the first embodiment, the number of defect location detections determined in step S1404 is used as the evaluation indicator.

[0138] Finally, in step S1406, the host control device 3 stores the defect detection parameter with the largest number of defect position detections in the defect detection parameter storage unit 211 and outputs it. Figure 14 The parameter adjustment process is completed.

[0139] [1-14. Estimation Processing of High-Quality Inspection Images]

[0140] use Figure 16The estimation process for the high-quality inspection image described above will now be described. First, the host control device 3 determines a clipping region 1502, indicated by a dotted line, centered on the defect candidate in the inspection image 1501 and performs image clipping step S1503. In this example, the defect candidate is the portion that slightly protrudes laterally from the vertical line, as shown in the figure. Furthermore, the host control device 3 uses the high-quality image estimation unit 217 to perform high-quality inspection image estimation step S1505 on the clipped image 1504, thereby estimating a high-quality inspection image 1506.

[0141] [1-15. Method for estimating a high-quality reference image]

[0142] use Figure 17 , the method of estimating the above-mentioned high-quality reference image is described. First, the upper control device 3 uses the inspection image 1601 to perform step S1602 of determining the reference image clipping area. The inspection image 1601 includes the area 1603 in which the high-quality inspection image was obtained in step S1401. Here, when the selection process of the high-quality inspection image is performed, the shooting area of the selected high-quality image is determined to be the area 1603, and when the estimation process of the high-quality inspection image is performed, Figure 16 The clipped area in step S1503 of clipping the image is determined to be area 1603.

[0143] In step S1602, the upper control device 3 selects a clipped area 1604 corresponding to the area 1603 of the high-quality inspection image obtained in step S1401 from the reference image 1605 corresponding to the inspection image 1601 in the defect candidate (k) of the inspection image 1601 at the defect position (i).

[0144] After the cutout area 1604 is determined, the host control device 3 performs step S1606 of image cutting. Step S1606 is Figure 16 The host control device 3 performs the same processing as step S1503. Then, the host control device 3 uses the high-quality image estimation unit 217 to perform step S1608 of estimating a high-quality reference image on the clipped reference image 1607, thereby obtaining a high-quality reference image 1609.

[0145] [1-16. Method for synthesizing high-quality reference images]

[0146] use Figure 18, a method for synthesizing the above-mentioned high-quality reference image will be described. 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 similarly apply the reference image synthesis processing described in Patent Document 1, for example. In this way, a high-quality reference image 1702 that does not contain defects is synthesized. In addition, "synthesis" here refers to the creation of a database of reference images collected in advance, and the implementation of the process of replacing the characteristic area in the inspection image with a similar area in the database, but the application is not limited to this process.

[0147] [1-17. User Interface]

[0148] exist Figure 1 In the embodiment 1, the user operates the sample observation apparatus 1 using the input / output terminal 6. The user operates by observing the screen of the GUI provided by the user interface control unit 106 of the host control device 3.

[0149] Figure 19 This is an example of a GUI screen related to the image acquisition of step S702 in embodiment 1. In this screen, in the inspection image list area 1801, the IDs of the inspection images for which image acquisition has been completed are displayed in a list. The user can select an ID from the list. In the inspection image area 1802, the inspection image of the ID selected in the list is displayed. At this time, when the "Show high-quality image shooting area" in the check box area 1803 is checked, the shooting area of the high-quality image is displayed in the area 1802, for example, in the form of a dotted rectangle. In addition, in the high-quality image area 1804, a high-quality inspection image related to the inspection image of the ID selected in the list is displayed (corresponding to the high-quality image shooting area of the area 1802). For example, multiple high-quality inspection images are displayed in parallel. In the detector area 1805, the detection object and image type corresponding to the detector 111 can be selected.

[0150] Figure 20This figure shows an example of a GUI screen related to the high-quality image estimation model learning in step S703 and the defect detection parameter adjustment in step S704. This screen generally includes an area 1901 for high-quality image estimation and an area 1910 for parameter adjustment. Area 1901 is used for learning the high-quality image estimation model. Area 1902 is a button for instructing model learning, which is a button for manually calling the model learning unit 216 to execute the learning process of the high-quality image estimation model. Area 1903 is a button for instructing high-quality image estimation, which is a button for manually calling the high-quality image estimation unit 217. An inspection image is displayed in area 1904. A high-quality image estimated by the high-quality image estimation unit 217 is displayed in area 1905. A list of inspection image IDs is displayed in area 1906. In area 1907, the acquired high-quality image is displayed for comparison with the estimated high-quality image in area 1905. The host control device 3 reads and acquires the high-quality image from the image storage unit 210 and displays it in the area 1907 .

[0151] Area 1910 is used to adjust defect detection parameters. Area 1911, located above area 1910, displays the defect detection parameters (multiple parameters) using sliders and other components. In this example, the user manually adjusts the values of each parameter (#1 through #4) by changing the slider position. The default values for each parameter are displayed with gray sliders.

[0152] Area 1912 is a button that indicates automatic adjustment. When area 1912 is pressed, the host control device 3 executes the processing of the parameter adjustment unit 218 to automatically adjust the defect detection parameters. The host control device 3 automatically changes the displayed values (positions) of the sliders for each parameter in the upper area 1911 based on the adjusted defect detection parameter values.

[0153] In area 1915, an overview of the inspection images is displayed in the form of a list of IDs, and the success or failure (success / failure) of the detection of actual defects in the inspection images is displayed according to the execution items such as "Run 1" and "Run 2". In area 1916, the defect area (the area containing the defective part) in the inspection image is displayed as an area surrounded by a dotted line. In addition, in area 1916, the user can also manually specify the defect area. Area 1913 is a button for indicating the addition of a defect area, which is a button for calling the process of adding a defect area. In addition, area 1914 is a button for indicating the deletion of a defect area, which is a button for calling the process of deleting the defect area on the inspection image. Area 1918 is an area that displays the capture rate of defects. Area 1919 is a button for indicating parameter output, which is a button for calling the process of storing the defect detection parameters set in area 1910 in the defect detection parameter storage unit 211.

[0154] [1-18. Effects, etc.]

[0155] As described above, in the sample observation device and method of Embodiment 1, using the SEM 101 as the imaging device 2, an inspection image, a reference image, and multiple high-quality images are acquired according to the defect coordinates indicated by the defect location information 8. The inspection image and the high-quality image are used to learn a high-quality image estimation model, and defect detection parameters are adjusted using the high-quality inspection image and the high-quality reference image. Embodiment 1 allows users to easily improve the accuracy of defect detection. Embodiment 1 improves the accuracy of the high-quality image estimation model and also improves the usability of the ADR function in manufacturing processes. Embodiment 1 allows multiple high-quality images to be captured according to the defect locations indicated by the defect location information 8. This increases the probability that defects in the inspection image will be captured in the high-quality image, thereby improving the accuracy of the high-quality estimation model. This also improves the accuracy of defect detection processing parameter adjustment. Furthermore, the accuracy of the high-quality image estimated during defect observation is improved. Furthermore, Embodiment 1 provides the user with the aforementioned functionality, allowing the user to perform operations while viewing the GUI screen, thereby improving usability during defect observation.

[0156] In addition, regarding step S901, the user can Figure 10 The number of shots (N) is set in the screen. The following effects are listed as the effects brought about by this. Regarding the number of images used in the ADR function of the model using machine learning, generally speaking, there is a trade-off between accuracy and processing time. If the number of images is increased, the accuracy may be improved, but the processing time becomes longer. Therefore, in embodiment 1, the user can adjust the number of shots (N), thereby easily adjusting to the accuracy and processing time preferred by the user. It is not limited to a structure in which the user can variably set the number of shots (N). As a modification, it is also possible to have a structure in which the system including the sample observation device 1 automatically determines and sets the number of shots (N).

[0157] <Implementation Method 2>

[0158] use Figure 21 Hereinafter, Embodiment 2 will be described. The basic structure of Embodiment 2 and the like is the same as that of Embodiment 1, and the following mainly describes the structural parts of Embodiment 2 and the like that are different from those of Embodiment 1.

[0159] In the second embodiment, when generating a reference image based on 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 from a low-quality inspection image by synthesis, and a high-quality reference image is generated from a high-quality inspection image by synthesis. When acquiring images in defect observation, the fewer the number of images captured, the shorter the acquisition time and the higher the throughput. In the second embodiment, by generating a reference image from an inspection image, the capture of the reference image is omitted and defect observation is implemented. The GUI in the second embodiment can be similarly applied to the method described in the first embodiment. Figure 10 、 Figure 19 、 Figure 20 That kind of GUI.

[0160] [2-1. Reference Image Acquisition Unit 2001]

[0161] Figure 21 FIG. 1 shows a functional block structure implemented by the control unit 102 and the operation unit 104 of the host control device 3 in the second embodiment. The structure of the second embodiment is mainly different from that of the first embodiment in the reference image acquisition unit. In the second embodiment, as shown in FIG. Figure 21 As shown, the reference image acquisition unit 2001 is configured in the calculation unit 104. The reference image acquisition unit 2001 does not control the SEM 101 but only performs calculation processing, and is therefore configured as the calculation unit 104. The reference image acquisition unit 2001 obtains the reference image from the storage unit 103 (not shown). Figure 2 The image storage unit 210 (same as the one described in Patent Document 1) reads the inspection image acquired under the first imaging condition by the inspection image acquisition unit 205 and performs reference image synthesis processing on the inspection image to acquire a reference image that does not contain defects. The reference image synthesis processing can be similar to the reference image synthesis processing described in Patent Document 1, for example.

[0162] [2-2. Overall processing]

[0163] A defect observation method performed in the sample observation apparatus 1 according to the second embodiment will be described. Figure 22 The overall processing and operation of the sample observation device 1 according to the second embodiment are shown. Figure 21 The process has steps S2101 to S2108. When defect observation begins, in step S2101 (equivalent to Figure 7 In S701), it is determined whether the process related to the defect detection process is set. If not set, it is transferred to step S2102. If it is set, it is transferred to step S2105.

[0164] In step S2102, the host control device 3 uses the reference image acquisition unit 2001, the inspection image acquisition unit 205, and the high-quality image acquisition unit 206 to acquire images used in high-quality image estimation model learning and defect detection parameter adjustment. Figure 7 In step S703), model learning is performed. Figure 7 In S704), parameter adjustment is performed.

[0165] When the process setting is completed, or after the setting is completed through steps S2102 to S2104, the loop processing of steps S2105 to S2108 is performed according to the defect position (i) of the defect position information 8. In step S2105 (equivalent to Figure 7 In step S706), the upper control device 3 uses the inspection image acquisition unit 205 to obtain the inspection image at the defect position (i). In step S2106, the upper control device 3 uses the inspection image read from the image storage unit 210 in the reference image acquisition unit 2001 to perform reference image synthesis processing to obtain a reference image. The obtained reference image is stored in the image storage unit 210. In step S2107 (equivalent to Figure 7 In step S707), defect detection is performed. Figure 7 In step S708), high-quality image estimation is performed. After the processing of steps S2105 to S2108 is performed at all defect positions (i), this defect observation process ends.

[0166] [2-3. Step S2102 Image Acquisition]

[0167] Figure 23 The flow related to the image acquisition process of step S2102 is shown. The host control device 3 uses the inspection image acquisition unit 205, the reference image acquisition unit 2001 and the high-quality image acquisition unit 206 to perform the image acquisition process. First, in step S2201 (equivalent to Figure 9 In step S901), the host control device 3 determines the number of shots (N) (the maximum number of shots of high-quality images). After step S2201, the loop processing of steps S2202 to S2206 is performed according to the defect position (i) of the defect position information 8.

[0168] In step S2202 (equivalent to Figure 9In step S903 of the preceding text, the host control device 3 obtains an inspection image of the defect location (i) under the first imaging conditions stored in the imaging parameter storage unit 213. In step S2203, the host control device 3 uses the inspection images read from the image storage unit 210 in the reference image acquisition unit 2001 to synthesize a reference image of the defect location (i). Step S2203 is the same as the aforementioned step S2106. The synthesized reference image is stored in the image storage unit 210.

[0169] Next, in step S2204 (equivalent to Figure 9 In step S904), the upper control device 3 performs defect detection, and in step S2205 (equivalent to Figure 9 In step S905), the shooting point (j) of the high-quality image is determined. After the shooting point (j) is determined, in step S2206 (equivalent to Figure 9 In step S906 of the preceding step, the host control device 3 captures a high-quality image at the imaging point (j) at the defect location (i) under the second imaging conditions stored in the imaging parameter storage unit 213. The captured high-quality image is stored in the image storage unit 210. Once steps S2202 to S2206 are completed for all defect locations, the image acquisition process ends.

[0170] [2-4. Effects, etc.]

[0171] As described above, according to the second embodiment, it is possible to synthesize the reference image by capturing only the inspection image. Therefore, according to the second embodiment, it is possible to omit the capturing of the reference image and improve the throughput of defect observation.

[0172] <Implementation Method 3>

[0173] use Figure 24 Implementation 3 will be described below. Implementations 1 and 2 describe methods for using reference images to detect defects in inspection images during high-quality image estimation model learning and defect detection parameter adjustment. Implementation 3 captures the entire inspection image area with high-quality images, enabling defect observation without the need for reference images. Implementation 3 omits the need for reference images, improving throughput related to defect observation. ADR processing in Implementation 3 is implemented using machine learning. The GUI in Implementation 3 can be similarly applied to the GUI in Implementation 1.

[0174] [3-1. Functional blocks]

[0175] Figure 24The functional block structure of the control unit 102 and the calculation unit 104 of the host control device 3 in Embodiment 3 is shown. The main structural differences in Embodiment 3 are in the structures of the control unit 102 and the calculation unit 104. First, neither the control unit 102 nor the calculation unit 104 includes a reference image acquisition unit. Furthermore, the structure of the high-quality image acquisition unit 2301 is also different, lacking a number-of-shot determination unit. Furthermore, in Embodiment 3, to eliminate the need for a number-of-shot determination unit, a shooting point determination unit 2302 is included within the high-quality image acquisition unit 2301 of the control unit 102. Furthermore, the processing and operation of the defect detection unit 2303 differ from those in Embodiments 1 and 2, and therefore, the processing and operation of the parameter adjustment unit 2304 also differ.

[0176] [3-2. Defect Observation]

[0177] A defect observation method performed in the sample observation apparatus 1 according to the third embodiment will be described. Figure 25 The overall processing and operation of the sample observation device 1 according to the third embodiment are shown. When defect observation starts, in step S2501 (equivalent to Figure 7 In step S701), it is determined whether the process related to the defect detection process is set. If it is not set, it is transferred to step S2502, and if it is set, it is transferred 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 obtain the image used in the high-quality image estimation model learning and defect detection parameter adjustment. Then, in step S2503 (equivalent to Figure 7 Then, in step S2504 (equivalent to Figure 7 Parameter adjustment is performed in step S704 of the embodiment. Furthermore, in the third embodiment, the same model (here, the first model) as in the first embodiment is used in the model learning in step S2503. The parameter adjustment in step S2504 is the adjustment of parameters related to the defect detection process in step S2506.

[0178] When the process setting is completed, or after the setting is completed through steps S2502 to S2504, the loop processing of steps S2505 to S2507 is performed according to the defect position (i) of the defect position information 8. In step S2505 (equivalent to Figure 7In S706 of Implementation Method 1, 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 shooting condition stored in the shooting 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 Implementation Method 3, in the defect detection process of step S2506, a model different from the first model of step S2503 (here set as the second model) is used. The defect detection process of step S2506 is different from the defect detection process based on image processing in Implementation Method 1. It is a defect detection process based on machine learning, and therefore, the second model of machine learning is used. In step S2507 (equivalent to Figure 7 In S708), the host control device 3 estimates a high-quality image. When the processing of steps S2505 to S2507 is performed at all defect positions, this defect observation process ends.

[0179] [3-3. Step S2502 Image Acquisition]

[0180] Figure 26 This figure shows the image acquisition process flow for step S2502. The host 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 Embodiment 3 requires, for each defect location (i) in the defect location information 8, an inspection image containing the defect location and multiple high-quality images whose imaging ranges are determined to capture the entire inspection image imaging range. Therefore, these images are acquired in step S2502.

[0181] In the third embodiment, the number of images (N) is determined by the ratio of the inspection image to the high-quality image capture range (in other words, the field of view) (described later). Therefore, in step S2502 of image acquisition, the number of images (N) does not need to be determined by the user. In step S2502, the loop processing of steps S2601 to S2603 is executed for each defect location (i) in the defect location information 8.

[0182] In step S2601 (equivalent to step S2602) of obtaining the inspection image Figure 9 In step S903 of the above, the host control device 3 uses the inspection image acquisition unit 205 to acquire an inspection image of the defect position (i) under the first imaging conditions stored in the imaging parameter storage unit 213. Step S2601 is the same as step S706. The acquired image is stored in the image storage unit 210.

[0183] Next, in step S2602, the host control device 3 uses the imaging point determination unit 2302 to determine the imaging points (j) for the high-quality image. Unlike in Embodiments 1 and 2, in step S2602, the host control device 3 determines the minimum number of imaging points (j) required to capture the entire imaging range of the inspection image. Step S2602 will be described in detail later.

[0184] After the shooting point (j) is determined, in step S2603 (equivalent to Figure 9 In S906 of FIG. 1 , the host control device 3 uses the high-quality image capturing unit 209 to capture a high-quality image of the imaging point (j) corresponding to the defect location (i) under the second imaging conditions stored in the imaging parameter storage unit 213. The captured high-quality image is stored in the image storage unit 210. Once steps S2601 to S2603 have been completed for all defect locations, the image acquisition process ends.

[0185] [3-4. Step S2602: Shooting Point Determination]

[0186] Figure 27 This is an explanatory diagram of the photographing point determination process in step S2602 in the third embodiment. Figure 27 shows an example where the size ratio of the captured range (field of view) between the inspection image and the high-quality image is 2. In this example, the inspection image 2701 is a square, and its horizontal size (in pixels) is represented by SX1. The high-quality image (indicated by the dotted line) is also a square, and its 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.

[0187] In use Figure 27 When determining imaging points for an inspection image 2701, the host control device 3 divides the inspection image 2701 vertically and horizontally into regions 2702 equal in number to the size ratio (=2). In this example, four regions 2702 are formed, covering the entire area of the inspection image 2701. Furthermore, the host control device 3 determines a plurality (four) of imaging points (j) 2703, indicated by x marks, so that these regions 2702 can be captured as high-quality image areas.

[0188] [3-5. Step S2506 Defect Detection]

[0189] Figure 28 This is an explanatory diagram of the defect detection process of step S2506. In the defect detection process in Implementation 3, a machine learning model is used to detect defects in the inspection image. As the implementation method of machine learning for defect detection in Implementation 3, well-known deep learning is used. Specifically, Figure 28 In the defect detection of the third embodiment, the output is a defect detection result image (G(Z)). The defect detection result image is a binary image representing the area detected as a defect. 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.

[0190] exist Figure 28 In this example, Z represents the input inspection image, and G(Z) represents the defect detection result image, which is the output of the estimation result. Furthermore, G1(Z) and G2(Z) represent the intermediate data between the input and the estimation result. Intermediate data G1(Z) and G2(Z), along with the defect detection result image G(Z), which is the estimation result, are calculated using the following equations 4 to 6.

[0191] Formula 4: G1(Z)=max(0, V1*Z+A1)

[0192] Formula 5: G2(Z)=max(0,V2*G1(Z)+A2)

[0193] Formula 6: G(Z) = V3*G2(Z)+A3

[0194] Here, the symbol * represents a convolution operation, V1 represents m1 filters of size d0×g1×g1, d0 represents the number of channels in the input image, and g1 represents the size of the spatial filter. By convolving the input image Z with m1 filters of size d0×g1×g1, an m1-dimensional feature map is obtained. A1 is an m1-dimensional vector and is the bias component corresponding to the m1 filters. Similarly, V2 is m2 filters of size m1×g2×g2, A2 is an m2-dimensional vector, and V3 is one filter of size m2×g3×g3, A3 is a 1-dimensional vector. g1, g2, m1, and m2 are values determined by the user before defect detection. For example, g1=9, g2=5, m1=128, and m2=64 can be sufficient.

[0195] In the parameter adjustment unit 2304 in the third embodiment, the model parameters to be adjusted are V1, V2, V3, A1, A2, and A3. However, the present invention is not limited to the above-mentioned CNN structure.

[0196] [3-6. Step S2504 Parameter Adjustment]

[0197] Figure 29 Represents the process related to the defect detection parameter adjustment processing of step S2504. Figure 29There are steps S2801 to S2805. In the parameter adjustment in the third embodiment, first, in order to determine the detailed defect position in the inspection image of the defect position (i), the high-quality inspection image selection step S2801 and the defect identification step S2802 are performed for the shooting point (j) of the high-quality image. In step S2801, since the high-quality image corresponding to the entire area of the inspection image is obtained in the image acquisition step S2502, only the high-quality image selection process is performed. In step S2802, the upper control device 3 uses the high-quality inspection image to perform defect identification processing. In the defect identification processing in the third embodiment, the defect detection unit 2303 uses the high-quality inspection image to obtain 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 manufacturing process of the object. Therefore, for example, the defect detection parameters that have been adjusted in a different manufacturing process are used to apply the defect detection processing to the high-quality inspection image.

[0198] In step S2803 (equivalent to Figure 14 In step S1404 of the above process, the host control device 3 uses the defect detection result image of the high-quality inspection image obtained in step S2802 to determine the defect position in the inspection image. In step S2804, the host control device 3 uses the detailed defect position on the inspection image determined in step S2803 to create a defect detection result image corresponding to the inspection image. The created defect detection result image is stored in the image storage unit 210.

[0199] After the corresponding defect detection result images are produced in all the inspection images, finally, in step S2805 (equivalent to Figure 14 In step S1406), the upper control device 3 uses the model, takes the inspection image as input, takes the defect detection result image as the inference result, and adjusts the defect detection parameters. Figure 28 The CNN processing is the same as that used in the Figure 13 Similarly, various methods can be applied to the processing of CNN to achieve efficiency, etc. After the learning of the above-mentioned neural network is completed, the upper control device 3 stores the defect detection parameters as the adjustment results in the defect detection parameter storage unit 211, and the process of this parameter adjustment is completed.

[0200] [3-7. Effects, etc.]

[0201] As described above, according to the third embodiment, by performing learning and parameter adjustment of the high-quality image estimation model without acquiring a reference image, acquisition of a reference image can be omitted, and the throughput of the sample observation apparatus can be improved.

[0202] As mentioned above, although this invention was demonstrated concretely based on embodiment, this invention is not limited to the said embodiment, Various changes are possible within the range which does not deviate from the summary.

Claims

1. A sample observation device comprising: Filming device; as well as A processor that executes a learning process A for learning a high-quality image estimation model and a sample observation process B for performing defect detection, characterized in that: In the learning process A, the following processes are performed: Process A1: obtaining one or more learning defect positions related to the learning sample; Process A2: obtaining a low-quality learning image under a first shooting condition according to the learning defect position; Process A3: obtaining a first setting value related to the number of high-quality learning images to be captured; Process A4: Perform the following processing according to the defect location of the learning method: Process A4a: determining the number of high-quality learning images to be captured based on the first setting value. In step A4b, based on the number of shots determined in step A4a, one or more shooting points are determined as positions where the high-quality learning image is to be shot. Process A4c of acquiring the high-quality learning image under a second imaging condition for each of the one or more imaging points determined in process A4b; Process A5: learning the high-quality image estimation model using the low-quality learning image and the high-quality learning image; and Process A6, using the high-quality image estimation model to adjust defect detection parameters, In the sample observation process B, the following processes are performed according to the adjusted defect detection parameters: Process B1: obtaining a first inspection image of a defect position of an observation target sample under the first imaging condition; and In process B2, defect candidates of the observation target sample are detected based on the first inspection image.

2. The sample observation device according to claim 1, wherein The processor sets the first setting value related to the number of shots according to a user input.

3. The sample observation device according to claim 1, wherein When the processor determines the shooting point, Determine the priority of each defect candidate based on the feature quantity of the defect candidate in the low-quality learning image, selecting, based on the priority of each defect candidate, a plurality of defect candidates photographed under the second photographing condition from a plurality of defect candidates in the low-quality learning image in such a manner that the number of photographed images is less than the number of photographed images, The imaging points for imaging the selected plurality of defect candidates are determined.

4. The sample observation device according to claim 3, wherein When the processor determines the shooting point, When a plurality of defect candidates are close to each other in the low-quality learning image, the imaging point is determined so that a region centered on the imaging point becomes one region including the plurality of defect candidates.

5. The sample observation device according to claim 1, wherein In the learning process, According to the defect position for learning, under the first shooting condition, a reference image for learning that does not contain a defect and corresponds to the low-quality image for learning is obtained; Using the learning reference image to learn the model, During the sample observation process, According to the defect position of the observation target sample, under the first shooting condition, a first reference image that does not contain the defect and corresponds to the first inspection image is obtained; The position and feature value of the defect candidate are calculated based on the comparison between the first inspection image and the first reference image.

6. The sample observation device according to claim 5, characterized in that When acquiring the reference image for learning, the processor acquires the reference image for learning by photographing it with the photographing device under the first photographing condition. When acquiring the first reference image, the processor acquires the first reference image by capturing the image with the capturing device under the first capturing condition.

7. The sample observation device according to claim 5, characterized in that When acquiring the reference image for learning, the processor synthesizes the reference image for learning from the low-quality image for learning. When acquiring the first reference image, the processor synthesizes the first reference image from the first inspection image.

8. The sample observation device according to claim 5, wherein: When the processor adjusts the defect detection parameters, From the low-quality learning image, an image is captured for the defect candidate. Using the high-quality image estimation model, the high-quality learning image is estimated from the clipped image, or a high-quality learning image including the defect candidate is selected from the acquired high-quality learning images. Based on the learning reference image, the high-quality image estimation model is used to estimate the learning reference image for the position corresponding to the defect candidate, or the learning reference image is synthesized from the learning high-quality image. Using the high-quality learning image and the reference learning image, the defect candidate is judged to be a defect, thereby determining the defect position in the high-quality learning image. Based on the evaluation, parameters capable of detecting the determined defect position are selected.

9. The sample observation device according to claim 8, characterized in that The processor displays on the screen the low-quality learning image, the high-quality learning image, the estimation result of the high-quality inspection image based on the high-quality image estimation model according to the first inspection image, the adjustment result of the defect detection parameters, and the defect position corresponding to the detected defect candidate in the first inspection image after the defect detection parameters are adjusted.

10. A sample observation method in a sample observation device, the sample observation device comprising: a photographing device, a processor for executing a learning process A for learning a high-quality image estimation model and a sample observation process B for performing defect detection, characterized in that: As a step performed by the sample observation device, In the learning process A, the following steps are performed: Step A1, obtaining one or more learning defect positions related to a learning sample; Step A2, obtaining a low-quality learning image under a first shooting condition according to the learning defect position; Step A3, obtaining a first setting value related to the number of high-quality learning images to be captured; Step A4: Perform the following steps according to the defect location of the learning method: Step A4a, determining the number of high-quality learning images to be taken based on the first setting value, Step A4b, based on the number of shots determined in step A4a, determines the positions where the high-quality learning image is to be shot, that is, one or more shooting points. Step A4c, acquiring the high-quality learning image under a second imaging condition for each of the one or more imaging points determined in Step A4b; Step A5, using the low-quality learning image and the high-quality learning image to learn the high-quality image estimation model; as well as Step A6: using the high-quality image estimation model to adjust defect detection parameters. In the sample observation process B, the following steps are performed according to the adjusted defect detection parameters: Step B1, obtaining a first inspection image of a defect position of an observation target sample under the first shooting condition; as well as In step B2, defect candidates of the observation target sample are detected based on the first inspection image.

Images