Defect inspection of semiconductor samples

Abstract

A system and method for runtime defect inspection of a semiconductor sample is provided, comprising: obtaining a first image representing at least a portion of the semiconductor sample, the first image being acquired by an inspection tool configured with a first focal plane; estimating whether the first image is in focus using a machine learning (ML) model, wherein the ML model is pre-trained to classify images into in-focus images and out-of-focus images; when the first image is estimated to be out of focus, performing focus calibration on the inspection tool to select a second focal plane associated with an optimal focus score; and obtaining a second image acquired by the inspection tool configured with a second focal plane, and estimating whether the second image is in focus using the ML model. When the second image is estimated to be in focus, the second image can be used to inspect the sample for defects.

Description

Technical Field

[0001] The disclosed subject matter relates generally to the field of inspection of semiconductor samples, and more particularly, to focus calibration for inspection of semiconductor samples. Background Art

[0002] Currently, the high density and high performance requirements associated with ultra-large scale integration of manufacturing devices require sub-micron features, increased transistor and circuit speeds, and increased reliability. As semiconductor processes advance, pattern dimensions such as lining widths and other types of critical dimensions continue to shrink. This demand requires the formation of device features with high precision and high uniformity, which in turn requires careful monitoring of the manufacturing process, including automated inspection of the device while it is still in semiconductor wafer form.

[0003] Inspection can be provided by using nondestructive inspection tools during or after the manufacture of the sample to be inspected. Inspection generally involves generating a specific output (e.g., an image, a signal, etc.) for the sample by directing light or electrons onto a wafer and detecting the light or electrons from the wafer. As non-limiting examples, various nondestructive inspection tools include scanning electron microscopes, atomic force microscopes, optical inspection tools, etc.

[0004] The inspection process may include multiple inspection steps. During the manufacturing process, the inspection step may be performed multiple times, for example, after manufacturing or processing a specific layer, the inspection step may be performed multiple times, etc. Additionally or alternatively, each inspection step may be repeated multiple times, for example, for different wafer positions, or for the same wafer position with different inspection settings.

[0005] Inspection processes are used at various steps during semiconductor manufacturing to detect and classify defects on samples and perform metrology-related operations. The effectiveness of inspection can be improved by automating the process(es) such as defect detection, automatic defect classification (ADC), automatic defect review (ADR), automated metrology-related operations, etc.

[0006] The inspection process is typically performed on output images acquired by the inspection tool. Images that are degraded in various aspects, such as defocused images, may affect the inspection results when used to inspect semiconductor samples. Summary of the invention

[0007] According to certain aspects of the disclosed subject matter, there is provided a computerized system for performing runtime defect inspection on a semiconductor sample, the system comprising a processing and memory circuit system (PMC) configured to: obtain a first image representing at least a portion of the semiconductor sample, the first image acquired by an inspection tool configured with a first focal plane; estimate whether the first image is in focus using a machine learning (ML) model, wherein the ML model is pre-trained to classify images into in-focus images and out-of-focus images; when the first image is estimated to be out of focus, perform focus calibration on the inspection tool to select a second focal plane associated with an optimal focus score; and obtain a second image acquired by the inspection tool configured with a second focal plane, and estimate whether the second image is in focus using the ML model, wherein when the second image is estimated to be in focus, the second image can be used to perform defect inspection on the sample.

[0008] In addition to the features described above, the system according to this aspect of the disclosed subject matter may include one or more of the features (i) to (ix) listed below, in any desired combination or arrangement that is technically possible:

[0009] (i) The inspection tool is an electron beam tool.

[0010] (ii) the ML model is a binary classifier, which is previously trained using a training dataset, wherein the training dataset includes a subset of in-focus images and a subset of defocused images, each of which is associated with a corresponding ground truth label.

[0011] (iii) The training dataset is obtained by: for each of one or more samples having a corresponding design pattern, collecting an image set of the sample scanned under different focal plane configurations of an inspection tool, and providing true value labels respectively associated with the image sets.

[0012] (iv) The ground truth labels are provided by applying a focus threshold to the image set. The focus threshold is determined based on the focus sensitivity requirement for a specific inspection application.

[0013] (v) One or more defocused images in the defocused image subset are artificially generated using a generative adversarial network (GAN) based on image patterns indicated in corresponding in-focus images in the in-focus image subset.

[0014] (vi) The focus calibration is performed in the following manner: adjusting a focal plane of the inspection tool along a focal axis to a plurality of focal points along the focal axis; acquiring a plurality of images corresponding to the plurality of focal points; calculating a focus score for each of the plurality of images, thereby generating a plurality of focus scores; selecting an image having an optimal focus score among the plurality of focus scores from the plurality of images; and configuring the inspection tool with a second focal plane associated with the selected image.

[0015] (vii) The focal plane is adjusted by mechanically moving a stage for placing the semiconductor sample or electrically changing the landing energy of the electron beam of the inspection tool.

[0016] (viii) When estimating the defocus of the second image, the PMC is further configured to perform focus calibration on the inspection tool to select a third focal plane associated with an optimal focus score, obtain a third image acquired by the inspection tool configured with the third focal plane, and estimate whether the third image is in focus.

[0017] (ix) When estimating the focus of the first image, the PMC is further configured to determine to use the first image to perform defect inspection on the sample.

[0018] According to other aspects of the disclosed subject matter, a method for performing runtime defect inspection on a semiconductor sample is provided, the method being performed by a processing and memory circuit system (PMC) and comprising: obtaining a first image representing at least a portion of the semiconductor sample, the first image being acquired by an inspection tool configured with a first focal plane; estimating whether the first image is in focus using a machine learning (ML) model, wherein the ML model is pre-trained for classifying images into in-focus images and out-of-focus images; when the first image is estimated to be out-of-focus, performing focus calibration on the inspection tool to select a second focal plane associated with an optimal focus score; and obtaining a second image acquired by the inspection tool configured with a second focal plane, and estimating whether the second image is in focus using the ML model, wherein when the second image is estimated to be in focus, the second image can be used to perform defect inspection on the sample.

[0019] This aspect of the disclosed subject matter may comprise, mutatis mutandis, one or more of the features (i) to (ix) listed above with respect to the system, in any desired combination or arrangement technically possible.

[0020] According to other aspects of the disclosed subject matter, there is provided a non-transitory computer-readable medium comprising instructions, which, when executed by a computer, cause the computer to perform a method for runtime defect inspection of a semiconductor sample, the method comprising: obtaining a first image representing at least a portion of the semiconductor sample, the first image acquired by an inspection tool configured with a first focal plane; estimating whether the first image is in focus using a machine learning (ML) model, wherein the ML model is pre-trained to classify images into in-focus images and out-of-focus images; when the first image is estimated to be out of focus, performing focus calibration on the inspection tool to select a second focal plane associated with an optimal focus score; and obtaining a second image acquired by the inspection tool configured with a second focal plane, and estimating whether the second image is in focus using the ML model, wherein when the second image is estimated to be in focus, the second image can be used to perform defect inspection on the sample.

[0021] This aspect of the disclosed subject matter may comprise, mutatis mutandis, one or more of the features (i) to (ix) listed above with respect to the system, in any desired combination or arrangement technically possible. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to understand the present disclosure and to see how it may be implemented in practice, embodiments will now be described, by way of non-limiting example only, with reference to the accompanying drawings, in which:

[0023] Figure 1 A general block diagram of an inspection system according to certain embodiments of the disclosed subject matter is shown.

[0024] Figure 2 A general flow chart for performing runtime defect inspection on semiconductor samples according to certain embodiments of the disclosed subject matter is shown.

[0025] Figure 3 A general flow chart for training an ML model according to certain embodiments of the disclosed subject matter is shown.

[0026] Figure 4 A general flow chart for focus calibration according to certain embodiments of the disclosed subject matter is shown.

[0027] Figure 5 Examples of image sets scanned at different focus plane configurations are shown in accordance with certain embodiments of the disclosed subject matter. Specific embodiments

[0028] In the detailed description below, many specific details are set forth to provide a thorough understanding of the present disclosure. However, it should be understood by those skilled in the art that the present disclosure may be implemented without these specific details. In other cases, well-known methods, procedures, components and circuits are not described in detail to avoid obscuring the present disclosure.

[0029] Unless otherwise specifically stated, it will be apparent from the following discussion that in the discussion throughout the specification, terms such as "inspect", "obtain", "estimate", "determine", "acquire", "perform", "calibrate", "train", "collect", "provide", "apply", "generate", "adjust", "acquire", "calculate", "select", "set", "move", "change", etc., refer to (multiple) operations and / or (multiple) processes by which a computer manipulates and / or transforms data into other data, where the data is represented as physical (such as electronic) quantities and / or where the data represents physical objects. The term "computer" should be broadly interpreted to cover any kind of hardware-based electronic device with data processing capabilities, including, as non-limiting examples, the inspection system, focus estimation and calibration system disclosed in the present application, and their respective parts.

[0030] The term "inspection" as used in this specification should be broadly interpreted to cover any kind of operation related to defect detection and / or classification and metrology operations performed on samples during sample manufacturing. Inspection is provided by using a non-destructive inspection tool during or after the manufacture of the sample to be inspected. As a non-limiting example, the inspection process may include one or more of the following operations: runtime scanning (in a single or multiple scans), sampling, review, measurement, classification and / or other operations of the sample or part thereof using the same or different inspection tools. Similarly, inspection can be provided before the manufacture of the sample to be inspected, and may include, for example, generating (multiple) inspection recipes and / or other setup operations. It should be noted that unless otherwise specifically stated, the term "inspection" or its derivatives used in this specification are not limited in terms of the resolution or size of the inspection area. As non-limiting examples, various non-destructive inspection tools include scanning electron microscopes, atomic force microscopes, optical inspection tools, etc.

[0031] The term "metrology operation" as used in this specification should be broadly interpreted to cover any metrology operation procedure for extracting metrology information related to one or more structural elements on a semiconductor sample. For example, the metrology information to be extracted may indicate one or more of the following: size (e.g., lining width, lining spacing, contact diameter, component size, edge roughness, grayscale statistics, etc.), component shape, distance within or between components, related angles, coverage information associated with components corresponding to different design levels, etc. In some embodiments, the metrology operation may include measurement operations, such as, for example, critical dimension (CD) measurements performed on certain structures on the sample. For example, by using image processing techniques to analyze measurement results such as measurement images. It should be noted that unless otherwise specifically stated, the term "metrology" or its derivatives used in this specification are not limited by measurement technology, measurement resolution, or inspection area size.

[0032] As used herein, the terms "non-transitory memory" and "non-transitory storage medium" should be broadly interpreted to cover any volatile or non-volatile computer memory suitable for use with the presently disclosed subject matter.

[0033] The term "sample" used in this specification should be broadly interpreted to cover any kind of wafers, masks and other structures, combinations and / or components thereof used to manufacture semiconductor integrated circuits, magnetic heads, flat panel displays and other semiconductor manufacturing articles.

[0034] The term "defect" as used in this specification should be interpreted broadly to cover any kind of abnormal or undesirable feature formed on or within a sample.

[0035] It should be understood that, unless otherwise specifically stated, certain features of the disclosed subject matter described in the context of separate embodiments may also be provided in a single embodiment in combination. Conversely, the various features of the disclosed subject matter described in the context of a single embodiment may also be provided individually or in any suitable sub-combination. In the detailed description below, many specific details are set forth to provide a thorough understanding of the method and apparatus.

[0036] With this in mind, attention turns to Figure 1 , Figure 1 A functional block diagram of an inspection system according to certain embodiments of the disclosed subject matter is shown.

[0037] Figure 1The inspection system 100 shown in FIG. 1 can be used to inspect semiconductor samples (e.g., wafers and / or portions thereof) as part of a sample manufacturing process. According to certain embodiments of the disclosed subject matter, the inspection system 100 shown includes a computer-based system 101 that is capable of performing focus estimation and calibration on an inspection tool so as to obtain sample images in optimal focus during image acquisition for runtime sample inspection. The images acquired with optimal focus can be used for inspection of semiconductor samples, such as, for example, defect inspection and detection, and defect review and / or classification, etc. The system 101 is also referred to herein as a focus estimation and calibration system.

[0038] As mentioned above, inspection referred to herein may be interpreted as covering any kind of defect-related operations as well as metrology operations in a sample during or after sample fabrication. As an example, the acquired images may be used for inspection (such as, for example, defect inspection and detection, and / or automatic defect review (ADR), and / or automatic defect classification (ADC)) and / or metrology-related operations (such as, for example, critical dimension (CD) measurements on a sample or portion thereof).

[0039] As an example, in some embodiments, during defect inspection and detection, the sample can be moved in steps relative to a microscope of an inspection tool (or the sample and tool can be moved in opposite directions relative to each other), and the sample can be scanned step by step by the tool imaging only a portion / section of the sample at a time (referred to as a field of view (FOV)). For example, at each step, a signal can be detected from a rectangular FOV of the sample, and such detected signal can be converted into a plurality of intensity values ​​at a plurality of points in the FOV, thereby forming an image corresponding to the sample FOV. Thus, during sequential scanning of the sample, a plurality of images of the sample can be sequentially obtained, each image representing a corresponding portion / section (FOV) of the sample.

[0040] During sequential scanning and image acquisition, each FOV should be in focus so that images are acquired at optimal focus to provide sufficient quality and resolution for inspection of the sample. System 101 is designed and configured to check the focus of each FOV during scanning and trigger a focus calibration process if an image is found to be out of focus, as described in detail below.

[0041] The system 101 may be operably connected to one or more inspection tools 120 configured to scan a semiconductor sample and capture an image thereof for inspection of the sample.

[0042] The term "inspection tool(s)" as used herein should be interpreted broadly to encompass any tool that may be used in an inspection-related process including, by way of non-limiting example, imaging, scanning (in a single or multiple scans), sampling, reviewing, measuring, classification and / or other processes provided with respect to a sample or portion thereof.

[0043] In some embodiments, one or more inspection tools 120 may include one or more test tools and / or one or more review tools. In some cases, at least one of the inspection tools 120 may be an inspection tool (e.g., an optical inspection tool, a scanning electron microscope (SEM), etc.) configured to scan a sample (e.g., an entire wafer, an entire die, or a portion thereof) to capture an inspection image to detect potential defects on the sample.

[0044] In some cases, at least one of the inspection tools 120 may be a review tool configured to capture review images of at least some of the defects detected by the inspection tool for determining whether the potential defects are actually defects. Such inspection tools (e.g., relatively high resolution scanning electron microscopes (SEMs)) are typically configured to inspect fragments of the die one at a time (typically at relatively slow speeds and / or high resolutions). In some cases, the review tool may have defect classification capabilities. The inspection tool and the review tool may be different tools located in the same or different locations, or a single tool operating in two different modes. In some cases, at least one inspection tool may have metrology capabilities, such as, for example, a critical dimension scanning electron microscope (CD-SEM) for measuring critical dimensions of structural features in an image.

[0045] The resulting image data (either low-resolution image data or high-resolution image data) may be transmitted directly or via one or more intermediate systems to system 101. In some cases, the same inspection tool may provide both low-resolution image data and high-resolution image data.

[0046] According to certain embodiments, one of the inspection tools is an electron beam tool, such as, for example, a scanning electron microscope (SEM). A SEM is an electron microscope that produces an image of a sample by scanning the surface with a focused electron beam. The electrons interact with the atoms in the sample, thereby producing various signals that include information about the sample surface. The position of the beam is combined with the intensity of the detected signals to produce an image. SEMs are capable of accurately detecting and measuring features during semiconductor wafer manufacturing.

[0047] Image acquisition for sample inspection is typically preferably performed at an optimal focus that produces the highest contrast of the image pattern (i.e., the sample pattern being imaged). The optimal focus point can be found by performing a focus calibration, during which the inspection tool is scanned through different focal planes ("focus ramping"), the focus scores for each image at the different focal planes are calculated, and the focal plane with the highest score is found. However, once the focus of the current scan is found, keeping the acquired images in focus over long time spans of inspection and different patterns is an important and useful task for a variety of reasons, some of which are listed below.

[0048] First, as the inspection tool continues to navigate and acquire images at different locations on the wafer, certain physical factors (such as, for example, the relative distance between the wafer stage and the detector, wafer orientation, etc.) may vary from scan to scan, causing defocus issues.

[0049] Second, for image acquisition performed by high-resolution SEM, additional physical effects (such as charging effects) may occur during scanning, resulting in an increased defocus rate and the task of keeping the acquired image in focus becomes more challenging.

[0050] Specifically, semiconductor samples collect charge as they are scanned by an electron beam, and the accumulation of surface charge on the sample caused by the electron beam can affect the focus of the acquired image. This is because when charging occurs, some electrons may be buried inside the sample, resulting in a surface potential that is different from the potential at neighboring locations. This potential can affect the energy of the incident beam, thereby changing the focus. When the acquired image is not at its optimal focus, the image becomes blurred and the resolution is reduced, which negatively affects defect detection in advanced semiconductor processes. This defocusing effect can lead to an increasing inability to accurately detect and classify defects, as well as measure critical integrated device dimensions in the semiconductor industry. These inaccuracies can cause significant problems in the manufacture of integrated circuits.

[0051] Furthermore, the calculation of the focus score depends largely on the image pattern. In other words, the focus score calculated for one pattern cannot be used as an indicator of the focal plane for capturing an image with a different pattern. Therefore, focus ramping must be performed for each different pattern.

[0052] Different approaches have been taken to address these issues. One attempt to maintain focus is to apply a real-time focus tracking module (full-time continuous focus tracking) in the inspection tool. However, the implementation of such a module is complex and requires dedicated hardware and significant system modifications to integrate it into the image acquisition process.

[0053] An alternative approach is to trigger a periodic focus calibration for every N images to be acquired, where N is a predetermined integer in a range. This approach can be time-consuming, as multiple calibrations will inevitably increase the scan duration significantly, thus affecting the inspection throughput (TpT). Another disadvantage is the uncertainty in choosing the right value of N, resulting in no guarantee that all acquired images are in focus.

[0054] According to certain embodiments of the disclosed subject matter, novel methods are proposed for accurately determining whether to trigger focus calibration for each acquired image during runtime inspection based on a machine learning model that is pre-trained to estimate whether the acquired image is in focus. In particular, according to some embodiments, the proposed methods and systems can be applied to defect inspection and detection applications using electron beam tools such as SEMs.

[0055] like Figure 1 As shown, system 101 includes a processor and memory circuit system (PMC) 102 operatively connected to a hardware-based I / O interface 126. PMC 102 is configured to provide processing required for operating the system, as described with reference to FIG. Figure 2 , Figure 3 and Figure 4 102 and includes a processor (not shown separately) and a memory (not shown separately). The processor of PMC 102 can be configured to execute several functional modules according to computer-readable instructions implemented on a non-transitory computer-readable storage medium included in the PMC. Such functional modules are hereinafter referred to as being included in the PMC.

[0056] The processor referred to herein may represent one or more general processing devices, such as a microprocessor, a central processing unit, etc. More specifically, the processor may be a complex instruction set computing (CISC) microprocessor, a reduced instruction set computing (RISC) microprocessor, a very long instruction word (VLIW) microprocessor, or a processor implementing other instruction sets, or a plurality of processors implementing a combination of instruction sets. The processor may also be one or more special processing devices, such as an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), a digital signal processor (DSP), a network processor, etc. The processor is configured to execute instructions for executing the operations and steps discussed herein.

[0057] The memory referred to in this article may include main memory (e.g., read-only memory (ROM), flash memory, dynamic random access memory (DRAM) (such as synchronous DRAM (SDRAM) or bus DRAM (RDRAM)), etc.) and static memory (e.g., flash memory, static random access memory (SRAM), etc.).

[0058] According to some embodiments, the functional modules included in the PMC 102 may include a focus estimation module 104 and a focus calibration module 106. The focus estimation module 104 may include a machine learning model 105. The PMC 102 may be configured to obtain a first image representing at least a portion of a semiconductor sample via an I / O interface 126. The first image may be acquired by an inspection tool 120 (e.g., a SEM tool) configured with a first focal plane. The focus estimation module 104 may be configured to use a machine learning (ML) model 105 to estimate whether the first image is in focus. The ML model 105 is pre-trained to classify images into in-focus images and out-of-focus images.

[0059] Upon estimating the first image defocus (via the focus estimation module 104 ), the focus calibration module 106 may be configured to perform a focus calibration on the inspection tool in order to select a second focal plane associated with an optimal focus score.

[0060] PMC 102 can be configured to acquire a second image acquired by inspection tool 120 configured with a second focal plane via I / O interface 126, and focus estimation module 104 can be configured to use the ML model to estimate whether the second image is in focus. When the second image is estimated to be in focus, the second image can be used to inspect the semiconductor sample as described above.

[0061] The operation of the system 101, PMC 102 and the functional modules therein will refer to Figure 2 , Figure 3 and Figure 4 Further details.

[0062] According to certain embodiments, the system 101 may include a storage unit 122. The storage unit 122 may be configured to store any data required to operate the system 101 (e.g., data related to the input and output of the system 101), as well as intermediate processing results generated by the system 101. As an example, the storage unit 122 may be configured to store images generated by the inspection tool 120 and / or its derivatives. Thus, the images may be retrieved from the storage unit 122 and provided to the PMC 102 for further processing. In some cases, the storage unit 122 may be configured to store a training data set for training a machine learning model. Optionally, the storage unit 122 may also store the original ML model before training and / or the trained ML model.

[0063] In some embodiments, the system 101 may optionally include a computer-based graphical user interface (GUI) 124 that is configured to enable user-specified input related to the system 101. For example, a visual representation of a sample may be presented to the user (e.g., via a display forming part of the GUI 124), including an acquired sample image. The user may be provided with the option of defining certain operating parameters (such as, for example, a focus sensitivity requirement and / or a focus threshold) via the GUI. In some cases, the user may also view the results of the operation (such as whether the image is estimated to be in focus or out of focus), and / or further inspection results (e.g., detection results and / or review results and / or measurements of in-focus images) on the GUI.

[0064] As described above, the system 101 is configured to receive one or more images of the sample via the I / O interface 126. The image may include image data (and / or derivatives thereof) generated by the inspection tool 120 and / or image data stored in the storage unit 122 or one or more data repositories. In some cases, the image data may refer to images captured by the inspection tool during the manufacturing process, and / or pre-processed images derived from captured images obtained through various pre-processing stages, etc. It should be noted that in some cases, the image may include associated digital data (e.g., metadata, artificial attributes, etc.). It should also be noted that the image data may include data related to a layer of interest in the sample and / or data related to multiple layers of the sample.

[0065] The system 101 is also configured to process the received image and send the results or parts thereof (eg, images estimated to be in focus and / or further inspection results) to the storage unit 122 and / or the inspection tool 120 via the I / O interface 126 .

[0066] In some embodiments, in addition to the inspection tool 120, the inspection system 100 may include one or more inspection modules, such as, for example, a defect inspection and detection module, and / or an automatic defect review module (ADR), and / or an automatic defect classification module (ADC), and / or a metrology-related module, and / or other inspection modules that can be used to inspect semiconductor samples. One or more inspection modules can be implemented as independent computers, or the functions of one or more inspection modules (or at least some of them) can be integrated with the inspection tool 120. In some embodiments, the images obtained from the system 101 can be used by the inspection tool 120 and / or one or more inspection modules (or parts thereof) to further inspect the sample.

[0067] Those skilled in the art will readily appreciate that the teachings of the disclosed subject matter are not limited to Figure 1Constraints of the systems shown; equivalent and / or modified functionality may be combined or divided in another manner and implemented in any suitable combination of software and firmware and / or hardware.

[0068] It should be noted that Figure 1 The inspection system shown can be implemented in a distributed computing environment, wherein the aforementioned functional modules included in the PMC 102 can be distributed on several local and / or remote devices and can be linked via a communication network. It should also be noted that in other embodiments, at least some of the (multiple) inspection tools 120, storage units 122, and / or GUIs 124 can be external to the inspection system 100 and communicate data with the system 101 via the I / O interface 126. The system 101 can be implemented as (multiple) independent computers used in conjunction with the inspection tools. Alternatively, the various functions of the system 101 can be at least partially integrated with one or more inspection tools 120, thereby facilitating and enhancing the functions of the inspection tools 120 in the inspection-related processes.

[0069] Although not necessarily so, the operation of systems 101 and 100 may correspond to the operation of Figures 2 to 4 Some or all of the stages of the method described. Figures 2 to 4 The described methods and possible implementations thereof may be implemented by the systems 101 and 100. Therefore, it should be noted that Figures 2 to 4 The described method-related embodiments may also be implemented, mutatis mutandis, as various embodiments of the systems 101 and 100 , and vice versa.

[0070] Now refer to Figure 2 , Figure 2 A general flow chart for performing runtime defect inspection on semiconductor samples according to certain embodiments of the disclosed subject matter is shown.

[0071] A first image representing at least a portion of a semiconductor sample may be obtained (202) (e.g., by PMC 102 via I / O interface 126). The first image may be acquired by an inspection tool (e.g., by inspection tool 120) configured with a first focal plane. As an example, the first image may be acquired by an electron beam tool (such as a SEM tool) during runtime inspection of the semiconductor sample. It should be noted that the terms "runtime" or "runtime inspection" as used herein refer to a production phase during a manufacturing process of a FAB where actual in-line inspection of the semiconductor sample is performed, as compared to a recipe setup phase or a test phase or a training phase, which is an offline setup phase in preparation for runtime inspection.

[0072] It should also be noted that the term "first" used throughout the specification does not necessarily mean the first item, e.g., the first image captured by the inspection tool, or the first focal plane configured; rather, "first" is used to distinguish between a "second" or "third" (different) item. For example, a first image is an image different from a second or third image, and a first focal plane is a focal plane different from a second or third focal plane.

[0073] The focal plane (also called the plane of focus, or focal plane) is the plane perpendicular to the lens axis and passing through the focal point. The focal plane provides the best (sharpest) focus at a specific depth of field. The distance from the focal plane to the lens is equal to the focal length of the lens.

[0074] The first focal plane refers to the current focal plane set in the current configuration of the inspection tool before capturing the first image (for example, as the focal plane used in acquiring a previous image at a previous position of the sample, or, where the first image is actually the first image captured by the tool at the start of the inspection, as the focal plane initially configured in the tool).

[0075] As discussed above, it is recognized in the art that it is challenging for images acquired by an inspection tool to remain in focus over long inspection spans and changing patterns for a variety of reasons, such as, for example, focus scores being pattern dependent, charging effects built up on the tool, relative physical position changes, etc. Therefore, continuing with the description of block 202, even if a first focal plane was indeed optimally focused for capturing a previous image at a previous location on the sample, there is no guarantee that a first image captured at a current location using the same focal plane will still be in focus.

[0076] As used herein, an image pattern or design pattern may correspond to any primitive object or any combination of one or more primitive objects having a geometric shape or geometric structure with an outline on a sample, thereby forming a pattern. Examples of patterns may include general shape features / objects, including, for example, contacts, linings, etc., and / or any composite shapes formed by a combination of general shape features.

[0077] Certain embodiments of the disclosed subject matter propose an online focus estimation and calibration method to estimate whether an acquired image is in focus by accurately determining whether to trigger focus calibration based on a pre-trained machine learning model for each acquired image during runtime inspection.

[0078] The proposed method ensures that focus calibration is performed only when it is actually needed, thus not significantly increasing the scan duration and not affecting the inspection throughput (TpT). The method is easy to implement and does not require dedicated hardware or major system modifications to integrate with the current image acquisition process. In addition, the method can ensure that the images are all acquired in the optimal focus plane, thus guaranteeing the focus quality of the acquired images.

[0079] Specifically, whether the first image is in focus is estimated (204) (e.g., by the focus estimation module 104 included in the PMC 102) using a machine learning (ML) model (e.g., the ML module 105 included in the focus estimation module 104). The ML model is previously trained to classify images into in-focus images (also referred to as focused images) and defocused images (also referred to as out-of-focus images).

[0080] According to certain embodiments, the ML model 106 referred to herein may be implemented as various types of machine learning models, such as, for example, a decision tree, a support vector machine (SVM), a regression model, a neural network, or a collection or combination thereof, etc. The learning algorithm used by the ML model may be any of the following: supervised learning, unsupervised learning, or semi-supervised learning, etc. The disclosed subject matter is not limited to a specific type of ML model or the use of a specific type or learning algorithm by the ML model.

[0081] In some embodiments, the ML model 106 may be implemented as a deep neural network (DNN) that includes layers organized according to a corresponding DNN architecture. As non-limiting examples, the layers of the DNN may be organized according to a convolutional neural network (CNN) architecture, a recurrent neural network architecture, a recursive neural network architecture, a generative adversarial network (GAN) architecture, or other means. An exemplary implementation of a DNN that may be used to employ an ML model is now described.

[0082] Each layer of a DNN may include a plurality of basic computing elements (CEs), which are generally referred to in the art as dimensions, neurons, or nodes. Typically, a CE of a given layer may be connected to a CE of a previous layer and / or a subsequent layer. Each connection between a CE of a previous layer and a CE of a subsequent layer is associated with a weight value. A given CE may receive input from a CE of a previous layer via a corresponding connection, and each given connection is associated with a weight value of the input that may be applied to the given connection. The weight value may determine the relative strength of the connection, thereby determining the relative influence of each input on the output of a given CE. A given CE may be configured to calculate an activation value (e.g., a weighted sum of inputs), and further derive an output by applying an activation function to the calculated activation. The activation function may be, for example, an identity function, a deterministic function (e.g., linear, sigmoid, threshold, etc.), a random function, or other suitable function. The output from a given CE may be transmitted to a CE of a subsequent layer via a corresponding connection. Similarly, as described above, each connection at the output of a CE may be associated with a weight value, which may be applied to the output of the CE before being received as an input to a CE of a subsequent layer. In addition to the weight value, there may also be a threshold value (including a limiting function) associated with a connection and a CE.

[0083] The ML model (e.g., DNN) has a set of model parameters (such as, for example, weights and / or thresholds of the DNN) that are optimized during a training phase. The set of DNN input data used to adjust the DNN model parameters is hereinafter referred to as a training set, or training data set, or training data. The training of the ML model can be performed in advance by a training engine during the training phase, as will be described below with reference to Figure 3 Detailed.

[0084] Now turn to Figure 3 , Figure 3 A general flow chart for training an ML model according to certain embodiments of the disclosed subject matter is shown.

[0085] In some embodiments, the ML model may be implemented as a binary classifier, and a training dataset may be obtained (302) including a subset of in-focus images and a subset of out-of-focus images, each of which is associated with a corresponding true value label and used to train the binary classifier. The training of the ML model may be performed by a training engine, which in some cases may be a Figure 1 The training engine may be implemented as part of system 101 in the PMC 102 (e.g., as a training module included in PMC 102) or as part of system 100 (e.g., as a training unit included in system 100 and operably connected to system 101). Alternatively, in some other cases, the training engine may be implemented as a separate training system in a separate computer that may (or may not) be in communication with system 100 by wire or wirelessly.

[0086] According to some embodiments, a training data set may be obtained by, for each sample of one or more semiconductor samples having a corresponding design pattern, collecting (304) a set of images of the sample scanned at different focal plane configurations of an inspection tool, and providing (306) a ground truth label associated with the set of images.

[0087] Figure 5 An example of a collection of images scanned at different focus plane configurations according to certain embodiments of the disclosed subject matter is shown. As shown, image 500 is labeled as an in-focus image, while images 502, 504, and 506 are labeled as corresponding defocused images with respective degrees of defocus.

[0088] As mentioned above, in Figure 3In this exemplary embodiment of the present invention, the ML model can be implemented as a binary classifier. Therefore, it is of interest to identify which images in a collection of images are considered to be out of focus and which images are considered to be in focus, rather than to determine the absolute degree of focus of each image in the collection. Therefore, it is necessary to determine a focus threshold that can be used to distinguish between out-of-focus images and in-focus images. As an example, a focus threshold can be applied to a collection of images to provide a true value label (e.g., in-focus or out-of-focus) associated with an image.

[0089] In some embodiments, the focus threshold may be determined based on a focus sensitivity requirement. Focus sensitivity refers to the sensitivity that triggers the focus calibration process, as described below with respect to block 206. The focus sensitivity requirement may be predetermined by a customer and / or application engineer for a specific inspection application (e.g., defect detection of a specific sample, or defect review, etc.). As an example, for a defect detection application, the focus sensitivity requirement may be determined as a 10% resolution loss relative to an in-focus image.

[0090] In order to meet the predefined focus sensitivity requirement, the focus threshold needs to be determined accordingly so that the defocused image and the focused image can be correctly identified. However, there are various influencing factors, such as, for example, different image patterns and / or SEM working points, which may result in different focus thresholds corresponding to the predefined focus sensitivity requirement.

[0091] As an example, as described above, the calculation of the focus score of an image depends largely on the image pattern. Therefore, the focus scores indicating the same degree of focus calculated for different patterns may be different, which inevitably leads to different focus thresholds.

[0092] As another example, the SEM operating point may be defined for at least one of the following operating parameters: electron beam energy, beam current, extraction voltage, etc. Different operating points may affect the focus depth of the inspection tool, which in turn will affect the focus threshold.

[0093] Therefore, the same focus sensitivity requirement (e.g., 10% resolution loss) may result in different focus thresholds corresponding to different image patterns and / or different working points. Therefore, while taking into account the influencing factors (such as, for example, SEM working point, etc.) as described above, a focus threshold associated with a given sample with a specific design pattern (for which a set of images is obtained as described above with reference to block 304) can be calculated based on the focus sensitivity requirement.

[0094] Once the focus threshold is derived, the focus threshold can be applied to a set of images obtained for a given sample, for example, by comparing the focus scores of the images in the set to the focus threshold. As an example, assume that the focus scores are normalized and a score of 1 represents an image with optimal focus. Images with focus scores below the focus threshold can be labeled with a true value label of defocus (or out of focus), while images with focus scores above (or equal to) the focus threshold can be labeled with a true value label of in focus (or focused). As another example, if the focus score is defined in other ways, for example, a smaller score indicates a better degree of focus, images with focus scores above (or equal to) the focus threshold can be labeled with a true value label of defocus (or out of focus), while images with focus scores below the focus threshold can be labeled with a true value label of in focus (or focused). The present disclosure is not limited to a particular manner in which the focus threshold is applied to the focus score of an image.

[0095] According to certain embodiments, a focus score for an image may be calculated using different focus measurements that assess the clarity or focus level of an image, and the present disclosure is not limited to a particular focus score calculation. As an example, a gradient-based focus measurement based on a gradient or approximation of a first-order derivative of an image may be used to calculate a focus score. Such a focus measurement follows the assumption that a focused image presents sharper edges than a blurred image. Therefore, the energy of the gradient may be used to estimate the focus level. Similarly, a Laplacian-based focus measurement based on a second-order derivative of the image may also be used. As another example, a statistics-based focus measurement based on a textual descriptor of the image may be used. This focus measurement follows the assumption that a defocused image can be interpreted as a texture whose smoothness increases with increasing defocus level.

[0096] It should be noted that the above focus measures are for exemplary purposes only and should not be considered to limit the present disclosure in any way. In addition to or in place of the above, other suitable focus measures may be used, such as, for example, wavelet-based focus measures or image contrast-based focus measures.

[0097] The training data set obtained with reference to block 302 may be used to train (308) an ML model. As an example, initial values ​​of model parameters of an ML model (e.g., a DNN) may be selected prior to training, and may be further iteratively adjusted or modified during training to achieve an optimal set of weight values ​​and / or thresholds in the trained DNN. After each iteration, a difference between an actual output produced by the DNN and a target output associated with the corresponding training data set may be determined. The difference may be referred to as an error value. Training may be determined to be complete when a loss function indicating an error value is less than a predetermined value, or when a limited performance change is achieved between iterations.

[0098] In some cases, the above-described training data collection process as described with respect to blocks 304 and 306 may be performed as an offline process in the training phase. In order to meet relatively high focus sensitivity, the ML model should be trained on a training data set that includes as many image patterns and structures as possible at different focal planes. Collecting such extensive training data offline may be very time-consuming. In addition, capturing multiple images with different focus configurations at the same location of a sample by a SEM tool may cause certain physical effects on the sample (such as, for example, burn marks), thereby affecting the content of the captured images.

[0099] To address the training data collection problem described above, a generic training dataset may first be created that can be enriched each time a new sample (i.e., a manufactured sample to be inspected) is scanned during in-line runtime inspection of the new sample. Specifically, such scanning may be incorporated into an in-line focus calibration process that occurs prior to the actual inspection (as discussed below with respect to Figure 2 The samples will be scanned at different focus configurations, and the scanned images can be manually labeled and added to the training dataset. In this way, the training dataset can be enriched in both offline and online processes, and the ML model can be continuously updated based on the continuously enriched training dataset (e.g., by continuously retraining the ML model using the enriched training dataset).

[0100] According to certain embodiments, in addition to or in lieu of the online data collection process described above, data augmentation may be used to address the problem of insufficient size and diversity of the training dataset. As an example, data augmentation may be used to artificially generate defocused images (e.g., by learning defocus properties and their impact on image quality). In one embodiment, one or more defocused images in the defocused image subset are artificially generated using a generative adversarial network (GAN) based on image patterns indicated in corresponding in-focus images in the in-focus image subset.

[0101] GAN refers to a neural network system consisting of two neural networks competing with each other in a zero-sum game framework. The generator network G is trained to generate samples / instances by transforming the input data into a specific data distribution of interest, while the discriminator network D is trained to distinguish between instances from the real data distribution and instances generated by the generator. The training goal of the generator network is to increase the error rate of the discriminator network (i.e., to "fool" the discriminator network by generating new synthetic instances that appear to be from the real data distribution), while the training goal of the discriminator network is to accurately distinguish between instances generated by the generator and real data.

[0102] According to certain embodiments, a generator network G used in the present disclosure may be trained to generate one or more artificial defocused images based on an image pattern indicated in a corresponding in-focus image. As an example, the input to the generator network G may be a design pattern corresponding to the in-focus image (e.g., as obtained from CAD design data of the sample). As another example, the input to the generator network G may be the in-focus image itself. Optionally, in some cases, in addition to the image pattern indication, a random data vector may be provided to the generator, and the random data vector may be used by the generator to reconstruct and create a specific defocus representation to be combined / embedded into the reconstructed image of the in-focus image.

[0103] The input to the discriminator network includes artificial defocused images generated by the generator and real defocused images. The discriminator learns to distinguish between artificial defocused images generated by the generator and real defocused images. The generator and the discriminator can be trained together, thereby improving the performance of each other. With the help of the discriminator, the generator can be trained to generate more realistic and realistic defocused images (with different degrees of focus) at any desired resolution.

[0104] Additionally or alternatively, in some cases, data augmentation can also be used to enrich the training dataset by artificially creating images for samples with different pattern types. This can be performed in a similar manner using, for example, a GAN.

[0105] continue Figure 2 In the description of FIG. 2 , the training process of the ML model for estimating ( 204 ) whether the first image is in focus has been described. Now, the focus calibration process based on the estimation result is described for block 206 .

[0106] Specifically, when the first image defocus is estimated, a focus calibration may be performed (206) on the inspection tool (e.g., by a focus calibration module 106 included in the PMC 102) to select a second focus plane associated with an optimal focus score. In other words, the focus calibration process is only triggered when the first image defocus is estimated.

[0107] Now refer to Figure 4 , Figure 4 A general flow chart for focus calibration according to certain embodiments of the disclosed subject matter is shown.

[0108] Specifically, a focal plane of the inspection tool can be adjusted (402) to a plurality of focal points along a focal axis. As an example, the focal plane can be adjusted by mechanically moving a stage for the semiconductor sample (e.g., to a different height relative to a detector of the tool). As another example, the focal plane can be adjusted by electrically changing the landing energy of an electron beam of the inspection tool (e.g., to different intensities that can result in different degrees of focus in the image).

[0109] A plurality of images corresponding to a plurality of focal points may be acquired (404). A focus score may be calculated (406) for each of the plurality of images, resulting in a plurality of focus scores corresponding to the plurality of images. As described above, different focus measures may be used to calculate the focus scores, and the present disclosure is not limited to employing a particular focus measure. An image having an optimal focus score among the plurality of focus scores may be selected (408) from the plurality of images, and an inspection tool may be configured (410) with a second focus plane associated with the selected image (i.e., the focus plane at which the selected image was acquired).

[0110] A second image may be acquired (208) by an inspection tool configured with a second focal plane (e.g., by PMC 102 via I / O interface 126) and whether the second image is in focus may be estimated (210) using the ML model (e.g., by focus estimation module 104 included in PMC 102). When the second image is estimated to be in focus, the second image may be used (212) to perform defect inspection (such as defect detection and / or defect classification) on the sample.

[0111] On the other hand, if the second image is estimated to be out of focus, the process returns to block 206 and may (according to reference Figure 4 The process described in 206 to 210 may be repeated until the image acquired by the inspection tool is estimated by the ML model to be in focus.

[0112] According to certain embodiments, as described above with reference to Figure 2 , Figure 3 and Figure 4 The described focus calibration process can be included as part of an inspection recipe that can be used by the system 101 and / or the inspection tool 120 to inspect a sample at run time, such as, for example, defect detection and / or defect classification. In this case, the disclosed subject matter also includes systems and methods for generating an inspection recipe during a recipe setup phase, wherein the recipe includes the following: Figure 2 , Figure 3 and Figure 4 (and its various embodiments) described herein. It should be noted that the term "inspection recipe" should be interpreted broadly to cover any recipe that can be used by an inspection tool to perform operations related to any kind of inspection as described above.

[0113] It should be noted that the examples set forth in the present disclosure (such as, for example, ML model types, focus measurements, data enhancement techniques, etc.) are set forth for exemplary purposes and should not be considered to limit the present disclosure in any way. Other examples may be used in addition to or in place of the above.

[0114] One of the advantages of certain embodiments of the disclosed subject matter as described herein is to provide an online focus estimation and calibration process that is capable of accurately determining whether to trigger a focus calibration for each acquired image during runtime inspection based on a machine learning model that is pre-trained to estimate whether the acquired image is in focus. Since the focus calibration is triggered only when it is actually needed, such a process, although performed during the online inspection process, does not significantly increase the scan duration or affect the inspection throughput (TpT).

[0115] Furthermore, the online focus estimation and calibration process is easy to implement and does not require dedicated hardware or major system modifications to integrate with the current image acquisition process.

[0116] One of the advantages of certain embodiments of the focus estimation and calibration process as described herein is to ensure that the output images are acquired at the optimal focus plane, thereby ensuring the focus quality of the acquired images. The resulting images have reduced image artifacts in terms of defocus, which can provide better results with higher accuracy when the images are used for inspection of semiconductor samples (such as, for example, defect inspection).

[0117] It is to be understood that the disclosure is not limited in its application to the details set forth in the description contained herein or illustrated in the drawings.

[0118] It should also be understood that the system according to the present disclosure can be implemented at least in part on a suitably programmed computer. Likewise, the present disclosure contemplates a computer program that can be read by a computer for performing the method of the present disclosure. The present disclosure further contemplates a non-transitory computer-readable memory that tangibly embodies an instruction program that can be executed by a computer for performing the method of the present disclosure.

[0119] The present disclosure can have other embodiments and can be practiced and implemented in various ways. Therefore, it should be understood that the wording and terminology adopted herein are for descriptive purposes and should not be considered as limiting. Therefore, it will be understood by those skilled in the art that the concepts on which the present disclosure is based can be easily used as the basis for designing other structures, methods and systems for realizing the multiple purposes of the disclosed subject matter.

[0120] It will be readily appreciated by those skilled in the art that various modifications and changes may be applied to the embodiments of the present disclosure as described above, without departing from the scope thereof as defined in and by the appended claims.

Claims

1. A computerized system for performing runtime defect inspection on semiconductor samples, the system comprising: Processing and Memory Circuitry (PMC), Wherein the PMC is configured to perform a method comprising the following steps: obtaining, by the PMC, a first image representing at least a portion of the semiconductor sample, wherein the first image is acquired by an inspection tool configured with a first focal plane; estimating, by the PMC, whether the first image is in focus using a machine learning (ML) model, wherein the ML model is pre-trained to classify images into in-focus images and out-of-focus images; performing, by the PMC, focus calibration on the inspection tool to select a second focal plane associated with an optimal focus score when estimating the first image defocus; and A second image acquired by the inspection tool configured with the second focal plane is obtained by the PMC, and whether the second image is in focus is estimated by the PMC using the ML model, wherein when the second image is estimated to be in focus, the second image is used to perform defect inspection on the semiconductor sample.

2. The computerized system of claim 1, wherein the inspection tool is an electron beam tool.

3. The computerized system of claim 1 , wherein the ML model is a binary classifier that was previously trained using a training dataset, the training dataset comprising a subset of in-focus images and a subset of out-of-focus images, the subset of in-focus images and the subset of out-of-focus images each being associated with a corresponding true value label.

4. The computerized system of claim 3 , wherein the training data set is obtained by collecting, for each of one or more samples having a corresponding design pattern, a set of images of the sample scanned under different focal plane configurations of the inspection tool, and providing true value labels respectively associated with the set of images.

5. The computerized system of claim 4, wherein the truth label is provided by applying a focus threshold to the set of images, wherein the focus threshold is determined based on a focus sensitivity requirement for a particular inspection application.

6. The computerized system of claim 3, wherein one or more defocused images in the subset of defocused images are artificially generated using a generative adversarial network (GAN) based on image patterns indicated in corresponding in-focus images in the subset of in-focus images.

7. The computerized system of claim 1 , wherein the focus calibration is performed by: adjusting a focal plane of the inspection tool to a plurality of focal points along a focal axis; acquiring a plurality of images corresponding to the plurality of focal points; calculating a focus score for each image in the plurality of images, thereby generating a plurality of focus scores; selecting, from the plurality of images, an image having an optimal focus score among the plurality of focus scores; and The inspection tool is configured with a second focal plane associated with the selected image.

8. The computerized system of claim 7, wherein the focal plane is adjusted by mechanically moving a stage for placing the semiconductor sample or electrically changing a landing energy of an electron beam of the inspection tool.

9. A computerized system according to claim 1, wherein when estimating the defocus of the second image, the PMC is further configured to perform focus calibration on the inspection tool to select a third focal plane associated with an optimal focus score, obtain a third image acquired by the inspection tool configured with the third focal plane, and estimate whether the third image is in focus.

10. The computerized system of claim 1, wherein upon estimating the first image is in focus, the PMC is further configured to determine to use the first image to inspect the sample for defects.

11. A computerized method for performing runtime defect inspection on a semiconductor sample, the method being performed by a processing and memory circuit system (PMC), the method comprising: obtaining a first image representing at least a portion of the semiconductor sample, wherein the first image is acquired by an inspection tool configured with a first focal plane; estimating whether the first image is in focus using a machine learning (ML) model, wherein the ML model is pre-trained to classify images into in-focus images and out-of-focus images; performing a focus calibration on the inspection tool to select a second focus plane associated with an optimal focus score when estimating the first image defocus; and A second image acquired by the inspection tool configured with the second focal plane is obtained, and the second image is estimated using the ML model to determine whether the second image is in focus, wherein the second image is used to perform defect inspection on the semiconductor sample when the second image is estimated to be in focus.

12. The computerized method of claim 11, wherein the inspection tool is an electron beam tool.

13. The computerized method of claim 11, wherein the ML model is a binary classifier that was previously trained using a training dataset that includes a subset of in-focus images and a subset of out-of-focus images, each of the subset of in-focus images and the subset of out-of-focus images being associated with a corresponding true value label.

14. The computerized method of claim 13, wherein the training data set is obtained by collecting, for each of one or more samples having a corresponding design pattern, a set of images of the sample scanned under different focal plane configurations of the inspection tool, and providing true value labels respectively associated with the set of images.

15. The computerized method of claim 14, wherein the truth labels are provided by applying a focus threshold to the set of images, wherein the focus threshold is determined based on a focus sensitivity requirement for a particular inspection application.

16. The computerized method of claim 13, wherein one or more defocused images in the subset of defocused images are artificially generated using a generative adversarial network (GAN) based on image patterns indicated in corresponding in-focus images in the subset of in-focus images.

17. The computerized method of claim 11, wherein the focus calibration is performed by: adjusting a focal plane of the inspection tool to a plurality of focal points along a focal axis; acquiring a plurality of images corresponding to the plurality of focal points; calculating a focus score for each image in the plurality of images, thereby generating a plurality of focus scores; selecting, from the plurality of images, an image having an optimal focus score among the plurality of focus scores; and The inspection tool is configured with a second focal plane associated with the selected image.

18. The computerized method of claim 17, wherein the focal plane is adjusted by mechanically moving a stage for placing the semiconductor sample or electrically changing a landing energy of an electron beam of the inspection tool.

19. The computerized method of claim 11, further comprising: When the second image is estimated to be defocused, focus calibration is performed on the inspection tool to select a third focal plane associated with an optimal focus score, a third image acquired by the inspection tool configured with the third focal plane is obtained, and whether the third image is estimated to be in focus.

20. A non-transitory computer-readable storage medium tangibly embodying a program of instructions, which, when executed by a computer, causes the computer to perform a method of performing runtime defect inspection on a semiconductor sample, the method comprising: obtaining a first image representing at least a portion of the semiconductor sample, wherein the first image is acquired by an inspection tool configured with a first focal plane; estimating whether the first image is in focus using a machine learning (ML) model, wherein the ML model is pre-trained to classify images into in-focus images and out-of-focus images; performing a focus calibration on the inspection tool to select a second focus plane associated with an optimal focus score when estimating the first image defocus; and A second image acquired by the inspection tool configured with the second focal plane is obtained, and the second image is estimated using the ML model to determine whether the second image is in focus, wherein the second image is used to perform defect inspection on the semiconductor sample when the second image is estimated to be in focus.

Images