Defect detection using computationally efficient segmentation methods

Abstract

An inspection system of a semiconductor sample is provided, the system comprising a processor and a memory circuit configured to, for each given candidate defect of a plurality of candidate defects in an image of the sample, obtain a given region of the given candidate defect in the image, obtain a reference image, segment at least a portion of the reference image to determine, for each given candidate defect, a first reference region in the reference image that matches a given reference region corresponding to the given region, select a plurality of second reference regions in the first reference region, obtain a plurality of corresponding second regions in the image, and determine whether the given candidate defect corresponds to a defect using data providing pixel intensity information of the second regions and data providing pixel intensity information of the given region.

Description

Technical Field

[0001] The subject matter of this disclosure generally relates to the field of sample inspection, and more specifically, to the automated inspection of samples. Background Technology

[0002] Current demands for high density and high performance associated with the ultra-large-scale integration of manufactured devices require submicron feature structures, higher transistor and circuit speeds, and improved reliability. These demands require the formation of device feature structures with high precision and uniformity, which in turn necessitates careful monitoring of the manufacturing process, including automated inspection of devices while they are still in semiconductor wafer form.

[0003] Inspection processes are used at various stages of semiconductor manufacturing to detect and classify defects on samples. The effectiveness of inspection can be improved through process automation, such as Automatic Defect Classification (ADC) and Automatic Defect Review (ADR).

[0004] The applicant's US9,235,885 (the contents of which are incorporated herein by reference in their entirety) describes a system capable of inspecting for defects in articles. The system includes: a patch comparator configured to determine a similarity level for each of a plurality of reference patches in a reference image based on a predefined patch-similarity criterion and source patches defined in a reference image; an evaluation module configured to score each of the inspected pixels in the inspected image, excluding a plurality of inspected pixels, with a representative score based on the similarity level of the reference patches associated with a reference pixel corresponding to the inspected pixel; a selection module configured to select a plurality of selected inspected pixels based on the representative scores of the plurality of inspected pixels; and a defect detection module configured to determine whether the candidate pixels are defective based on the test values ​​of the candidate pixels and the test values ​​of the selected inspected pixels. Summary of the Invention

[0005] According to certain aspects of the subject matter of this disclosure, an inspection system for a semiconductor sample is provided, the system including a processor and memory circuit (PMC) configured to: for each given candidate defect among a plurality of candidate defects in an image of the semiconductor sample acquired by an inspection tool, obtain a given region of the given candidate defect in the image; obtain a reference image; and segment at least a portion of the reference image, the segmentation comprising: for each given candidate defect: obtaining a given reference region in the reference image corresponding to the given region of the given candidate defect in the image according to a correspondence criterion; determining a first reference region in the reference image according to a first similarity criterion, wherein for each given first reference region, a pixel intensity of the given first reference region is provided. First data matching of intensity information provides first data for pixel intensity information of the given reference region; for each given candidate defect: for each given first reference region, second data providing pixel intensity information of the given first reference region is compared with second data providing pixel intensity information of the given reference region, and multiple second reference regions are selected in the first reference region according to a second similarity criterion, wherein the comparison indication is a match for the second reference region, and multiple second regions in the image are obtained according to the correspondence criterion, wherein each given second region corresponds to one of the second reference regions, and the data providing pixel intensity information of the second region and the data providing pixel intensity information of the given region are used to determine whether the given candidate defect corresponds to a defect.

[0006] According to some implementations, for each given candidate defect, the number of the second reference regions is less than the number of the first reference regions.

[0007] According to some implementations, the PMC is configured to: apply at least one filter to the image to obtain at least one processed image; and for each given first reference region, use the at least one processed image to determine first data providing pixel intensity information for the given first reference region, and use the at least one processed image to determine the first data providing pixel intensity information for the given reference region.

[0008] According to some implementations, the at least one processed image is calculated once for all candidate defects.

[0009] According to some embodiments, the second data providing pixel intensity information of the given reference region and the second data providing pixel intensity information of each given first reference region are determined using at least one of the following: the reference image; or one or more second-processed images obtained by filtering at least a portion of the following: the reference image, or an image generated using the reference image. According to some embodiments, the second data providing pixel intensity information of the given reference region includes a reference patch providing one or more pixel information of the given reference region; for each given first reference region: the second data providing pixel intensity information of the given first reference region includes a patch providing one or more pixel information of the given first reference region, and wherein the system is configured to compare the patch of the given first reference region with the reference patch.

[0010] According to some implementations, the PMC is configured to select a limited subset of pixels of the reference patch, and for each given first reference region, data providing pixel intensity information of the limited subset of pixels of the reference patch is compared with data providing pixel intensity information of the corresponding pixels in the patch of the given first reference region.

[0011] According to some implementations, selecting the finite subset of pixels includes at least one of the following: selecting pixels in the image whose corresponding pixel is associated with a pixel intensity above a threshold, or selecting pixels in the image whose corresponding pixel is associated with a probability of the presence of a defect above a threshold.

[0012] According to some implementations, for each given candidate defect, selecting the second reference region in the first reference region is more computationally efficient than selecting the second reference region throughout the entire reference image.

[0013] According to some implementations, for each given first reference region, the first data providing pixel intensity information of the given first reference region is different from the second data providing pixel intensity information of the given first reference region.

[0014] According to some implementations, the PMC is configured to, for each given candidate defect among a plurality of candidate defects: for each given candidate defect among a plurality of candidate defects, obtain: a first given region of the given candidate defect in a first image of the semiconductor sample acquired by the inspection tool, and a second given region of the given candidate defect in a second image of the semiconductor sample acquired by the inspection tool, wherein the second image differs from the first image in at least one parameter; the first image is used only if the candidate pixel satisfies a correlation criterion in the first image to determine whether the candidate defect is a defect, and / or the second image is used only if the candidate pixel satisfies a correlation criterion in the second image to determine whether the candidate defect is a defect.

[0015] According to certain aspects of the subject matter of this disclosure, a method for inspecting a semiconductor sample is provided, the method comprising: a processor and memory circuit (PMC) for: for each given candidate defect among a plurality of candidate defects in an image of the semiconductor sample acquired by an inspection tool, obtaining a given region of the given candidate defect in the image; obtaining a reference image; segmenting at least a portion of the reference image, the segmentation comprising: for each given candidate defect: obtaining a given reference region in the reference image corresponding to the given region of the given candidate defect in the image according to a correspondence criterion; determining a first reference region in the reference image according to a first similarity criterion, wherein for each given first reference region, a first data matching providing pixel intensity information of the given first reference region is provided. The first data of pixel intensity information of a given reference region; for each given candidate defect: for each given first reference region, comparing the second data providing pixel intensity information of the given first reference region with the second data providing pixel intensity information of the given reference region; selecting a plurality of second reference regions in the first reference region according to a second similarity criterion, wherein the comparison indication is a match for the second reference regions; obtaining a plurality of second regions in the image according to the correspondence criterion, wherein each given second region corresponds to one of the second reference regions; and using the data providing pixel intensity information of the second region and the data providing pixel intensity information of the given region to determine whether the given candidate defect corresponds to a defect.

[0016] According to some implementations, the method can achieve one or more features as described above with reference to the system.

[0017] According to certain other subjects in the subject matter of this disclosure, a non-transitory computer-readable medium is provided, the medium including instructions that, when executed by a processor, cause the processor to perform the operations described above with respect to the methods.

[0018] According to some implementations, the proposed solution enables an increase in the number of candidate defects provided to the defect detection algorithm without increasing the computation time required for the defect detection algorithm to analyze the candidate defects.

[0019] According to some implementations, the proposed solution enables an increase in the number of candidate defects provided to the defect detection algorithm while maintaining the same throughput.

[0020] According to some implementation methods, the proposed solution enables the improvement of the computational efficiency of defect detection algorithms while maintaining similar sensitivity.

[0021] According to some implementation methods, the proposed solution enables the detection of more defects than existing algorithms while maintaining the same computational cost.

[0022] According to some implementation methods, the proposed solution enables a reduction in the computation time required to analyze each candidate defect.

[0023] According to some implementation methods, the proposed solution optimizes the processing of multi-perspective images. Attached Figure Description

[0024] To understand this disclosure and how it can be implemented in practice, embodiments will now be described by way of non-limiting example only, with reference to the accompanying drawings, in which:

[0025] Figure 1 A general block diagram of an inspection system according to certain embodiments of the subject matter of this disclosure is shown.

[0026] Figure 2A A general flowchart of the method for determining whether each candidate defect is a defect in the sample image is shown.

[0027] Figure 2B A non-limiting example of a candidate defect in a sample image is shown.

[0028] Figure 2C A non-limiting example of reference regions identified in a reference image of a sample is shown, each reference region corresponding to a different candidate defect.

[0029] Figure 3AA non-limiting example of a first reference region identified in a reference image of a sample is shown, the first reference region being associated with a given reference region.

[0030] Figure 3B A non-limiting method for processing a reference image to obtain multiple processed reference images is shown.

[0031] Figure 3C A general flowchart of a method for selecting a finite number of first reference regions in a reference image is shown.

[0032] Figure 4A A non-limiting example of a mosaic piece given a first reference region is shown.

[0033] Figure 4B Various possible puzzle types are shown in the reference image.

[0034] Figure 4C A non-limiting example of a mosaic that can be used for mosaic comparison in a reference image is shown.

[0035] Figure 4D Non-limiting examples of piece comparisons in various second-processed images are shown.

[0036] Figure 5A A general flowchart of a method for comparing mosaics in a reference image or a second processed image is shown.

[0037] Figure 5B and Figure 5C It shows Figure 5A Non-restrictive examples of methods.

[0038] Figure 6A and Figure 6B A non-limiting example is shown of determining a second region in a sample image that corresponds to a second reference region in a reference image.

[0039] Figure 7A A general flowchart is shown for a method of using multi-perspective images to identify defects in an image.

[0040] Figure 7B It shows Figure 7A The implementation method of the method.

[0041] Figure 8A A general flowchart is shown for a method to determine whether a candidate defect is a defect by using a given region of a candidate defect in an image and a second region in the image associated with that region.

[0042] Figure 8B It shows Figure 8ANon-restrictive examples of methods.

[0043] Figure 9 It shows Figure 8A Another non-restrictive example of the method. Detailed Implementation

[0044] In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the contents of this disclosure. However, those skilled in the art will understand that the subject matter of this disclosure can be practiced without these specific details. In other instances, well-known methods, processes, components, and circuits have not been described in detail so as not to obscure the subject matter of this disclosure.

[0045] Unless otherwise specifically stated, as will be apparent from the following discussion, it is understood that the use of terms such as “processing,” “obtaining,” “selecting,” “determining,” “generating,” “using,” “comparing,” “judging,” or similar terms throughout the discussion of this specification refers to the actions and / or processes by which a computer processes data and / or converts data into other data, said data being represented as physical quantities, such as electronic quantities, and / or said data representing physical objects. The term “computer” should be broadly interpreted to encompass any kind of hardware-based electronic device with data processing capabilities, including, as a non-limiting example, system 103 and its corresponding portions disclosed in this application.

[0046] The terms “non-transitory memory” and “non-transitory storage medium” as used herein should be broadly interpreted to cover any volatile or non-volatile computer memory suitable for the subject matter of this disclosure.

[0047] The term “sample” as used in this specification should be broadly interpreted to encompass any kind of wafer, mask, and other structure, combination of and / or part thereof, used in the manufacture of semiconductor integrated circuits, magnetic heads, flat panel displays and other articles of semiconductor manufacturing.

[0048] The term "inspection" as used in this specification should be broadly interpreted to encompass any type of metrologically related operation as well as operations related to the detection and / or classification of defects in a sample during sample fabrication. Inspection is provided during or after the fabrication of the sample to be inspected, using non-destructive testing (NDT) tools. As a non-limiting example, the inspection process may include scanning (single or multiple scans), sampling, examination, measurement, classification, and / or other operations performed on a sample or a portion thereof using the same or different inspection tools in operation. Similarly, inspection may be provided prior to the fabrication of the sample to be inspected, and said inspection may include, for example, generating an inspection formulation and / or other setup operations. It should be noted that, unless specifically stated otherwise, the term "inspection" or its derivatives as used in this specification are not limited to the resolution or size of the inspection area. Various NDT tools include, as non-limiting examples, scanning electron microscopes, atomic force microscopes, optical inspection tools, etc.

[0049] As a non-limiting example, runtime inspection can employ a two-stage process, such as inspecting a sample and then examining the sampling locations for potential defects. During the first stage, the sample surface is inspected at high speed and relatively low resolution. In this first stage, a defect map is generated to display suspicious locations on the sample with a high probability of defect occurrence. During the second stage, at least some of the suspicious locations are analyzed more thoroughly at a relatively high resolution. In some cases, both stages can be implemented using the same inspection tool, and in other cases, they are implemented using different inspection tools.

[0050] The term “defect” as used in this specification should be interpreted broadly to encompass any type of abnormality or undesirable feature that forms on or within the sample.

[0051] It should be understood that, unless otherwise specifically stated, certain features of the subject matter of this disclosure described in the context of separate embodiments may also be provided in a combined manner in a single embodiment. Conversely, various features of the subject matter of this disclosure described in the context of a single embodiment may also be provided separately or in any suitable sub-combination. In the following detailed description, numerous specific details are set forth in order to provide a thorough understanding of these methods and apparatuses.

[0052] Therefore, please note Figure 1 This diagram illustrates a functional block diagram of an inspection system according to certain embodiments of the subject matter of this disclosure. Figure 1The inspection system 100 shown can be used for the inspection of samples (e.g., wafers and / or wafer components) as part of a sample manufacturing process. The illustrated inspection system 100 includes a computer-based system 103 capable of automatically determining metrologically relevant and / or defect-related information using images acquired during sample manufacturing. System 103 is operatively connected to one or more low-resolution inspection tools 101 and / or one or more high-resolution inspection tools 102 and / or other inspection tools. The inspection tools are configured to capture and / or view captured images and / or enable or provide measurements associated with the captured images. System 103 can be further operatively connected to a CAD server 110 and a data repository 109.

[0053] System 103 includes a processor and memory circuitry (PMC) 104 operatively connected to a hardware-based input interface 105 and a hardware-based output interface 106. The PMC 104 is configured to provide all the processing required by the operating system 103, as described in further detail below (see [link to documentation]). Figure 2A , Figure 5A , Figure 7A , Figure 7B and Figure 8A The method described herein (which may be performed at least in part by system 103) includes a processor (not shown separately) and a memory (not shown separately).

[0054] System 103 is configured to receive input data via input interface 105. Input data may include data generated by the inspection tool (and / or derived data and / or associated metadata), and / or data stored or generated in one or more data repositories 109 and / or in CAD server 110 and / or another related data repository. Note that input data may include images (e.g., captured images, images derived from captured images, simulated images, synthetic images, etc.) and associated digital data (e.g., metadata, hand-crafted attributes, etc.). It should also be noted that image data may include data associated with the layer of interest of the sample and / or with one or more other layers of the sample.

[0055] System 103 is further configured to process at least a portion of the received input data and send results (or a portion thereof) via output interface 106 to storage system 107, to inspection tools, to a computer-based graphical user interface (GUI) 108 for presenting the results, and / or to an external system (e.g., FAB's Yield Management System (YMS)). GUI 108 may be further configured to enable user-defined input associated with operating system 103.

[0056] As a non-limiting example, the sample can be examined by one or more low-resolution inspection machines 101 (e.g., optical inspection systems, low-resolution SEMs, etc.). The resulting data (hereinafter referred to as low-resolution image data 121) provides low-resolution image information of the sample and can be transmitted directly or via one or more intermediate systems to system 103. Alternatively or additionally, the sample can be examined by a high-resolution machine 102 (e.g., by examining a subset of potential defect locations selected for examination using a scanning electron microscope (SEM) or an atomic force microscope (AFM)). The resulting data (hereinafter referred to as high-resolution image data 122) provides high-resolution image information of the sample and can be transmitted directly or via one or more intermediate systems to system 103.

[0057] It should be noted that images of desired locations on a sample can be captured at different resolutions. As a non-limiting example, a so-called "defect image" of the desired location can be used to distinguish between defects and false alarms, while a so-called "class image" of the desired location is obtained at a higher resolution and can be used for defect classification. In some implementations, images of the same location (having the same or different resolutions) may include several images registered among them (e.g., an image captured from a given location and one or more reference images corresponding to that given location).

[0058] It should be noted that image data can be received and processed together with metadata associated with the image data (e.g., pixel size, textual description of defect type, parameters of the image capture process, etc.).

[0059] When processing input data (e.g., low-resolution image data and / or high-resolution image data, optionally together with other data such as design data, synthetic data, etc.), system 103 can send results (e.g., instruction-related data 123 and / or 124) to any inspection tool, store results (e.g., defect attributes, defect classification, etc.) in storage system 107, present results through GUI 108, and / or send them to external systems (e.g., YMS).

[0060] Those skilled in the art will readily understand that the teachings of this disclosure are not limited to... Figure 1 The system shown is subject to constraints; equivalent and / or modified functions may be combined or separated in another manner and may be implemented in any suitable combination of software and firmware and / or hardware.

[0061] Without limiting the scope of this disclosure in any way, it should also be noted that the inspection tool can be implemented as various types of inspection machines, such as optical imaging machines, electron beam inspection machines, etc. In some cases, the same inspection tool can provide both low-resolution and high-resolution image data. In some cases, at least one inspection tool can have metrological functions.

[0062] It should be noted that Figure 1 The inspection system shown can be implemented in a distributed computing environment, where Figure 1 The functional modules shown above can be distributed across several local and / or remote devices and can be linked via a communication network. It should be further noted that in other embodiments, at least some of the inspection tools 101 and / or 102, data repository 109, storage system 107, and / or GUI 108 can be located outside the inspection system 100 and operate by communicating with the system 103 via input interface 105 and output interface 106. The system 103 can be implemented as a standalone computer and used in conjunction with the inspection tools. Alternatively, the corresponding functions of the system can be at least partially integrated with one or more inspection tools.

[0063] Please note now. Figures 2A to 2C . Figure 2AThe method includes (operation 200) obtaining data (such as inspection tool 101 and / or 102) of multiple candidate defect information in an image 201 providing a semiconductor sample, acquired by an inspection tool. For each given candidate defect, the data includes a region in the image representing the location of the candidate defect within the image. This region may include at least one pixel, or in some embodiments, multiple pixels. For each region, the data further includes the pixel intensity of each pixel in that region. Pixel intensity corresponds to, for example, grayscale intensity. However, this is not limiting; pixel intensity may include various different components (e.g., pixel intensities of different channels, such as brightfield and grayfield, RGB, etc.), depending on the inspection tool and the acquisition type. Figure 2B In the non-restricted example, candidate defect 202 is associated with region 203, candidate defect 204 is associated with region 205, candidate defect 206 is associated with region 207, and candidate defect 208 is associated with region 209.

[0064] The data providing information on multiple candidate defects in image 201 can be provided by a system configured to perform a first determination of candidate defects in the image (based on, for example, the pattern and / or pixel intensity of the candidate defects in the image). In some embodiments, the first determination of candidate defects can be performed using an image acquired by low-resolution inspection machine 101. Since the number of candidate defects can be very large, additional methods are required (e.g., ...). Figure 2A The method is used to identify the actual defect among these candidate defects.

[0065] Figure 2A The method further includes obtaining (operation 210) a reference image 301. According to some embodiments, a plurality of reference images 301 are obtained. Each reference image 301 is a reference image of image 201.

[0066] Image 201 provides information about a first region of the semiconductor sample, and reference image 301 is an image of a reference region of the sample (e.g., die, cell, etc.). The first region is configured to satisfy a similarity criterion regarding the reference region and may belong to the same or different semiconductor samples. The similarity criterion may be defined, for example, that the first region and the reference region correspond to similar regions of the same physical component or semiconductor sample (e.g., similar dies, cells, etc.).

[0067] Image 201 and reference image 301 are comparable (e.g., die-to-die, cell-to-cell, die-to-database, etc.).

[0068] According to some embodiments, reference image 301 is an image of a reference region of a sample (e.g., a die, cell, etc.), wherein individual image data are verified to represent a defect-free reference region. The reference image may be an image captured from a reference (gold) die, reference cell, or other region verified to be defect-free. Alternatively or additionally, the reference image may be modeled using CAD data and / or may be augmented after capture to exclude defects (if any) in the reference region.

[0069] According to some implementations, in order to ensure compatibility between images, at least one reference image 301 and image 201 undergo a registration procedure.

[0070] Figure 2A The method further includes performing a first segmentation on at least a portion of the reference image 301.

[0071] As described above, for each candidate defect, a given region in image 201 has been obtained. The method includes determining (operation 220) a given reference region in reference image 301 that corresponds to the given region in image 201 according to a correspondence criterion.

[0072] Assuming image 201 and reference image 301 are aligned, then for each pixel in image 201, there is a corresponding pixel in reference image 301 that has a similar position. It should be noted that if image 201 and reference image 301 are not perfectly aligned (or due to other factors), a pixel at a given position in image 201 may correspond to a pixel in reference image 301 that has a different position in reference image 301.

[0073] In some implementations, a given reference region is selected as the region of reference image 301 that has the same location in reference image 301 as the given region in image 201.

[0074] Assume that corresponding pixels in different images represent corresponding locations in a sample (or similar locations in a sample). For example, assume that pixel P in image 201... 201 And the pixel P of reference image 301 301 They correspond to each other, and are composed of pixel P. 201 The location of the represented sample region relative to the first bare die (cell or sample) including that region can be determined by pixel P. 301 The location of the sample region is substantially the same relative to the second die (cell or sample) containing that region.

[0075] According to some implementations, a given region in image 201 and a given reference region in reference image 301 should provide at least a portion of the information of the same elements. For example, assuming that a given region provides information on a metal line at a specific location within a sample, the given reference region is expected to provide information on a similar metal line at the same specific location within a reference sample (e.g., in another sample comparable to this sample).

[0076] exist Figure 2B and Figure 2C In a non-limiting example, given regions 203, 205, 207 and 209 in image 201 correspond to reference regions 303, 305, 307 and 309 in reference image 301, respectively.

[0077] The reference image 301 is processed to identify (operation 225) a first reference region of the reference image 301 for each given reference region in the reference image 301. According to a first similarity criterion, a first reference region is selected in the reference image 301 such that, for each given first reference region, first data providing pixel intensity information for that given first reference region matches the first data providing pixel intensity information for that given reference region. Examples of the first data and examples of the first similarity criterion are provided below. As can be seen from the examples provided below, the first data providing pixel intensity information for that given first reference region is selected to be of the same type as the first data providing pixel intensity information for that given reference region, so that they are comparable.

[0078] In other words, for each given reference region, at least some first reference regions are considered to provide one or more element information similar to one or more elements present in that given reference region. For example, if a given reference region provides contact information, it is expected that at least some first reference regions will also provide similar contact information.

[0079] exist Figure 3A In the illustrated example, for reference region 307 (identified in operation 220), the first reference region 3071 to 307... N1 It has already been identified in reference image 301. For reference region 303 (identified in operation 220), the first reference regions 3031 to 303... N2 It has been identified in reference image 301.

[0080] According to some embodiments, in operation 225, the reference image 301 is processed to generate a processed reference image. According to some embodiments, a filter (also called a "kernel") is applied to the reference image 301 to obtain the processed reference image (this may include a convolution operation between the reference image 301 and the filter). In some embodiments, the filter is applied to the entire reference image 301 to obtain the processed reference image. In some embodiments, this operation is computationally efficient because the entire reference image 301 is processed together for a given filter.

[0081] According to some implementations, a finite number of filters are used to process the entire reference image 301 to obtain a finite number of processed images. This is in Figure 3B The description is as follows, wherein filters 350, 351, and 352 are applied to reference image 301 to obtain processed images 361, 371, and 381, respectively. Although reference image 301 includes multiple reference regions (one reference region for each candidate defect), the same processed image can be used to identify a first reference region similar to that given reference region for each given reference region.

[0082] Since the processed image has already been calculated once for all candidate defects, it is not necessary to recalculate the processed image for each different candidate defect each time. Instead, the same processed image obtained from a single calculation can be used to identify the first reference region associated with each given reference region. As a result, operation 225 is computationally efficient.

[0083] According to some implementations, the filter is a differential filter (derivative kernel) that enables the acquisition of the derivative of the pixel intensity of the reference image 301. This filter can be convolved with the reference image 301. Two non-limiting examples of this filter are provided below:

[0084]

[0085]

[0086] The first filter G x (Also known as the Sobel kernel along the X-axis) can be used in particular to determine the derivative along the horizontal axis (X-axis) and the second filter G y (Also known as the Sobel kernel along the Y-axis) can be used in particular to determine the derivative along the vertical axis (Y-axis).

[0087] According to some implementations, the filter is a high-pass filter (high-pass kernel).

[0088] Since the reference image 301 is a two-dimensional image, pixel positions can be represented using a two-axis (x, y) coordinate system. According to some embodiments, a first filter Dx is applied to the reference image 301 to obtain a processed image (the reference image 301 has been processed along the "x" axis), a second filter Dy is applied to the reference image 301 to obtain another processed image along the "y" axis (the reference image 301 has been processed along the "y" axis), and a third filter D is applied to the reference image 301 to obtain yet another processed image. In some embodiments, Dx is a differential filter that enables the acquisition of the derivative of the reference image 301 along the "x" axis, Dy is a differential filter that enables the acquisition of the derivative of the reference image 301 along the "y" axis, and D is a differential filter that enables the acquisition of the derivatives of the reference image 301 along both the "x" and "y" axes.

[0089] In the unrestricted example, Dx is defined as follows:

[0090]

[0091] Dy can be defined as the transpose of Dx. These values ​​are not restrictive.

[0092] Based on this segmentation, for each given reference region, a first reference region similar to that given reference region is determined in the reference image.

[0093] Operation 225 may include using a processed reference image. In particular, for each given reference region, operation 225 may include the following steps: selecting a first reference region having a pixel intensity distribution (in the reference image and / or in the processed reference image—this corresponds to the first data providing pixel intensity information for the first reference region) that is similar to the pixel intensity distribution of the given reference region (in the reference image and / or in the processed reference image—this corresponds to the first data providing pixel intensity information for the given reference region).

[0094] Similarity can be evaluated based on a first similarity criterion. For example, suppose a given reference region includes a given pixel in the reference image (and in the processed image, respectively) that has a given pixel intensity. The given pixel intensity is compared with the pixel intensities of other pixels in the reference image (and the processed image, respectively). Pixels can be sorted (classified) based on this comparison. Pixels showing the highest level of similarity in the comparison can be selected as the first reference region. The first similarity criterion defines, for example, a threshold for what constitutes a sufficient level of similarity, and / or can define a maximum number N of pixels that can be selected as similar pixels (this corresponds to the top N pixels after sorting, which are the pixels most similar to the given pixel).

[0095] If multiple processed images are obtained, each pixel of reference image 301 is associated with a vector containing multiple values ​​(one value per processed image—this vector may also include the pixel intensity value in the reference image). This vector is another example of the first data providing pixel intensity information described above. The vector of pixels in each given reference region is compared with the vectors of pixels in other regions of reference image 301. The comparison between vectors can use, for example, cross-correlation, cross-entropy, magnitude comparison, orientation comparison, etc. However, this is not limiting. The pixels of reference image 301 can be sorted based on the comparison results. According to the first similarity criterion, for each given reference region, this sorting can be used to identify multiple first reference regions, which include pixels with a vector that is similar to the vector of pixels in the given reference region. As mentioned above, the first similarity criterion can define the level of similarity required to be considered similar and / or the maximum number of first reference regions selected as a result of the sorting operation.

[0096] It should be noted that if a given reference region is a region of multiple pixels, it can be compared with other regions of the same size in the reference image. The given reference region is associated with a matrix comprising the intensity vectors of the pixels belonging to that given reference region. The matrix of the given reference region is compared with the matrix of pixel intensities of other regions of the same size in the reference image. This comparison can be performed using, for example, cross-correlation or other suitable methods (e.g., cross-entropy, etc.) that can be used to compare matrices. However, this is not limiting. Based on this comparison, the regions of the reference image are sorted, and those regions most similar to the given reference region are selected as the first reference regions.

[0097] Figure 2A The method further includes (operation 230): for each candidate defect, selecting multiple second reference regions within these first reference regions. The number of second reference regions is less than the number of first reference regions.

[0098] In the non-restrictive example, the number of the first region is approximately 2000, while the number of the second reference region is approximately 100. However, this is not restrictive.

[0099] Operation 230 can correspond to a second segmentation of the reference image, for example, following the first segmentation performed in operation 225. The second segmentation benefits from the fact that the first segmentation has reduced the number of regions to be examined in order to identify regions similar to each given reference region. Therefore, the computational efficiency of the second segmentation is improved because it is not necessary to segment the entire reference image, but only a finite number of first reference regions. For a given processor and memory circuitry, selecting the second region only within the first reference regions is computationally more efficient than determining the second reference region across the entire reference image 301 (for a given reference image, the time required to perform the determination of the second region is reduced due to the reduced number of computational operations).

[0100] According to some implementations, the first segmentation is less accurate than the second segmentation (the first segmentation provides a first estimate of the first reference region, and the second segmentation reduces that amount to obtain a more accurate subset of the first reference region), but is more computationally efficient than the second segmentation (the first segmentation relies on, for example, one or more processed images that are computed once for all defects—while the second segmentation may rely on patch comparison, which requires repeated patch calculations and patch comparisons for each candidate defect).

[0101] Figure 3C A method for implementing operation 230 is illustrated. As shown, operation 230 may include: for each candidate defect, calculating (operations 390 and 391) second data providing pixel intensity information of a given reference region and second data providing pixel intensity information of a first reference region. A comparison may be performed between the second data providing pixel intensity information of the given reference region and the second data providing pixel intensity information of the first reference region (operation 392). According to a second similarity criterion, the comparison indicates that a matching first reference region is selected as the second reference region (operation 393). The second similarity criterion may be defined, for example, by an algorithm and / or an operator, and the second similarity criterion indicates that the comparison is considered to indicate a similarity level of a match.

[0102] In some implementations, the first data providing pixel intensity information of a first reference region (as used in operation 225) and the second data providing pixel intensity information of the first reference region (as used in operation 230) are different / different types. Similarly, the first data providing pixel intensity information of a given reference region (as used in operation 225) and the second data providing pixel intensity information of a given reference region (as used in operation 230) are different / different types. In other words, a multi-segmentation method of the reference image is used, where each segmentation is performed using different types of data.

[0103] According to some embodiments, the second data providing pixel intensity information for a given reference region includes a patch (or kernel) providing pixel information for that given reference region. A patch includes multiple pixels arranged according to a predefined shape / distribution. Similarly, according to some embodiments, for each first reference region, the second data providing pixel intensity information for the first reference region includes a patch (or kernel) providing pixels for that first reference region.

[0104] According to some implementations, a reference image or at least one of one or more second-processed images is used to obtain second data that provides pixel intensity information for a given reference region and second data that provides pixel intensity information for a first reference region.

[0105] It should be noted that each given reference region may include a group of one or more pixels in a reference image, or a group of one or more corresponding pixels having the same position (the same position as the pixels in that group in the reference image) in each of one or more second-processed images. Similarly, each first reference region may include a group of one or more pixels in a reference image, or a group of one or more corresponding pixels having the same position (the same position as the pixels in that group in the reference image) in each of one or more second-processed images.

[0106] For each given reference region, second data can be determined using pixel intensities in the reference image and / or one or more second-processed images. For example, a patch can be determined for that given reference region in the reference image and / or one or more second-processed images. Similarly, for each first reference region, a patch can be determined for that first reference region in the reference image and / or one or more second-processed images.

[0107] Operation 392 may include: for each given first reference region (of the plurality of first reference regions), comparing second data providing pixel intensity information of the given first reference region in the second processed image with second data providing pixel intensity information of the corresponding given reference region in the second processed image. The comparison of operation 392 can therefore be performed for each second processed image (and also for the reference image). Operation 393 includes: using the results of these respective comparisons to select a finite number of second reference regions among these first reference regions.

[0108] One or more second-processed images can be obtained by processing at least a portion of a reference image (or an image generated using that reference image). This processing may include filtering (smoothing, noise reduction, segmentation, contrast enhancement, etc.) the reference image or an image generated using that reference image. In some embodiments, the processing used to obtain the second-processed image differs from the processing used to obtain the processed image (the acquisition of the processed image has been described with reference to operation 225). For example, the filter used to obtain the second-processed image may be selected differently from the filter used to obtain the processed image.

[0109] In some implementations, the filter used to obtain the second processed image may be applied only to a portion of the reference image, which includes the region of interest (ROI) of the reference image (ROI includes each given reference region and a corresponding first reference region identified in operation 225). In other words, it is sufficient not to apply the filter to the entire reference image, but only to process the ROI.

[0110] According to some implementations, for all candidate defects of a given reference image, each second-processed image is calculated once.

[0111] Figure 4A A non-limiting example of a patch 410 for a given reference region 403 is shown. As shown, patch 410 includes the given reference region 403 and a finite number of pixels 411 located in the vicinity of the given reference region 403. Figure 4A A first reference region 4031 to 403 associated with the given reference region 403 is also depicted. N2 .

[0112] In this non-limiting example, the shape of puzzle piece 410 is rectangular. However, this is not restrictive, and the shape of the puzzle piece can be any suitable shape (square, polygon, etc.).

[0113] In some implementations, the puzzle pieces are symmetrical. This is not limiting, and in some implementations, the puzzle pieces are asymmetrical.

[0114] A limited number of pixels 411 selected within the patch 410 surrounding a given reference region 403 may be adjacent. This is not mandatory, and in some implementations, the limited number of pixels 411 may include isolated pixels or isolated groups of pixels.

[0115] Generally, the size of the piece 410 is chosen to be smaller than the size of the reference image 401.

[0116] exist Figure 4AIn the example, tile 410 covers an area of ​​5×3 pixels (15 pixels). Tile 410 can be defined as including all pixels covered by the shape of tile 410 (15 pixels in this example) or only including some pixels covered by the shape of tile 410 (e.g., N pixels less than 15 pixels).

[0117] Figure 4B It describes all possible puzzle types. Figure 4B The examples described are not limiting. For example, tile 420 comprises two distinct groups of pixels (see gray pixels) that are not adjacent to each other. In tile 421, the first reference region 4033 is not located at the center of tile 421. In tile 422, the first reference region 4034 is located at the center of tile 422. Tile 422 is symmetrical, while tiles 420 and 421 are asymmetrical.

[0118] For each first reference region (and for each given reference region in the reference image, multiple first reference regions have already been identified in the reference image), the second data providing pixel intensity information for the first reference region may also include a mosaic piece. The size and shape of the mosaic piece defined for a given reference region may be the same as the size and shape of the mosaic piece defined for each first reference region. However, this is not limiting, and in some implementations, the mosaic piece defined for a given reference region may differ from the mosaic piece defined for each first reference region.

[0119] Figure 4C The diagram illustrates a non-limiting example of tile comparison. Tile 425 has been defined for reference region 403. Tiles 426, 427, and 428 are defined for first reference regions 4031, 4033, and 4034, respectively.

[0120] Piece 425 of reference region 403 is compared with pieces (426, 427, and 428) of each of the first reference regions (4031, 4033, and 4034). This can be done for each reference region of a reference image associated with a different subset of the first reference region.

[0121] The comparison may include comparing the pixel intensity of one or more pixels in the patch of a given reference region with the pixel intensity of a corresponding pixel in the patch of a first reference region. Each pixel in the first patch has a corresponding pixel in the second patch being compared with the first patch. This corresponding pixel is, for example, a pixel in the second patch that is at the same position as a pixel in the first patch.

[0122] The comparison may include: performing a difference between corresponding pixel intensity values ​​and comparing that difference with a threshold.

[0123] Based on this comparison, a similarity score can be generated. Consider an example: for a comparison between two mosaic pieces, M pixel intensity differences have been obtained. In some implementations, the similarity score can be calculated based on the maximum absolute value of these M values, the minimum absolute value of these M values, the average of all M values, etc.

[0124] The second similarity criterion can be defined as follows: in operation 230, only the first reference region with the highest similarity score (or the lowest similarity score, depending on the definition of similarity score) is selected as the second reference region.

[0125] Piece comparison can be performed for each candidate defect. Therefore, for each candidate defect, a finite number of second reference regions are obtained in the reference image (selected from the larger number of first reference regions obtained in operation 225).

[0126] Figure 4C A patch comparison in the reference image was described. However, as mentioned above, a similar patch comparison can be performed in one or more second-processed images (in addition to, or instead of, the patch comparison performed in the reference image). Therefore, a similarity score can be defined based on the results of patch comparisons in one or more second-processed images. Thus, a second reference region can be selected in the second-processed images.

[0127] Figure 4D An example is shown. In the second processed image 470, the patchwork of reference region 450 is compared with the patchwork of first reference region 451 and the patchwork of first reference region 452. In the second processed image 471, the patchwork of reference region 450 is compared with the patchwork of first reference region 451 and the patchwork of first reference region 452. In the second processed image 472, the patchwork of reference region 450 is compared with the patchwork of first reference region 451 and the patchwork of first reference region 452. A similarity score can be generated for each first reference region based on these comparisons. A second reference region can be selected from the first reference regions using the similarity score of each first reference region. In this example, the second reference region is determined within the second processed image.

[0128] Figure 5A A specific implementation of the piece comparison that can be performed in operation 230 is shown.

[0129] Suppose that for a given reference region, multiple first reference regions have been identified in the reference image during operation 225.

[0130] A tile can be defined for a given reference region (located within the shape of the tile), which includes multiple pixels within the tile's shape (operation 505). Assume the tile comprises N pixels. In a reference image (e.g., reference image 301) or in the second-processed image, the tile covers pixel P. 1,ref To P N,ref Pixel P in the mosaic of the reference image (and in the mosaic of the second processed image respectively) 1,ref To P N,ref Pixels in the same location in the same image (e.g., image 201) are labeled as P. 1,img To P N,img Therefore, pixel P 1,img To P N,img It is the pixel P in this image that corresponds to the pixel in the reference image (in the second processed image respectively). 1,ref To P N,ref Pixels.

[0131] The method may include: selecting only a finite number N1 (N1 < N) pixels (P) within the pixels located in the patch shape (operation 510) 1,ref To P N1,ref For these pixels, the corresponding pixel (P) in image 201 1,img To P N1,img —That is to say, the pixel P in image 201 is the same as that in the reference image or the second processed image. 1,ref To P N1,ref Pixels at the same location satisfy the correlation criterion. In reality, a defect in image 201 may only be located in a sub-part of the pattern, so it may be necessary to find similar locations in image 201 that specifically correspond to this sub-part of the pattern.

[0132] According to some implementation methods, the corresponding pixel P of image 201 that satisfies the correlation criterion... 1,img To P N1,img The pixel with the highest intensity in image 201 (among those pixels located in the region of image 201 corresponding to the patch shape defined in reference image 301) or the pixel with an intensity equal to or higher than an intensity threshold (the intensity threshold may be defined by the operator, for example). Therefore, only pixels with the same intensity as the corresponding pixel P in reference image 301 (in the second processed image respectively) are considered. 1,img To P N1,img Pixel P at the same location (identified as the most relevant pixel in image 201) 1,ref To P N1,ref These pixels were selected as part of the mosaic (so only these pixels will be used for mosaic comparison later).

[0133] According to some implementations, defect detection algorithms have been used to detect defects in image 201 (e.g., the defect detection algorithm described in US9,235,885, or another defect detection algorithm). The defect detection algorithm provides a probability of the presence of a defect for each pixel (or group of pixels) of image 201. The corresponding pixel P of image 201 that satisfies the correlation criterion... 1,img To P N1,img This can correspond to the pixels in image 201 with the highest probability of having a defect (those pixels located in the region of image 201 corresponding to the patch shape defined in reference image 301) or those pixels with a probability higher than a threshold. Therefore, only pixels in reference image 301 (in the second processed image, respectively) that have a corresponding pixel P in image 201 are considered. 1,img To P N1,img Pixel P at the same position 1,ref To P N1,ref These pixels were selected as part of the mosaic (so only these pixels will be used for mosaic comparison later).

[0134] Figure 5B and Figure 5C Non-restrictive examples are shown in the figure.

[0135] Suppose that a tile 525 is defined for a reference region 503 in a reference image 501. The shape of tile 525 covers an area of ​​5*3=15 pixels.

[0136] The pixels of patch 525 have corresponding pixels in image 500, which are located in region 502 of image 500. If the correlation criterion is defined about pixel intensity levels, then pixels 5021, 5022, 5023, and 5024 in region 502 have the highest pixel intensity within region 502.

[0137] Pixel 5021 in image 500 corresponds to pixel 560 in reference image 501. Pixel 5022 in image 500 corresponds to pixel 561 in reference image 501. Pixel 5023 in image 500 corresponds to pixel 562 in reference image 501. Pixel 5024 in image 500 corresponds to pixel 563 in reference image 501. Therefore, only pixels 560, 561, 562, and 563 of image 501 are selected as part of the mosaic for subsequent mosaic comparison. Selecting this limited number of pixels further reduces the computational cost when performing mosaic comparison. Since these pixels best represent the region of interest, good accuracy can be maintained. As mentioned above, pixel intensity is just an example and can serve as a correlation criterion for selecting a limited number of pixels in the mosaic of the reference image.

[0138] Figure 5AThe method further includes: performing a patch comparison between the selected pixel of the patch in the reference region and the corresponding pixel of the patch in each first reference region in the first reference region (operation 520).

[0139] This is in, for example Figure 5C As shown in the diagram, when patch 525 of reference region 503 is compared with patch 527 of first reference region 5033, only a limited number of pixel intensity comparisons are performed: the pixel intensity of pixel 560 is compared with the pixel intensity of pixel 5603, the pixel intensity of pixel 561 is compared with the pixel intensity of pixel 5613, the pixel intensity of pixel 562 is compared with the pixel intensity of pixel 5623, and the pixel intensity of pixel 563 is compared with the pixel intensity of pixel 5624. The same type of comparison can be performed when patch 525 is compared with other patches of the other first reference regions.

[0140] The second reference region can be selected as a subset of the first reference region, for which the comparison performed in operation 393 or 520 indicates the highest matching level.

[0141] exist Figure 5A In a variation of the method, for a given reference region, different puzzle pieces of different shapes are defined for that given reference region. For each shape, the puzzle piece for that given reference region is compared with the puzzle pieces of the same shape associated with each first reference region. A similarity score can be obtained for each shape. A total similarity score can be calculated, which can be used to select the second reference region that receives the highest score.

[0142] Once the second reference region has been identified in the reference image (or in the second processed image), Figure 2A The method includes (operation 240) identifying a second region in the image that corresponds to a second reference region in the reference image (or the second processed image). Correspondence can be evaluated according to a correspondence criterion. Such a correspondence criterion has been defined above for operation 220. The same correspondence criterion can be used in operation 240 (but this is not mandatory). As mentioned above, the correspondence criterion can be defined, for example, as: the position of the second region in the image matches the position of the second reference region in the reference image (or the second processed image).

[0143] Figure 6A Non-restrictive examples are shown in the figure.

[0144] Assume that in operation 230, second reference regions 600 to 607 have been obtained in reference image 611 (some second reference regions are associated with a first given reference region, some second reference regions are associated with a second given reference region, etc.).

[0145] Operation 240 includes: identifying second regions 6001 to 6071 in image 621 that respectively correspond to second reference regions 600 to 607 in reference image 611, according to a correspondence criterion. As described above, in some embodiments, second regions in the image that correspond to the second reference regions identified in the second processed image can be identified.

[0146] After operation 240, for each candidate defect, there exists a given region corresponding to the location of the candidate defect in the sample image and multiple second regions (in the image) associated with the given region.

[0147] Figure 2A The method further includes, for each candidate defect, using (operation 250) data providing pixel intensity information of these second regions and data providing pixel intensity information of a given region (corresponding to the location of the candidate defect in image 201) to determine whether the candidate defect corresponds to a defect. In some implementations, the pixel intensity of pixels located near these second regions is also considered (e.g., for each given second region, using a mosaic including the given second region).

[0148] In some embodiments, operation 250 may use a comparison between the pattern of a given region and the pattern of each second region to determine whether the candidate defect is a defect. Operation 250 includes classifying the candidate pixels (e.g., classifying them as defective or not defective). In some embodiments, classifying the candidate defect as a defect may be a binary judgment (i.e., presence or absence, defective or not defective, etc.). In some embodiments, a score representing the foreground (probability) that the candidate defect is a defect is generated.

[0149] According to some implementations, operation 250 is based on the pixel intensity in a given region and a threshold determined based on the pixel intensity of a second region associated with the given region. For example, the difference between the pixel intensity in the given region and the pixel intensity in a first reference region (which is in the reference image and corresponds to the given region) can be determined, and this difference can be compared with the threshold determined based on the pixel intensity of the second region. If the difference is higher than the threshold, operation 250 can determine that the candidate defect corresponds to a defect, while if the difference is lower than the threshold, the candidate defect can be determined not to be a defect.

[0150] It should be noted that more than one threshold can be determined based on a second region associated with a given region of a candidate defect. For example, two thresholds can be set to define a range of pixel intensity values, such that if the pixel intensity of a given region is outside this range, the candidate defect can be determined to be a defect, but if the pixel intensity of a given region is within this range, the candidate defect can be determined not to be a defect.

[0151] In some implementations, in addition to defects, operation 250 may include classifying candidate defects based on another type of classification system (e.g., defect categories that require further processing, defect categories that require modification of inspection or testing criteria, etc.).

[0152] The output of Operation 250 can be provided, for example, to another system and / or to a person using the display. For each candidate defect, the output may include determining whether the candidate defect is a defect, as well as other data, such as the location of the candidate defect.

[0153] The following describes various other implementations of operation 250.

[0154] exist Figure 2A In the implementation of the method, multiple reference images are used. For each given reference image, operations 220 to 250 can be performed. Therefore, for each reference image, a score representing the foreground of each candidate defect as a defect can be obtained. A total score can be obtained based on each score to provide a final decision on whether each candidate defect is a defect.

[0155] See Figure 7A .

[0156] According to some implementation methods Figure 7AThe method includes acquiring (operation 700) multiple images of a sample (or at least a region of the sample) acquired by an imager. These multiple images differ from each other in at least one parameter, which is selected from a list including at least the following parameters: illumination spectrum (the spectral response of the illumination portion of the imager), collection spectrum (the spectral response of the collection portion of the imager), illumination polarization (polarization applied by the illumination portion of the imager), collection polarization (polarization applied by the collection portion of the imager), illumination angle (the illumination angle of the illumination portion of the imager relative to the region), incident angle (the incident angle of the illumination portion relative to the region), collection angle, one or more focus offsets of the illumination beam, phase of the collected beam, collection channel (e.g., bright field channel, gray field channel), Fourier transform of the returned light, and sensing type (e.g., sensing amplitude and / or sensing phase). Such multiple images may also be referred to as multi-perspective images. As shown, this parameter can typically correspond to an optical configuration modified between different acquisitions.

[0157] Since at least two images of the sample have been obtained, it is possible to perform the procedure on each image. Figure 2A The method (in) Figure 2A In China, it is called an "image"— Figure 7A In this method, this corresponds to each image acquired from different viewpoints. Therefore, for each viewpoint, a score can be calculated that provides foreground information indicating whether a given defect corresponds to a defect. The total score can be calculated to output a final decision indicating whether a candidate defect actually corresponds to a defect.

[0158] For example, suppose we have obtained bright-field and gray-field images of the sample. For each candidate defect, a first score can be determined indicating whether the candidate defect obtained in the bright-field image corresponds to a defect (using, for example...). Figure 2A The method), and a second score indicating whether a candidate defect obtained in the grayscale image corresponds to a defect (using, for example) Figure 2A (The method). The total score can be calculated based on the first and second scores.

[0159] See Figure 7B , Figure 7B An alternative implementation of the method in Figure 7 is described.

[0160] According to some implementation methods Figure 7B The method includes: acquiring (operation 750) multiple images of a sample (or at least one region of the sample). The multiple images differ from each other in at least one parameter (as described above with reference to operation 700). It is assumed, for example, that a first image and a second image have already been acquired.

[0161] Assume that multiple candidate defects have been identified. For each candidate defect, obtain a first given region of the candidate defect in a first image (in the first image, the first given region corresponds to the location of the candidate defect) and a second given region of the candidate defect in a second image (in the second image, the second given region corresponds to the location of the candidate defect).

[0162] For each candidate defect, it is then determined whether the candidate defect satisfies the correlation criterion in the first image (an example of the correlation criterion is provided below). Similarly, for each candidate defect, it is determined whether the candidate defect satisfies the correlation criterion in the second image.

[0163] In some implementations, when the data providing pixel intensity information for a first given region (and a second given region, respectively) satisfies an intensity criterion, candidate defects in the first image (and in the second image, respectively) satisfy a correlation criterion (see operations 760 and 7601). The intensity criterion may define, for example, a minimum pixel intensity that must be satisfied for each pixel.

[0164] In some implementations, a correlation criterion is satisfied for a candidate defect in the first image (and in the second image, respectively) when the probability that the candidate defect corresponds to a defect in the first image (and in the second image, respectively) is higher than a threshold. This probability can be provided by a defect detection algorithm (e.g., the defect detection algorithm described in US 9,235,885), which was previously used to detect the presence of defects in the first image (and in the second image, respectively).

[0165] In some implementations, the correlation criterion used for the first image may differ from the correlation criterion used for the second image (for example, the correlation criterion used to determine the correlation of candidate defects in the first image is based on the pixel intensity of the candidate defects in the first image, while the correlation criterion used to determine the correlation of candidate defects in the second image is based on the probability that the pixel of the candidate defect corresponds to a defect in the second image). However, this is not limiting.

[0166] Figure 2A The method is then applied only to the first image of those candidate defects that satisfy the correlation criterion (operation 770). For example, if the data providing pixel intensity information for a given first region does not satisfy the intensity criterion for a candidate defect, then the candidate defect in the first image can be considered insufficiently obvious and therefore can be ignored when processing the candidate defects of the first image.

[0167] Similarly, Figure 2AThe method is then applied only to the second image of candidate defects that satisfy the correlation criterion (see Operation 7701). For example, if the data providing pixel intensity information for a given second region does not satisfy the intensity criterion for a candidate defect, the candidate defect in the second image can be considered insufficiently obvious and therefore can be ignored when processing candidate defects in the second image. Thus, the computational work is optimized.

[0168] For each candidate defect, a score indicating whether the candidate defect is a defect can be generated. For some candidate defects, scores are generated only based on the first image (since the candidate defect only has sufficient pixel intensity in the first image, the first image is used only for that candidate defect). Figure 2A The method). For some candidate defects, scores are generated only based on the second image (since the candidate defect only has sufficient pixel intensity in the second image, the second image is used only for that candidate defect). Figure 2A The method involves using a first image and a second image to generate scores for some candidate defects. This is because the candidate defect has sufficient pixel intensity in both images; therefore, the first image is used to obtain a first score, and the second image is used to obtain a second score for that candidate defect. Figure 2A (Methods).

[0169] Some candidate defects were ignored because they did not meet the correlation criterion in either the first or second image.

[0170] The following describes an unrestricted method for implementing operation 250.

[0171] According to some implementations, for a candidate defect, the pixel intensity of a given region is compared with the pixel intensity of the pixel with the highest (or lowest) value in the pixel group of a second region. If the difference between these values ​​exceeds a threshold, the candidate defect can be considered defective.

[0172] In another example (see) Figure 8A (Operation 800) For a given candidate defect, determine the ratio between (a) and (b): (a) the difference between the pixel intensity of a given region and the average pixel intensity value of all pixels in a second region associated with that given region; (b) the difference between the pixel intensity of the pixel in the second region with the highest (or lowest) pixel intensity and the average pixel intensity of the second region. This ratio can be compared with a threshold (which can be determined based on the pixel intensity of the second region, but is not required) (Operation 805), and the candidate defect can be classified as a defect or a non-defect based on this comparison (Operation 806).

[0173] It should be noted that different types of averages can be achieved, such as arithmetic mean, median, geometric median, geometric mean, harmonic mean, quadratic mean, weighted mean, truncated mean, interquartile mean, midrange, winsorized mean, etc.

[0174] refer to Figure 8B Provided Figure 8A Non-restrictive examples of methods, Figure 8B A histogram depicts the pixel intensity. This histogram depicts the pixel intensity of a given region of a candidate defect and the pixel intensity of pixels in a second region associated with that given region. The horizontal axis 810 of the histogram represents the pixel intensity value, and the vertical axis 801 represents the number of pixels with each pixel intensity value.

[0175] The pixel intensity of a given region of a candidate defect is depicted as reference 850. In a non-limiting example, the given region comprises a single pixel. In other implementations, the given region may comprise multiple pixels and the pixel intensity value may correspond to, for example, the average of these multiple pixels, or the maximum of these multiple pixels, or the centroid of these multiple pixels.

[0176] The pixel intensity of the pixels in the second region corresponds to part 860 of the histogram.

[0177] The value 830 corresponds to the difference between the pixel in the second region with the highest pixel intensity and the average pixel intensity of the second region (see reference 820).

[0178] The value 840 corresponds to the difference between the pixel intensity 850 of a given region and the average pixel intensity value 820 of all pixels in a second region associated with that given region.

[0179] If the ratio between values ​​840 and 830 is higher than the threshold, the candidate defect can be determined to be a defect. If the ratio between values ​​840 and 830 is lower than the threshold, the candidate defect can be determined to be a non-defect.

[0180] although Figure 8B The example depicts a symmetrical bar chart, but this is not mandatory.

[0181] Figure 9Another histogram depicts the pixel intensity. This histogram depicts the pixel intensity 950 of a given region of a candidate defect and the pixel intensity of pixels in a second region associated with that given region (see section 960 of the histogram). The horizontal axis 910 of this histogram represents the pixel intensity value, and the vertical axis 901 of this histogram represents the number of pixels with each pixel intensity value.

[0182] Because the bar chart is not symmetrical, two different rules can be set.

[0183] Value 930 corresponds to the difference between the pixel in the second region with the highest pixel intensity and the average pixel intensity of the second region (see reference 920).

[0184] The value 940 corresponds to the difference between the pixel intensity 950 of a given region and the average pixel intensity value 920 of all pixels in a second region associated with that given region.

[0185] The value 931 corresponds to the difference between the average pixel intensity of the second region (see reference 920) and the pixel in the second region with the lowest pixel intensity.

[0186] If the value 940 is higher than the value 930, then a defect can be identified.

[0187] If the value 940 is lower than the value 931, then a defect can be identified.

[0188] If a value of 940 falls between values ​​930 and 931, it is unlikely to be a defect. A threshold can be set to determine whether a value of 940 is classified as corresponding to a defect.

[0189] exist Figure 9 In the example, since the absolute value of 940 is different from the absolute value of 931, the asymmetric decision rule is used.

[0190] A possible generalized example of an asymmetric determination rule is that determining the existence of a defect includes: determining that a candidate defect is a defect if and only if (a) the pixel intensity of a given region of a candidate pixel is greater than the average pixel intensity of pixels in a second region by at least a first threshold X or (b) the pixel intensity of a given region of a candidate pixel is less than the average pixel intensity of pixels in a second region by at least a second threshold Y, where X is different from Y.

[0191] It should be understood that the application of this invention is not limited to the details set forth in the description or shown in the accompanying drawings.

[0192] It should also be understood that the system according to the invention can be implemented, at least in part, on a suitably programmed computer. Similarly, the invention contemplates a computer-readable computer program for performing the methods of the invention. The invention further contemplates a non-transitory computer-readable memory tangibly embodying an instruction program executable by a computer to perform the methods of the invention.

[0193] This invention can have other embodiments and can be practiced and performed in various ways. Therefore, it should be understood that the expressions and terminology used herein are intended for description and should not be considered limiting. Consequently, those skilled in the art will understand that the concepts upon which this disclosure is based can be readily used as the basis for designing other structures, methods, and systems to achieve several of the objectives of this disclosure.

[0194] Those skilled in the art will readily understand that various modifications and alterations can be applied to the embodiments of the invention as described above without departing from the scope defined by the appended claims.

Claims

1. A semiconductor sample inspection system, the system comprising a processor and a memory circuit (PMC), configured to: ---For each given candidate defect among a plurality of candidate defects in an image of the semiconductor sample acquired by the inspection tool, a given region of the given candidate defect in the image is obtained; ---Obtain a reference image; ---Segmenting at least a portion of the reference image, the segmentation comprising: for each given candidate defect: o. Based on the correspondence criterion, obtain a given reference region in the reference image that corresponds to a given region of the given candidate defect in the image; and o. Based on a first similarity criterion, a first reference region in the reference image is determined, wherein for each given first reference region, first data providing pixel intensity information of the given first reference region is matched with first data providing pixel intensity information of the given reference region; and ---For each given candidate defect: o For each given first reference region, the second data providing pixel intensity information of the given first reference region is compared with the second data providing pixel intensity information of the given reference region; o Select a plurality of second reference regions in the first reference region, wherein the comparison indicates that, according to a second similarity criterion, there is a sufficient level of similarity between the pixel intensity of the plurality of second reference regions and the pixel intensity of the given reference region; o According to the correspondence criterion, obtain a plurality of second regions in the image, wherein each given second region corresponds to a second reference region in the second reference region; and o Use the data providing pixel intensity information of the second region and the data providing pixel intensity information of the given region to determine whether the given candidate defect corresponds to a defect.

2. The system of claim 1, wherein for each given candidate defect, the number of the second reference regions is less than the number of the first reference regions.

3. The system of claim 1, wherein the PMC is configured to: At least one filter is applied to the image to obtain at least one processed image; For each first given reference region, the at least one processed image is used to determine first data providing pixel intensity information for the given first reference region; and The first data providing pixel intensity information for the given reference region is determined using the at least one processed image.

4. The system of claim 3, wherein the at least one processed image is calculated once for all candidate defects.

5. The system according to claim 1, wherein: The second data providing pixel intensity information for the given reference region and the second data providing pixel intensity information for each given first reference region are determined using at least one of the following: The reference image; or One or more second-processed images, which are obtained by filtering at least a subset of the following: The reference image; or The image generated using the reference image.

6. The system according to claim 1, wherein: The second data providing pixel intensity information for the given reference region includes a reference mosaic providing one or more pixel information for the given reference region; For each given first reference region: The second data providing pixel intensity information for the given first reference region includes a mosaic providing pixel information for one or more of the given first reference region; and The system is configured to compare the patch of the given first reference region with the reference patch.

7. The system of claim 6, configured to: Select a finite subset of pixels from the reference mosaic; and For each given first reference region, data providing pixel intensity information of the finite subset of pixels in the reference patch is compared with data providing pixel intensity information of the corresponding pixels in the patch of the given first reference region.

8. The system of claim 7, wherein selecting the finite subset of pixels comprises at least one of the following: Select the pixels in the image whose corresponding pixel intensity is associated with a pixel intensity higher than a threshold; or Select the pixels in the image whose corresponding pixels are associated with the probability of the presence of defects above a threshold.

9. The system of claim 1, wherein for each given candidate defect, selecting the second reference region in the first reference region is more computationally efficient than selecting the second reference region throughout the entire reference image.

10. The system of claim 1, wherein for each given first reference region, the first data providing pixel intensity information of the given first reference region is different from the second data providing pixel intensity information of the given first reference region.

11. The system of claim 1, configured for each given candidate defect among a plurality of candidate defects: ---For each given candidate defect among a plurality of candidate defects, we obtain: O is a first given region of the given candidate defect in a first image of the semiconductor sample obtained by the inspection tool; and O is a second given region of the given candidate defect in a second image of the semiconductor sample obtained by the inspection tool, wherein the second image differs from the first image in at least one parameter. ---Do at least one of the following: The first image is used to determine whether a candidate defect is a defect only if the candidate pixel satisfies the correlation criterion in the first image; or The second image is used to determine whether a candidate defect is a defect only if the candidate pixel satisfies the correlation criterion in the second image.

12. A method for inspecting a semiconductor sample, the method comprising: Composed of processor and memory circuitry (PMC): ---For each given candidate defect among a plurality of candidate defects in an image of the semiconductor sample acquired by the inspection tool, a given region of the given candidate defect in the image is obtained; ---Obtain a reference image; ---Segmenting at least a portion of the reference image, the segmentation comprising: for each given candidate defect: O, according to the correspondence criterion, obtain a given reference region in the reference image that corresponds to a given region of the given candidate defect in the image; and O. Based on the first similarity criterion, a first reference region in the reference image is determined, wherein For each given first reference region, a first data match providing pixel intensity information of the given first reference region; and ---For each given candidate defect: O For each given first reference region, the second data providing pixel intensity information of the given first reference region is compared with the second data providing pixel intensity information of the given reference region; O Select a plurality of second reference regions in the first reference region, wherein the comparison indicates that, according to a second similarity criterion, there is a sufficient level of similarity between the pixel intensity of the plurality of second reference regions and the pixel intensity of the given reference region; O, according to the correspondence criterion, obtains a plurality of second regions in the image, wherein each given second region corresponds to a second reference region in the second reference region; and O uses data providing pixel intensity information of the second region and data providing pixel intensity information of the given region to determine whether the given candidate defect corresponds to a defect.

13. The method of claim 12, wherein for each given candidate defect, the number of the second reference regions is less than the number of the first reference regions.

14. The method of claim 12, comprising: At least one filter is applied to the image to obtain at least one processed image; For each given first reference region, the at least one processed image is used to determine first data that provides pixel intensity information for the given first reference region; and The first data providing pixel intensity information for the given reference region is determined using the at least one processed image.

15. The method of claim 14, wherein the at least one processed image is calculated once for all candidate defects.

16. The method of claim 12, comprising: The second data providing pixel intensity information for the given reference region includes a reference mosaic providing one or more pixel information for the given reference region; For each given first reference region: The second data providing pixel intensity information for the given first reference region includes a mosaic providing pixel information for one or more of the given first reference region; and The method includes comparing the patch of the given first reference region with the reference patch.

17. The method of claim 16, comprising: Select a finite subset of pixels from the reference mosaic; as well as For each given first reference region, data providing pixel intensity information of the finite subset of pixels in the reference patch is compared with data providing pixel intensity information of the corresponding pixels in the patch of the given first reference region.

18. The method of claim 17, wherein selecting the finite subset of pixels comprises at least one of the following: Select the pixels in the image whose corresponding pixel intensity is associated with a pixel intensity higher than a threshold; or Select the pixels in the image whose corresponding pixels are associated with the probability of the presence of defects above a threshold.

19. The method of claim 12, comprising: For each given candidate defect among a plurality of candidate defects: ---For each given candidate defect among a plurality of candidate defects, we obtain: O is a first given region of the given candidate defect in a first image of the semiconductor sample obtained by the inspection tool; as well as O is a second given region of a given candidate defect in a second image of the semiconductor sample obtained by the inspection tool, wherein the second image differs from the first image in at least one parameter; as well as ---Do at least one of the following: The first image is used to determine whether a given candidate defect is a defect only if the candidate pixel satisfies the correlation criterion in the first image. or The second image is used to determine whether a candidate defect is a defect only if the candidate pixel satisfies the correlation criterion in the second image.

20. A non-transitory computer-readable medium comprising instructions that, when executed by a processor and memory circuitry (PMC), cause the PMC to perform operations including: ---For each given candidate defect among multiple candidate defects in an image of a semiconductor sample acquired by an inspection tool, obtain a given region of the given candidate defect in the image; ---Obtain a reference image; ---Segmenting at least a portion of the reference image, the segmentation comprising: for each given candidate defect: O, according to the correspondence criterion, obtain a given reference region in the reference image that corresponds to a given region of the given candidate defect in the image; and O. Based on the first similarity criterion, a first reference region in the reference image is determined, wherein For each given first reference region, a first data match providing pixel intensity information of the given first reference region; and ---For each given candidate defect: O For each given first reference region, the second data providing pixel intensity information of the given first reference region is compared with the second data providing pixel intensity information of the given reference region; O Select a plurality of second reference regions in the first reference region, wherein the comparison indicates that, according to a second similarity criterion, there is a sufficient level of similarity between the pixel intensity of the plurality of second reference regions and the pixel intensity of the given reference region; O, according to the correspondence criterion, obtains a plurality of second regions in the image, wherein each given second region corresponds to a second reference region in the second reference region; and O uses data providing pixel intensity information of the second region and data providing pixel intensity information of the given region to determine whether the given candidate defect corresponds to a defect.

Images