Inspection of semiconductor samples

Abstract

Systems and methods are provided for inspecting semiconductor samples using an inspection recipe. The methods include the steps of obtaining a registered image pair; for each design-based structural element associated with a given layer: calculating edge properties; using a trained classifier to determine the class of the design-based structural element; and generating a layer score that can be used to determine the validity of the registered image pair. Systems and methods are also provided for generating an inspection recipe that can be used to inspect semiconductor samples.

Description

Technical Field

[0001] The presently disclosed subject matter relates generally to the field of inspecting samples, and more particularly to inspecting semiconductor samples using inspection recipes. Background Art

[0002] The current demand for high density and high performance associated with ultra-large-scale integration of manufactured devices requires submicron features, increased transistor and circuit speeds, and improved reliability. Such demands require that device features be formed with high precision and high uniformity, which in turn requires careful monitoring of the manufacturing process, including automated inspection of the device while it is still in the form of a semiconductor wafer.

[0003] By way of non-limiting example, runtime inspection can employ a two-stage process, such as inspecting a sample and then reviewing sample locations for potential defects. During the first stage, the surface of the sample is inspected at high speed and relatively low resolution. In the first stage, a defect map is generated to indicate suspect locations on the sample with a high probability of defects. During the second stage, at least some of the suspect locations are more fully analyzed at relatively high resolution. In some cases, both stages can be performed by the same inspection tool, and in some other cases, the two stages are performed by different inspection tools.

[0004] Inspection processes are used at various steps during semiconductor manufacturing to detect and classify defects on samples, as well as to perform metrology-related operations. Inspection efficiency can be increased by automating (multiple) processes such as automatic defect classification (ADC), automatic defect review (ADR), image segmentation, and so on. Summary of the Invention

[0005] According to certain aspects of the presently disclosed subject matter, a system for inspecting a semiconductor sample using an inspection recipe is provided, the system comprising a processor and memory circuitry (PMC) configured to: obtain a registered image pair, the registered image pair comprising a first image and a second image, the first image captured by an inspection tool and representing at least a portion of the semiconductor sample, the second image generated based on design data characterizing the at least a portion of the sample, the second image providing information of one or more design-based structural elements present in the design data and a corresponding layer associated with each design-based structural element; and for each design-based structural element associated with a given layer: calculate edge properties, the edge properties an estimated presence of an edge of an image-based structural element in the first image associated with the design-based structural element and indicating a location of the edge of the design-based structural element in the second image; and determining, using a trained classifier included in the inspection recipe, a category of the design-based structural element based on the edge properties associated with the design-based structural element, the category indicating the validity of the registration between the first image and the second image at the location of the design-based structural element; and generating a layer score for the given layer based on the category of each design-based structural element associated with the given layer, the score being usable for determining the validity of the registered image pair based on a layer threshold predetermined in the inspection recipe.

[0006] In addition to the features described above, the system according to this aspect of the presently disclosed subject matter may also include one or more of the features (i) to (vi) listed below, in any desired combination or arrangement that is technically possible:

[0007] (i) The first image may be a high-resolution image captured by a review tool.

[0008] (ii) The second image may be generated by performing simulation on the design data.

[0009] (iii) the PMC may be further configured to, for each design-based structural element associated with the given layer: calculate one or more grayscale profiles along one or more specific directions at a position in the first image corresponding to the position of the design-based structural element in the second image; and calculate one or more profile attributes associated with the design-based structural element, wherein each profile attribute indicates a difference between a corresponding grayscale profile of the design-based structural element and a corresponding baseline grayscale profile included in the inspection recipe, the corresponding baseline grayscale profile being calculated along the corresponding specific direction for the family of design-based structural elements to which the design-based structural element belongs; wherein the step of determining the category using the trained classifier is performed based on the edge attribute and the one or more profile attributes associated with the design-based structural element.

[0010] (iv) The PMC may be configured to calculate the edge properties by applying a statistical test between two groups of pixels on either side of the edge of the design-based structural element from the first image, and determining a separation between the two groups of pixels based on a result of the statistical test.

[0011] (v) The layer fraction may be the percentage of the design-based structural elements that are classified as validly registered.

[0012] (vi) the second image may provide information of one or more design-based structural elements associated with a plurality of layers, and the calculating step, the using step, and the generating step are performed for each of the plurality of layers to generate a plurality of layer scores, and wherein the PMC is further configured to determine the validity of the registered image pair based on the plurality of layer scores and a plurality of layer thresholds predetermined in the inspection recipe.

[0013] According to other aspects of the presently disclosed subject matter, a method for inspecting a semiconductor sample using an inspection recipe is provided, the method being executed by a processor and memory circuitry (PMC), the method comprising the steps of: obtaining a registered image pair, the registered image pair comprising a first image and a second image, the first image being captured by an inspection tool and representing at least a portion of the semiconductor sample, the second image being generated based on design data characterizing the at least a portion of the sample, the second image providing information on one or more design-based structural elements present in the design data and corresponding layers associated with each design-based structural element; and for each design-based structural element associated with a given layer: calculating edge properties, the edge properties being correlated with the design-based structural element. The invention relates to a method for obtaining a registration between the first image and the second image according to the present invention; wherein the registration is associated with a designed structural element and indicates an estimated presence of an edge of the image-based structural element in the first image at a location of the edge of the designed structural element in the second image; and determining a category of the design-based structural element based on the edge attributes associated with the design-based structural element using a trained classifier included in the inspection recipe, the category indicating the validity of the registration between the first image and the second image at the location of the design-based structural element; and generating a layer score for the given layer based on the category of each design-based structural element associated with the given layer, the score being usable for determining the validity of the registered image pair based on a layer threshold predetermined in the inspection recipe.

[0014] This aspect of the disclosed subject matter may include, mutatis mutandis, one or more of the features (i) to (vi) listed above with respect to the system, in any desired combination or permutation technically possible.

[0015] According to other aspects of the presently disclosed subject matter, a non-transitory computer-readable medium comprising instructions is provided, the instructions, when executed by a computer, causing the computer to perform a method for inspecting a semiconductor sample using an inspection recipe, the method being performed by a processor and memory circuitry (PMC), the method comprising the steps of: obtaining a registered image pair, the registered image pair comprising a first image and a second image, the first image being captured by an inspection tool and representing at least a portion of the semiconductor sample, the second image being generated based on design data characterizing the at least a portion of the sample, the second image providing information of one or more design-based structural elements present in the design data and a corresponding layer associated with each design-based structural element; and for each design-based structural element associated with a given layer, : calculating an edge attribute associated with the design-based structural element and indicating an estimated presence of an edge of the image-based structural element in the first image at a location of the edge of the design-based structural element in the second image; and determining, using a trained classifier included in the inspection recipe, a class of the design-based structural element based on the edge attribute associated with the design-based structural element, the class indicating the validity of the registration between the first image and the second image at the location of the design-based structural element; and generating a layer score for the given layer based on the class of each design-based structural element associated with the given layer, the score being usable for determining the validity of the registered image pair based on a layer threshold predetermined in the inspection recipe.

[0016] This aspect of the disclosed subject matter may include, mutatis mutandis, one or more of the features (i) to (vi) listed above with respect to the system, in any desired combination or permutation technically possible.

[0017] According to certain aspects of the presently disclosed subject matter, a system for generating an inspection recipe that can be used to inspect a semiconductor sample is provided, the system comprising a processor and memory circuitry (PMC), the processor and memory circuitry (PMC) being configured to: obtain a training set, the training set comprising: i) a first subset comprising one or more first image pairs, each first image pair comprising a first image and a second image, the first image being captured by an inspection tool and representing at least a portion of a sample, the second image being generated based on design data characterizing the at least portion of the sample, the first image and the second image being effectively registered, the second image providing information of one or more design-based structural elements presented in the design data and corresponding layers associated with each design-based structural element; and ii) a second subset comprising one or more second image pairs, each second image pair being generated by: modifying at least one of the first image and the second image in a corresponding first image pair in the first subset so that the generated second image pair is invalidly registered; and modifying one or more of the second images of each image pair in the training set based on the design data. Each of the designed structural elements is associated with a label indicating the validity of the registration; for each designed structural element, an edge attribute is calculated, the edge attribute being associated with the designed structural element and indicating the estimated presence of an edge of the image-based structural element in the first image at the location of the edge of the designed structural element in the second image; a classifier is trained using the edge attribute associated with each designed structural element associated with a given layer and the label of the designed structural element to determine the validity of the registration between the first image and the second image at the location of the designed structural element; a classifier is determined using the trained classifier for each designed structural element associated with the given layer of each image pair, the determination being based on the edge attribute associated with the designed structural element, and a layer score for the given layer is generated based on the determined category of each designed structural element; a layer threshold is generated based on the layer score of the given layer for each image pair of the training set; and the trained classifier and the layer threshold associated with the given layer are included in the inspection recipe.

[0018] In addition to the features described above, the system according to this aspect of the presently disclosed subject matter may also include one or more of the features (i) to (vi) listed below, in any desired combination or arrangement that is technically possible:

[0019] (i) At least one second image pair may be generated by modifying the content of the first image in the first image pair.

[0020] (ii) At least one second image pair may be generated by modifying the relative position between the first image and the second image in the first image pair.

[0021] (iii) The PMC may be configured to perform the associating step based on at least one of: the registration validity of the corresponding image pair, and the modification if the corresponding image pair is a second image pair.

[0022] (iv) The PMC may be further configured to: calculate a grayscale profile of each design-based structural element along a specific direction at a position in the first image corresponding to a position of the design-based structural element in the second image; calculate a baseline grayscale profile of each family of design-based structural elements among the effectively registered design-based structural elements along the specific direction, the calculation being based on the grayscale profile of each design-based structural element in the family; and calculate a profile attribute for each design-based structural element, the profile attribute being associated with the design-based structural element and indicating a difference between the grayscale profile of the design-based structural element and the baseline grayscale profile, wherein the steps of training a classifier and using the trained classifier are performed based on the edge attribute and the profile attribute associated with the design-based structural element. The inspection recipe may further include the baseline grayscale profile.

[0023] (v) A plurality of baseline grayscale profiles may be calculated for each family along a plurality of specific directions and included in the inspection recipe.

[0024] (vi) The second image in each image pair can provide information of one or more design-based structural elements associated with multiple layers, and the associating step, the calculating step, the training step, the using step, and the including step can be performed for each of the multiple layers to generate multiple inspection recipes corresponding to the multiple layers.

[0025] According to other aspects of the presently disclosed subject matter, a method for generating an inspection recipe that can be used to inspect a semiconductor sample is provided, the method comprising the steps of: obtaining a training set, the training set comprising: i) a first subset, the first subset comprising one or more first image pairs, each first image pair comprising a first image and a second image, the first image being captured by an inspection tool and representing at least a portion of the sample, the second image being generated based on design data characterizing the at least portion of the sample, the first image and the second image being effectively registered, the second image providing information of one or more design-based structural elements presented in the design data and corresponding layers associated with each design-based structural element; and ii) a second subset, the second subset comprising one or more second image pairs, each second image pair being generated by the steps of: modifying at least one of the first image and the second image in a corresponding first image pair in the first subset such that the generated second image pair is not effectively registered; registering each of the one or more design-based structural elements in the second image of each image pair of the training set with an indication of registration. a label of validity; for each design-based structural element, calculating an edge attribute associated with the design-based structural element and indicating the estimated presence of an edge of the image-based structural element in the first image at the location of the edge of the design-based structural element in the second image; training a classifier using the edge attribute associated with each design-based structural element associated with a given layer and the label of the design-based structural element to determine the validity of the registration between the first image and the second image at the location of the design-based structural element; using the trained classifier to determine the category of each design-based structural element associated with the given layer of each image pair, the determination being based on the edge attribute associated with the design-based structural element, and generating a layer score for the given layer based on the determined category of each design-based structural element; generating a layer threshold based on the layer score for the given layer of each image pair of the training set; and including the trained classifier and the layer threshold associated with the given layer in the inspection recipe.

[0026] This aspect of the disclosed subject matter may include, mutatis mutandis, one or more of the features (i) to (vi) listed above with respect to the system, in any desired combination or permutation technically possible.

[0027] According to other aspects of the presently disclosed subject matter, there is provided a non-transitory computer-readable medium comprising instructions that, when executed by a computer, cause the computer to perform a method for generating an inspection recipe that can be used to inspect a semiconductor sample, the method comprising the steps of: obtaining a training set, the training set comprising: i) a first subset, the first subset comprising one or more first image pairs, each first image pair comprising a first image and a second image, the first image being captured by an inspection tool and representing at least a portion of the sample, the second image being generated based on design data characterizing the at least portion of the sample, the first image and the second image being effectively registered, the second image providing information of one or more design-based structural elements present in the design data and corresponding layers associated with each design-based structural element; and ii) a second subset, the second subset comprising one or more second image pairs, each second image pair being generated by: modifying at least one of the first image and the second image in a corresponding first image pair in the first subset such that the generated second image pair is invalidly registered; and Each of the designed-based structural elements is associated with a label indicating the validity of the registration; for each designed-based structural element, an edge attribute is calculated, the edge attribute being associated with the designed-based structural element and indicating the estimated presence of an edge of the image-based structural element in the first image at the location of the edge of the designed-based structural element in the second image; a classifier is trained using the edge attribute associated with each designed-based structural element associated with a given layer and the label of the designed-based structural element to determine the validity of the registration between the first image and the second image at the location of the designed-based structural element; a classifier is determined using the trained classifier for each of the designed-based structural elements associated with the given layer of each image pair, the determination being based on the edge attribute associated with the designed-based structural element, and a layer score for the given layer is generated based on the determined category of each designed-based structural element; a layer threshold is generated based on the layer score of the given layer for each image pair of the training set; and the trained classifier and the layer threshold associated with the given layer are included in the inspection recipe.

[0028] This aspect of the disclosed subject matter may include, mutatis mutandis, one or more of the features (i) to (vi) listed above with respect to the system, in any desired combination or permutation technically possible. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to understand the present disclosure and to see how it may be implemented in practice, embodiments will now be described, by way of non-limiting example only, with reference to the accompanying drawings, in which:

[0030] Figure 1A A functional block diagram of a recipe generation system according to certain embodiments of the presently disclosed subject matter is shown.

[0031] Figure 1B A functional block diagram of an inspection system according to certain embodiments of the presently disclosed subject matter is shown.

[0032] Figure 2 A generalized flow chart for generating an inspection recipe that can be used to inspect semiconductor samples in accordance with certain embodiments of the presently disclosed subject matter is shown.

[0033] Figure 3 A generalized flow chart illustrating the use of inspection recipes for inspecting semiconductor samples in accordance with certain embodiments of the presently disclosed subject matter is shown.

[0034] Figure 4 An exemplary image registration process in accordance with certain embodiments of the presently disclosed subject matter is shown.

[0035] Figure 5 Several examples of invalidly registered image pairs are shown in accordance with certain embodiments of the presently disclosed subject matter.

[0036] Figure 6 Several examples of second image pairs artificially generated based on first image pairs are shown in accordance with certain embodiments of the presently disclosed subject matter.

[0037] Figure 7 Schematically illustrates how edge properties may be used to test for edge existence in accordance with certain embodiments of the presently disclosed subject matter.

[0038] Figure 8 An example of calculating a GL profile and a baseline GL profile in accordance with certain embodiments of the presently disclosed subject matter is shown.

[0039] Figure 9 Several examples of curve properties in accordance with certain embodiments of the presently disclosed subject matter are shown.

[0040] Figure 10 is a schematic diagram of training a classifier using attributes and labels of design-based structural elements according to certain embodiments of the presently disclosed subject matter.

[0041] Figure 11 A schematic diagram illustrating generation of layer thresholds according to certain embodiments of the presently disclosed subject matter.

[0042] Figure 12An example of using a check recipe to perform runtime checking on a sample according to some embodiments of the presently disclosed subject matter is schematically illustrated. DETAILED DESCRIPTION

[0043] In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the present disclosure. However, those skilled in the art will appreciate that the presently disclosed subject matter may be practiced without these specific details. In other instances, well-known methods, procedures, components, and circuits are not described in detail so as not to obscure the presently disclosed subject matter.

[0044] Unless specifically stated otherwise, as will be clear from the following discussion, it should be understood that throughout this specification, discussions utilizing terms such as "examine," "obtain," "calculate," "use," "generate," "determine," "associate," "train," "include," "modify," "execute," and the like refer to computer actions and / or processes that manipulate data and / or transform data into other data, the data being represented as physical (such as electronic) quantities and / or the 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, by way of non-limiting example, the inspection system, recipe generation system, and corresponding portions thereof disclosed in this application.

[0045] 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 presently disclosed subject matter.

[0046] The term "sample" as used in this specification should be broadly interpreted to encompass any kind of wafers, masks and other structures, and combinations and / or portions thereof, used to manufacture semiconductor integrated circuits, magnetic heads, flat panel displays, and other semiconductor manufactured articles.

[0047] The term "inspection" as used in this specification should be interpreted broadly to cover any kind of metrology-related operations and operations related to detecting and / or classifying defects in the sample during the manufacture of the sample. Inspection is provided by using a non-destructive inspection tool during or after the manufacture of the sample to be inspected. By way of non-limiting example, the inspection process may include runtime scanning (in the form of a single or multiple scans), sampling, review, measurement, classification, and / or other operations provided with respect to the sample or part thereof using the same or different inspection tools. Similarly, inspection can be provided before the sample to be inspected is manufactured, and the inspection may include, for example, the generation of (multiple) inspection recipes and / or other setup operations. It should be noted that, unless specifically stated otherwise, the term "inspection" or its derivatives used in this specification has no limitation on the resolution or size of the inspection area. By way of non-limiting example, various non-destructive inspection tools include scanning electron microscopes, atomic force microscopes, optical inspection tools, and the like.

[0048] The term "defect" as used in this specification should be interpreted broadly to encompass any kind of abnormal or undesirable feature formed on or in a sample.

[0049] The term "design data" as used in this specification should be broadly interpreted to encompass any data indicative of the hierarchical physical design (layout) of a sample. The design data may be provided by a corresponding designer and / or may be derived from the physical design (e.g., through complex simulations, simple geometric and Boolean operations, etc.). The design data may be provided in various formats, such as, by way of non-limiting example, GDSII format, OASIS format, etc. The design data may be presented in a vector format, a grayscale intensity image format, or in other ways.

[0050] It should be understood that, unless specifically stated otherwise, certain features of the presently disclosed subject matter that are described in the context of separate embodiments may also be provided in combination in a single embodiment. Conversely, various features of the presently disclosed subject matter that are described in the context of individual embodiments may also be provided individually or in any suitable subcombination. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the methods and apparatus.

[0051] With this in mind, attention turns to Figure 1A , which shows a functional block diagram of a recipe generation system according to certain embodiments of the presently disclosed subject matter.

[0052] Figure 1A The recipe generation system 100 shown in FIG. 1 may be used to generate a recipe that may be used by an inspection system 130 (e.g., Figure 1B) for inspecting a semiconductor sample (eg, a wafer and / or a portion thereof). As will be described in more detail below with reference to Figure 2 As described herein, the illustrated system 100 can be a computer-based system capable of automatically generating an inspection recipe based on a training set comprising image pairs representing at least a portion of a sample. According to certain embodiments of the presently disclosed subject matter, the system 100 can be operatively connected to one or more inspection tools 120 configured to capture one or more images of a sample. Figure 1B The details of the inspection tool and the images captured thereby are described in more detail in the following. According to certain embodiments, the system 100 may be further operatively connected to a design data server 110 (e.g., a CAD server) configured to store and provide design data representing the sample. The design data of the sample may be in any of the following formats: a physical design layout of the sample (e.g., a CAD fragment), a raster image, and a simulated image derived from the design layout. The images and design data of the sample may be pre-collected and stored in the storage unit 122 and may be used to generate a training set.

[0053] System 100 includes a processor and memory circuitry (PMC) 102 operatively connected to a hardware-based I / O interface 126. PMC 102 is configured to provide Figure 2 The PMC 102 includes all processing required for operating the system 101 as further described, and includes a processor (not separately shown) and memory (not separately shown). The processor of the PMC 102 can be configured to execute several functional modules according to computer-readable instructions implemented on a non-transitory computer-readable memory included in the PMC. Such functional modules are hereinafter referred to as being included in the PMC.

[0054] According to certain embodiments of the presently disclosed subject matter, functional modules included in PMC 102 may include a training set generator 103, an attribute generator 104, a training module 105, a classifier 106, and a recipe generator 107. Training set generator 103 may be configured to obtain a training set comprising a first subset comprising one or more first image pairs and a second subset comprising one or more second image pairs. Each first image pair comprises two correctly / validly registered images: a first image captured by an inspection tool and representing at least a portion of a sample, and a second image generated based on design data representing at least a portion of the sample. Each second image pair is artificially / synthetically generated based on the corresponding first image pair (e.g., by modifying at least one of the first and second images in the first image pair), such that the two images in the generated second image pair are invalidly registered. In some cases, the training set may be pre-generated and stored in storage unit 122 and may be retrieved from storage unit 122 by training set generator 103. In other cases, the training set may be generated by system 100. For example, the training set generator 103 may be configured to receive captured images from the inspection tool 120 and corresponding design data from the design data server 110 via the I / O interface 126 and perform image registration therebetween. The training set generator 103 may be further configured to generate a second subset of image pairs. Figure 2 Image registration and generation of the second subset are described in more detail.

[0055] Each second image may provide information of one or more design-based structural elements present in the design data and corresponding layers associated with each design-based structural element. The training set generator 103 may be further configured to associate each of the one or more design-based structural elements in the second image of each image pair of the training set with a label indicating registration validity.

[0056] For each design-based structural element, the attribute generator 104 may be configured to calculate an edge attribute associated with the design-based structural element, the edge attribute indicating an estimated presence of an edge of the image-based structural element in the first image at a location of the edge of the design-based structural element in the second image. The training module 105 may be configured to train the classifier 106 using the label of each design-based structural element and the edge attribute associated with the design-based structural element to determine the validity of the registration between the first image and the second image at the location of the design-based structural element.

[0057] The recipe generator 107 can be configured to: use a trained classifier to determine a category of each of one or more design-based structural elements associated with a given layer of each image pair, the determination being based on edge attributes associated with the design-based structural elements; and generate a layer score for the given layer based on the determined category of the one or more design-based structural elements. The recipe generator 107 can be further configured to generate a layer threshold based on the layer score of the given layer of each image pair of the training set, and include the trained classifier and layer threshold in the inspection recipe.

[0058] Will refer to Figure 2 The operation of the system 100, the PMC 102, and the functional modules therein are further described in detail.

[0059] According to certain embodiments, system 100 may include a storage unit 122. Storage unit 122 may be configured to store any data required to operate system 100 (e.g., data related to input and output of system 100), as well as intermediate processing results generated by system 100. For example, storage unit 122 may be configured to store training images and / or derivatives thereof generated by inspection tool 120. Storage unit 122 may also be configured to store design data representing samples and / or derivatives thereof. In some embodiments, storage unit 122 may be configured to store pre-generated training sets as described above. Thus, stored data may be retrieved from storage unit 122 and provided to PMC 102 for further processing.

[0060] In some embodiments, the system 100 may optionally include a computer-based graphical user interface (GUI) 124 configured to implement user-specified input related to the system 100. For example, a visual representation of a sample including image data and / or design data of the sample may be presented to the user (e.g., via a display forming part of the GUI 124). The user may be provided with options for defining certain operational parameters (e.g., training configuration parameters) via the GUI. The user may also view operational results (such as, for example, training results, etc.) on the GUI.

[0061] Those skilled in the art will readily appreciate that the teachings of the presently disclosed subject matter are not limited to Figure 1A The present invention is not limited to the systems shown in FIG; equivalent and / or modified functionality may be combined or divided in another manner and may be implemented in any suitable combination of software and firmware and / or hardware.

[0062] Note that you can Figure 1A The recipe generation system shown in is implemented in a distributed computing environment in which Figure 1AThe above-described functional modules shown in can be distributed across several local devices and / or remote devices and can be linked via a communication network. It is further noted that in other embodiments, at least a portion of the storage unit 122 and / or the GUI 124 can be external to the system 100 and operable to communicate data with the system 100 via the I / O interface 126. The system 100 can be implemented as a (multiple) stand-alone computer to be used in conjunction with an inspection tool. Alternatively, in some cases, the corresponding functionality of the system 100 can be at least partially integrated with one or more inspection tools 120.

[0063] Attention shifts Figure 1B , which shows a functional block diagram of an inspection system according to certain embodiments of the presently disclosed subject matter.

[0064] Figure 1B The inspection system 130 shown in FIG. 1 can be used to inspect semiconductor samples (e.g., semiconductor samples of wafers and / or portions thereof) as part of a sample manufacturing process. The system 130 can be used by Figure 1A The inspection recipe generated by the recipe generation system 100 shown in the embodiment of the present invention is used to inspect the sample. The inspection system 130 shown may include a computer-based system 131 that can automatically determine inspection-related information using images obtained during sample manufacturing (hereinafter referred to as manufacturing process (FP) images or images) and design data characterizing the sample. Generally speaking, the system 131 can be referred to as an FPEI (manufacturing process inspection information) system. According to certain embodiments of the presently disclosed subject matter, as will be described in more detail below with reference to Figure 3 As described above, system 131 can be configured to determine the registration validity of an image pair. System 131 can be operatively connected to one or more inspection tools 120. Inspection tool 120 is configured to capture FP images and / or review captured FP images and / or perform or provide measurements related to the captured images. System 131 can be further operatively connected to design data server 110 and storage unit 142.

[0065] By way of example, an FP image (also referred to herein as an image) may be selected from a plurality of images of a sample (e.g., a wafer or portion thereof) captured during a manufacturing process, derivatives of captured images obtained by various pre-processing stages (e.g., an image of a portion of a wafer or photomask captured by an SEM or optical inspection system, an SEM image approximately centered on a defect to be classified by an ADC, an SEM image of a larger area in which a defect to be located by an ADR is located, registered images of different inspection modalities corresponding to the same mask position, segmented images, height map images, etc.). It should be noted that in some cases, an image may include image data (e.g., a captured image, a processed image, etc.) and associated digital data (e.g., metadata, handcrafted attributes, etc.). It should be further noted that the image data may include data related to a layer of interest and / or one or more layers of the sample.

[0066] The term "inspection tool" as used herein should be broadly interpreted to encompass any tool that can be used in an inspection-related process, including, by way of non-limiting example, imaging, scanning (performed in a single or multiple scans), sampling, reviewing, measuring, sorting, and / or other processes provided with respect to a sample or portion thereof. One or more inspection tools 120 may include one or more inspection tools and / or one or more review tools. In some cases, at least one of the inspection tools 120 may be an inspection tool configured to scan a sample (e.g., an entire wafer, an entire die, or a portion thereof) to capture inspection images (typically, captured at a relatively high speed and / or low resolution) for detecting potential defects. In some cases, at least one of the inspection tools 120 may be a review tool configured to capture review images of at least some of the defects detected by the inspection tool for determining whether the potential defects are actually defects. Such review tools are typically configured to inspect segments of the die one at a time (typically, inspected at a relatively low speed and / or high resolution). The inspection tool and review tool may be different tools located at the same or different locations, or a single tool operating in two different modes.In some cases, at least one inspection tool may have metrology capabilities.

[0067] Without limiting the scope of the present disclosure in any way, it should also be noted that the inspection tool 120 can be implemented as various types of inspection machines, such as an optical imaging machine, an electron beam inspection machine, etc. In some cases, the same inspection tool can provide low-resolution image data and high-resolution image data.

[0068] System 131 includes a processor and memory circuitry (PMC) 132 operatively connected to a hardware-based I / O interface 136. PMC 132 is configured to provide Figure 3The PMC 132 includes all processing required by the operating system 131, which is further described in detail, and includes a processor (not separately shown) and memory (not separately shown). The processor of the PMC 132 can be configured to execute several functional modules according to computer-readable instructions implemented on a non-transitory computer-readable memory included in the PMC. Such functional modules are hereinafter referred to as being included in the PMC.

[0069] According to certain embodiments, functional modules included in PMC 132 may include an attribute generator 134, a classifier 135, and a validity determination module 138. PMC 132 may be configured to obtain a registered image pair at runtime (i.e., production time / stage) via I / O interface 136, the registered image pair comprising a first image captured by an inspection tool and representing at least a portion of a semiconductor sample and a second image generated based on design data representing at least a portion of the sample. The second image may provide information about one or more design-based structural elements presented in the design data and corresponding layers associated with each design-based structural element. In some embodiments, after receiving the captured image from inspection tool 120 and the corresponding design data from design data server 110 via I / O interface 136, system 131 may perform image registration between the first and second images. Alternatively, image registration or at least a portion of its functionality may be performed externally (e.g., by a system external to system 131), and the registration results (i.e., the registered image pair) may be provided to system 131 for further processing.

[0070] For each design-based structural element associated with a given layer, the attribute generator 134 can be configured to calculate an edge attribute associated with the design-based structural element, the edge attribute indicating the estimated presence of an edge of the image-based structural element in the first image at the location of the edge of the design-based structural element in the second image. A classifier 135 (i.e., a trained classifier) ​​included in the inspection recipe can be used to determine the category of the design-based structural element based on the edge attribute associated with the design-based structural element. The category indicates the effectiveness of the registration between the first image and the second image at the location of the design-based structural element. The effectiveness determination module 138 can be configured to generate a layer score for the given layer based on the category of each design-based structural element associated with the given layer. The score can be used to determine the effectiveness of the registered image pair based on a threshold predetermined in the inspection recipe.

[0071] Will refer to Figure 3 The operations of system 130, system 131, PMC 132 and the functional modules therein are further described in detail.

[0072] According to certain embodiments, system 130 may include a storage unit 142. Storage unit 142 may be configured to store any data required to operate system 130 (e.g., data related to input and output of system 130), as well as intermediate processing results generated by system 130. For example, storage unit 142 may be configured to store images and / or derivatives thereof generated by inspection tool 120. Storage unit 142 may also be configured to store design data representing a sample and / or derivatives thereof. In some embodiments, storage unit 142 may be configured to store pre-generated registered image pairs as described above. Thus, stored data may be retrieved from storage unit 142 and provided to PMC 132 for further processing.

[0073] In some embodiments, the system 130 may optionally include a computer-based graphical user interface (GUI) 144 configured to implement user-specified input related to the system 130. For example, a visual representation of a sample including image data and / or design data of the sample may be presented to the user (e.g., via a display forming part of the GUI 144). Options for defining certain operating parameters may be provided to the user via the GUI. The user may also view the results of the operation (such as, for example, inspection results, etc.) on the GUI. In some cases, the system 131 may be further configured to send the results (or portions thereof) to the storage unit 142, and / or to the inspection tool(s) 120, and / or to an external system (e.g., a yield management system (YMS) of the FAB) via the I / O interface 126.

[0074] According to certain embodiments, the system 131 may be operatively connected to a segmentation network 112 configured to perform segmentation on inspection results of the system 131. The segmentation network 112 may be a deep learning model, such as a deep neural network (DNN), comprising layers organized according to a corresponding DNN architecture. By way of non-limiting example, the layers of the DNN may be organized according to a convolutional neural network (CNN) architecture, a recurrent neural network architecture, a recursive neural network architecture, a generative adversarial network (GAN) architecture, or in other ways. Optionally, at least some of the layers may be organized in multiple DNN subnetworks. Each layer of the DNN may include multiple basic computational elements (CEs), which are generally referred to in the art as dimensions, neurons, or nodes.

[0075] In general, the computational elements of a given layer can be connected to the CEs of the previous layer and / or the next layer. Each connection between the CE of the previous layer and the CE of the next layer is associated with a weighted value. A given CE can receive input from the CE of the previous layer via the corresponding connection, and each given connection is associated with a weighted value, which can be applied to the input of the given connection. The weighted value can determine the relative strength of the connection and therefore the relative influence of the corresponding input on the output of the given CE. A given CE can be configured to calculate an activation value (e.g., a weighted sum of the inputs) and further derive an output by applying an activation function to the calculated activation. It should be noted that the teachings of the presently disclosed subject matter are not constrained by a specific architecture of the DNN.

[0076] In some embodiments, the system 131 may further include the segmented network 112 or a portion thereof. In other words, the corresponding functions of the segmented network 112 may be at least partially integrated into the system 131.

[0077] In addition to the system 131, the inspection system 100 may also include one or more inspection modules, such as, for example, a defect detection module, an automatic defect review module (ADR), an automatic defect classification module (ADC), a metrology-related module, and / or other inspection modules. Such inspection modules may use the output of the system 131 and / or the output of the segmentation network 112 to inspect semiconductor samples. In some cases, the one or more inspection modules may be at least partially integrated with the one or more inspection tools 120.

[0078] Those skilled in the art will readily appreciate that the teachings of the presently disclosed subject matter are not limited to Figure 1B The present invention is not limited to the systems shown in FIG; equivalent and / or modified functionality may be combined or divided in another manner and may be implemented in any suitable combination of software and firmware and / or hardware.

[0079] Note that you can Figure 1B The recipe generation system shown in is implemented in a distributed computing environment in which Figure 1B The above-mentioned functional modules shown in can be distributed on several local devices and / or remote devices and can be linked through a communication network. It is further noted that in other embodiments, at least some of the inspection tool 120, the storage unit 142, and / or the GUI 144 can be external to the system 130 and operate to communicate data with the system 130 via the I / O interface 126. The system 130 can be implemented as (a plurality of) independent computers to be used in conjunction with the inspection tool. Alternatively, in some cases, the corresponding functions of the system 130 can be at least partially integrated with one or more inspection tools 120, thereby facilitating and enhancing the functionality of the inspection tool 120 in the inspection-related process.

[0080] As mentioned above Figure 1B As described, an inspection system 130 including inspection tool(s) 120 may be configured to capture one or more images of a sample and perform inspection-related operations, such as, for example, defect detection, and / or defect review, and / or metrology-related operations, etc. According to certain embodiments, the inspection system may be configured to obtain the images captured by the inspection tool(s) and corresponding design data (e.g., CAD data), and perform registration between the images and the CAD data to perform further inspection operations, such as, for example, segmentation and metrology measurements, etc.

[0081] Now refer to Figure 4 , illustrating an exemplary image registration process in accordance with certain embodiments of the presently disclosed subject matter.

[0082] FP image 402 (also referred to herein as a first image) can be obtained from an inspection tool and represents a portion of a wafer. Design data 404 (e.g., a CAD fragment) representing the same portion of the wafer can be obtained from a CAD server. As shown in this example, CAD fragment 404 presents two types of design-based structural elements (e.g., polygons): cylindrical polygons in a first layer, and square polygons in a second layer. A design-based image 406 (also referred to herein as a second image) can be generated based on the CAD data. For example, CAD fragment 404 can be simulated to generate a simulated CAD image 406. In some cases, such simulation can account for differences between the design-based structural elements and the corresponding image-based structural elements. For example, in some cases, due to factors such as manufacturing process variations (e.g., printing the design pattern on the wafer using an optical lithography tool) and / or design tool constraints, a design-based structural element (such as a square polygon) may actually appear as an ellipse in the image, as shown in simulated CAD image 406. The simulation can account for such differences and generate a simulated image that includes the structural elements with simulated shapes as they would appear in the image. As can be seen, the polygon shapes in simulated CAD image 406 are more consistent with the polygon shapes in captured image 402 than with CAD data 404 .

[0083] FP image 402 and simulated CAD image 406 can be registered to align the CAD image with the FP image. For example, global registration can be performed to globally align the two images based on their relative position. The relative position between the CAD image and the FP image may be caused by, for example, navigation issues with the inspection tool when capturing the FP image. As another example, in some cases, registration can be performed in two stages: global registration as described above, and per-layer registration, which aligns polygons in each layer of the specimen. Figure 4 Such a two-stage registration process is shown. As shown, after global registration, a globally registered image pair is shown in 408, where the cylindrical polygons are correctly registered (e.g., corresponding polygons in the two images match), while there is a difference (e.g., a relative offset) between the positions of the elliptical polygons in the two images. In this example, this may be caused by, for example, an overlay problem during the manufacturing process, which results in a relative offset between the two layers in the manufactured sample compared to the design.

[0084] After each layer registration is performed, the elliptical polygons in the second layer of the simulated CAD image 406 are shifted accordingly to match the corresponding polygons in the FP image, thereby producing a registered image pair 410 .

[0085] According to certain embodiments of the present disclosure, after registration (e.g., global registration or two-stage registration as described above), an image pair in which the FP image and the corresponding design-based image are aligned and the structural elements of each layer between the two images match (e.g., satisfying predefined matching criteria) is referred to herein as effectively / correctly registered (such an image pair is also referred to as a "good image pair"). When further inspection operations (such as, for example, segmentation and metrology measurements, etc.) are performed on the effectively registered image pair, the user can be assured and confident that the inspection results (e.g., measurement results such as overlay, CD, etc.) obtained are accurate and can be trusted.

[0086] However, in some cases, even after the registration process as described above, the image pairs are not correctly registered, i.e., after registration, there are still mismatches or differences between the FP image and the corresponding design-based image (e.g., predefined matching criteria are not met). Such image pairs are referred to herein as invalid / incorrectly registered image pairs (also referred to as "erroneous image pairs"). Now referring to Figure 5 , showing several examples of invalidly registered image pairs in accordance with certain embodiments of the presently disclosed subject matter.

[0087] In 502, a mismatch between an FP image and CAD due to a navigation error is shown. For example, the inspection tool may be navigated incorrectly, thereby capturing an FP image that does not correspond to the CAD data from the wrong location. In 504, a mismatch between an FP image and CAD due to a focus problem is shown. As shown, the FP image in 504 is captured out of focus (e.g., the image polygons have blurred edges) and therefore do not match the CAD polygons. 506 shows an example of misregistration of polygons in a layer. Specifically, the elliptical polygons in the FP image and CAD are not correctly registered. 508 and 510 show two different examples of incorrect patterning, in which the patterns in the FP image and CAD do not match. For example, due to possible manufacturing errors, in 508, some polygons are missing from the FP image compared to the CAD data, while in 510, the pattern in the image is completely different from the design. 512 shows an example of a bare FP image, which does not capture any pattern compared to the CAD data.

[0088] Mismatches / discrepancies between the images and the CAD data, as exemplified above, may be caused by a variety of reasons, such as, for example, inspection system problems (e.g., focus problems, navigation errors), algorithmic difficulties (e.g., registration failures), and / or manufacturing process failures (e.g., incorrect patterning, bare images). Performing further inspection operations (such as, for example, segmentation and metrology measurements, etc.) on such invalidly registered image pairs without knowing that a mismatch exists may result in meaningless results being reported to the user. Such results may provide misleading information for adjusting the manufacturing process, thereby causing serious damage and reducing the user's confidence in the inspection system in some cases. For example, incorrect overlay measurements (caused by invalid registration between the image pairs) may mislead future manufacturing processes, thereby resulting in faulty production wafers.

[0089] According to certain statistics, it is shown that about 5% of the sites on the wafer may be associated with such mismatch / difference problems between the image and the corresponding design data. Such sites are also referred to as low-confidence sites in this article. Some inspection systems do not provide a filtering mechanism to detect such problems during the inspection process. To alleviate such problems, users can manually check all sites on the wafer before inspection, or check the inspection results of the entire wafer after inspection to find outliers. Both measures are extremely time-consuming, inefficient, and ineffective. In some cases, users must iterate the calibration process of the inspection recipe many times until these cases become rare, which significantly increases the recipe time (Time-to-Recipe, TtR).

[0090] According to certain embodiments of the presently disclosed subject matter, as shown in FIG. Figure 2 and Figure 3Described in more detail, methods are proposed for generating inspection recipes capable of detecting such invalid registration problems, and methods for inspecting samples using the inspection recipes thus generated. These methods are machine learning-based and require minimal user calibration and interaction.

[0091] Reference Figure 2 , shows a generalized flow chart for generating an inspection recipe that can be used to inspect semiconductor samples according to certain embodiments of the presently disclosed subject matter.

[0092] A training set (202) can be obtained (e.g., by the training set generator 103 via the I / O interface 126 or from the storage unit 122). The training set can include a first subset 204 and a second subset 206, the first subset 204 including one or more first image pairs, and the second subset 206 including one or more second image pairs. Each first image pair in the first subset includes two effectively registered images: a first image captured by an inspection tool and representing at least a portion of the sample, and a second image generated based on design data characterizing the at least portion of the sample. For example, the first image can be a FP image captured by the inspection tool. In some cases, the first image is a high-resolution image captured by a review tool, such as, for example, a scanning electron microscope (SEM) image captured by a SEM machine.

[0093] The second image may be a simulated image generated by performing a simulation on the design data, such as, for example, a CAD simulated image. As described above, the simulation may account for the differences between the design-based structural elements and the corresponding image-based structural elements and generate a simulated image including the structural elements having the simulated shapes as they would appear in the image.

[0094] The second image may include one or more layers corresponding to the design layers in the design data. The second image may provide information on one or more design-based structural elements present in the design data and corresponding layers associated with each design-based structural element. Each layer may be viewed as a binary image in which the design-based structural element is separated from the background. The first image is effectively registered with the second image. Figure 4 Image registration between the first image and the second image is performed in a similar manner to that described.

[0095] As used herein, a structural element may refer to any primitive object having a geometric shape or geometric structure with an outline on image data or design data. In some cases, a structural element may refer to a plurality of combined objects forming a pattern. Structural elements positioned / presented on image data may be referred to as image-based structural elements (also referred to as image structural elements or image polygons). Structural elements positioned / presented on design data may be referred to as design-based structural elements (also referred to as design structural elements or design polygons or CAD polygons). Figure 4 As illustrated in , the structural elements may be presented in the form of polygons, for example. The structural elements may be defined by a user, or may be defined automatically, for example, using rule-based techniques or machine learning techniques.

[0096] Each design-based structural element can be associated with the corresponding layer in which the element is located. In some cases, a given layer can contain more than one type of structural element (e.g., each type corresponding to a respective design pattern). In such cases, each design-based structural element can be further associated with a tag indicating its type.

[0097] The first image pair included in the first subset is a validly registered image pair pre-selected (e.g., by a user) from a training database. As described above, the user can select and confirm the validity of multiple locations to ensure that there are no undesirable differences between the images and the corresponding CAD data.

[0098] Because invalid / erroneous image pairs are generally rare in a wafer, the present disclosure proposes a method for artificially / synthetically generating such image pairs (i.e., second image pairs as referred to herein) to form part of a training set. According to certain embodiments, the second image pairs can be artificially generated based on the corresponding first image pairs. For example, the second image pair can be generated by modifying at least one of the first image and the second image in the first image pair so that the two images in the generated second image pair are invalidly aligned. For example, at least one second image pair can be generated by modifying the content of the first image in the first image pair. As another example, at least one second image pair can be generated by modifying the relative position between the first image and the second image in the first image pair. In some embodiments, one or more second image pairs can be generated for each first image pair.

[0099] Now refer to Figure 6 , showing several examples of second image pairs artificially generated based on first image pairs in accordance with some embodiments of the presently disclosed subject matter.

[0100] In the second image pair shown in 602, the content of the original FP image of the first image pair is modified (e.g., replaced by a bare image (e.g., an image consisting only of random noise)), while the CAD simulation image remains intact. In 604 and 606, the relative position between the FP image and the CAD simulation image (or parts thereof) is modified. Specifically, the design-based structural elements in a certain layer of the CAD simulation image are offset relative to the FP image, while the FP image remains unchanged. For example, in 604, the cylindrical polygons in one layer are offset toward the left, while the elliptical polygons in another layer do not move. In 606, the elliptical polygons are offset toward the left, while the elliptical polygons in another layer do not move. Therefore, in this example, three different second image pairs are generated for one first image pair.

[0101] It should be noted that Figure 6 The examples shown in are shown for illustrative purposes and should not be considered as limiting the present disclosure in any way. In addition to or in lieu of the above, other possible ways of modifying the first image and / or the second image may be used (such as, for example, modifying the size and / or shape of one or more of the design-based structural elements in a layer, degrading the focus of the FP image, for example by convolving the image with a Gaussian kernel, replacing a portion of the FP image with random noise, etc.).

[0102] In some embodiments, the training set may be pre-generated and stored in the storage unit 122 and may be retrieved by the training set generator 103 during training. In such cases, the training set may be generated externally to the system 100 and stored in a manner such as Figure 2 The functionality of performing image registration and / or generating the second subset (or at least a portion thereof) is performed prior to the presently proposed recipe generation method described in

[0046] . In some other embodiments, such functionality may be performed by the system 100 (e.g., by the training set generator 103) and thus may be considered as Figure 2 part of the current process.

[0103] continue Figure 2 As described, once a training set including a first subset and a second subset is obtained, each of the one or more design-based structural elements in the second image of each image pair of the training set can be associated (e.g., by the training set generator 103) with a label indicating the validity of the alignment (208).

[0104] According to some embodiments, labeling may be performed based on at least one of the following factors: registration validity of the corresponding image pair, and modification if the corresponding image pair is a second image pair. For example, for each of the first image pairs in the first subset that are selected by the user and confirmed as validly registered image pairs, a design-based structural element in the second image of the first image pair is labeled as a validly registered structural element.

[0105] As another example, for a second image pair in the second subset that was artificially generated as an invalidly registered image pair, the labeling depends on the specific modification applied to the second image pair. For example, in the case where the content of the FP image is modified (such as, for example, replacing the FP image with a bare image), all design-based structural elements in its second image are labeled as invalidly registered structural elements. In another example, in the case where the relative position between the first image and the second image is modified (such as, for example, offsetting polygons in a specific layer of the second image relative to the first image), the offset polygons in the specific layer are labeled as invalidly registered structural elements, while polygons in other layers are still labeled as validly registered structural elements. In this way, all design-based structural elements in the training set are associated with corresponding labels indicating the validity of their registration.

[0106] One or more properties of the design-based structural element may be calculated and used together with the label associated with the design-based structural element to train a classifier. The calculation of the properties is now described below with reference to block 210 .

[0107] For each design-based structural element, an edge attribute associated with the design-based structural element may be calculated (e.g., by the attribute generator 104) 210. The edge attribute may indicate an estimated presence of an edge of the image-based structural element in the first image at a location of an edge of the design-based structural element in the second image.

[0108] The edge attribute is used as a measure of the presence of a grayscale edge in the FP image (e.g., an edge of an image polygon) at the location of an edge of a CAD polygon in the simulated CAD image. According to certain embodiments, the edge attribute can be calculated by applying a statistical test between two pixel groups on either side of an edge of a design-based structural element in the first image and determining a separation between the two pixel groups based on the result of the statistical test. The separation can indicate whether an edge actually exists in the FP image at the estimated location (e.g., the location of a CAD polygon edge).

[0109] Reference Figure 7 , schematically illustrating how edge attributes may be used to test for edge existence in accordance with certain embodiments of the presently disclosed subject matter.

[0110] As shown in the examples of 702, 704, and 706, a strip of pixels is taken from either side of the estimated edge in the FP image (shown as two dashed lines from the inside and outside of the estimated edge location), and a statistical test is applied between the two pixel populations to test for separation between the two pixel populations: the better the separation, the higher the probability that an edge exists in the estimated location. One possible test that can be applied is the Kolmogorov-Smirnov (KS) test. The KS test quantifies the maximum distance between distribution functions (such as the normalized cumulative distribution function (CDF) shown in this example), and the distance can indicate the separation between the two populations.

[0111] For example, image 702 shows that, in the case where the image polygon edges match the CAD polygon edges (i.e., edges do exist in the FP image at the locations of the CAD polygon edges), the edge properties calculated by the KS test show a large separation between the two populations (e.g., the vertical line 701 shown corresponds to the maximum distance between the two curves in the corresponding graph 703). In contrast, in image 704 (which shows a misalignment between the estimated edges and the actual edges in the FP image) and image 706 (which shows a bare image without polygons and edges), the edge properties calculated by the KS test show little or no separation between the two populations (e.g., the vertical line 708 corresponding to the maximum distance between the two curves shows a slight separation between the two curves in the corresponding graph 705, while the two curves almost overlap in the corresponding graph 707).

[0112] Therefore, by using the edge attributes thus calculated, it is possible to estimate whether an edge actually exists at the position of the CAD polygon edge in the FP image.

[0113] According to some embodiments, optionally, for each design-based structural element associated with a given layer, in addition to the edge properties, one or more contour properties associated with the design-based structural element may be calculated.

[0114] Specifically, in some embodiments, a grayscale (GL) profile of each design-based structural element may be calculated along a specific direction at a position in the first image that corresponds to a position of the design-based structural element in the second image. Once the GL profile is obtained for each design-based structural element, a baseline GL profile may be calculated for each family of design-based structural elements in the effectively registered design-based structural elements. Based on the GL profile of each design-based structural element in the family, a baseline may be calculated along a specific direction in the first image. For example, the baseline may be calculated by averaging the GL profiles of each design-based structural element in the family. A family of design-based structural elements includes equivalent design-based structural elements that share the same design pattern. For example, Figure 4 The elliptical polygons in 406 belong to one family, while the cylindrical polygons belong to another family. In some cases, a given layer may contain one or more families of structural elements based on the design.

[0115] Once the baseline GL profile is generated, a profile property may be calculated for each design-based structural element, the profile property indicating the difference between the grayscale profile of the design-based structural element and the baseline grayscale profile.

[0116] Now refer to Figure 8 , which shows an example of calculating a GL profile and a baseline GL profile according to certain embodiments of the presently disclosed subject matter.

[0117] In 802, an FP image of a single elliptical polygon is illustrated. Because the position in the FP image used to calculate the GL profile should be determined based on the position of the design-based structural elements in the CAD simulation image, it is assumed in this example that the polygon is effectively registered between the image and the CAD data, that is, the position of the image polygon matches the position of the CAD polygon. Therefore, the GL profile should be calculated at the position of the image polygon. As shown in the figure, a specific direction / angle 804 (shown as a parallel chord) is selected, along which the profile attributes are calculated. It should be noted that the image of the polygon has been processed by masking the overlapping portion overlapped by another polygon (e.g., a cylindrical polygon) from the upper layer in the original FP image. The GL profile is calculated for the elliptical polygon along direction 804, for example, by averaging the GL values ​​within the polygon along the parallel chords. The resulting GL profile is shown in chart 806 (after ignoring the uninformative chords (i.e., the masked portion)).

[0118] Once the GL profiles are calculated for all equivalent polygons from a given family (such as shown in graph 808, where the GL profiles for three polygons are obtained), a baseline GL profile can be calculated based on the GL profiles of the equivalent polygons. For example, as shown in graph 810, the baseline GL profile can be obtained by averaging the GL profiles of the three polygons. In some cases, the GL profiles can be independently normalized in terms of size (i.e., number of points) and intensity (i.e., grayscale) before averaging.

[0119] Now refer to Figure 9 , showing several examples of profile attributes in accordance with some embodiments of the presently disclosed subject matter.

[0120] Once the baseline GL profile is generated, profile properties can be calculated for each design-based structural element. For example, the profile property can be calculated as the difference between its grayscale profile and the baseline grayscale profile. For example, the difference can be the L2 distance between the GL profile and the baseline GL profile. Continuing with the baseline GL profile example above, Figure 9 In FIG. 3 , three polygons are shown, for which corresponding contour attributes are calculated. In 902 (where the polygons are effectively registered), the GL contours calculated for the polygons are compared with the GL contours calculated for the polygons. Figure 8 The baseline GL profiles obtained as described in [ 1 ] are shown together in graph 903. As can be seen, the difference between the two curves is relatively small, as evidenced by the profile attribute calculated as 0.13. In 904 (where the polygon is misregistered between the image and the CAD), as shown in graph 905, the calculated GL profile is far from the baseline GL profile, and the profile attribute indicating the difference is relatively large (e.g., 3.46). Similarly, in 906 (where the FP image is a bare image with random noise), as shown in graph 907, the calculated GL profile is a random curve, and the profile attribute is calculated as 3.80, which is also relatively large.

[0121] It should be noted that in some cases, multiple grayscale profiles can be calculated for each design-based structural element along multiple specific directions. The multiple specific directions can be defined by the user. Therefore, multiple baseline GL profiles can be calculated for each polygon family along the multiple specific directions. The multiple baseline GL profiles are included as part of the inspection recipe.

[0122] continue Figure 2As illustrated, once the property(ies) of the design-based structural element are obtained (e.g., edge properties and / or one or more contour properties), a classifier (212) can be trained (e.g., by the training module 105) using the label of each design-based structural element associated with a given layer (as obtained with reference to block 208) and the property(ies) associated with the design-based structural element to determine the validity of the registration between the first image and the second image at the location of the design-based structural element.

[0123] The term "classifier" or "classification model" as used herein should be interpreted broadly to encompass any learning model that can identify which of a set of types / categories a new instance belongs to based on a training data set. Classifiers can be implemented as various types of machine learning models, such as, for example, linear classifiers, support vector machines (SVMs), neural networks, decision trees, and the like. In some cases, the learning algorithm used by the classifier can be supervised learning. In such cases, the training set can contain training instances having labels indicating their types. Figure 2 An example of such a training process is shown. In some other cases, the classifier may use an unsupervised or semi-supervised learning algorithm. The presently disclosed subject matter is not limited to the particular type of classifier and / or learning algorithm for the classifier implemented therewith.

[0124] Each design-based structural element can be represented by its associated attribute(s), such as, for example, an edge attribute and / or one or more contour attributes as described above. The design-based structural element can be projected into an attribute space characterized by the attribute(s). A classifier can be trained based on the attributes of the design-based structural elements in a training set and the labels associated therewith. For example, the classifier can be implemented as a binary classifier (such as, for example, an SVM model), which can be used to determine a decision boundary in the attribute space, thereby separating effectively registered structural elements from ineffectively registered structural elements.

[0125] Now refer to Figure 10 , which shows a schematic diagram of training a classifier using attributes and labels of design-based structural elements according to certain embodiments of the presently disclosed subject matter.

[0126] For each structural element based on the design, it is assumed that two attributes are associated with it: edge attribute and contour attribute. The attribute space is generated and placed in Figure 10is shown as a coordinate system with edge attributes as the X-axis and contour attributes as the Y-axis. Each point in the attribute space represents a design-based structural element characterized by its specific attributes. As illustrated, the points marked by circles represent effectively registered structural elements, while the remaining points represent invalidly registered structural elements, including points marked by pentagons representing invalid situations caused by bare images and points marked by triangles representing invalid situations caused by incorrect registration. The attributes of the design-based structural elements can be fed into the SVM model, and the predictions of the SVM model can be compared with the ground truth (e.g., the labels of the design-based structural elements) to iteratively adjust and modify the parameters of the SVM model until the trained model is able to determine a decision boundary 1002 in the attribute space, which can separate the effectively registered structural elements (e.g., the points marked by circles) from the invalidly registered structural elements (e.g., the remaining points) that meet predefined criteria.

[0127] It should be noted that the classifiers described above are trained based on the design-based structural elements in a given layer. Therefore, the classifiers are also referred to as per-layer classifiers. A per-layer classifier for each layer is included as part of the inspection recipe.

[0128] Because the classifier is trained to classify at the polygon level and a validity decision must be made for each layer, a layer score and layer threshold are calculated. Specifically, the trained classifier can be used (214) to determine the category of each design-based structural element associated with a given layer for each image pair, the determination being made based on attributes associated with the design-based structural element. A layer score for a given layer can be generated based on the determined category of each design-based structural element. A layer threshold (216) can be generated based on the layer score for a given layer for each image pair of the training set (e.g., by the recipe generator 107). The layer threshold for each layer can be included as part of the inspection recipe.

[0129] Now refer to Figure 11 , showing a schematic diagram of generating layer thresholds according to certain embodiments of the presently disclosed subject matter.

[0130] In this example, the trained classifier for a given layer can be applied to the polygons of the given layer for the ten first image pairs in the first subset (i.e., effectively registered image pairs) and the corresponding artificial image pairs generated in the second subset (i.e., invalidly registered image pairs) (assuming that one or more artificial image pairs are generated for each first image pair). For example, the layer score (L) of layer i can be calculated as the percentage of polygons that pass (i.e., are classified as effectively registered) out of the total number of polygons in the given layer: Li = N 通过 / N多边形

[0131] A layer score is generated for a given layer in each of the image pairs. The layer scores of ten effectively registered image pairs should be relatively high and close to 1, as given by Figure 11 On the other hand, the layer scores of invalidly registered image pairs should be relatively low, as indicated by Figure 11 The values ​​are represented by the squares in the figure, and range from 0% to 20%.

[0132] A layer threshold can be generated based on the layer score of the image pair. For example, the layer threshold can be determined so as to maintain the accuracy of the classification. In this example, the threshold is determined to be slightly larger than the layer score of invalid image pairs so as not to miss any validly registered image pairs.

[0133] The trained classifier and layer threshold associated with a given layer may be included in the inspection recipe (eg, by the recipe generator 107 ) ( 218 ).

[0134] According to some embodiments, the second image (e.g., CAD simulation image) in each image pair provides information of one or more design-based structural elements associated with the plurality of layers. Thus, polygon labeling, attribute calculation, classifier training, generation of layer scores and layer thresholds, and recipe generation are performed for each of the plurality of layers, thereby generating a plurality of inspection recipes corresponding to the plurality of layers. As will be described below with reference to Figure 3 As described, the inspection recipe(s) thus generated can be used to inspect samples.

[0135] Now refer to Figure 3 , shows a generalized flow chart for inspecting semiconductor samples using inspection recipes in accordance with certain embodiments of the presently disclosed subject matter.

[0136] A registered image pair including a first image and a second image can be obtained at runtime. According to some embodiments, the first image (302) can be obtained, for example, by being captured by an inspection tool at runtime. The first image can represent at least a portion of a semiconductor sample. A second image (304) can be generated based on design data representing the at least a portion of the sample. The second image can provide information about one or more design-based structural elements presented in the design data and corresponding layers associated with each design-based structural element.

[0137] As mentioned above Figure 2As described, the first image may be an FP image captured by an inspection tool. For example, the first image may be a high-resolution image captured by a review tool, such as, for example, a SEM image captured by a scanning electron microscope (SEM) machine. The second image may be a simulated image generated by performing a simulation on the design data, such as, for example, a CAD simulated image. The second image may include one or more layers corresponding to design layers in the design data. Each layer may be viewed as a binary image in which design-based structural elements are separated from the background. Each design-based structural element may be associated with a corresponding layer in which the element is located. Figure 4 The first image and the second image are registered in a manner similar to that described (306).

[0138] It should be noted that although blocks 302, 304, and 306 are shown as Figure 3 , but in some embodiments may also be external to system 131 and in Figure 3 The functionality of these processes (or at least a portion thereof) is performed prior to the presently proposed inspection method described in . In some other embodiments, such functionality may be performed by system 131 and thus may be considered part of the present inspection process.

[0139] For each design-based structural element associated with a given layer, an edge attribute associated with the design-based structural element may be calculated (e.g., by the attribute generator 134) (308). As described above, the edge attribute may indicate the estimated presence of an edge of the image-based structural element in the first image at the location of the edge of the design-based structural element in the second image. For example, the edge attribute may be calculated by applying a statistical test between two groups of pixels from either side of an edge of the design-based structural element in the first image, and determining a separation between the two groups of pixels based on the result of the statistical test. Figure 2 and Figure 7 Further details of edge attribute calculations and their use to determine edge presence are described.

[0140] In some embodiments, one or more grayscale profiles may additionally be calculated along one or more specific directions at locations in the first image corresponding to locations of the design-based structural elements in the second image. One or more profile attributes associated with the design-based structural elements may be calculated. Each profile attribute indicates the difference between its corresponding grayscale profile and a corresponding baseline grayscale profile included in the inspection recipe. As described above with reference to Figure 2 and Figure 8 as well as Figure 9As described, during the training phase, a respective baseline grayscale profile is calculated along a respective specific direction for the design-based structural element family to which the design-based structural element belongs.

[0141] The trained classifier trained during the training phase and included in the inspection recipe can be used (e.g., by the validity determination module 138) to determine a category of the design-based structural element based on the attribute(s) associated with the design-based structural element (e.g., edge attributes and / or one or more contour attributes) (310). The category can indicate the validity of the registration between the first image and the second image at the location of the design-based structural element (also referred to as the registration validity of the design-based structural element).

[0142] As mentioned above, the term classifier as used herein should be broadly interpreted to encompass any learning model that can identify which of a set of types / categories a new instance belongs to based on a training data set. Classifiers can be implemented as various types of machine learning models and can utilize different kinds of learning algorithms. In some embodiments, the trained classifier used in block 310 can be a supervised learning model, such as, for example, Figure 2 In some other embodiments, the trained classifier used in block 310 may also be an unsupervised or semi-supervised learning model, such as, for example, a clustering model, an anomaly detection model, etc. The presently disclosed subject matter is not limited to a particular type of classifier and / or learning algorithm implemented therewith.

[0143] A layer score (312) can be generated for a given layer based on the classification of each design-based structural element associated with the given layer (e.g., by the validity determination module 138). For example, the layer score can be calculated as the percentage of design-based structural elements that are classified as validly registered. The layer score can be used to determine the validity of the registered image pair based on a layer threshold predetermined in the inspection recipe. For example, in the case of a sample having only one layer, the validity of the registered image pair can be determined based on the layer score relative to the layer threshold.

[0144] As another example, where the specimen has multiple layers and the second image provides information about one or more design-based structural elements associated with the multiple layers, calculation of attribute(s), classification using a trained classifier, and generation of layer scores are performed for each of the multiple layers, thereby generating multiple layer scores. The validity of the registered image pair can be determined based on the multiple layer scores and multiple layer thresholds predetermined in the inspection recipe. For example, one determination criterion can be that if any of the layers is not validly registered (i.e., if the layer score of any layer does not meet the criteria regarding the layer threshold), the validity of the registered image pair can be rejected. In other words, if any layer score is less than its layer threshold, the site represented by the image pair is rejected as a whole. On the other hand, if all layers are validly registered (i.e., if the layer scores of all layers meet the criteria regarding the corresponding layer threshold), the validity of the registered image pair can be confirmed. The layer results and / or site results can be stored in a storage unit and / or presented on a GUI for user review.

[0145] Now refer to Figure 12 , schematically illustrating an example of using a check recipe to perform runtime checking on a sample according to some embodiments of the presently disclosed subject matter.

[0146] Assume that the sample has multiple layers L as shown 1 、L 2 ,…,L l .and Figure 10 Similar to the example of , two attributes (edge ​​attribute and contour attribute) are obtained for each design-based structural element in each layer, thereby generating an attribute space for each layer (shown in Figure 12 , with edge attributes as the X-axis and contour attributes as the Y-axis). For exemplary purposes, two attribute spaces 1202 and 1204 corresponding to two layers are shown. A point in the attribute space represents a corresponding design-based structural element in a given layer, which is characterized by its specific attributes.

[0147] The properties of the design-based structural elements in a given layer can be fed into a classifier trained for the given layer. The trained classifier can determine the category of each design-based structural element based on the properties associated with the design-based structural element. The category is a binary category indicating the registration effectiveness of the design-based structural element. Decision boundaries 1206 and 1208 represent the separation between effectively registered structural elements and ineffectively registered structural elements.

[0148] Correspondingly for each layer L 1 、L 2 ,…,L lA layer score is calculated and a decision can be made for each layer based on the layer score and a layer threshold predetermined in the inspection recipe. The validity of the registered image pair can be determined based on the decision for the corresponding layer.

[0149] It should be noted that the examples shown are described herein for illustrative purposes and should not be considered to limit the present disclosure in any way. Other suitable examples may be used in addition to or in place of the above.

[0150] One of the advantages of certain embodiments of the inspection process as described herein is the ability to automatically detect mismatches / differences between FP images and corresponding design data before performing further inspection operations (such as, for example, segmentation and metrology measurements, etc.), thereby allowing trustworthy inspection results (e.g., measurement results such as overlay, CD, etc.) to be provided to the user and avoiding providing misleading inspection information to the manufacturing process.

[0151] One of the advantages of certain embodiments of the recipe generation process as described herein is providing an inspection recipe that can be used to automatically detect mismatches / differences between FP images and corresponding design data before performing further inspection operations as described above.

[0152] One advantage of certain embodiments of the inspection process described herein is an automated inspection process for detecting discrepancies between FP images and corresponding design data, which can save a user the effort of manually inspecting all sites on a wafer before inspection or manually inspecting the inspection results of the entire wafer after inspection to look for outliers. This is achieved by at least the following steps: using a trained classifier to perform classification on a structural element based on one or more attributes of the structural element, and generating an indication of the site based on the classification results.

[0153] One of the advantages of certain embodiments of the recipe generation process as described herein is generating a training set that includes the steps of generating an artificial / synthetic erroneous training subset by modifying a valid / good training subset so that properly labeled training data is provided.

[0154] The inspection process as presently disclosed herein may be used for different applications such as, for example, ADC, ADR, defect detection, matching, metrology, and other inspection tasks.

[0155] It is to be understood that the disclosure is not limited in its application to the details set forth in the description contained herein or shown in the drawings.

[0156] It will also be understood that a system according to the present disclosure can be implemented at least in part on a suitably programmed computer. Likewise, the present disclosure contemplates a computer program readable by a computer for performing the method of the present disclosure. The present disclosure further contemplates a non-transitory computer-readable memory tangibly embodying a computer-executable program of instructions for performing the method of the present disclosure.

[0157] The present disclosure is capable of other embodiments and can be practiced and implemented in various ways. Therefore, it is to be understood that the words and terms used herein are for illustrative purposes and should not be construed as limiting. Therefore, those skilled in the art will appreciate that the concepts upon which the present disclosure is based can be readily used as a basis for designing other structures, methods, and systems for achieving the several purposes of the presently disclosed subject matter.

[0158] It will be readily appreciated by those skilled in the art that various modifications and changes may be applied to the embodiments of the present disclosure as described above without departing from the scope of the present disclosure as defined in and by the appended claims.

Claims

1. A computerized system for inspecting semiconductor samples using an inspection recipe, the system comprising a processor and memory circuitry (PMC) configured to: obtaining a registered image pair, the registered image pair comprising a first image and a second image, the first image being captured by an inspection tool and representing at least a portion of the semiconductor sample, the second image being generated based on design data characterizing the at least a portion of the sample, the second image comprising one or more layers corresponding to design layers in the design data, the second image providing information of one or more design-based structural elements present in the design data, each design-based structural element being associated with a respective layer in which the design-based structural element is located; For each design-based structural element associated with a given layer of the one or more layers: calculating an edge attribute associated with the design-based structural element and indicating an estimated presence of an edge of the image-based structural element in the first image at a location of an edge of the design-based structural element in the second image; as well as determining, using a trained classifier included in the inspection recipe, a class of the design-based structural element based on the edge properties associated with the design-based structural element, the class indicating validity of registration between the first image and the second image at the location of the design-based structural element; as well as A layer score is generated for the given layer based on the category of each design-based structural element associated with the given layer, the score being usable to determine the validity of the registered image pair based on a layer threshold predetermined in the inspection recipe.

2. The computerized system of claim 1, wherein the first image is a high-resolution image captured by a review tool.

3. The computerized system of claim 1, wherein the second image is generated by performing a simulation on the design data.

4. The computerized system of claim 1 , wherein the PMC is further configured to: for each design-based structural element associated with the given layer: calculating one or more grayscale profiles along one or more specific directions at locations in the first image corresponding to locations of the design-based structural element in the second image; as well as calculating one or more profile attributes associated with the design-based structural element, each profile attribute indicating a difference between a corresponding grayscale profile of the design-based structural element and a corresponding baseline grayscale profile included in the inspection recipe, the corresponding baseline grayscale profile being calculated for a family of design-based structural elements to which the design-based structural element belongs along a corresponding specific direction; Wherein the step of determining a category using a trained classifier is performed based on the edge attribute and the one or more contour attributes associated with the design-based structural element.

5. The computerized system of claim 1 , wherein the PMC is configured to calculate the edge property by applying a statistical test between two groups of pixels on either side of the edge of the design-based structural element from the first image, and determining a separation between the two groups of pixels based on a result of the statistical test.

6. The computerized system of claim 1, wherein the layer score is a percentage of the design-based structural elements that are classified as validly registered.

7. A computerized system as described in claim 1, wherein the second image provides information of one or more design-based structural elements associated with a plurality of layers, and the calculating step, the using step, and the generating step are performed for each of the plurality of layers to generate a plurality of layer scores, and wherein the PMC is further configured to determine the validity of the registered image pair based on the plurality of layer scores and a plurality of layer thresholds predetermined in the inspection recipe.

8. A computerized method for inspecting a semiconductor sample using an inspection recipe, the method being executed by a processor and memory circuitry (PMC), the method comprising the steps of: obtaining a registered image pair, the registered image pair comprising a first image and a second image, the first image being captured by an inspection tool and representing at least a portion of the semiconductor sample, the second image being generated based on design data characterizing the at least a portion of the sample, the second image comprising one or more layers corresponding to design layers in the design data, the second image providing information of one or more design-based structural elements present in the design data, each design-based structural element being associated with a respective layer in which the design-based structural element is located; For each design-based structural element associated with a given layer of the one or more layers: calculating an edge attribute associated with the design-based structural element and indicating an estimated presence of an edge of the image-based structural element in the first image at a location of an edge of the design-based structural element in the second image; as well as determining, using a trained classifier included in the inspection recipe, a class of the design-based structural element based on the edge properties associated with the design-based structural element, the class indicating validity of registration between the first image and the second image at the location of the design-based structural element; as well as A layer score is generated for the given layer based on the category of each design-based structural element associated with the given layer, the score being usable to determine the validity of the registered image pair based on a layer threshold predetermined in the inspection recipe.

9. The computerized method of claim 8, wherein the first image is a high-resolution image captured by a review tool.

10. The computerized method of claim 8, wherein the second image is generated by performing a simulation on the design data.

11. The computerized method of claim 8, further comprising the steps of: For each design-based structural element associated with the given layer: calculating one or more grayscale profiles along one or more specific directions at locations in the first image corresponding to locations of the design-based structural element in the second image; as well as calculating one or more profile attributes associated with the design-based structural element, each profile attribute indicating a difference between a corresponding grayscale profile of the design-based structural element and a corresponding baseline grayscale profile included in the inspection recipe, the corresponding baseline grayscale profile being calculated for a family of design-based structural elements to which the design-based structural element belongs along a corresponding specific direction; Wherein the step of determining a category using a trained classifier is performed based on the edge attribute and the one or more contour attributes associated with the design-based structural element.

12. The computerized method of claim 8, wherein the edge attribute is calculated by applying a statistical test between two groups of pixels on either side of the edge of the design-based structural element from the first image, and determining a separation between the two groups of pixels based on a result of the statistical test.

13. The computerized method of claim 8, wherein the layer score is a percentage of the design-based structural elements that are classified as validly registered.

14. The computerized method of claim 8, wherein the second image provides information of one or more design-based structural elements associated with a plurality of plies, and the calculating step, the using step, and the generating step are performed for each of the plurality of plies to generate a plurality of ply scores, and the method further comprises the steps of: The validity of the registered image pair is determined based on the plurality of layer scores and a plurality of layer thresholds predetermined in the inspection recipe.

15. A non-transitory computer-readable storage medium tangibly embodying a program of instructions, which, when executed by a computer, causes the computer to perform a method of inspecting a semiconductor sample using an inspection recipe, the method comprising the steps of: obtaining a registered image pair, the registered image pair comprising a first image and a second image, the first image being captured by an inspection tool and representing at least a portion of the semiconductor sample, the second image being generated based on design data characterizing the at least a portion of the sample, the second image comprising one or more layers corresponding to design layers in the design data, the second image providing information of one or more design-based structural elements present in the design data, each design-based structural element being associated with a respective layer in which the design-based structural element is located; For each design-based structural element associated with a given layer of the one or more layers: calculating an edge attribute associated with the design-based structural element and indicating an estimated presence of an edge of the image-based structural element in the first image at a location of an edge of the design-based structural element in the second image; as well as determining, using a trained classifier included in the inspection recipe, a class of the design-based structural element based on the edge properties associated with the design-based structural element, the class indicating validity of registration between the first image and the second image at the location of the design-based structural element; as well as A layer score is generated for the given layer based on the category of each design-based structural element associated with the given layer, the score being usable to determine the validity of the registered image pair based on a layer threshold predetermined in the inspection recipe.

16. A computerized system for generating an inspection recipe that can be used to inspect a semiconductor sample, the system comprising a processor and memory circuitry (PMC), the processor and memory circuitry (PMC) configured to: A training set is obtained, wherein the training set includes: i) a first subset, the first subset comprising one or more first image pairs, each first image pair comprising a first image and a second image, the first image being captured by an inspection tool and representing at least a portion of a sample, the second image being generated based on design data characterizing the at least a portion of the sample, the first image and the second image being effectively registered, the second image comprising one or more layers corresponding to design layers in the design data, the second image providing information of one or more design-based structural elements present in the design data, each design-based structural element being associated with a respective layer in which the design-based structural element is located; and ii) a second subset comprising one or more second image pairs, each second image pair generated by modifying at least one of the first image and the second image in a corresponding first image pair in the first subset such that the generated second image pair is invalidly registered; associating each of the one or more design-based structural elements in the second image of each image pair of the training set with a label indicating registration validity; calculating, for each design-based structural element, an edge attribute associated with the design-based structural element and indicating an estimated presence of an edge of the image-based structural element in the first image at a location of the edge of the design-based structural element in the second image; training a classifier using the edge attributes associated with each design-based structural element associated with a given layer of the one or more layers and the label of the design-based structural element to determine validity of registration between the first image and the second image at the location of the design-based structural element; determining a category of each design-based structural element associated with the given layer of each image pair using the trained classifier, the determination being based on the edge attributes associated with the design-based structural element, and generating a layer score for the given layer based on the determined category of each design-based structural element; generating a layer threshold based on the layer score for the given layer for each image pair of the training set; as well as The trained classifier and the layer threshold associated with the given layer are included in the inspection recipe.

17. The computerized system of claim 16, wherein at least one second image pair is generated by modifying content of the first image in the first image pair.

18. The computerized system of claim 16, wherein at least one second image pair is generated by modifying a relative position between the first image and the second image in a first image pair.

19. The computerized system of claim 16, wherein the PMC is configured to perform the associating step based on at least one of: the registration validity of the corresponding image pair, and the modification if the corresponding image pair is a second image pair.

20. The computerized system of claim 16, wherein the PMC is further configured to: calculating a grayscale profile of each design-based structural element along a specific direction at a position in the first image corresponding to a position of the design-based structural element in the second image; calculating a baseline grayscale profile of each family of design-based structure elements among the effectively registered design-based structure elements along the specific direction, the calculating being performed based on the grayscale profile of each design-based structure element in the family; as well as for each design-based structural element, calculating a profile attribute, the profile attribute being associated with the design-based structural element and indicating a difference between the grayscale profile of the design-based structural element and the baseline grayscale profile, and wherein the steps of training a classifier and using the trained classifier are performed based on the edge attribute and the profile attribute associated with the design-based structural element, Wherein the inspection recipe further comprises the baseline grayscale profile.

21. The computerized system of claim 20, wherein a plurality of baseline grayscale profiles are calculated for each family along a plurality of specific directions and included in the inspection recipe.

22. The computerized system of claim 16, wherein the second image in each image pair provides information of one or more design-based structural elements associated with a plurality of layers, and the associating step, the calculating step, the training step, the using step, and the including step are performed for each of the plurality of layers to generate a plurality of inspection recipes corresponding to the plurality of layers.

23. A non-transitory computer-readable storage medium tangibly embodying a program of instructions, which, when executed by a computer, causes the computer to perform a method of generating an inspection recipe that can be used to inspect a semiconductor sample, the method comprising the steps of: A training set is obtained, the training set comprising: i) a first subset comprising one or more first image pairs, each first image pair comprising a first image and a second image, the first image being captured by an inspection tool and representing at least a portion of a sample, the second image being generated based on design data characterizing the at least a portion of the sample, the first image and the second image being effectively registered, the second image comprising one or more layers corresponding to design layers in the design data, the second image providing information of one or more design-based structural elements present in the design data, each design-based structural element being associated with a corresponding layer in which the design-based structural element is located; and ii) a second subset comprising one or more second image pairs, each second image pair being generated by modifying at least one of the first image and the second image in a corresponding first image pair in the first subset such that the generated second image pair is not effectively registered; associating each of the one or more design-based structural elements in the second image of each image pair of the training set with a label indicating registration validity; calculating, for each design-based structural element, an edge attribute associated with the design-based structural element and indicating an estimated presence of an edge of the image-based structural element in the first image at a location of the edge of the design-based structural element in the second image; training a classifier using the edge attributes associated with each design-based structural element associated with a given layer of the one or more layers and the label of the design-based structural element to determine validity of registration between the first image and the second image at the location of the design-based structural element; determining a category of each design-based structural element associated with the given layer of each image pair using the trained classifier, the determination being based on the edge attributes associated with the design-based structural element, and generating a layer score for the given layer based on the determined category of each design-based structural element; generating a layer threshold based on the layer score for the given layer for each image pair of the training set; and The trained classifier and the layer threshold associated with the given layer are included in the inspection recipe.

24. The non-transitory computer-readable storage medium of claim 23, wherein at least one second image pair is generated by modifying content of the first image in the first image pair.

25. The non-transitory computer-readable storage medium of claim 23, wherein at least one second image pair is generated by modifying a relative position between the first image and the second image in the first image pair.

26. The non-transitory computer-readable storage medium of claim 23, wherein the associating step is performed based on at least one of: a validity of the registration of a corresponding image pair, and a modification if the corresponding image pair is a second image pair.

27. The non-transitory computer-readable storage medium of claim 23, wherein the method further comprises the steps of: calculating a grayscale profile of each design-based structural element along a specific direction at a position in the first image corresponding to a position of the design-based structural element in the second image; calculating a baseline grayscale profile of each family of design-based structure elements among the effectively registered design-based structure elements along the specific direction, the calculating being performed based on the grayscale profile of each design-based structure element in the family; as well as for each design-based structural element, calculating a profile attribute associated with the design-based structural element and indicating a difference between the grayscale profile of the design-based structural element and the baseline grayscale profile, and wherein the steps of training a classifier and using the trained classifier are performed based on the edge attribute and the profile attribute associated with the design-based structural element, and Wherein the inspection recipe further comprises the baseline grayscale profile.

28. The non-transitory computer-readable storage medium of claim 27, wherein a plurality of baseline grayscale profiles are calculated for each family along a plurality of specific directions and included in the inspection recipe.

29. The non-transitory computer-readable storage medium of claim 23, wherein the second image in each image pair provides information of one or more design-based structural elements associated with a plurality of layers, and the associating step, the calculating step, the training step, the using step, and the including step are performed for each of the plurality of layers to generate a plurality of inspection recipes corresponding to the plurality of layers.

Images