Context-based defect inspection

Abstract

A context-based inspection system is disclosed. The system may include an optical imaging subsystem. The system may further include one or more controllers communicatively coupled to the optical imaging system. The one or more controllers may be configured to: receive one or more reference images; receive one or more test images of a sample; generate one or more probabilistic context maps using an unsupervised classifier during an inspection run; provide the generated probabilistic context maps to a supervised classifier during the inspection run; and apply the supervised classifier to the received one or more test images to identify one or more DOIs on the sample.

Description

[0001] Cross-reference of related applications

[0002] This application claims the benefit of U.S. Provisional Application No. 63 / 326,268, filed March 31, 2022, and U.S. Provisional Application No. 63 / 443,050, filed February 3, 2023, pursuant to 35 U.S.SC §119(e), the entire contents of which are incorporated herein by reference. Technical Field

[0003] This disclosure generally relates to defect inspection, and more specifically, to context-based defect inspection. Background Technology

[0004] Inspection systems are commonly used in semiconductor manufacturing processes to identify manufacturing defects that can lead to performance degradation or failure of the manufactured equipment. Increasingly complex structures result in a greater number of parameters that must be monitored and controlled to maintain equipment integrity. Conventional inspection techniques require considerable expertise to operate effectively. Therefore, as the number of parameters increases, the quality of inspection kits largely depends on the operator's skill and experience in setting up the kits.

[0005] Therefore, there is a need to develop systems and methods to address the aforementioned deficiencies. Summary of the Invention

[0006] According to one or more embodiments of this disclosure, a context-based inspection system is disclosed. In one embodiment, the system includes one or more controllers, each controller including one or more processors for executing program instructions stored in memory. In one embodiment, the one or more program instructions are configured to cause the one or more processors to receive one or more reference images. In one embodiment, the one or more program instructions are configured to cause the one or more processors to receive one or more test images of samples, the one or more test images containing one or more defects of interest (DOIs). In one embodiment, the one or more program instructions are configured to cause the one or more processors to generate one or more probabilistic context maps using an unsupervised classifier during an inspection run. In one embodiment, the one or more program instructions are configured to cause the one or more processors to provide the generated one or more probabilistic context maps to a supervised classifier during the inspection run. In one embodiment, the one or more program instructions are configured to cause the one or more processors to apply the supervised classifier to the received one or more test images to identify the one or more DOIs on the samples.

[0007] According to one or more embodiments of this disclosure, a system is disclosed. In one embodiment, the system includes an optical imaging subsystem. In another embodiment, the system includes one or more controllers communicatively coupled to the optical imaging subsystem. In another embodiment, the one or more controllers include one or more processors for executing program instructions stored in a memory. In another embodiment, the one or more program instructions are configured to cause the one or more processors to receive one or more reference images. In another embodiment, the one or more program instructions are configured to cause the one or more processors to receive one or more test images of samples, the one or more test images containing one or more defects of interest (DOIs). In another embodiment, the one or more program instructions are configured to cause the one or more processors to generate one or more probabilistic context maps using an unsupervised classifier during an inspection run. In another embodiment, the one or more program instructions are configured to cause the one or more processors to provide the generated one or more probabilistic context maps to a supervised classifier during the inspection run. In an embodiment, the one or more program instructions are configured to cause the one or more processors to apply the supervised classifier to the received one or more test images to identify the one or more DOIs on the samples.

[0008] According to one or more embodiments of this disclosure, a context-based verification system is disclosed. In one embodiment, the system includes one or more controllers, each controller including one or more processors for executing program instructions stored in memory. In one embodiment, the one or more program instructions are configured to cause the one or more processors to receive empirical data obtained from samples from a characterization subsystem. In one embodiment, the one or more program instructions are configured to cause the one or more processors to receive design decomposition data from a design decomposition database. In one embodiment, the one or more program instructions are configured to cause the one or more processors to generate one or more probabilistic context maps based on the received empirical data and the received data from the design database prior to verification runtime. In one embodiment, the one or more program instructions are configured to cause the one or more processors to store the generated one or more probabilistic context maps in the memory. In one embodiment, the one or more program instructions are configured to cause the one or more processors to receive one or more reference images. In one embodiment, the one or more program instructions are configured to cause the one or more processors to receive one or more test images of the sample, the one or more test images containing one or more defects of interest (DOIs). In one embodiment, the one or more program instructions are configured to cause the one or more processors to retrieve the stored one or more probabilistic context maps from the memory. In one embodiment, the one or more program instructions are configured to cause the one or more processors to provide the one or more probabilistic context maps to a supervised classifier. In one embodiment, the one or more program instructions are configured to cause the one or more processors to apply the supervised classifier to the received one or more test images to identify the one or more DOIs on the sample.

[0009] According to one or more embodiments of this disclosure, a system is disclosed. In one embodiment, the system includes an optical imaging subsystem. In one embodiment, the system includes one or more controllers communicatively coupled to the optical imaging subsystem. In one embodiment, the one or more controllers include one or more processors for executing program instructions stored in a memory. In one embodiment, the one or more program instructions are configured to cause the one or more processors to receive empirical data acquired from samples from a characterization subsystem. In one embodiment, the one or more program instructions are configured to cause the one or more processors to receive design decomposition data from a design decomposition database. In one embodiment, the one or more program instructions are configured to cause the one or more processors to generate one or more probabilistic context maps based on the received empirical data and the received data from the design database before verifying runtime. In one embodiment, the one or more program instructions are configured to cause the one or more processors to store the generated one or more probabilistic context maps in the memory. In one embodiment, the one or more program instructions are configured to cause the one or more processors to receive one or more reference images. In one embodiment, the one or more program instructions are configured to cause the one or more processors to receive one or more test images of the sample, the one or more test images containing one or more defects of interest (DOIs). In one embodiment, the one or more program instructions are configured to cause the one or more processors to retrieve the stored one or more probabilistic context maps from the memory. In one embodiment, the one or more program instructions are configured to cause the one or more processors to provide the one or more probabilistic context maps to a supervised classifier. In one embodiment, the one or more program instructions are configured to cause the one or more processors to apply the trained supervised classifier to the received one or more test images to identify the one or more DOIs on the sample.

[0010] According to one or more embodiments of this disclosure, a context-based inspection system is disclosed. In one embodiment, the system includes one or more controllers, each controller including one or more processors for executing program instructions stored in memory. In one embodiment, the one or more program instructions are configured to cause the one or more processors to receive empirical data from a characterization subsystem. In one embodiment, the one or more program instructions are configured to cause the one or more processors to generate one or more context maps based on the received empirical data prior to inspection runtime. In one embodiment, the one or more program instructions are configured to cause the one or more processors to train a supervised classifier based on the generated one or more context maps and a set of labeled training images. In one embodiment, the one or more program instructions are configured to cause the one or more processors to receive one or more reference images. In one embodiment, the one or more program instructions are configured to cause the one or more processors to receive one or more test images of samples, the one or more test images containing one or more defects of interest (DOIs). In an embodiment, the one or more program instructions are configured to cause the one or more processors to apply the trained supervised classifier to the received one or more test images to identify the one or more DOIs on the samples.

[0011] According to one or more embodiments of this disclosure, a system is disclosed. In one embodiment, the system includes an optical imaging subsystem. In another embodiment, the system includes one or more controllers communicatively coupled to the optical imaging subsystem. In another embodiment, the one or more controllers include one or more processors for executing program instructions stored in a memory. In another embodiment, the one or more program instructions are configured to cause the one or more processors to receive empirical data from a characterization subsystem. In another embodiment, the one or more program instructions are configured to cause the one or more processors to generate one or more context maps based on the received empirical data prior to a test runtime. In another embodiment, the one or more program instructions are configured to cause the one or more processors to train a supervised classifier based on the generated one or more context maps and a set of labeled training images. In another embodiment, the one or more program instructions are configured to cause the one or more processors to receive one or more reference images. In another embodiment, the one or more program instructions are configured to cause the one or more processors to receive one or more test images of samples, the one or more test images containing one or more defects of interest (DOIs). In an embodiment, the one or more program instructions are configured to cause the one or more processors to apply the trained supervised classifier to the received one or more test images to identify the one or more DOIs on the samples.

[0012] It should be understood that the foregoing overview and the following detailed description are merely illustrative and explanatory and do not necessarily limit the invention as claimed. The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and, together with the overview, serve to explain the principles of the invention. Attached Figure Description

[0013] Those skilled in the art will better understand the many advantages of this disclosure by referring to the accompanying drawings.

[0014] Figure 1A A simplified block diagram illustrating a system for context-based verification according to one or more embodiments of the present disclosure.

[0015] Figure 1B A simplified schematic diagram illustrating an image-based inspection system according to one or more embodiments of the present disclosure.

[0016] Figure 2 The illustration depicts a flowchart of a method for performing context-based tests using an in-situ generated probability context graph, according to one or more embodiments of the present disclosure.

[0017] Figure 3 The illustration describes a process flowchart of a method for performing context-based testing using an in-situ generated probability context graph according to one or more embodiments of the present disclosure.

[0018] Figure 4 The illustration depicts a flowchart of a method for performing context-based testing using a non-in-situ generated probability context graph, according to one or more embodiments of the present disclosure.

[0019] Figure 5 The illustration describes a process flowchart of a method for performing context-based testing using a probabilistic context graph generated in non-in-situ, according to one or more embodiments of the present disclosure.

[0020] Figure 6 The illustration depicts a flowchart of a method for performing context-based verification using a generated training context graph, according to one or more embodiments of the present disclosure.

[0021] Figure 7 The illustration depicts a process flowchart of a method for performing context-based verification using a generated training context graph, according to one or more embodiments of the present disclosure. Detailed Implementation

[0022] Reference will now be made in detail to the disclosed subject matter illustrated in the accompanying drawings. This disclosure has been particularly shown and described with respect to certain embodiments and their specific features. The embodiments set forth herein should be considered illustrative rather than restrictive. It will be readily apparent to those skilled in the art that various changes and modifications in form and detail may be made without departing from the spirit and scope of this disclosure.

[0023] Embodiments of this disclosure relate to systems and methods for context-based defect inspection.

[0024] Semiconductor devices are typically manufactured using a series of process steps, such as, but not limited to, depositing process layers and / or photoresist, exposing the photoresist to a desired pattern, etching the photoresist and / or underlying process layers, polishing, or the like. Variations in any of these processes can, individually or in combination, result in variations in the characteristics of the manufactured sample. In a general sense, some sample variations may be acceptable and considered nuisances, while others may lead to unacceptable performance degradation or device failure and are considered defects of concern (DOI).

[0025] Inspection tools can be used after selected process steps to inspect samples for defects. For example, an inspection tool can generate an inspection image of a sample after a specific process step. This inspection image is then typically compared to a reference image to identify defects, where defects are correlated with differences between the inspection and reference images. For example, a difference image can be generated by subtracting the reference image from the inspection image, where features in the difference image correspond to sample variations. These sample variations can then be classified as DOIs or nuisances. Although various types of inspection tools have been developed, optical inspection tools based on illumination using a light source to generate images are commonly used for online inspection due to their relatively high throughput.

[0026] Some embodiments of this disclosure relate to methods for developing inspection formulations for optical inspection based on contextual data. For example, developing an inspection formulation may include, but is not limited to, generating a context graph, as will be discussed further. Inspection recipe generation is substantially discussed in the following patents: U.S. Patent No. 7,877,722, issued January 25, 2011, entitled "Systems and methods for creating inspection recipes"; U.S. Patent No. 8,126,255, issued February 28, 2012, entitled "Systems and methods for creating persistent data for a wafer and for using persistent data for inspection-related functions"; U.S. Patent No. 8,611,639, issued December 17, 2013, entitled "Semiconductor device property extraction, generation, visualization, and monitoring methods"; and U.S. Patent No. 9,262,821, issued February 16, 2016, entitled "Inspection recipe setup from reference image variation," the entire contents of which are incorporated herein by reference.

[0027] More specifically, embodiments of this disclosure relate to systems and methods for generating sample-specific context maps that can be generated using sample-dependent context data (i.e., data related to the sample being tested), wherein the generated sample-specific pixel-level probabilistic context map can be fed into a supervised classifier that serves as the primary detection mechanism for detecting DOIs. In this regard, the sample-specific context map can be added to channels of the primary detection stage (e.g., a supervised classifier) ​​based on data obtained from the sample being tested to improve sensitivity, reduce sample size, and mitigate the risk of missing novel defect types that are not understood by a single detection mechanism (e.g., a deep learning model). Defect detection using machine learning is substantially discussed in the following patents: U.S. Patent No. 11,551,348, issued January 10, 2023, entitled "Learnable defect detection for semiconductor applications"; and U.S. Patent No. 10,346,740, issued July 9, 2019, entitled "Systems and methods incorporating a neural network and a forward physical model for semiconductor applications," the entire contents of which are incorporated herein by reference.

[0028] Some embodiments of this disclosure relate to systems and methods for generating in-situ sample-specific context maps during test run time, wherein the in-situ sample-specific context maps are fed into a supervised classifier. For example, the in-situ sample-specific context maps may be generated using at least an unsupervised classifier.

[0029] Any suitable unsupervised classifier may be used within the spirit and scope of this disclosure. In some embodiments, the unsupervised classifier comprises a neural network (e.g., an unsupervised neural network). For example, the neural network may generate a probability that each pixel can be classified as either background or defect. In this way, the output of the neural network may have substantially the same type as other techniques (e.g., soft clustering based on point clouds, or similar). In some embodiments, the unsupervised classifier comprises soft clustering techniques, such as, but not limited to, Gaussian mixture modeling. For example, a point cloud may be generated using a test image, wherein soft clustering techniques generate a probability that each pixel (or an associated location on the sample of interest) can be classified as a particular category (e.g., background or defect).

[0030] Some embodiments of this disclosure relate to systems and methods for generating in-situ sample-specific context maps prior to test run time, wherein the in-situ sample-specific context maps are stored prior to run time and then fed into a supervised classifier during test run time. For example, the in-situ context maps can be generated prior to test based on a database and empirical data obtained from the samples to be tested.

[0031] Any suitable database may be used within the spirit and scope of this disclosure. In some embodiments, the database includes a geometric pattern center database containing the behavior of product-specific patterns conditioned on process metrology data (e.g., lithography focus, lithography dose, critical dimensions, stacking, and the like). In some embodiments, the database includes a patterning simulation database containing pattern shape probabilities conditioned on process conditions.

[0032] Some embodiments of this disclosure relate to systems and methods for generating context graphs that can be used to train a supervised classifier. For example, the context graph may be generated based on empirical data prior to test run time and used to train a supervised classifier, wherein the supervised classifier may be configured to perform primary detections based on the generated context graph. For example, a sample-specific context graph may be generated based on sample-dependent context data from a metrology tool, testing tool, or process tool. In this regard, the supervised classifier may be trained using the generated context graph until the supervised classifier is sufficiently trained for defect detection of samples not included in the training process.

[0033] Any suitable metrology tool may be used within the spirit and scope of this disclosure. In some embodiments, the metrology tool includes, but is not limited to, optical metrology tools, electron beam metrology tools, X-ray-based metrology tools, spectroscopic metrology tools, or the like.

[0034] Any suitable inspection tool may be used within the spirit and scope of this disclosure. In some embodiments, the inspection tool includes an image-based inspection tool. For example, an image-based inspection tool may include the same inspection tool used to perform context-based inspections.

[0035] Any suitable process tool may be used within the spirit and scope of this disclosure. In some embodiments, the process tool includes a scanner.

[0036] For reference Figures 1A to 7 Systems and methods for context-based verification are described in more detail according to one or more embodiments of the present disclosure.

[0037] Figure 1A This is a block diagram of a context-based verification system 100 according to one or more embodiments of the present disclosure.

[0038] In an embodiment, the inspection system 100 includes one or more optical imaging subsystems 102 (e.g., optical imaging tools) configured to generate one or more images of sample 104, wherein the one or more optical imaging subsystems 102 can be configured to image sample 104. For example, the optical imaging subsystem 102 may include an illumination subsystem 106 configured to illuminate sample 104 using illumination 108 from illumination source 110 and a light-collecting subsystem 112 configured to generate an image of sample 104 using detector 116 in response to light emitted from the sample (e.g., sample light 114).

[0039] Sample 104 may comprise a substrate (e.g., a wafer or the like) formed of a semiconductor or non-semiconductor material. For example, semiconductor or non-semiconductor materials may include, but are not limited to, single-crystal silicon, gallium arsenide, and indium phosphide. The sample may further comprise a mask, lens (e.g., a superlens), photomask, or the like formed of a semiconductor or non-semiconductor material. Sample 104 may further comprise one or more layers disposed on the substrate. For example, such layers may include, but are not limited to, resists, dielectric materials, conductive materials, and semiconducting materials. Many different types of such layers are known in the art, and the term "sample" as used herein is intended to cover samples on which all types of such layers can be formed. The one or more layers formed on the sample may be patterned or unpatterned. For example, the sample may comprise multiple dies, each having repeatable patterned features. The formation and processing of such material layers ultimately produce a finished device. Many different types of devices may be formed on the sample, and the term "sample" as used herein is intended to cover samples on which any type of device known in the art is manufactured.

[0040] The optical imaging subsystem 102 may generate one or more images of the sample 104 using any techniques known in the art. In some embodiments, the optical imaging subsystem 102 is an optical imaging subsystem 102 in which the illumination source 110 is an optical source configured to generate illumination 108 in the form of light, and in which the light-collecting subsystem 112 images the sample 104 based on light emitted from the sample 104. Figure 1B This is a simplified schematic diagram of an optical inspection system 100 according to one or more embodiments of the present disclosure.

[0041] The illumination source 110 may comprise any type of illumination source known in the art suitable for generating optical illumination 108 in the form of one or more illumination beams. Furthermore, the illumination 108 may have any spectrum, such as, but not limited to, extreme ultraviolet (EUV) wavelengths, ultraviolet (UV) wavelengths, visible light wavelengths, or infrared (IR) wavelengths. Additionally, the illumination source 110 may be a broadband source, a narrowband source, and / or a tunable source.

[0042] In an embodiment, illumination source 110 comprises a broadband plasma (BBP) illumination source. In this respect, illumination 108 may comprise radiation emitted by the plasma. For example, BBP illumination source 110 may include, but does not necessarily include, one or more pump sources (e.g., one or more lasers) configured to be focused into a gas volume, causing energy to be absorbed by the gas to generate or sustain a plasma suitable for emitting radiation. Furthermore, at least a portion of the plasma radiation may be used as illumination 108.

[0043] In an embodiment, the illumination source 110 may include one or more lasers. For example, the illumination source 110 may include any laser system known in the art capable of emitting radiation in the infrared, visible, or ultraviolet portions of the electromagnetic spectrum.

[0044] The illumination source 110 may further generate illumination 108 having any time profile. For example, the illumination source 110 may generate continuous wave (CW) illumination 108, pulsed illumination 108, or modulated illumination 108. In addition, illumination 108 may be delivered from the illumination source 110 via free space propagation or guided light (e.g., optical fiber, optical guide, or the like).

[0045] The illumination subsystem 106 and / or the optical imaging subsystem 102 may include various components for guiding the illumination 108 to the sample 104, such as, but not limited to, a lens 118, a mirror, or the like. Furthermore, such components may be reflective or transmissive elements. In this way, Figure 1B The description of lens 118 as a transmissive element is illustrative and not limiting. The illumination subsystem 106 may further include one or more optical elements 120 for modifying and / or adjusting light in the associated optical path, such as, but not limited to, one or more polarizers, one or more filters, one or more beam splitters, one or more diffusers, one or more homogenizers, one or more apodizers, or one or more beam shapers.

[0046] In an embodiment, the inspection system 100 includes a translation stage 122 for fixing and / or positioning the sample 104 during imaging. For example, the translation stage 122 may include any combination of linear actuators, rotary actuators, or angular actuators to position the sample 104 using any number of degrees of freedom.

[0047] The optical imaging subsystem 102 may include various components for collecting at least a portion of the sample light 114 and directing at least a portion of the sample light 114 to the detector 116 for generating an image.

[0048] The inspection system 100 can further image the sample 104 using any techniques known in the art. In some embodiments, the inspection system 100 generates an image of the sample 104 in a scanning mode by: focusing illumination 108 as a spot or line onto the sample 104; capturing a spot or line image; and scanning the sample 104 to establish a two-dimensional image. In this configuration, scanning can be achieved by moving the sample 104 relative to illumination 108 (e.g., using a translation stage 122), by moving illumination 108 relative to the sample 104, or a combination thereof (e.g., using an actuated mirror or the like). In some embodiments, the inspection system 100 generates an image of the sample 104 in a static mode by directing illumination 108 onto the sample 104 in a two-dimensional field of view and directly capturing a two-dimensional image using detector 116.

[0049] The images generated by the inspection system 100 can be any type of image known in the art, such as, but not limited to, bright-field images, dark-field images, phase-contrast images, or similar. Furthermore, the images can be stitched together to form a composite image of sample 104 or a portion thereof.

[0050] Detector 116 may comprise any type of sensor known in the art suitable for measuring sample light 114. For example, detector 116 may comprise a multi-pixel sensor, such as, but not limited to, a charge-coupled device (CCD), a complementary metal-oxide-semiconductor (CMOS) device, a line sensor, or a time-delay integration (TDI) sensor. As another example, detector 116 may comprise two or more single-pixel sensors, such as, but not limited to, a photodiode, a collapsing photodiode, a photomultiplier tube, or a single-photon detector.

[0051] In an embodiment, the inspection system 100 includes a controller 124. The controller 124 may include one or more processors 126 configured to execute program instructions held on a memory 128 (e.g., a memory medium). In this respect, the one or more processors 126 of the controller 124 may perform any of the various process steps described herein.

[0052] Refer again Figure 1AIn an embodiment, the context-based verification system 100 may be communicatively coupled to an offline controller 130. The offline controller 130 may include one or more processors 132 configured to execute program instructions held on memory 134 (e.g., memory media). In this regard, the one or more processors 132 of the controller 130 may perform any of the various process steps described throughout this disclosure. For example, as will be further described herein, the offline controller 130 may be configured to generate one or more context graphs. As another example, the offline controller 130 may be configured to train a supervised classifier stored in memory 134 based on the generated context graphs.

[0053] One or more processors 126, 132 of controllers 124, 130 may comprise any processing element known in the art. In this sense, one or more processors 126, 132 may comprise any microprocessor-type device configured to execute algorithms and / or instructions. In one embodiment, one or more processors 126 may comprise a desktop computer, mainframe computer system, workstation, graphics computer, parallel processor, or any other computer system (e.g., a networked computer) configured to execute a program configured to operate verification system 100, as described throughout this disclosure. It should be further appreciated that the term "processor" may be broadly defined to encompass any device having one or more processing elements that execute program instructions from non-transitory memories 128, 134.

[0054] Memory 128, 134 may comprise any storage medium known in the art suitable for storing program instructions executable by one or more associated processors 126, 132. For example, memory 128, 134 may comprise non-transitory memory media. As another example, memory 128, 134 may comprise, but is not limited to, read-only memory, random access memory, magnetic or optical storage devices (e.g., disks), magnetic tape, solid-state drives, and the like. It should be further noted that memory 128 may be housed together with one or more processors 126 in a common controller housing. In one embodiment, memory 128, 134 may be remotely located relative to the physical location of one or more processors 126, 132 and controller 124, 130. For example, one or more processors 126, 132 of controller 124, 130 may access remote memory (e.g., a server) accessible via a network (e.g., the Internet, intranet, and the like). Therefore, the above description should not be construed as limiting the invention but is merely illustrative.

[0055] Figure 2The illustration depicts a flowchart of a method 200 for performing context-based testing using an in-situ generated probability context graph, according to one or more embodiments of this disclosure. Applicants should note that the embodiments and feasible techniques previously described herein in the context of testing system 100 should be interpreted as extended to method 200. However, it should be further noted that method 200 is not limited to the architecture of testing system 100.

[0056] In step 202, one or more reference images may be received. For example, controller 124 may be configured to receive one or more reference images from optical imaging subsystem 102. As another example, controller 124 may be configured to receive one or more reference images from an additional optical imaging subsystem.

[0057] One or more reference images may represent a paradigm or reference. This reference image can be generated using any suitable technique. For example, a reference image can be generated using a single image of a reference region having a design common to the test region and known or expected to be defect-free. As another example, a reference image can be generated by combining (e.g., averaging or similar) multiple images of multiple reference regions.

[0058] In step 204, one or more test images may be received. For example, controller 124 may be configured to receive one or more test images from optical imaging subsystem 102. The one or more test images may contain various patterned features on the sample. In some embodiments, the one or more test images contain one or more DOIs.

[0059] In step 206, one or more in-situ probability context maps are generated. For example, during the test run, controller 124 may be configured to generate one or more in-situ probability maps.

[0060] Figure 3 This illustration depicts a process flowchart of a method 300 for generating one or more in-situ generated probability context graphs according to one or more embodiments of the present disclosure.

[0061] In step 302, target context data may be generated based on the received test image and provided to an unsupervised classifier. For example, controller 124 may be configured to perform input shifting to generate target context data based on the received test image. For example, controller 124 may be configured to remove a portion from the received test image, wherein the removed portion may correspond to a defect on the test image.

[0062] In optional step 304, the reference image may be provided to the unsupervised classifier.

[0063] In step 306, an unsupervised classifier may be used to generate one or more generated reference images.

[0064] For example, using an unsupervised classifier, controller 124 may be configured to generate one or more generated reference images 307. For the purposes of this disclosure, the “generated reference image” of step 306 is different from the reference image received in step 202, unless otherwise indicated herein.

[0065] In an embodiment, the unsupervised classifier is configured to generate one or more generated reference images (i.e., “TestRef” images) based on target context data (from step 302) and, in some cases, a received reference image (from optional step 304). For example, the unsupervised classifier may output Gaussian mixture model (GMM) parameters for the intensity of the generated reference images (“TestRef” images) to compare with the actual observed pixels 309 from the received test image (“TestActual” image) derived from step 204.

[0066] In step 308, the entropy score and the negative log-likelihood (NLL) score can be calculated. For example, the entropy score can be calculated as the expected negative log-likelihood (NLL) per pixel, which can be calculated by integrating over all possible intensity values ​​of the GMM parameters for a given pixel.

[0067] The negative log-likelihood (NLL) score can be calculated by comparing the generated test reference image (“TestRef”) from an unsupervised classifier with the actual test image (“TestActual”). For example, controller 124 can be configured to calculate the NLL score based on contextual data and the reference image, as shown and described by Equation 1:

[0068]

[0069] in This is the parametric probability density given the following clues: target context and reference image. The expectation-maximization algorithm can then adjust the parameters of the probabilistic model to maximize the NLL of the associated data given the following clues: target context and reference image.

[0070] In step 310, one or more probabilistic context graphs (or fractional graphs) 311 may be generated based on the entropy score and the NLL score. For example, the entropy score from step 308 may be used to normalize the NLL score by subtracting the expected value to generate one or more probabilistic context graphs (or fractional graphs).

[0071] Return to reference Figure 2In step 208, one or more generated probabilistic context maps may be provided to the supervised classifier. For example, controller 124 may be configured to provide one or more generated probabilistic context maps (or score maps) to the supervised classifier during test run time. It should be noted that the context map (or score map) fed to the supervised classifier is a measure of deficiencies. Hints provided by the score map reduce the size of the labeled training data required for authorization by more than two orders of magnitude because the burden of removing the vast majority of normal variations has been implemented.

[0072] Supervised classifiers can contain any type of deep learning classifier or algorithm, including, but not limited to, deep convolutional neural networks (CNNs) or similar. For example, a supervised classifier can contain a supervised discriminative CNN.

[0073] In the context of supervised learning, training images may include images that are false positives or true negatives. Controller 124 may be further configured to store the training images and the supervised classifier in memory 128.

[0074] In step 210, a supervised classifier may be applied to one or more test images to identify one or more DOIs on the sample. For example, using a supervised classifier, controller 124 may be configured to identify one or more DOIs on sample 104.

[0075] Figure 4 The illustration depicts a flowchart of a method 400 for performing context-based testing using a non-in-situ generated probability context graph, according to one or more embodiments of the present disclosure. Applicants should note that the embodiments and feasible techniques previously described herein in the context of testing system 100 should be interpreted as extending to method 400. However, it should be further noted that method 400 is not limited to the architecture of testing system 100.

[0076] In step 402, one or more probability context maps may be generated prior to the test run. For example, prior to the test run, controller 124 may be configured to generate one or more probability context maps based on empirical data obtained from the samples to be tested. In this regard, controller 124 may be configured to store one or more probability context maps in memory prior to testing and to provide the generated probability maps to the supervised classifier during the run (as will be discussed further herein).

[0077] Figure 5 The illustration depicts a process flowchart of a method 500 for generating a probability context graph prior to testing, according to one or more embodiments of the present disclosure.

[0078] For example, one or more probability context diagrams can indicate the parametric probability density given empirical data (y), sample-dependent context data (z), and design data, such as by The expression, where empirical data (y) indicates the probability density given sample context data (z) of a given pattern, is as follows: Express.

[0079] In step 502, data from the sample to be tested may be received. For example, controller 124 may be configured to receive metrological data from a metrological instrument. The metrological instrument may include, but is not limited to, optical metrological instruments, electron beam metrological instruments, X-ray-based metrological instruments, spectroscopic metrological instruments, or the like. The metrological data may include, but is not limited to, critical dimensions, stacking, film thickness, pattern profile, and the like.

[0080] In step 504, data may be received from a design decomposition database. For example, controller 124 may be configured to receive data from a design decomposition database that reflects, by location, the primitive patterns contained in the design of the sample under test, along with the predicted behavior of the primitive patterns varying according to some measured characteristics of the sample. In one example, the design decomposition database may be segmented into a database based on the location of the primitive patterns. In another example, the database may be segmented into separate databases reflecting the predicted behavior.

[0081] Design decomposition databases are generally discussed in U.S. Patent No. 9,183,624, issued November 10, 2015, entitled "Detecting defects on a wafer with run time use of design data," the entire contents of which are incorporated herein by reference.

[0082] In step 506, one or more probability context maps (or fractional maps) are generated based on the received data from steps 502 through 504 and stored in memory. For example, one or more probability context maps are generated before testing and provided to the optical imaging subsystem 102 during runtime. For example, the offline controller 130 may be configured to generate one or more probability context maps and store the generated probability context maps in memory 134 (or a remote database) and provide the stored probability context maps 136 to the context-based testing system 100 during runtime.

[0083] Return to reference Figure 4In step 404, one or more reference images may be received. For example, controller 124 may be configured to receive one or more reference images from optical imaging subsystem 102. The one or more reference images may represent a paradigm or reference. This reference image may be generated using any suitable technique. For example, a reference image may be generated using a single image of a reference region having a design common to the test region and known or expected to be defect-free. As another example, a reference image may be generated by combining (e.g., averaging or the like) multiple images of multiple reference regions.

[0084] In step 406, one or more test images may be received. For example, controller 124 may be configured to receive one or more test images from optical imaging subsystem 102.

[0085] One or more test images may contain various patterned features on the sample. In some embodiments, one or more test images contain one or more DOIs.

[0086] In step 408, one or more generated probability context maps may be provided to the supervised classifier. For example, the offline controller 130 may be configured to provide one or more generated probability context maps 136 stored in memory 134 to the supervised classifier.

[0087] Supervised classifiers can contain any type of deep learning classifier or algorithm, including, but not limited to, deep convolutional neural networks (CNNs) or similar. For example, a supervised classifier can contain a supervised discriminative CNN.

[0088] In the context of supervised learning, training images may include images that are false positives or true negatives. Offline controller 130 may be further configured to store training images in memory 134, and controller 124 may be configured to store the trained supervised classifier in memory 128.

[0089] In step 410, a supervised classifier may be applied to one or more test images to identify one or more DOIs on the sample. For example, using a supervised classifier, controller 124 may be configured to identify one or more DOIs on sample 104.

[0090] Figure 6 The description depicts a flowchart of a method 600 for performing a context-based test using a context graph containing empirical data obtained from a sample prior to being tested, according to one or more embodiments of this disclosure. Applicants should note that the embodiments and feasible techniques previously described herein in the context of testing system 100 should be interpreted as extending to method 600. However, it should be further noted that method 600 is not limited to the architecture of testing system 100. Figure 7This describes a process flow diagram of a method 700 for generating a context diagram containing empirical econometric data obtained from a sample prior to testing.

[0091] In step 602, empirical data from the sample to be tested can be acquired. For example, controller 124 may be configured to receive metrological data from an optical metrology tool. For example, controller 124 may be configured to receive optical metrological data generated by the optical metrology tool and provided to controller 124, including but not limited to critical dimensions, stacking, film thickness, pattern profile, or the like. In another example, controller 124 may be configured to receive metrological data from an electron beam metrology tool.

[0092] As another example, controller 124 may be configured to receive inspection data from optical imaging subsystem 102. As another example, controller 124 may be configured to receive inspection data from a simulated inspection source.

[0093] As another example, controller 124 may be configured to receive process tool data from process tools (including, but not limited to, scanners or the like).

[0094] In step 604, one or more context graphs may be generated. For example, prior to the test run, the offline controller 130 may be configured to generate one or more context graphs based on empirical data obtained from the samples prior to the test. One or more probabilistic context graphs may indicate the parametric probability density given the sample-dependent context data (z), such as... Express.

[0095] In step 606, a supervised classifier can be trained based on one or more generated context graphs and a set of labeled training images. When a supervised classifier (e.g., a detection network) is trained using measurement data (context), the supervised classifier can learn the dependence of defect probabilities on the measurement data.

[0096] Supervised classifiers can contain any type of deep learning classifier or algorithm, including, but not limited to, deep convolutional neural networks (CNNs) or similar. For example, a supervised classifier can contain a supervised discriminative CNN.

[0097] In step 608, one or more reference images may be received. For example, controller 124 may be configured to receive one or more reference images from optical imaging subsystem 102. As another example, controller 124 may be configured to receive one or more reference images from an additional optical imaging subsystem. The one or more reference images may represent a paradigm or reference. This reference image may be generated using any suitable technique. For example, a reference image may be generated using a single image of a reference region having a design common to the test region and known or expected to be defect-free. As another example, a reference image may be generated by combining (e.g., averaging or the like) multiple images of multiple reference regions.

[0098] In step 610, one or more test images may be received. For example, controller 124 may be configured to receive one or more test images from optical imaging subsystem 102.

[0099] One or more test images may contain various patterned features on the sample. In some embodiments, one or more test images contain one or more DOIs.

[0100] In step 612, a supervised classifier may be applied to one or more test images to identify one or more DOIs on the sample. For example, using a supervised classifier, controller 124 may be configured to identify one or more DOIs on sample 104.

[0101] The topics described herein sometimes indicate that other components contain or are connected to other components. It should be understood that such depicted architectures are merely exemplary, and many other architectures can in fact be implemented to achieve the same functionality. Conceptually, any arrangement of components used to achieve the same functionality is effectively “associated” to achieve the desired functionality. Therefore, any two components combined in this document to achieve a particular functionality can be considered “associated” with each other to achieve the desired functionality, regardless of the architecture or intermediate components. Similarly, any two such associated components can also be considered “connected” or “coupled” to each other to achieve the desired functionality, and any two components that can be suchly associated can also be considered “coupleable” to each other to achieve the desired functionality. Specific examples of coupleability include, but are not limited to, physically interactive and / or physically interactive components and / or wirelessly interactive and / or logically interactive components.

[0102] It is believed that this disclosure and many of its accompanying advantages will be understood from the foregoing description, and it will be appreciated that various changes can be made to the form, construction, and arrangement of the components without departing from the disclosed subject matter or sacrificing all its substantial advantages. The forms described are merely illustrative, and the appended claims are intended to cover and encompass such changes. Furthermore, it should be understood that the invention is defined by the appended claims.

Claims

1. An inspection system comprising: One or more controllers include one or more processors for executing program instructions stored in memory, said program instructions being configured to cause said one or more processors to: Receive one or more reference images; Receive one or more test images of the sample, the one or more test images containing one or more DOIs of the defect of interest; During the test run, an unsupervised classifier is used to generate one or more probabilistic context graphs; During the test run time, the generated one or more probability context maps are provided to the supervised classifier; and The supervised classifier is applied to the received test images to identify the one or more DOIs on the samples.

2. The inspection system of claim 1, wherein generating one or more probabilistic context graphs using an unsupervised classifier during the inspection run further comprises: Target context data is generated by removing defective portions from one or more test images; The generated target context data is provided to the unsupervised classifier; and Using the unsupervised classifier, one or more generated reference images are generated based on the generated target context data and the received one or more reference images; Calculate the entropy fraction; The negative log-likelihood score is calculated by comparing one or more generated reference images from the unsupervised classifier with one or more actual images from the one or more received test images. and The one or more probability context graphs are generated based on the calculated entropy score and the negative log-likelihood score.

3. The inspection system according to claim 2, wherein the unsupervised classifier comprises an unsupervised generative neural network.

4. The inspection system of claim 3, wherein the unsupervised generative neural network uses soft clustering techniques to generate the one or more generated reference images.

5. The inspection system according to claim 4, wherein the soft clustering technique comprises: Gaussian mixture model.

6. The inspection system according to claim 1, wherein the supervised classifier comprises a supervised discriminative neural network.

7. The inspection system according to claim 1, further comprising: An optical imaging subsystem communicatively coupled to one or more controllers.

8. The testing system according to claim 1, wherein the sample includes a substrate.

9. The testing system of claim 1, wherein the sample comprises a wafer.

10. The inspection system of claim 1, wherein the sample includes a mask.

11. An inspection system comprising: One or more controllers include one or more processors for executing program instructions stored in memory, said program instructions being configured to cause said one or more processors to: Receive empirical data from samples obtained from the characterization subsystem; Receive design breakdown data from the design breakdown database; Before verifying the runtime, one or more probabilistic context graphs are generated based on the received empirical data and the received data from the design decomposition database; The generated one or more probabilistic context graphs are stored in the memory; Receive one or more reference images; Receive one or more test images of the sample, wherein the one or more test images contain one or more defects of interest (DOIs); Retrieve the stored one or more probability context graphs from the memory; Provide the one or more probabilistic context graphs to the supervised classifier; and The supervised classifier is applied to the received test images to identify the one or more DOIs on the samples.

12. The testing system of claim 11, wherein the characterization subsystem comprises: The metering subsystem is configured to generate metering data.

13. The inspection system according to claim 12, wherein the measurement data includes at least one of the following: Critical dimension data, overlay data, film thickness data, or pattern contour data.

14. The inspection system of claim 11, wherein the design decomposition database comprises primitive patterns of the sample at sample locations and predicted behavior of the primitive patterns varying according to one or more measured characteristics of the sample, the primitive patterns being incorporated into the design of the sample.

15. The inspection system of claim 11, wherein the supervised classifier comprises a supervised discriminative neural network.

16. The inspection system according to claim 11, further comprising: An optical imaging subsystem communicatively coupled to one or more controllers.

17. The testing system of claim 11, wherein the sample comprises a substrate.

18. The inspection system of claim 17, wherein the sample comprises a wafer.

19. The inspection system of claim 11, wherein the sample includes a mask.

20. An inspection system comprising: One or more controllers include one or more processors for executing program instructions stored in memory, said program instructions being configured to cause said one or more processors to: Receive empirical data from the characterization subsystem; One or more context graphs are generated based on the received empirical data prior to the runtime verification; A supervised classifier is trained based on one or more context graphs generated and a set of labeled training images. Receive one or more reference images; Receive one or more test images of the sample, the one or more test images containing one or more DOIs of the defect of interest; and The trained supervised classifier is applied to the received test images to identify the one or more DOIs on the samples.

21. The testing system of claim 20, wherein the characterization subsystem comprises: The metering subsystem is configured to generate metering data.

22. The inspection system according to claim 21, wherein the measurement data includes at least one of the following: Critical dimension data, overlay data, film thickness data, or pattern contour data.

23. The inspection system of claim 20, wherein the supervised classifier comprises a supervised discriminative neural network.

24. The inspection system according to claim 20, further comprising: An optical imaging system communicatively coupled to one or more controllers.

25. The testing system of claim 20, wherein the sample comprises a substrate.

26. The inspection system of claim 25, wherein the sample comprises a wafer.

27. The inspection system of claim 20, wherein the sample includes a mask.

Images