Machine learning based metrology for semiconductor samples

Abstract

A system and method for inspecting a semiconductor sample is provided. The method includes: obtaining a runtime image of a semiconductor sample acquired by an inspection tool; processing the runtime image to create one or more image stripes, each image stripe containing an edge, and for each image stripe, extracting a sequence of topological points representing a contour of the edge in the image stripe; providing the topological point sequence of each image strip to a trained machine learning (ML) model for processing, and obtaining an updated topological point sequence as the output of the ML model; and obtaining measurement data on the runtime image using the updated sequence of topological points, wherein the measurement data has improved performance with respect to at least one metrology metric.

Description

Technical Field

[0001] The presently disclosed subject matter relates generally to the field of inspection of semiconductor samples, and more particularly to machine learning-based metrology applications of the samples. Background Art

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

[0003] During or after fabrication of the sample to be inspected, inspection may be provided by using non-destructive inspection tools. Inspection generally involves generating some output (e.g., image, signal, etc.) of the sample by directing light or electrons to a wafer and detecting the light or electrons from the wafer. As non-limiting examples, various non-destructive inspection tools include scanning electron microscopes, atomic force microscopes, optical inspection tools, and the like.

[0004] The inspection process may include multiple inspection steps. The manufacturing process of the semiconductor device may include various procedures, such as etching, deposition, planarization, growth such as epitaxial growth, implantation, etc. The inspection step may be performed multiple times, for example after certain process procedures and / or after manufacturing certain layers, etc. Additionally or alternatively, each inspection step may be repeated multiple times, for example for different wafer positions, or for the same wafer position with different inspection settings.

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

[0006] Automated inspection systems ensure that manufactured parts meet expected quality standards and provide useful information regarding adjustments that may be needed to manufacturing tools, equipment and / or components depending on the type of defects identified.

[0007] In some cases, machine learning techniques can be used to assist in automated inspection processes to facilitate increased throughput. For example, supervised machine learning can be used to provide accurate and efficient solutions for automating specific inspection applications based on well-annotated training images. Summary of the invention

[0008] According to certain aspects of the presently disclosed subject matter, there is provided a computerized metrology system for inspecting a semiconductor sample, the system comprising a processing circuit system configured to: obtain a runtime image of the semiconductor sample acquired by an inspection tool; process the runtime image to create one or more image strips, each image strip containing an edge, and for each image strip, extract a topological point sequence representing a contour of the edge in the image strip; provide the topological point sequence of each image strip to a trained machine learning (ML) model for processing, and obtain an updated topological point sequence as an output of the ML model; and obtain measurement data on the runtime image using the updated topological point sequence, wherein the measurement data has improved performance in at least one metrology metric.

[0009] In addition to the above features, the system according to this aspect of the presently disclosed subject matter may include one or more of the features (i) to (x) listed below, in any desired combination or arrangement that is technically possible:

[0010] (i) At least one metrology metric is from the group consisting of: matching, accuracy, correlation, and sensitivity.

[0011] (ii). The processing includes: identifying one or more edges from the runtime image; and for each edge, cropping the image tile set along a set of perpendicular lines relative to the edge; and combining the image tile set to form an image strip containing the edge.

[0012] (iii). The extraction includes: for each image strip, generating a plurality of grayscale (GL) distributions on the image strip; and for each GL distribution, identifying a position with a maximum derivative along the GL distribution, the position corresponding to a topological point on the contour of the edge.

[0013] (iv). Pre-training the ML model using a training set during a training phase, the training set comprising a plurality of training images collected from at least one inspection tool.

[0014] (v). Training the ML model comprises: for each training image: processing the training image to create one or more training image strips, each training image strip comprising an edge, and for each training image strip, extracting a topological point sequence representing a contour of the edge; providing the topological point sequence to the ML model for processing, and obtaining a predicted topological point sequence; using the predicted topological point sequence to obtain predicted measurement data; and evaluating the predicted measurement data using a loss function representing at least a metrological metric, and optimizing the ML model until the loss function satisfies a predefined criterion.

[0015] (vi). The processing includes: performing image registration between a runtime image and a training image, the training image being associated with one or more locations of one or more edges identified during training; and creating one or more image strips from the runtime image based on the one or more locations in the training image.

[0016] (vii). The improved performance of the measurement data is relative to the measurement data obtained using a topological point sequence.

[0017] (viii). Improved performance on measurement data further includes robustness and interpretability with respect to measurement data obtained using the end-to-end learning model.

[0018] (ix). Precision indicates the repeatability of predicted measurement data for different training images acquired by one metrology tool for a given feature on a sample. Correlation is the correlation between the predicted measurement data for a training image and the corresponding ground truth measurement data associated with the training image. Match indicates the repeatability of predicted measurement data for different training images acquired by different metrology tools for a given feature. Sensitivity indicates how sensitive the predicted measurement data is to changes in the size of a given feature.

[0019] (x). ML models are trained for specific metrology applications from a group including: critical dimension (CD) metrology, overlay (OVL), measurement-based inspection (MBI), critical dimension uniformity (CDU), CAD awareness (CADA), and lithography process control.

[0020] According to other aspects of the presently disclosed subject matter, a computerized method for inspecting a semiconductor sample is provided, the method comprising: obtaining a runtime image of the semiconductor sample acquired by an inspection tool; processing the runtime image to create one or more image strips, each image strip containing an edge, and for each image strip, extracting a topological point sequence representing a contour of the edge in the image strip; providing the topological point sequence of each image strip to a trained machine learning (ML) model for processing, and obtaining an updated topological point sequence as an output of the ML model; and obtaining measurement data on the runtime image using the updated topological point sequence, wherein the measurement data has improved performance in at least one metrology metric.

[0021] According to other aspects of the presently disclosed subject matter, there is provided a computerized method for training a machine learning model that can be used to inspect semiconductor samples, the method comprising: obtaining a plurality of training images collected from at least one metrology tool; for each training image, processing the training image to create one or more training image strips, each training image strip containing an edge, and for each training image strip, extracting a sequence of topological points representing a contour of the edge; providing the sequence of topological points to an ML model for processing, and obtaining a sequence of predicted topological points; using the predicted sequence of topological points to obtain predicted measurement data; and evaluating the predicted measurement data using a loss function representing at least a metrology metric, and optimizing the ML model until the loss function satisfies a predefined criterion.

[0022] These aspects of the disclosed subject matter may include, mutatis mutandis, one or more of the features (i) to (x) listed above with respect to the system, in any desired combination or arrangement technically possible.

[0023] According to other aspects of the presently disclosed subject matter, a non-transitory computer-readable medium is provided, the non-transitory computer-readable medium comprising instructions that, when executed by a computer, cause the computer to perform a computerized metrology method for inspecting a semiconductor sample, the method comprising: obtaining a runtime image of the semiconductor sample acquired by an inspection tool; processing the runtime image to create one or more image strips, each image strip containing an edge, and for each image strip, extracting a topological point sequence representing a contour of the edge in the image strip; providing the topological point sequence of each image strip to a trained machine learning (ML) model for processing, and obtaining an updated topological point sequence as an output of the ML model; and obtaining measurement data on the runtime image using the updated topological point sequence, wherein the measurement data has improved performance in at least one metrology metric.

[0024] According to other aspects of the presently disclosed subject matter, a non-transitory computer-readable medium is provided, the non-transitory computer-readable medium comprising instructions that, when executed by a computer, cause the computer to perform a computerized method for training a machine learning model that can be used to inspect a semiconductor sample, the method comprising: obtaining a plurality of training images collected from at least one metrology tool; for each training image, processing the training image to create one or more training image strips, each training image strip containing an edge, and for each training image strip, extracting a sequence of topological points representing a contour of the edge; providing the sequence of topological points to an ML model for processing, and obtaining a sequence of predicted topological points; using the predicted sequence of topological points to obtain predicted measurement data; and evaluating the predicted measurement data using a loss function representing at least a metrology metric, and optimizing the ML model until the loss function satisfies a predefined criterion.

[0025] These aspects of the disclosed subject matter may include, mutatis mutandis, one or more of the features (i) to (x) listed above with respect to the system, in any desired combination or arrangement technically possible. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to understand the present disclosure and to see how it may be carried out in practice, embodiments will now be described, by way of non-limiting example only, with reference to the accompanying drawings, in which:

[0027] Figure 1 Depicted is a generalized block diagram of an inspection system in accordance with certain embodiments of the presently disclosed subject matter.

[0028] Figure 2 A generalized flow chart for runtime inspection of semiconductor samples using a trained ML model, according to certain embodiments of the presently disclosed subject matter, is depicted.

[0029] Figure 3 A generalized flow diagram for generating image strips according to certain embodiments of the presently disclosed subject matter is depicted.

[0030] Figure 4 A generalized flow chart for extracting a set of topological points according to certain embodiments of the presently disclosed subject matter is depicted.

[0031] Figure 5 A generalized flow chart for training an ML model that can be used for runtime checking, according to certain embodiments of the presently disclosed subject matter, is shown.

[0032] Figure 6 Depicted are examples of semiconductor sample images including identification of different structures and their edges and extraction of topological points, according to some embodiments of the presently disclosed subject matter.

[0033] Figure 7 A schematic diagram of a process of extracting a sequence of topological points according to some embodiments of the presently disclosed subject matter is depicted. DETAILED DESCRIPTION

[0034] The process of semiconductor manufacturing generally requires multiple sequential processing steps and / or layers, each of which may produce errors that may result in yield loss. Examples of various processing steps may include lithography, etching, deposition, planarization, growth (such as, for example, epitaxial growth), and implantation, etc. Various inspection operations such as defect-related inspections and / or metrology-related inspections may be performed at different processing steps / layers during the manufacturing process to monitor and control the process. Inspection operations may be performed multiple times, for example, after certain processing steps and / or after manufacturing certain layers, etc.

[0035] As an example, a metrology operation can be used to measure one or more characteristics of a sample, such as, for example, CD measurements of features formed on the sample during a processing step (e.g., line width, thickness, etc.), so that the performance of the processing step can be evaluated based on the measurements. For example, if some of the measurements of the sample are unacceptable (e.g., exceed a predetermined range or threshold), such measurements can be used to change one or more parameters of the processing step so that subsequent samples made by the processing step can have acceptable characteristics.

[0036] Conventionally, metrology systems typically process captured images of samples to identify edges, and derive measurements based on the edge information. The measurements are evaluated relative to certain metrology benchmarks or metrics, such as, for example, accuracy, matching, sensitivity, and / or correlation (which may be predetermined by the customer). Such conventional metrology systems are also referred to herein as traditional metrology systems.

[0037] With the continuous advancement of semiconductor manufacturing processes, semiconductor devices are developed into increasingly complex structures with gradually shrinking feature sizes, which makes it more difficult for conventional metrology systems to meet metrology requirements.

[0038] Taking matching as an example, matching refers to a metrological metric / benchmark that represents the measurement variance between different tools, so matching is also called tool-to-tool matching. Therefore, matching is related to the repeatability of measurement data from different images of the same given feature acquired by different metrology tools. As an example, the matching specification requirements for a given current technology node (also called a process node, i.e., a specific generation of chips manufactured according to specific design rules) may be in the sub-nanometer range, and the matching specification requirements may be further reduced in future generations of technology nodes, such as, for example, reduced to the sub-angstrom range. Conventional metrology systems as described above cannot meet such high matching requirements.

[0039] Therefore, certain embodiments of the presently disclosed subject matter propose to use a machine learning-based metrology system to replace or optimize at least a portion of a conventional metrology system, which does not have one or more of the above-mentioned disadvantages. The present disclosure proposes preprocessing a runtime image to extract topological points representing edges in the runtime image, and processing the topological points using a trained machine learning (ML) model, and obtaining updated topological points. The updated topological points, when used to obtain measurement data on the runtime image, can result in improved measurement performance in at least one metrology metric, which will be described in detail below.

[0040] With this in mind, attention turns to Figure 1 , Figure 1 Depicted is a functional block diagram of an inspection system according to certain embodiments of the presently disclosed subject matter.

[0041] exist Figure 1 The inspection system 100 shown in the can be used to inspect semiconductor samples (e.g., wafers, dies, and / or portions thereof) as part of a sample manufacturing process. As described above, inspection referred to herein can be interpreted as covering any type of operation involving defect inspection / detection, defect classification, segmentation, and / or metrology operations (such as, for example, critical dimension (CD) measurement, roughness, coverage, etc.) on the sample. The system 100 includes one or more inspection tools 120, which are configured to scan the sample and capture images of the sample for further processing for various inspection applications.

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

[0043] 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 inspection machine, an electron beam inspection machine (e.g., a scanning electron microscope (SEM), an atomic force microscope (AFM), or a transmission electron microscope (TEM), etc.), etc. In some cases, the same inspection tool can provide low-resolution image data and high-resolution image data. The resulting image data (low-resolution image data and / or high-resolution image data) can be transmitted to the system 101 directly or via one or more intermediate systems. The present disclosure is not limited to any particular type of inspection tool and / or the resolution of the image data generated by the inspection tool.

[0044] In some embodiments, at least one of the inspection tools 120 has metrology capabilities and can be configured to capture images and perform metrology operations on the captured images. Such inspection tools are also referred to herein as metrology tools.

[0045] According to some embodiments, the metrology tool may be an electron beam tool, such as, for example, a scanning electron microscope (SEM). A SEM is an electron microscope that generates an image of a sample by scanning the sample with a focused electron beam. The electrons interact with the atoms in the sample, thereby generating various signals containing information about the surface topography and / or composition of the sample. The SEM is capable of accurately measuring features during the manufacture of semiconductor wafers. As an example, the metrology tool may be a critical dimension scanning electron microscope (CD-SEM) for measuring critical dimensions of structural features in an image.

[0046] According to certain embodiments of the presently disclosed subject matter, inspection system 100 includes a computer-based system 101 operably connected to inspection tool 120 and capable of implementing automated metrology operations with respect to semiconductor samples at runtime based on runtime images obtained during sample fabrication. System 101 is also referred to as a metrology system.

[0047] System 101 includes processing circuitry 102 that is operably connected to hardware-based I / O interface 126 and configured to provide processing required by an operating system, such as reference Figures 2 to 5 Detailed description. Processing circuit system 102 may include one or more processors (not shown separately) and one or more memories (not shown separately). One or more processors of processing circuit system 102 may be configured to execute several functional modules individually or in any appropriate combination according to computer-readable instructions implemented on a non-transitory computer-readable memory included in the processing circuit system. Such functional modules are hereinafter referred to as being included in the processing circuit system.

[0048] The one or more processors mentioned in this article may represent one or more general-purpose processing devices, such as microprocessors, central processing units, etc. More specifically, a given processor may be one of the following: a complex instruction set computing (CISC) microprocessor, a reduced instruction set computing (RISC) microprocessor, a very long instruction word (VLIW) microprocessor, a processor implementing other instruction sets, or a processor implementing a combination of instruction sets. One or more processors may also be one or more special-purpose processing devices, such as application-specific integrated circuits (ASICs), field programmable gate arrays (FPGAs), digital signal processors (DSPs), network processors, etc. One or more processors are configured to execute instructions for executing the operations and steps discussed herein.

[0049] The memory mentioned in this document may include one or more of the following: internal memory such as, for example, processor registers and cache, and main memory such as, for example, read-only memory (ROM), flash memory, dynamic random access memory (DRAM) (such as synchronous DRAM (SDRAM) or Rambus DRAM (RDRAM)), etc.

[0050] According to certain embodiments of the presently disclosed subject matter, system 101 may be a metrology system configured to perform metrology operations using a trained machine learning (ML) model based on runtime images obtained during sample fabrication. In this case, one or more functional modules included in processing circuitry 102 of system 101 may include an image processing module 104, a pre-trained machine learning (ML) model 106, and a measurement module 108.

[0051] Specifically, the processing circuit system 102 can be configured to obtain a runtime image of a semiconductor sample acquired by an inspection tool via an I / O interface 126. The image processing module 104 can be configured to process the runtime image to create one or more image strips, each image strip containing an edge, and for each image strip, extract a topological point sequence representing the contour of the edge. The topological point sequence of each image strip can be processed by the trained ML model 106. When processed, an updated topological point sequence can be provided as an output of the ML model 106. The measurement module 108 can be configured to obtain measurement data on the runtime image using the updated topological point sequence. The measurement data can be obtained for a specific metrology application. The measurement data can have improved performance in at least one metrology metric.

[0052] In this case, the functional modules 104 to 108 can be viewed as being included in a metrology recipe that can be used to perform metrology operations for a particular metrology application based on the acquired runtime image. The system 101 can be viewed as a metrology system that can use the metrology recipe to perform runtime metrology operations. Figure 2 Describes the details of the runtime checking process.

[0053] In some embodiments, the system 101 can be configured as a training system capable of training an ML model using a specially generated training set and cost function during a training / setup phase. In this case, one or more functional modules included in the processing circuit system 102 of the system 101 may include a training module (not shown) and an ML model 106. Specifically, the training module can be configured to obtain a training set, which includes multiple training images of samples and optionally corresponding ground truth measurement data associated with the training images. The training module can be configured to train the ML model 106 using the training set and one or more loss functions. The cost function can be specifically configured to evaluate the predicted measurement data obtained using the ML learning model relative to one or more metrology benchmarks (such as, for example, accuracy, correlation, sensitivity and / or matching, etc.). As described above, after being trained, the ML model can be used to process runtime images and obtain measurement data from the runtime images. Refer to the following Figure 5 Describes the details of the training process.

[0054] According to certain embodiments, ML models may be trained for different metrology applications based on specific training images and ground truth data related to the respective applications. Various applications to which the present disclosure may be applied include, but are not limited to, the following: critical dimension (CD) metrology applications, overlay applications, etc., as described in detail below.

[0055] Reference Figures 2 to 5 The operation of system 100 and system 101 , processing circuitry 102 , and functional modules in processing circuitry 102 are described in further detail.

[0056] According to certain embodiments, the ML model 106 mentioned herein may be implemented as various types of machine learning models, such as, for example, a decision tree, a support vector machine (SVM), an artificial neural network (ANN), a regression model, a Bayesian network, etc., or a collection / combination of the above. The learning algorithm used by the ML model may be any of the following: supervised learning, unsupervised learning, or semi-supervised learning, etc. The presently disclosed subject matter is not limited to a specific type of ML model or a specific type of learning algorithm used by the ML model.

[0057] In some embodiments, the ML model can be implemented as a deep neural network (DNN). The DNN can include multiple layers organized according to the corresponding DNN architecture. As a non-limiting example, the layers of the DNN can 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, etc. Optionally, at least some of the layers can be organized into multiple DNN subnetworks. Each layer of the DNN may include multiple basic computing units (CEs), which are generally referred to as dimensions, neurons, or nodes in the art.

[0058] The weighted values ​​and / or thresholds associated with the CE of the deep neural network and its connections can be initially selected before training, and can be further iteratively adjusted or modified during training to achieve an optimal set of weighted values ​​and / or thresholds in the trained DNN. After each iteration, the difference between the actual output generated by the DNN module and the target output associated with the corresponding training data set can be determined. The difference can be referred to as an error value. When the loss / cost function indicating the error value is less than a predetermined value, or when a limited change in performance between iterations is achieved, the training can be determined to be complete. The input data set used to adjust the weights / thresholds of the deep neural network is called a training set.

[0059] Note that the teachings of the presently disclosed subject matter are not constrained by a particular architecture of ML or DNN as described above.

[0060] It should be noted that although certain embodiments of the present disclosure relate to a processing circuit system 102 configured to perform the above-described operations, the functions / operations of the aforementioned functional modules may be performed in various ways by one or more processors in the processing circuit system 102. As an example, the operations of each functional module may be performed by a specific processor or a combination of processors. Thus, the operations of various functional modules, such as processing runtime images, extracting a sequence of topological points, providing a sequence of topological points for each image strip to a trained ML model for processing, and obtaining measurement data, etc., may be performed by a corresponding processor (or combination of processors) in the processing circuit system 102, and optionally, these operations may be performed by the same processor. The present disclosure should not be limited to a single processor that is interpreted as always performing all operations.

[0061] In some cases, in addition to the system 101, the inspection system 100 may include one or more inspection modules, such as, for example, a metrology operation module, a defect detection module, an automatic defect review module (ADR), an automatic defect classification module (ADC), and / or other inspection modules that can be used to inspect semiconductor samples. One or more inspection modules can be implemented as independent computers, or the functions of one or more inspection modules (or at least a portion of the functions) can be integrated with the inspection tool 120. In some cases, the output of the system 101 (e.g., a trained ML model, an updated topological point sequence, measurement data) can be provided to one or more inspection modules for further processing.

[0062] According to some embodiments, the system 100 may include a storage unit 122. The storage unit 122 may be configured to store any data required to operate the system 101, such as data related to the input and output of the system 101, and intermediate processing results generated by the system 101. As an example, the storage unit 122 may be configured to store images of samples generated by the inspection tool 120 and / or derivatives of the images, such as, for example, runtime images, training sets as described above. Therefore, these input images can be retrieved from the storage unit 122 and provided to the processing circuit system 102 for further processing. The output of the system 101 (such as, for example, a trained ML model, an updated topological point sequence and / or measurement data) can be sent to the storage unit 122 for storage.

[0063] In some embodiments, the system 100 may optionally include a computer-based graphical user interface (GUI) 124 that is configured to implement user-specified inputs related to the system 101. For example, a visual representation of the sample, including an image of the sample, etc., may be presented to the user (e.g., via a display that forms part of the GUI 124). Options to define certain operating parameters may be provided to the user via the GUI. The user may also view operating results or intermediate processing results, such as, for example, updated topological point sequences, measurement data, etc., on the GUI.

[0064] In some cases, the system 101 may be further configured to send the results of the operation to the inspection tool 120 via the I / O interface 126 for further processing. In some cases, the system 101 may be further configured to send the results to the storage unit 122 and / or an external system (e.g., a yield management system (YMS) of a manufacturing plant (semiconductor foundry)). In the context of semiconductor manufacturing, the yield management system (YMS) is a data management, analysis and tool system, especially during manufacturing upgrades, the system collects data from semiconductor foundries and helps engineers find ways to increase yields. YMS helps semiconductor manufacturers and semiconductor foundries manage a large amount of production analysis with fewer engineers. These systems analyze yield data and generate reports. Integrated device manufacturers (IMDs), semiconductor foundries, fabless semiconductor companies, and outsourced semiconductor assembly and testing (OSATs) can use YMS.

[0065] Those skilled in the art will readily appreciate that the teachings of the presently disclosed subject matter are not limited to Figure 1 Constraints of the system shown. Figure 1 Each system component and module in the present disclosure may be composed of any combination of related software, hardware and / or firmware, executed on an appropriate device or devices, which perform the functions defined and explained herein. Equivalent and / or modified functions described with respect to each system component and module may be combined or divided in another manner. Thus, in some embodiments of the presently disclosed subject matter, the system may include more than Figure 1 The components, modules and functions shown may include fewer, additional, modified and / or different components, modules and functions.

[0066] Figure 1Each component in the can represent multiple specific components, which are suitable for operating independently and / or collaboratively to process various data and electrical inputs and for implementing operations related to the computerized inspection system. In some cases, multiple instances of a component can be utilized for performance, redundancy and / or availability reasons. Similarly, in some cases, multiple instances of a component can be utilized for functional or application reasons. For example, different parts of a specific function can be placed in different instances of a component.

[0067] It should be noted that it can be implemented in a distributed computing environment Figure 1 The inspection system shown, wherein Figure 1 One or more of the aforementioned components and functional modules shown may be distributed on several local devices and / or remote devices. As an example, the inspection tool 120 and the system 101 may be located at the same entity (in some cases hosted by the same device), or distributed on different entities, depending on the specific system configuration and implementation needs. In some examples, certain components utilize a cloud implementation such as implemented in a private cloud or a public cloud. In the case where the various components of the inspection system are not completely located in one location or one physical entity, the communication between the various components of the inspection system may be achieved by any signaling system or communication component, module, protocol, software language, and drive signal, and the communication may be appropriately wired and / or wireless.

[0068] It should also be noted that in some embodiments, at least some of the inspection tools 120, storage unit 122, and / or GUI 124 may be external to the inspection system 100 and operate in data communication with the system 100 and the system 101 via the I / O interface 126. As described above, the system 101 may be implemented as a (multiple) stand-alone computer used in conjunction with the inspection tools and / or additional inspection modules. Alternatively, the corresponding functions of the system 101 may be at least partially integrated with one or more inspection tools 120, thereby improving and enhancing the functions of the inspection tools 120 in the inspection-related processes.

[0069] Although not necessarily so, system 101 and the operational processing of system 100 may be similar to that described with respect to Figures 2 to 5 Some or all of the steps of the method described herein may correspond to the following. Figures 2 to 5 The described methods and possible implementations thereof may be implemented by the system 101 and the system 100. Therefore, it should be noted that, mutatis mutandis, Figures 2 to 5 Embodiments discussed in connection with the described methods may also be implemented as various embodiments of system 101 and system 100 , and vice versa.

[0070] refer to Figure 2, depicts a generalized flow chart for runtime inspection of semiconductor samples using a trained ML model, according to certain embodiments of the presently disclosed subject matter.

[0071] As described above, semiconductor samples are typically made of multiple layers. During the manufacturing process of the sample, for example, after a processing step of a particular layer, the inspection process of the sample may be performed multiple times. In some cases, a sampling set of processing steps may be selected for in-line inspection based on the known impact of the processing steps on device characteristics or yield. An image of the sample or portion of the sample may be acquired at the sampling set of processing steps for inspection.

[0072] For purposes of illustration only, certain embodiments described below are described with respect to images of a given processing step / layer for a sample set of processing steps. Those skilled in the art will readily appreciate that the teachings of the presently disclosed subject matter, such as processes for machine learning-based inspection, may be performed after any layer and / or processing step of a sample. The present disclosure should not be limited to the number of layers included in a sample and / or the specific layer(s) to be inspected.

[0073] A runtime image of a semiconductor sample may be obtained (202) (e.g., acquired by an inspection tool 120) during runtime inspection of the sample. A semiconductor sample may refer herein to a portion of a semiconductor wafer, die, or the like that is manufactured and inspected in a semiconductor foundry during a manufacturing process. An image of a sample may refer to an image that captures at least a portion of the sample. As an example, an image may capture a target area of ​​interest or a target structure to be inspected on a semiconductor sample (e.g., a structural feature or pattern on a semiconductor sample). For example, an image may be an electron beam (e-beam) image acquired by an electron beam tool at runtime during in-line inspection of a semiconductor sample.

[0074] The runtime image may be processed (204) (e.g., by image processing module 104) to create one or more image strips, each image strip containing an edge. For each image strip, a sequence of topological points representing the outline of the edge may be extracted. The term "topological point" may refer to a specific point or location on the topography of a semiconductor sample at which a measurement or other metrological operation may be performed. It may be a point of interest selected for detailed inspection, such as the location of a specific feature of a semiconductor sample. The sequence of topological points extracted herein represents the outline of an edge in the image, including, for example, a bottom point and a top point of an edge. Various image processing approaches may be used to extract topological points representing an edge. Figure 3 and Figure 4 One possible way to extract topological points is shown.

[0075] Figure 3A generalized flow chart for generating an image strip according to some embodiments of the presently disclosed subject matter is depicted. When acquiring a runtime image, one or more edges can be identified (302) from the runtime image. For each identified edge, a set of image tiles can be cropped (304) along a set of perpendicular lines relative to the edge. The set of image tiles can be combined (306) to form an image strip containing the edge.

[0076] like Figure 6 , an image 602 capturing a plurality of line structures is shown. The edges of the line structures can be identified, such as shown in 604. Such identified edges are also referred to as rough edges. As an example, the rough edges 604 of the line structures can be identified by obtaining the derivative of the grayscale distribution of the line structures along the direction passing through the line structures, and placing the lines representing the edges based on the average derivative. Two image strips 606 are obtained, each image strip 606 containing the identified rough edges 604.

[0077] In the case of linear structures, the rough edges are typically identified as straight lines. In this case, the image strip can be cropped along the identified edge. In some other cases, such as shown in image 612 capturing a circular structure such as a contact, the rough edge 614 of the contact is identified as a circular shape. In this case, the image strip cannot be cropped directly along the edge. According to some embodiments, a set of perpendicular lines 616 relative to the edge can be placed around the rough edge 614, as shown, and the image tile set can be cropped along the set of perpendicular lines 616. The image tile set can be combined to form an image strip 618 that includes the edge 614.

[0078] The image strip 618 thus obtained can be considered as "unfolding" the circular edge 614 and the area around it into a relatively straight edge in the shape of a rectangular strip, similar to the image strip 606 obtained for the line structure. This achieves the purpose of easier processing of topological point extraction, as described below with reference to Figure 4 Images of other shaped structures, such as, for example, ellipses, curves, or free forms, etc., may be processed in a similar manner to create image strips containing edges.

[0079] In some embodiments, optionally, instead of identifying rough edges on a runtime image as described above, rough edges can be identified / found by image registration performed between a runtime image and a training image. As an example, a training image is associated with (e.g., marked) one or more locations of one or more rough edges identified during training. The training image can be used as a reference image. When registering between the runtime image and the training image, the location of the rough edges in the runtime image can be identified based on the location of the rough edges in the training image, and one or more image strips can be created from the runtime image based on the identified rough edges. As described above, this can save the runtime repetitive computational work used to perform edge identification and image strip creation.

[0080] Figure 4 A generalized flow chart for extracting a topological point set according to certain embodiments of the presently disclosed subject matter is depicted. As shown in the figure, for each image strip, a topological point set may be extracted along the image strip (e.g., along a direction perpendicular to the longitudinal axis of the structure in the image strip, an example of such a direction is represented by Figure 6 A plurality of grayscale (GL) distributions are generated (402) as shown by dashed lines 608 on a magnified view 607 of an image strip 606 in FIG. 4 ). The plurality of GL distributions may be generated at different locations on the image strip (e.g., parallel to dashed lines 608). For each GL distribution, a location with a maximum derivative may be identified (404) along the GL distribution (or a processed version of the GL distribution, such as a smoothed distribution). In some embodiments, the location with the maximum derivative may be used as the location of a topological point extracted on the contour of the edge.

[0081] In some embodiments, a range can be defined (406) along the GL distribution (or processed version of the GL distribution) that includes the location with the maximum derivative. A location on the GL distribution (or processed version of the GL distribution) corresponding to a topological point on the contour of the edge can be selected (408) within the range. In this way, a plurality of topological points corresponding to a plurality of GL distributions can be identified. The plurality of topological points form a sequence along the edge of the structure that represents a fine contour of the edge.

[0082] Figure 7 A schematic diagram illustrating a process of extracting a sequence of topological points according to some embodiments of the presently disclosed subject matter is shown.

[0083] A wire structure 702 such as captured in image 602 is depicted in the figure from a cross-sectional view. Figure 6A grayscale (GL) distribution 704 (illustrated as a waveform signal) is generated on the image strip 606 in the direction shown by the dotted line 608 in FIG. 7 . Based on the GL distribution 704, a smoothed version 706 of the waveform of the GL distribution 704 is generated (e.g., by noise filtering), and a derivative signal 708 of the smoothed waveform 706 is created. A peak 710 of the derivative signal 708 can be identified, indicating a position 714 with a maximum slope on the line structure 702. The peak 710 corresponds to a position 712 on the smoothed waveform 706. In some embodiments, the position 712 with the maximum derivative can be used as the position of a topological point extracted on the edge of the line structure. In some other embodiments, a range can be defined on the derivative signal 708 around the peak 710, and a position corresponding to the position of a topological point can be selected within the range.

[0084] Similarly, multiple topological points may be identified based on multiple GL distributions generated at different locations on the image strip 606. Figure 6 As shown, a plurality of topological points form a sequence 610 along the edge of the line structure representing the fine contour of the edge in the image strip 607. A sequence 620 of topological points representing the fine contour of the edge in the image strip 618 may be extracted in a similar manner.

[0085] It should be noted that the definitions of various parameters as described in the above examples (such as, for example, the selection of peaks, ranges, etc.) are for illustrative purposes only and should not be considered to limit the present disclosure in any way. Such parameters may be configured differently corresponding to different structures.

[0086] It should be noted that Figure 3 and Figure 4 One possible approach for topological point extraction is shown for exemplary purposes only and should not be considered to limit the present disclosure in any way.Instead of the above-described method for extracting a sequence of topological points representing edges, alternative image processing techniques may be used, such as, for example, image segmentation.

[0087] The term "image segmentation" generally refers to any process that divides an image into meaningful parts / segments (e.g., background and foreground, noisy and non-noise areas, various structural elements, defects and non-defects, etc.) while providing per-pixel or per-region values ​​indicative of such segmentation. In some cases, image segmentation can be based on machine learning. As an example, Figure 1The image processing module 104 shown may include a segmentation submodule configured to segment the runtime image, such as the line structure presented in the runtime image 602, based on one or more structural elements present in the image (structural elements or features used in this article may refer to any primitive object on the image, the object having a geometric shape or geometric structure with a contour, or a combination of such objects). The output of the segmentation module may be, for example, a segmentation map, in which the value of each pixel or sub-pixel in the segmentation map indicates the predicted probability that the corresponding pixel / sub-pixel in the image belongs to the structural element in the image or to the background in the image. In this way, the contours of the edges of the line structure can be identified based on the values ​​in the segmentation map.

[0088] continue Figure 2 As described with reference to block 204, once the topological point sequence for each image strip is obtained, the topological point sequence for each image strip may be provided as input (206) to a trained machine learning (ML) model (e.g., ML model 106) for processing. An updated topological point sequence may be obtained as an output of the ML model. As an example, the input to the ML model may include a topological point sequence identified on each of one or more image strips of the runtime image. In some cases, the location of the topological point sequence may be marked on the image strip, such as on Figure 6 A topological point sequence 610 is shown marked in an image strip 607, and the topological point sequence 610 is provided as input to the ML model. In some cases, the positions of the topological point sequence can be stored as a coordinate list and provided as input to the ML model together with the image strip.

[0089] As described above, various model architectures can be used to implement ML models. As an example, the ML model can be implemented as a CNN. CNN generally has a structure including an input layer and an output layer and multiple hidden layers. The hidden layers of a CNN generally include a series of convolutional layers, followed by additional layers such as pooling layers, fully connected layers, and normalization layers. In some cases, a CNN can be considered to consist of two main functions: feature extraction and classification. As an example, the feature extraction part may include several convolutional layers, followed by maximum pooling and activation functions. The classification part generally includes a fully connected layer. The convolutional layer in the feature extraction part includes a learnable filter set. Each filter of a particular layer can be convolved on the width and height of the input volume of a given input image, calculate the dot product between the entries of the filter and the input, and generate an activation map, which gives the response of the filter at each spatial position. The activation maps of all filters stacked along the depth dimension form the full output feature map of a given convolutional layer. In this way, CNN learns filters that are activated when a certain type of feature is detected at a certain spatial position in the input.

[0090] According to the following reference Figure 5 The process described in detail pre-trains the ML model used in runtime checking during the training phase.

[0091] Measurement data may be obtained (208) on the runtime image (e.g., by the measurement module 108) using the updated topological point sequence. The measurement data may be obtained for a particular metrology application. A metrology application refers to what a customer / user is typically interested in measuring for a sample. As a non-limiting example, a metrology application may be selected from the group of metrology applications, the group comprising: critical dimension (CD) metrology, overlay (OVL), measurement based inspection (MBI), critical dimension uniformity (CDU), and lithography process control.

[0092] CD metrology refers to measuring the critical dimensions of fine patterns formed on semiconductor wafers. CD measurements include, but are not limited to, the following: dimensions (e.g., line width, line spacing, contact diameter, size of components, edge roughness, grayscale statistics, etc.); shape of components; distance within or between components; related angles, etc. Overlay refers to the measurement of overlay offsets between multilayer patterns. An example of overlay is finding the nominal overlay error between two layers on the edge of a wafer. MBI refers to defect inspection using measurement. An example of MBI can be the detection of etch residues at the bottom of a trench. CDU refers to uniformity measurements associated with critical dimensions. An example of CDU can be creating a uniformity map of the offset between two contacts. Lithography process control refers to the control of lithography tools and material roughness.

[0093] The measurement data obtained using the updated topological point sequence has improved performance in at least one metrological metric. As an example, the at least one metrological metric is from the group including: accuracy, matching, correlation, and sensitivity. The improved performance is achieved via training of the ML model, as described below with reference to Figure 5 Described in detail.

[0094] Figure 5 A generalized flow chart for training an ML model that can be used for runtime checking, according to certain embodiments of the presently disclosed subject matter, is shown.

[0095] The ML model is trained using a training set that includes a plurality of training images of samples and optionally corresponding ground truth measurement data associated with the training images. Before feeding each training image into the ML model, the training set may be trained in the same manner as in the above reference. Figure 2 Each training image is processed in a similar manner as described in block 204 .

[0096] The training images may be "real world" images of semiconductor samples obtained by an inspection tool during the manufacturing process of the semiconductor samples. In some embodiments, in addition to the "real world" training images, the training set used to train the ML model may be enriched by one or more synthetic images simulated for the semiconductor samples.

[0097] The ground truth data is application specific. For example, where the metrology application is a CD measurement application, the ground truth measurement data may represent actual / true CD measurements derived from training images.

[0098] In some embodiments, the ground truth measurement data may include one or more of the following: measurement data generated by processing training images using a reference metrology system; measurement data obtained from a customer based on a test sample; and measurement data automatically generated for a simulated image (because the image is simulated based on design data, which is typically associated with corresponding ground truth data for structural elements of the test sample, the simulated image corresponding to the design data may also be associated with the corresponding ground truth data).

[0099] Specifically, a training image can be processed (502) to create one or more training image strips, each training image strip containing an edge, and for each training image strip, a topological point sequence representing the contour of the edge can be extracted. The topological point sequence of each training image strip can be provided (504) to an ML model for processing to obtain a predicted topological point sequence as an output of the ML model. Predicted measurement data can be obtained on the training image using the (506) predicted topological point sequence. The predicted measurement data can be evaluated (508) using a loss function (also referred to as a cost function) representing at least a metrological metric. The loss function refers to an overall loss function or a total loss function, which can include one or more loss functions specifically configured to evaluate one or more metrological metrics. The ML model can be optimized until the overall loss function meets a predefined criterion.

[0100] As described above, the weights and / or thresholds of the ML model may be initially selected prior to training, and may be further iteratively adjusted or modified during training to achieve an optimal set of weights and / or thresholds in the trained ML model. After each iteration, the difference between the actual output produced by the ML model and the target output associated with the corresponding training data may be determined according to a particular loss function. The difference may be referred to as an error value. Training may be determined to be complete when one or more loss functions indicating one or more error values ​​are less than corresponding predetermined values, or when a limited change in performance between iterations is achieved (the change may be predefined).

[0101] As an example, the at least one metrology metric is from the group including: accuracy, match, correlation, and sensitivity, etc., which will be explained in detail below.

[0102] Specifically, precision refers to the closeness of consistency between independent measurements (by the same metrology tool) on the same feature of the same area / location of the sample. As an example, high precision indicates that independent measurements of the same feature are repeatable (i.e., the measurements have a small variance from each other and the measurement distribution is relatively close). In some embodiments, precision can be regarded as measuring repeatability. In some other embodiments, precision may include two components: repeatability and reproducibility. Repeatability refers to a measure of the distribution of measurement results, in which continuous measurements are repeated on the same location of the sample without any operator intervention. The cause of the variation in repeated measurement results may be mainly due to the statistical properties of the tool signal (e.g., SEM signal) and the interpretation of the new signal set by the measurement algorithm included in the recipe. Reproducibility refers to another measure of the distribution of measurement results, in which measurements are obtained from different locations of the same sample at different times. Other sources of variation between independent measurements are considered: wafer alignment, SEM autofocus, pattern recognition, tool stability, etc.

[0103] When evaluating ML performance on training data, precision can indicate the repeatability of training measurement data for different training images acquired by one metrology tool for a given feature on a sample (e.g., a structural feature from a given region / location of a sample, or the same type of feature from different regions / locations of a sample).

[0104] The loss function may be configured as a criterion for evaluating accuracy. As an example, the loss function may be associated with an accuracy score obtained by calculating the variance between training measurement data of different training images based on the accuracy measure.

[0105] As an example, a loss function configured to evaluate accuracy may be instantiated as follows:

[0106] L 精确度 = difference measure (y1,y2,...y n )

[0107] y1 to y n represents the predicted measurements for n specific scanning runs for a given location or locations. The ML model is optimized by minimizing the value of a cost function, which represents a difference metric for evaluating the difference between the predicted measurements for the n specific runs. Similarly, as described above, the difference metric can be, for example, the variance of the distance of the measurements for each run to the average measurement of multiple runs or any Lp norm (p can be any integer), etc.

[0108] Match refers to a metrological metric / benchmark that represents the measurement variance between different tools, and thus match is also referred to as tool-to-tool match. Therefore, match is related to the repeatability of training measurement data of different training images of the same given feature acquired by different metrology tools. In order to evaluate such a benchmark, training images representing tool-to-tool variance, such as, for example, training images acquired by different metrology tools, should be collected during the training data collection / preparation phase (e.g., the inspection tool 120 may include multiple metrology tools). Various data collection schemes can be used to collect training images acquired by different metrology tools (such as, for example, ABBA, QC) to compensate for variations caused by physical effects (such as shrinkage and charging, etc.).

[0109] The loss function for evaluating the matching criterion may be configured similarly as described above, for example by calculating the difference between training measurement data of different training images acquired by different metrology tools based on the matching measure.

[0110] As an example, in the data collection scenario of ABBA, where two training subsets A and B are collected from two tools in an alternating manner, the loss function can be configured as follows:

[0111] L 匹配度 = difference measure (y i 预测(A) ,y i 预测(B) )

[0112] y i 预测(A) represents the predicted measurement value (i.e., training measurement data) of training image i from training subset A, and y i 预测(B) Represents the predicted measurement value of the corresponding training image i from the training subset B. The ML model is optimized by minimizing the value of a loss function, which represents a difference metric for evaluating the difference between the corresponding predicted measurement values ​​from two (or more) training subsets. In some cases, the difference metric can be based on distance. Examples of such difference metrics may include, for example, Lasso, Ridge regression, Lp norm (p can be any integer), etc., which also applies to the case of multiple tools (e.g., more than two tools). For example, in the case of multiple tools, the difference metric can be configured to minimize the distance from the measurement value of each tool to the average of the measurement values ​​of all tools.

[0113] Correlation refers to the relationship between training measurement data of a training image and corresponding ground truth measurement data associated with the training image. The loss function of the criterion for evaluating correlation may be configured to represent the difference between the training measurement data and the corresponding ground truth measurement data.

[0114] In order to evaluate the relevance, ground truth measurement data needs to be provided. As an example, the cost function configured to evaluate the relevance can be illustrated as follows:

[0115] L 相关性 = difference measure (y 真实 ,y 预测 )

[0116] y 预测 Represents the predicted measurement value of the ML model. 真实 Represents ground truth measurement data, such as provided by a reference metrology algorithm. The ML model can be optimized by minimizing the value of a loss function, which represents a difference metric for evaluating the difference between a predicted measurement value and a corresponding ground truth measurement value.

[0117] Sensitivity refers to how sensitive a measurement is to actual physical changes in the size of a feature of a sample. As an example, if a feature of a sample (e.g., the width of a structural element) changes from 10 nm to 10.1 nm, a high sensitivity indicates that the corresponding measurement should be sensitive to changes in this scale, and the measurement should reflect this change. Sensitivity can be assessed where the training set includes training images with changing sizes (such as synthetic training images that are specifically simulated with changing sizes of certain structural elements, as described above) and corresponding ground truth data for the training images.

[0118] The loss function of the criterion for evaluating the sensitivity may be configured, for example, by estimating a linear regression function between a plurality of training measurement data and associated ground truth data. As an example, the linear regression may be estimated as, for example, training measurement data = gain * ground truth + offset.

[0119] In some cases, a loss function for a specific metric may be used to optimize an ML model to satisfy a specific metrology metric. In some other cases, an ML model may be optimized to satisfy multiple metrology metrics, such as, for example, match, accuracy, sensitivity, and relevance. In this case, the total loss function of the ML model may include various components of a specific cost function configured for a specific metric, where corresponding weights may be applied for a specific loss function. For example, the total loss function may be expressed as follows:

[0120] L 总 =a1L 精确度 +a2L 匹配度 +a3L相关性 +a4L 灵敏度

[0121] Once the ML model is trained using the total loss function configured with the above components, the ML model can be used at runtime to process input images from any tool and provide runtime measurements without tool-to-tool variance while meeting the relevance, sensitivity, and accuracy requirements.

[0122] In some cases, one or more additional loss functions may be added to the total loss function, optionally in addition to or in lieu of the above exemplary components, and the present disclosure is not limited to a particular representation and / or number of components included in the total loss function.

[0123] In some embodiments, one or more metrology benchmark criteria may be predetermined according to customer specifications and / or based on previous inspection experience.

[0124] Once the ML model is trained, the trained ML model may be used at runtime to update topological points to locations that, when used to perform runtime metrology operations, may provide runtime measurement data having improved performance in one or more metrology metrics.

[0125] like Figure 6 As shown, the original topological point sequence 610 used as input to the ML model shows a slight deviation from the bottom portion of the edge of the line structure. After being processed, the updated topological point sequence shows that the topological points at the bottom portion of the sequence have been updated to a new position 611 that more closely matches the actual edge. Relative to the measurement data obtained using the original topological point sequence, the use of the updated topological point sequence can provide measurement data with better performance, for example, in terms of high accuracy, matching and / or correlation. This is at least because the ML model is trained using a total loss function that is configured to specifically evaluate measurement performance in terms of these metrological metrics. As an example, in the case where the total loss function includes a sub-loss function of matching, the ML model is trained to provide updated topological points, which can be used to obtain measurements that at least meet the matching metric.

[0126] The currently disclosed metrology system also provides improved explainability / interpretability compared to E2E type learning models where the input to the model is an image and the output is directly measurement data. As an example, the currently disclosed metrology system can visually demonstrate how measurement data is obtained via the location of topological points, thereby enabling a better understanding of the correlation between the runtime image and the topological points of the runtime image and the measurement data obtained from the runtime image. This also allows easier adjustment of the metrology algorithm if necessary.

[0127] On the other hand, E2E learning models work in a black-box manner. It is unlikely to find out how the measurement data is obtained from the input image and which image characteristics may affect the measurement values. This is especially important in cases where GL changes and / or the presence of potential pattern deviations in the runtime image may affect the measurement values ​​provided by the E2E model without the user knowing how. The currently disclosed metrology system can demonstrate the impact of such changes on the selection of topological points via pre-processing of the runtime image, which provides additional "visual" input to the user and enables the user to intervene if necessary. The proposed method can provide topological points for all edges on all sides in the image, which enables the collection of statistical values ​​over the entire image, such as, for example, STD for CD measurement.

[0128] It should be noted that the examples shown in the present disclosure (such as, for example, exemplary metering applications, exemplary metering metrics, various ways of extracting topological points, etc.) are described for exemplary purposes and should not be considered to limit the present disclosure in any way. Other appropriate examples / implementations may be used in addition to or in place of the above.

[0129] One of the advantages of certain embodiments of the presently disclosed subject matter as described herein is the provision of an optimized metrology system based on machine learning that is capable of providing measurement data having improved performance in at least one metrology metric (such as, for example, accuracy, match, sensitivity and / or correlation) relative to measurement data obtained using an original sequence of topological points.

[0130] As described above, this is accomplished by providing runtime pre-processing of the image of initial topological points representing edges in the image and the ability of the ML model to process the initial topological points and provide updated topological points that can provide measurement data with improved performance when used in metrology operations. As described above, the ML model is specifically trained to achieve this capability.

[0131] One of the further advantages of certain embodiments of the presently disclosed subject matter described herein is improved explainability / interpretability compared to E2E type learning models. The presently disclosed metrology system can visually demonstrate how measurement data is obtained via the location of topological points, thereby enabling a better understanding of the correlation between the runtime image and the topological points of the runtime image and the measurement data obtained from the runtime image. This also allows the user to intervene more easily if necessary and allows easier adjustment of the metrology algorithm.

[0132] One of the other advantages of certain embodiments of the presently disclosed subject matter as described herein is an optimized metrology system capable of providing runtime measurement data with reduced tool-to-tool variance. This is achieved by training an ML model in the metrology system to at least satisfy a tool-to-tool fit metric using a training set and at least one cost function specifically configured therefor. Optionally, in addition to tool-to-tool fit, one or more additional cost functions may be added to train the learning model to satisfy accuracy, sensitivity, and / or relevance metrics. The learning model trained in this manner can provide measurement data with reduced tool-to-tool variance while satisfying accuracy, sensitivity, and / or relevance criteria.

[0133] It is to be understood that the disclosure is not limited in its application to the details set forth in the description contained herein or illustrated in the drawings.

[0134] In this detailed description, many specific details are set forth in order to fully understand the present disclosure. However, it will be appreciated by those skilled in the art that the subject matter currently disclosed can be practiced without these specific details. In other cases, well-known methods, procedures, components and circuits are not described in detail to avoid obscuring the subject matter currently disclosed.

[0135] Unless otherwise specifically stated, as is apparent from this discussion, it should be understood that throughout the specification, discussions utilizing terms such as "obtain," "inspect," "provide," "train," "extract," "generate," "process," "use," "identify," "tailor," "combine," "define," "select," "evaluate," "optimize," "execute," "create," etc. 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 type of hardware-based electronic device with data processing capabilities, including, as a non-limiting example, the inspection systems, metrology systems, and corresponding portions thereof disclosed in the present application.

[0136] 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. The terms should be considered to include a single medium or multiple media (e.g., a centralized or distributed database, and / or associated caches and servers) that store one or more instruction sets. The terms should also be understood to include any medium capable of storing or encoding an instruction set for execution by a computer and causing the computer to perform any one or more of the methods of the present disclosure. Therefore, the terms should be considered to include, but are not limited to, read-only memory ("ROM"), random access memory ("RAM"), magnetic disk storage media, optical storage media, flash memory devices, and the like.

[0137] The term "sample" used in this specification should be broadly interpreted to cover any type of physical object or substrate used to manufacture semiconductor integrated circuits, magnetic heads, flat panel displays, and other semiconductor products, including wafers, masks, graticules, and other structures, as well as combinations and / or portions of the above. The sample is also referred to as a semiconductor sample in this article, and the sample can be produced by manufacturing equipment that performs the corresponding manufacturing process.

[0138] The term "inspection" used in this specification should be broadly interpreted to cover any type of operation related to various types of defect detection, defect review and / or defect classification, segmentation, and / or metrology operations during and / or after the sample manufacturing process. Inspection is provided by using a non-destructive inspection tool during or after the manufacture of the sample to be inspected. As a non-limiting example, the inspection process may include runtime scanning (in a single scan or multiple scans), imaging, sampling, detection, review, measurement, classification and / or other operations provided for the sample or part of the sample using the same or different inspection tools. Similarly, the inspection can be performed before the sample to be inspected is manufactured, and the inspection may include, for example, generating (multiple) inspection recipes and / or other setup operations. It should be noted that unless otherwise specifically stated, the term "inspection" or its derivatives used in this specification are not limited in terms of the resolution or size of the inspection area. As a non-limiting example, various non-destructive inspection tools include scanning electron microscopes (SEMs), atomic force microscopes (AFMs), optical inspection tools, etc.

[0139] The term "metrology operation" as used in this specification should be broadly interpreted to cover any metrology operation procedure for extracting metrology information related to one or more structural elements on a semiconductor sample. In some embodiments, the metrology operation may include measurement operations, such as, for example, critical dimension (CD) measurements performed on certain structural elements on the sample, including but not limited to the following: dimensions (e.g., line width, line spacing, contact diameter, size of elements, edge roughness, grayscale statistics, etc.); shape of elements; distances within or between elements; related angles; overlap information associated with elements corresponding to different design levels, etc. For example, by using image processing techniques to analyze measurement results such as measured images. It should be noted that unless otherwise specifically stated, the term "metrology" or its derivatives used in this specification are not limited to measurement techniques, measurement resolutions, or the size of inspection areas.

[0140] The term "defect" as used in this specification should be interpreted broadly to cover any type of abnormal or undesirable feature / function formed on a sample. In some cases, a defect may be a defect of interest (DOI), which is a real defect that has some impact on the function of the manufactured device, so it is in the interest of the customer to detect the DOI. For example, any "killer" defect that may cause yield loss may be indicated as a DOI. In some other cases, a defect may be a perturbable (also known as a "false positive" defect), which can be ignored because the perturbation has no effect on the function of the finished device and does not affect yield.

[0141] The term "design data" as used in this specification should be interpreted broadly to encompass any data indicative of a hierarchical physical design (layout) of a sample. The design data may be provided by a respective designer, and / or may be derived from a physical design (e.g., by complex simulations, simple geometric and Boolean operations, etc.). The design data may be provided in different formats, such as, as non-limiting examples, GDSII format, OASIS format, etc. The design data may be presented in a vector format, a grayscale intensity image format, or other formats.

[0142] The terms "image(s)" or "image data" as used in this specification should be interpreted broadly to cover any original image / frame of a sample captured by an inspection tool during a manufacturing process, derivatives of the captured image / frame obtained by various pre-processing stages, and / or computer-generated synthesized images (in some cases based on design data). Depending on the specific manner of scanning (e.g., one-dimensional scanning such as line scanning, two-dimensional scanning in both x- and y-directions, or point scanning at specific points, etc.), the image data can be represented in different formats, such as, for example, as a grayscale distribution, a two-dimensional image, or discrete pixels, etc. It should be noted that in some cases, in addition to the image (e.g., captured image, processed image, etc.), the image data mentioned in this document may also include digital data associated with the image (e.g., metadata, manual attributes, etc.). It should also be noted that the image or image data may include data related to the processing step / layer of interest or multiple processing steps / layers of the sample.

[0143] It should be understood that, unless otherwise specifically stated, certain features of the presently disclosed subject matter 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 described in the context of a single embodiment may also be provided individually or in any suitable subcombination. In this detailed description, many specific details are set forth in order to fully understand the method and apparatus.

[0144] It should also be understood that the system according to the present disclosure can be implemented at least in part on a suitably programmed computer. Likewise, the present disclosure contemplates a computer program 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 an instruction program executable by a computer for performing the method of the present disclosure.

[0145] The present disclosure is capable of other embodiments and can be practiced and executed in various ways. Therefore, it should be understood that the wording and terminology employed herein are for descriptive purposes and should not be considered restrictive. Therefore, those skilled in the art will appreciate that the concepts upon which the present disclosure is based can be easily used as the basis for designing other structures, methods and systems for achieving the several purposes of the presently disclosed subject matter.

[0146] Those skilled in the art will readily appreciate that various modifications and changes may be made to the embodiments of the present disclosure described above without departing from the scope of the present invention as defined by the appended claims.

Claims

1. A computerized metrology system for inspecting a semiconductor sample, the system comprising processing circuitry configured to: obtaining a runtime image of a semiconductor sample acquired by an inspection tool; processing the runtime image to create one or more image strips, each image strip containing an edge, and for each image strip, extracting a sequence of topological points representing a contour of the edge in the image strip; providing the topological point sequence of each image strip to a trained machine learning (ML) model for processing, and obtaining an updated topological point sequence as an output of the ML model; and Measurement data is obtained on the runtime image using the updated sequence of topological points, wherein the measurement data has improved performance in at least one metrological metric.

2. The computerized system of claim 1, wherein the at least one metrology metric is from the group consisting of: match, accuracy, correlation, and sensitivity.

3. The computerized system of claim 1, wherein the processing comprises: One or more edges are identified from the runtime image; and for each edge, a set of image tiles are cropped along a set of perpendicular lines relative to the edge; and the set of image tiles are combined to form an image strip containing the edge.

4. The computerized system of claim 1, wherein said extracting comprises: For each image strip, generating a plurality of gray level (GL) distributions on the image strip; And for each GL distribution, identifying a location along the GL distribution having a maximum derivative, the location corresponding to a topological point on the contour of the edge.

5. The computerized system of claim 1, wherein the ML model is pre-trained during a training phase using a training set, the training set comprising a plurality of training images collected from at least one inspection tool.

6. The computerized system of claim 5, wherein said training said ML model comprises: For each training image: processing the training image to create one or more training image strips, each training image strip containing an edge, and for each training image strip, extracting a sequence of topological points representing a contour of the edge; Providing the topological point sequence to the ML model for processing, and obtaining a predicted topological point sequence; Using the predicted topological point sequence to obtain predicted measurement data; as well as The predicted measurement data is evaluated using a loss function representing the at least metrology metric, and the ML model is optimized until the loss function satisfies a predefined criterion.

7. The computerized system of claim 1, wherein the processing comprises: performing image registration between the runtime image and a training image associated with one or more locations of one or more edges identified during training; and creating the one or more image strips from the runtime image based on the one or more locations in the training image.

8. The computerized system of claim 1, wherein the improved performance of the measurement data is relative to measurement data obtained using the sequence of topological points.

9. The computerized system of claim 1, wherein the improved performance of the measurement data further includes robustness and interpretability of the measurement data obtained using an end-to-end learning model.

10. A computerized method for training a machine learning model that can be used to inspect semiconductor samples, the method comprising: obtaining a plurality of training images collected from at least one metrology tool; For each training image, processing the training image to create one or more training image strips, each training image strip containing an edge, and for each training image strip, extracting a sequence of topological points representing a contour of the edge; Providing the topological point sequence to the ML model for processing, and obtaining a predicted topological point sequence; Using the predicted topological point sequence to obtain predicted measurement data; as well as The predicted measurement data is evaluated using a loss function representing the at least metrology metric, and the ML model is optimized until the loss function satisfies a predefined criterion.

11. The computerized method of claim 10, wherein the at least one metrological metric is from the group consisting of: match, accuracy, correlation, and sensitivity.

12. The computerized method of claim 11, wherein the precision indicates the repeatability of predicted measurement data for different training images acquired by one metrology tool for a given feature on the sample; the correlation is the correlation between the predicted measurement data for the training images and corresponding ground truth measurement data associated with the training images; the match indicates the repeatability of predicted measurement data for different training images acquired by different metrology tools for the given feature; and the sensitivity indicates how sensitive the predicted measurement data is to changes in the size of the given feature.

13. The computerized method of claim 10, wherein said processing comprises: identifying one or more edges from the training image; And for each edge, cropping the set of training image tiles along a set of perpendicular lines relative to the edge; and combining the set of training image tiles to form a training image strip containing the edge.

14. The computerized method of claim 10, wherein said extracting comprises: For each training image strip, generating a plurality of gray level (GL) distributions on the training image strip; And for each GL distribution, identifying a location along the GL distribution having a maximum derivative, the location corresponding to a topological point on the contour of the edge.

15. The computerized method of claim 10, wherein a training image is associated with one or more locations of the one or more edges identified during training, and the training image can be used as a reference image for image registration with a runtime image to create one or more image strips from the runtime image based on the one or more locations in the training image.

16. The computerized method of claim 10, wherein the ML model is trained for a specific metrology application from the group consisting of: critical dimension (CD) metrology, overlay (OVL), measurement based inspection (MBI), critical dimension uniformity (CDU), and lithography process control.

17. A non-transitory computer-readable storage medium tangibly embodying a program of instructions that, when executed by a computer, causes the computer to perform a method of inspecting a semiconductor sample, the method comprising: obtaining a runtime image of a semiconductor sample acquired by an inspection tool; processing the runtime image to create one or more image strips, each image strip containing an edge, and for each image strip, extracting a sequence of topological points representing a contour of the edge in the image strip; providing the topological point sequence of each image strip to a trained machine learning (ML) model for processing, and obtaining an updated topological point sequence as an output of the ML model; as well as Measurement data is obtained on the runtime image using the updated sequence of topological points, wherein the measurement data has improved performance in at least one metrological metric.

18. The non-transitory computer readable storage medium of claim 17, wherein the at least one metrology metric is from the group consisting of: match, accuracy, correlation, and sensitivity.

19. The non-transitory computer-readable storage medium of claim 17, wherein the processing comprises: One or more edges are identified from the runtime image; and for each edge, a set of image tiles are cropped along a set of perpendicular lines relative to the edge; and the set of image tiles are combined to form an image strip containing the edge.

20. The non-transitory computer-readable storage medium of claim 17, wherein the extracting comprises: For each image strip, generating a plurality of gray level (GL) distributions on the image strip; And for each GL distribution, identifying a location along the GL distribution having a maximum derivative, the location corresponding to a topological point on the contour of the edge.

Images