Method for converting measurement data

Abstract

A metrology system and method for transforming metrology data via a trained machine learning (ML) model are described herein. The method includes accessing a first scanning electron metrology (SEM) dataset (e.g., images, contour lights) acquired by a first SEM system and a second SEM dataset acquired by a second SEM system, wherein the first SEM dataset and the second SEM dataset are associated with a patterned substrate. Using the first SEM dataset and the second SEM dataset as training data, a machine learning (ML) model is trained such that the trained ML model is configured to convert a metrology dataset acquired by the second SEM system into a converted dataset having characteristics comparable to metrology data acquired by the first SEM system. Further, a measurement value may be determined based on the converted SEM data.

Description

Technical Field

[0001] This article describes, in general, the processing of measurement data acquired by a measurement system, and more specifically, the processing of measurement data using machine learning models. Background Technology

[0002] Photolithography projection equipment can be used, for example, in the fabrication of integrated circuits (ICs). In this case, a patterning apparatus (e.g., a mask) can contain or provide a pattern (“design layout”) corresponding to a single layer of the IC, and this pattern can be transferred onto a target portion (e.g., comprising one or more dies) on a substrate (e.g., a silicon wafer) by methods such as irradiating a target portion with the pattern on the patterning apparatus, which has been coated with a layer of radiation-sensitive material (“resist”). Typically, a single substrate contains multiple adjacent target portions, to which the pattern is sequentially transferred by the photolithography projection equipment, one target portion at a time. In one type of photolithography projection equipment, the pattern on the entire patterning apparatus is transferred onto one target portion at a time; this type of equipment is generally referred to as a stepper. In an alternative equipment, generally referred to as a step-scanning equipment, a projection beam scans across the patterning apparatus along a given reference direction (“scanning” direction) while the substrate moves parallel to or antiparallel to that reference direction. Different portions of the pattern on the patterning apparatus are gradually transferred to a target portion. Typically, since the photolithography projection apparatus will have a reduction ratio M (e.g., 4), the speed at which the substrate is moved, F, will be 1 / M times that of the projection beam scanning pattern forming apparatus. Further information about the photolithography apparatus described herein can be found, for example, in US 6,046,792, which is incorporated herein by reference.

[0003] Before a pattern is transferred from a patterning apparatus to a substrate, the substrate may undergo various procedures such as coating, resist coating, and soft baking. After exposure, the substrate can undergo other procedures (“post-exposure procedures”) such as post-exposure baking (PEB), development, hard baking, and measurement / inspection of the transferred pattern. This array of procedures is used as the basis for a single layer in the fabrication of a device (e.g., an IC). The substrate can then undergo various processes such as etching, ion implantation (doping), metallization, oxidation, chemical mechanical polishing, etc., all of which are designed to complete a single layer of the device. If multiple layers are required in the device, the entire procedure or its variations are repeated for each layer. Ultimately, the device will exist in each target portion on the substrate. These devices are then separated from each other using techniques such as dicing or sawing, so that individual devices can be mounted on a carrier, connected to pins, etc.

[0004] Therefore, manufacturing apparatuses such as semiconductor devices typically involve processing a substrate (e.g., a semiconductor wafer) using numerous manufacturing processes to form various features and multiple layers of the device. These layers and features are typically fabricated and processed using techniques such as deposition, photolithography, etching, chemical mechanical polishing, and ion implantation. Multiple devices can be fabricated on multiple dies on a substrate and then separated into individual devices. The device fabrication process can be considered a patterning process. A patterning process involves patterning steps, such as optical and / or nanoprinting lithography using patterning apparatus in a photolithography device, to transfer a pattern from the patterning apparatus to the substrate, and typically, but optionally, involves one or more associated patterning processing steps, such as resist development by a developing apparatus, baking of the substrate using a baking tool, etching of the pattern using an etching apparatus, etc. Summary of the Invention

[0005] In semiconductor manufacturing, multiple metrology tools operating on similar measurement principles or with similar mechanisms for generating signals from a patterned substrate can be employed. For example, multiple scanning electron microscopes (SEMs) that determine measurements based on the interaction of an electron beam with a patterned substrate can be used during semiconductor manufacturing. Multiple metrology systems can be used to improve measurement throughput, measurement accuracy, cost efficiency, or for other reasons. For example, measurements can be used to improve the patterning process. While operating similarly, one metrology system may differ from another in various aspects, such as mechanical subsystems, electrical subsystems, electronic devices, signal detection, image acquisition algorithms, image processing algorithms, contour extraction algorithms, or other structures and software. Therefore, using multiple metrology systems can lead to inconsistent measurement results for the same pattern on the same substrate. This disclosure provides a mechanism for transforming metrology data (e.g., SEM images or contours) obtained from a specific metrology tool using a trained machine learning model. This transformed metrology data is matched or equivalent to another metrology tool (e.g., a reference metrology system), thereby allowing measurement consistency between different metrology tools. For example, the converted signal and / or CD is associated with a feature of the patterned substrate, and it is matched with a signal and / or CD acquired by another measurement tool. In other words, the converted measurement data or the measurement derived from the converted measurement data is equivalent to the measurement data or measurement obtained using another measurement tool.

[0006] The measurement system can be, for example, multiple scanning electron microscopes (SEMs). In one embodiment, the mechanism described herein involves training a machine learning (ML) model to transform an SEM image acquired by one SEM system into an image as if acquired by another SEM system. Therefore, physical property measurements of patterned features performed on the transformed image will be similar to measurements performed on images acquired by other SEM systems. In one embodiment, differences between measurement data from different tools (e.g., CD mismatch and / or SEM image mismatch) can be incorporated into the cost function of training the ML model. In the example, measurement data including SEM signals can be used to guide the image transformation. Furthermore, the mechanism involves obtaining a measurement setup (e.g., a CD measurement setup) using training data and applying it to the transformed image to obtain measurements (e.g., CD) associated with the patterned substrate. These measurements are as if acquired by another SEM system (e.g., a reference SEM system).

[0007] In one embodiment, a method is provided for training a machine learning model and using the trained machine learning model to transform measurement data. The method includes: accessing a first SEM dataset acquired by a first scanning electron metrology (SEM) system and a second SEM dataset acquired by a second SEM system, wherein the first and second SEM datasets are associated with a patterned substrate. Using the first and second SEM datasets as training data, a machine learning (ML) model is trained such that the trained ML model is configured to transform the measurement dataset acquired by the second SEM system into a transformed dataset having characteristics equivalent to the measurement data acquired by the first SEM system.

[0008] In one embodiment, the first SEM dataset and the second SEM dataset may be an image set of a patterned substrate, the outline of a feature on the patterned substrate, a physical property associated with a pattern on the patterned substrate, or a combination thereof. In one embodiment, the physical property includes the critical dimension (CD) of the pattern on the patterned substrate.

[0009] In one embodiment, training an ML model involves: comparing a first SEM dataset and a second SEM dataset; and adjusting the parameters of the ML model based on the comparison to influence the cost function used to train the ML model.

[0010] In one embodiment, training an ML model involves signal-to-signal matching or CD-to-CD matching. For example, training an ML model involves: comparing a first CD value from a first SEM dataset with a second CD value from a second SEM dataset; and adjusting the parameters of the ML model based on the comparison to influence the cost function used to train the ML model, thereby improving the CD matching between the first and second SEM datasets, the cost function being a function of the first and second CD values.

[0011] In one embodiment, the method may further involve: receiving a measurement matching scheme of a measurement system based on a first SEM dataset and physical property measurements of a substrate from a first SEM system; and applying the measurement matching scheme to the transformed measurement data to determine another physical property measurement. In one embodiment, the measurement matching scheme includes a CD threshold, which indicates the location where a CD measurement value is acquired on the captured measurement data.

[0012] In one embodiment, determining the measurement selection scheme involves: extracting a contour from an image of a first SEM dataset via a first contour extraction algorithm; drawing a cutting line at a location on the contour to measure CD; and determining a CD threshold corresponding to the measured CD based on a signal along the cutting line.

[0013] In one embodiment, a measurement system is provided. The measurement system includes a process or computer system comprising one or more processors having a trained machine learning (ML) model stored thereon and programmed with computer program instructions that, when executed, cause the computer system to: capture measurement data of a patterned substrate; and convert the captured measurement data into transformed measurement data via the trained ML model, the transformed measurement data having characteristics similar to those captured by another measurement system.

[0014] According to an embodiment, a computer system is provided, comprising a non-transitory computer-readable medium on which instructions are recorded. When executed by a computer, these instructions perform the method steps described above. Attached Figure Description

[0015] The above aspects, as well as other aspects and features, will become apparent to those skilled in the art when examining the following description of specific embodiments in conjunction with the accompanying drawings, wherein:

[0016] Figure 1 A block diagram of various subsystems of the lithography system according to an embodiment is shown;

[0017] Figure 2This is an exemplary flowchart of a method for converting an image acquired by a particular measurement system into an image as if acquired by another measurement system, according to an embodiment.

[0018] Figure 3 The illustration shows exemplary measurement data acquired by a first measurement system and a second measurement system according to an embodiment;

[0019] Figure 4 The illustration shows an embodiment of the invention. Figure 3 Signals within a portion of the image obtained from the image;

[0020] Figure 5 The illustration shows an exemplary training of an ML model using a generator adversarial network (GAN) according to an embodiment;

[0021] Figure 6A The illustration shows the input of a second image acquired by a second measurement system into an input according to an embodiment. Figure 2 The trained model is used to generate the transformed image;

[0022] Figure 6B The illustration shows an exemplary first image acquired by a first measurement system according to an embodiment, the exemplary image being positioned to... Figure 6A The transformed images are adjacent for comparison;

[0023] Figure 7 The illustration shows an embodiment. Figure 6A and 6B The signals within a portion of the first image, the second image, and the transformed image, wherein the signals of the first image and the transformed image overlap each other, indicating that the transformed image has similar characteristics to the first image;

[0024] Figure 8 This is a block diagram of an exemplary training of a machine learning (ML) model according to an embodiment and the application of the trained ML model to determine measurements, the machine learning (ML) model being configured to convert an image acquired by a second measurement system into an image as acquired by a first measurement system;

[0025] Figure 9 An embodiment of a scanning electron microscope (SEM) according to an example is schematically depicted;

[0026] Figure 10 An embodiment of an electron beam inspection apparatus according to an embodiment is schematically depicted; and

[0027] Figure 11 This is a block diagram of an example computer system according to an embodiment. Detailed Implementation

[0028] Before describing the embodiments in detail, it is helpful to present an example environment in which the embodiments can be implemented.

[0029] While specific references to IC manufacturing may be made herein, it should be clearly understood that the description herein has many other possible applications. For example, it can be used to manufacture integrated optical systems, guide and inspection patterns for magnetic domain memories, liquid crystal display panels, thin-film magnetic heads, etc. Those skilled in the art will understand that, in the context of such alternative applications, any use of the terms “mask,” “wafer,” or “die” herein should be considered interchangeable with the more general terms “mask,” “substrate,” and “target portion,” respectively.

[0030] In this document, the terms “radiation” and “beam” can be used to cover all types of electromagnetic radiation, including ultraviolet radiation (e.g., wavelengths of 365, 248, 193, 157, or 126 nm) and EUV (extreme ultraviolet radiation, e.g., wavelengths in the range of about 5 to 100 nm).

[0031] Patterning apparatuses may include or can form one or more design layouts. Design layouts can be generated using CAD (Computer-Aided Design) programs, a process often referred to as EDA (Electronic Design Automation). Most CAD programs follow a predetermined set of design rules to create functional design layouts / patterning apparatuses. These rules are set by processing and design constraints. For example, design rules define spatial tolerances between devices (such as gates, capacitors, etc.) or interconnects to ensure that devices or lines do not interact with each other in undesirable ways. One or more design rule constraints may be referred to as “critical dimensions” (CDs). A critical dimension of an apparatus can be defined as the minimum width of a line or hole, or the minimum space between two lines or two holes. Therefore, CDs determine the overall size and density of the designed apparatus. Of course, one of the goals in apparatus fabrication is to faithfully reproduce the original design intent on the substrate (via the patterning apparatus).

[0032] As an example, pattern layout design may include the application of resolution enhancement techniques, such as optical proximity correction (OPC). OPC addresses the fact that the final size and placement of an image of the design layout projected onto a substrate will differ from, or depend solely on, the size and placement of the design layout on the patterning apparatus. It should be noted that the terms “mask,” “patterning plate,” and “patterning apparatus” are used interchangeably herein. Furthermore, those skilled in the art will recognize that the terms “mask,” “patterning apparatus,” and “design layout” can be used interchangeably because, in the context of RET, the physical patterning apparatus is not necessarily used, but the design layout can be used to refer to the physical patterning apparatus. Given the small feature size and high feature density present on a design layout, the position of a particular edge of a given feature will be influenced to some extent by the presence or absence of other adjacent features. These proximity effects are caused by minute radiation or non-geometric optical effects (such as diffraction and interference) coupling from one feature to another. Similarly, proximity effects may be caused by diffusion and other chemical effects during post-exposure baking (PEB), resist development, and etching, which typically occurs after photolithography.

[0033] To increase the chances of a projected image of a design layout conforming to the requirements of a given target circuit design, proximity effects can be predicted and compensated for using, for example, complex numerical models, design layout corrections, or pre-distortion. C. Spence's article "Full-Chip Lithography Simulation and Design Analysis - How OPC Is Changing IC Design," published in the 2005 SPIE proceedings, Volume 5751, pages 1-14, provides an overview of current "model-based" optical proximity correction processes. In typical high-end designs, almost every feature of the design layout undergoes some modification to achieve high fidelity of the projected image to the target design. These modifications can include offsets or biases of edge positions or linewidths, as well as the application of "auxiliary" features designed to assist the projection of other features.

[0034] Auxiliary features can be considered as the difference between features on the pattern-forming apparatus and features in the design layout. The terms "primary feature" and "auxiliary feature" do not mean that a specific feature on the pattern-forming apparatus must be labeled as one or the other.

[0035] The term "mask" or "patterning apparatus" as used herein can be broadly interpreted to refer to a general patterning apparatus that can be used to impart a patterned cross-section to an incoming radiation beam, corresponding to a pattern to be created in a target portion of a substrate; in this context, the term "optical valve" may also be used. Examples of other such patterning apparatuses besides classic masks (transmission or reflection masks; binary, phase-shifting, hybrid masks, etc.) include:

[0036] - Programmable mirror arrays. An example of such a device is a matrix-addressable surface with a viscoelastic control layer and a reflective surface. The underlying principle behind this device is that, for example, addressable regions of the reflective surface reflect incident radiation as diffracted radiation, while unaddressed regions reflect incident radiation as non-diffracted radiation. Using appropriate filters, the non-diffracted radiation can be filtered out from the reflected beam, leaving only the diffracted radiation; in this way, the beam is patterned according to the addressing pattern of the matrix-addressable surface. The desired matrix addressing can be performed using suitable electronic components.

[0037] - Programmable LCD array. An example of such a construction is given in U.S. Patent No. 5,229,872, which is incorporated herein by reference.

[0038] As a brief introduction Figure 1 An exemplary photolithography projection apparatus 10A is illustrated. The main component is a radiation source 12A, which can be a deep ultraviolet excimer laser source or other types of sources, including: extreme ultraviolet (EUV) sources (as discussed above, the photolithography projection apparatus itself does not need to have a radiation source); irradiation optics, which, for example, define partial coherence (denoted as σ), and may include optics 14A, 16Aa, and 16Ab for shaping the radiation from source 12A; a pattern forming apparatus 18A; and a transmission optics 16Ac for projecting an image of a pattern from the pattern forming apparatus onto a substrate plane 22A. An adjustable filter or aperture 20A at the pupil plane of the projection optics may limit the range of beam angles striking the substrate plane 22A, where the maximum possible angle defines the numerical aperture NA of the projection optics = n sin(Θ). max ), where n is the refractive index of the medium between the substrate and the last element of the projection optics, and Θ max It is the maximum angle of the beam leaving the projection optics, which can still strike the substrate plane 22A.

[0039] In a photolithography projection apparatus, a source provides illumination (i.e., radiation) to a patterning apparatus, and projection optics guide and shape the illumination onto a substrate via the patterning apparatus. The projection optics may include at least some of components 14A, 16Aa, 16Ab, and 16Ac. The spatial image (AI) is the radiation intensity distribution at the substrate level. A resist layer on the substrate is exposed, and the spatial image, as a potential “resist image” (RI), is transferred to the resist layer. The resist image (RI) can be defined as the spatial distribution of the solubility of the resist in the resist layer. Resist models can be used to calculate the resist image from the spatial image, examples of which can be found in U.S. Patent Application Publication No. 2009-0157360, the disclosure of which is incorporated herein by reference in its entirety. The resist model relates to the properties of the resist layer (e.g., the effects of chemical processes occurring during exposure, PEB, and development). The optical characteristics of the photolithography projection apparatus (e.g., the characteristics of the source, patterning apparatus, and projection optics) specify the spatial image. Since the pattern forming apparatus used in a photolithography projection device can be modified, it may be desirable to separate the optical characteristics of the pattern forming apparatus from the optical characteristics of the rest of the photolithography projection device, which includes at least the source and projection optics.

[0040] While specific references may be made herein to the use of lithography equipment in IC manufacturing, it should be understood that the lithography equipment described herein may have other applications, such as the fabrication of integrated optical systems, magnetic domain memory, liquid crystal displays (LCDs), thin-film magnetic heads, and the guidance and inspection of patterns. Those skilled in the art will appreciate that, in the context of such alternative applications, any use of the terms “wafer” or “die” herein may be considered synonymous with the more general terms “substrate” or “target portion,” respectively. The substrate referred to herein may be processed before or after exposure, for example in a track (a tool typically used to apply a resist layer to a substrate and develop the exposed resist) or in a measurement or inspection tool. Where applicable, this disclosure may be applied to such and other substrate processing tools. Furthermore, the substrate may be processed more than once, for example to create a multilayer IC, such that the term “substrate” as used herein may also refer to a substrate that already contains multiple processed layers.

[0041] The terms “radiation” and “beam” used in this article cover all types of electromagnetic radiation, including ultraviolet (UV) radiation (e.g., wavelengths of 365, 248, 193, 157, or 126 nm) and extreme ultraviolet (EUV) radiation (e.g., wavelengths in the range of 5 to 20 nm) as well as particle beams, such as ion beams or electron beams.

[0042] In semiconductor manufacturing, multiple measurement systems operating on similar principles can be used to measure one or more physical properties (e.g., CD, EPE, overlap, etc.) of a patterned substrate. While the operating principles may be similar, these measurement systems differ from each other in various aspects, such as mechanical subsystems, electrical subsystems, electronics, image processing algorithms, contour extraction algorithms, or other structural and software components that may cause variations in the measurements. However, it is desirable that these measurements match on patterns on the same patterned substrate or on different patterned substrates executed using different tools. To obtain consistent measurements between different measurement systems, measurement data that matches data acquired by a reference (or reference) measurement system may be required. For example, measurement data of a patterned substrate obtained using a first tool (e.g., a reference measurement system) should match measurement data of the same patterned substrate obtained using a second tool. In this way, variations caused by tool-to-tool differences are significantly reduced or eliminated from variations in measurement results, and these variations advantageously and accurately indicate substrate variations. In one embodiment, the reference measurement system may be from the same supplier or a different supplier than the second measurement system. For example, the tools may include different process models or designs supplied by different or the same manufacturers. In some cases, tools may include the same process model, but their behavior may still be different; that is, there is a tool-to-tool variation.

[0043] In existing methods, after acquiring measurement data (e.g., SEM images or contours of patterned features), measurement system settings (e.g., tunable parameters of a second measurement tool) can be determined to extract measurements (e.g., CD) from the measurement data. For example, the measurement system settings can be determined to obtain the best CD match result with the CD obtained using a reference measurement system. In many cases, even if multiple or all available parameters are tuned, a specification-matched measurement (e.g., CD) may not be achieved. In one embodiment, the measurement settings may include tunable parameters such as dose, field of view (FOV), or other parameters of the second measurement tool. In one embodiment, tunable parameters such as dose and FOV may not be modified to obtain a CD-to-CD match between measurements from the first and second tools. For example, the second measurement tool may use a higher dose, have a faster image acquisition speed, or a larger FOV compared to the first measurement tool. These tunable parameters facilitate faster measurement. Moreover, changing such parameters may result in different charging effects, leading to additional differences in the measurements. Therefore, in one embodiment, advantageous tunable parameters such as velocity and FOV may remain unchanged, while other parameters, such as the CD threshold applied during CD measurement via SEM images, may be modified. Thus, matching of measurement results from the first measurement tool can be achieved while maintaining the advantageous settings of the second measurement system.

[0044] This disclosure provides a mechanism for transforming measurement data (e.g., SEM images or contours) obtained from a second measurement tool using a machine learning model. The transformed measurement data is matched to, or equivalent to, measurement data obtained from, another measurement tool (e.g., a reference measurement system). For example, the mechanism herein includes training a machine learning (ML) model to transform an image obtained by one SEM system into an image to be obtained by another SEM system. In one embodiment, the ML model can transform signals, CD values, or other characteristics determined by the SEM system into those to be obtained by another SEM system. Thus, any measurement of the physical properties of patterned features performed on the transformed image will be equivalent to (e.g., matched to) a measurement performed on an image obtained by another SEM system. In one embodiment, differences between measurement data from different tools (e.g., CD mismatch and / or SEM image mismatch) can be incorporated into the cost function of training the ML model. As an example, measurement data including SEM image signals (e.g., intensity values) can be used to guide the transformation.

[0045] Furthermore, this mechanism involves obtaining measurement settings (e.g., measurement system settings, such as a CD threshold for obtaining CD measurements) using reference measurement data (e.g., from a reference measurement system). These settings can be applied to the transformed image to obtain measurements (e.g., CD) associated with the patterned substrate, such that the measurements are as if obtained by, for example, a reference measurement system. In one embodiment, the measurement system can be a different SEM system, such as a reference SEM system and another different SEM system. Example measurement system relative to Figure 9 and 10 Illustrations and discussion. In one embodiment, the mechanism described herein can be used with images captured by such a measurement system, such as SEM images.

[0046] The mechanism disclosed herein has several advantages. For example, consistent measurements associated with similar patterns can be obtained using different measurement tools. Even if the measurement tools use different algorithms to acquire measurements, extract contours from images, perform image enhancement to identify features or contours, segment images or contours, align images or contours with a reference, derive measurements, etc., the final measurement obtained after applying the mechanism of this invention will provide a close match with the measurement from the reference measurement system. In one embodiment, the match between the first measurement data and the second measurement data can be characterized by the difference between the first measurement data and the second measurement data, statistics associated with the measurement data, detection intensity values ​​within the measurement data, or other matching parameters. As an example, a close match refers to a match relative to a difference threshold, reference statistics, reference intensity values, or other methods used to characterize the matching reference data.

[0047] The mechanism described in this paper also enables the faster production of semiconductor chips using different measurement systems without causing significant variations in the measurements of the patterned substrate. Therefore, adjustments to the patterning process based on measurements from a reference measurement system can remain substantially the same, thus maintaining a similar production setup for the patterning process to achieve the desired throughput.

[0048] Figure 2 This is an exemplary flowchart of a method 300, according to an embodiment, for converting measurement data acquired by a particular measurement system into measurement data having characteristics similar to those acquired by another measurement system. In one embodiment, method 300 involves training an ML model to transform the measurement data. The ML model may be a convolutional neural network (CNN), a deep convolutional neural network (DNN), a generative adversarial network (GAN), or other types of neural networks. Training may be based on a measurement dataset of the same patterned substrate but acquired by different measurement systems. The ML model learns the differences in measurement data from different systems, which may be due to unknown algorithms, different measurement fitting schemes, different measurement tool configurations or operating principles, etc. During training, the trained ML model can predict measurement data measured using different systems (e.g., a reference measurement system). Therefore, when different measurement systems are used, the trained ML model can be advantageously used to improve the consistency of patterned substrate measurements. Exemplary implementations of method 300 involve processes P301 and P303 for training the ML model. Furthermore, processes P305, P307, P309, and P311 may be included to apply trained ML models to transform measurement data and predict measurements of the patterned substrate. These processes are discussed in more detail below.

[0049] Process P301 involves acquiring training data for the same patterned substrate (e.g., a patterned training substrate) from different metrology tools. In one embodiment, process P301 involves accessing a first metrology dataset MD1 acquired by a first metrology system TS1 and a second metrology dataset MD2 acquired by a second metrology system TS2. For example, accessing a first SEM dataset acquired by a first scanning electron metrology (SEM) system and a second SEM dataset acquired by a second SEM system. The first metrology dataset MD1 (e.g., the first SEM dataset) and the second metrology dataset MD2 (e.g., the second SEM dataset) are associated with the same patterned substrate. In one embodiment, the metrology dataset used to train the ML model may be referred to as training data, and the patterned substrate may be referred to as the training substrate.

[0050] In one embodiment, the first measurement dataset MD1 and the second measurement dataset MD2 may include sets of images (e.g., SEM images) of a patterned substrate (or training substrate) obtained from a first measurement system TS1 (e.g., a reference measurement system, such as SEM) and a second measurement system TS2 (e.g., another SEM), respectively. Therefore, the first measurement dataset MD1 may be referred to as the first SEM image set, and the second measurement dataset MD2 may be referred to as the second SEM image set. In one embodiment, the first measurement dataset MD1 and the second measurement dataset MD2 may include contours of features on the patterned substrate. In one embodiment, the contours may be extracted from images of the patterned substrate (e.g., SEM images). In one embodiment, the first contour extracted by the first measurement system TS1 may employ a first extraction algorithm (e.g., an unknown algorithm with fixed or untunable parameters), and the second contour extracted by the second measurement system TS2 may employ a second extraction algorithm different from the first algorithm (e.g., a known algorithm with tunable parameters). In one embodiment, the first measurement dataset MD1 and the second measurement dataset MD2 may include physical properties (e.g., CD, overlap, etc.) associated with the pattern on the patterned substrate.

[0051] In one embodiment, a first measurement system may output CD measurements, but the implementation details (e.g., algorithms, image processing, etc.) used to determine the CD measurements may be unknown. This disclosure provides a mechanism for adjusting one or more parameters of a second measurement system to match CD measurements from the first measurement system (e.g., generating a transformed image using a trained ML model). For example, if the SEM image captured using the second measurement system differs from that captured by the first measurement system, simply tuning the parameters of the second measurement system may not achieve satisfactory CD-to-CD matching results. However, transforming the captured image of the second measurement system using the trained ML model described herein, and tuning parameters (e.g., CD thresholds) based on this transformed image, can advantageously achieve the desired CD-to-CD matching or other measurement matching results.

[0052] Figure 3The illustration shows exemplary measurement data of a patterned substrate W acquired by a first measurement system TS1 and a second measurement system TS2 according to an embodiment. In this example, the first measurement system TS1 may be a reference measurement system, such as a first SEM. The first measurement system TS1 may capture a first image IMG1 (e.g., an SEM image) of the patterned substrate W or a portion thereof. As an example, the SEM image IMG1 includes a feature F1 (represented by a light-colored elliptical portion in the image IMG1) corresponding to a feature (e.g., F, not marked on the substrate W) patterned on the substrate W. In one embodiment, a cut line CL1 may be drawn across the feature F1 to measure the intensity value of a pixel along the cut line CL1. These intensity values ​​are represented as a first signal S1 (see [reference]). Figure 5 ).

[0053] Similarly, the second measurement system TS2 can be another measurement system different from the first SEM, such as a second SEM. The second measurement system TS2 can capture a second image IMG2 (e.g., another SEM image) of a patterned substrate W or a portion of the patterned substrate W. For example, the SEM image IMG2 includes a feature F1' (represented by a light-colored elliptical portion in image IMG2) corresponding to the same feature (e.g., F, not marked on the substrate W) patterned on the substrate W. In one embodiment, a cut line CL1 can be drawn on feature F1' to measure the intensity values ​​of pixels along the cut line CL1. These intensity values ​​are represented as a second signal S2 (see [link to documentation]). Figure 5 ).

[0054] exist Figure 5In this context, comparing a first signal S1 and a second signal S2 patterned on a substrate W for the same feature indicates that the first measurement system TS1 and the second measurement system TS2 can produce different measurements for the same feature. For example, when CD is measured based on the first signal S1 along a cut line CL1, the measured CD will differ from the CD measured based on the second signal S2. As an example, CD can be measured based on a CD threshold defined as a function of the distance between adjacent peaks in the signal. In some embodiments, the systems may have different design configurations (e.g., different product models) supplied by the same or different manufacturers. Differences in signals for the same feature may be associated with differences in tooling structure (such as system differences in design with mechanical, electrical, and electronic configurations), differences in algorithms employed by systems TS1 and TS2 (such as signal acquisition mechanisms, signal processing algorithms, image enhancement mechanisms, etc.), differences in the selection schemes employed by each system TS1 and TS2 during measurement (such as contour extraction algorithms, cut line placement algorithms, etc.), or other known or unknown differences between systems TS1 and TS2. In some embodiments, the systems may have the same design configuration (e.g., the same product model), and measurement differences may be attributed to unintended system variations caused by changes in manufacturing, assembly, and / or system drift, etc.

[0055] Return to reference Figure 3 A contour C1 corresponding to feature F1 in image IMG1 can be extracted. In one embodiment, contour C1 can be extracted using a first algorithm implemented in a first measurement system TS1. In one embodiment, the implementation details of the first contour extraction algorithm may be unknown or fixed, because such parameters within the algorithm may not be adjustable when extracting contour C1. In one embodiment, the measurement system configuration may be fixed, as it is considered a reference and knowledge of specific differences or the causes of differences may not be required. In one embodiment, such contour C1 can be used as first measurement data. Similarly, another contour C2 corresponding to feature F1' in image IMG2 can be extracted. In one embodiment, contour C2 can be extracted using a second algorithm implemented in a second measurement system TS2. In one embodiment, the implementation details of the extraction algorithm may be known, and one or more parameters of the second algorithm (e.g., intensity thresholds characterizing feature edges, extraction model parameters such as Gaussian σ, etc.) may be adjustable during contour adjustment.

[0056] In one embodiment, based on the extracted contour C1, characteristic values ​​such as CD of feature F1 can be determined. In one embodiment, CD can be determined based on the average distance between the two ends along the length of feature F1. Similarly, based on the extracted contour C2, values ​​of physical properties such as CD of feature F1' can be determined. In one embodiment, the same algorithm can be used to determine the CD values ​​of features F1 and F1'. This disclosure is not limited to a particular measurement method. For example, CD can be measured in a linear or threshold mode of the measurement system. In another example, the maximum, minimum, or average value of several neighboring locations of the feature can be used to determine the CD value. In yet another example, shape fitting of contours C1 and C2 can be performed. In yet another example, parameters such as intensity thresholds, signal smoothing window size, starting points for finding edge locations of features, or other tunable parameters can be applied to the desired CD-to-CD matching result.

[0057] This disclosure is not limited to extracting measurements using specific techniques. CD measurements determined based on contour extraction and adjustment algorithms are merely exemplary and do not limit the scope of this disclosure. In one embodiment, the measurement system may employ image processing algorithms and / or CD measurement algorithms to determine the CD of a feature without extracting the contours of the patterned features on the substrate from an image.

[0058] However, due to the inherent differences between the images IMG1 and IMG2, which derive contours C1 and C2 respectively, the measurements CD of the same features patterned on the substrate W will be different.

[0059] In this disclosure, the trained ML model is configured to transform the second measurement data (e.g., the second signal S2) such that it closely matches the first measurement data (e.g., the first signal S1). Therefore, any measurement performed using the transformed measurement data (e.g., a transformed version of the second signal S2) can have similar measurement values.

[0060] Return to reference Figure 2 Process P303 involves training an ML model based on a first measurement dataset MD1 and a second measurement dataset MD2. After the training process is completed, the trained ML model TML is configured to convert the measurement dataset obtained by the second measurement system into a transformed dataset, which has the same characteristics as that obtained by the first measurement system TS1 (e.g., a reference measurement system).

[0061] In one embodiment, training the ML model involves comparing a first measurement dataset MD1 and a second measurement dataset MD2 of a patterned training substrate. Based on the comparison, the parameters of the ML model can be adjusted to influence the cost function used to train the ML model. For example, the cost function could be a function of the difference between the first measurement dataset MD1 and the second measurement dataset MD2. As the ML model parameters (e.g., weights) are adjusted, the cost function value (e.g., the difference) is gradually reduced. In one embodiment, the cost function is minimized. In one embodiment, adjusting the ML model parameters is stopped when a given number of iterations is reached, when the cost function value is within a desired threshold, when the cost function value does not decrease significantly in subsequent iterations, or other stopping criteria.

[0062] In one embodiment, training an ML model is performed to obtain CD-to-CD or signal-to-signal matching between a first SEM dataset and a second SEM dataset. For example, training the ML model involves: comparing a first CD value of the first SEM dataset with a second CD value of the second SEM dataset; and adjusting the parameters of the ML model based on the comparison to influence the cost function used to train the ML model, thereby improving the CD matching between the first SEM dataset and the second SEM dataset, the cost function being a function of the first CD value and the second CD value.

[0063] In one embodiment, training the ML model involves using a training dataset that includes measurement data aligned with the same design layout. For example, a first image set (e.g., a first SEM image) of a first measurement dataset MD1 may be aligned with a design layout image. As another example, a first contour may be aligned with a design contour of the design layout. Similarly, a second image set (e.g., another SEM image) of a second measurement dataset MD2 may be aligned with a design layout image, or a second contour may be aligned with a design contour of the design layout. In one embodiment, the aligned first image set (or aligned first contour) and the aligned second image set (or aligned second contour) may be used as training data to train the ML model.

[0064] As an example, training an ML model involves comparing intensity values ​​from a first image set and a second image set; and adjusting the parameters of the ML model based on the comparison to influence the cost function used to train the ML model. In one embodiment, the intensity value of each pixel in the first image set may be compared with the intensity value of the corresponding pixel in the second image set.

[0065] In one embodiment, training the ML model is an iterative process. Each iteration involves (i) applying adjusted parameters to the ML model to transform the second measurement dataset MD2 into a transformed measurement dataset MD2'; (ii) comparing the intensity values ​​from the first measurement dataset MD1 and the transformed measurement dataset MD2'; (iii) further adjusting the parameters of the ML model based on the comparison to influence the cost function (e.g., within the desired threshold, or minimized); (iv) determining whether the cost function is within the desired threshold (or minimized); and (v) repeating steps (i) to (iv) in response to the cost function not being within the desired threshold (or minimized).

[0066] In one embodiment, the cost function may include the difference between measurement datasets MD1 and MD2. For example, the cost function may include the difference between a first intensity value from the first measurement dataset MD1 and a corresponding second intensity value from the second measurement dataset MD2 or a transformed dataset MD2'. In one embodiment, the intensity value may be determined by a cut line drawn on the feature using the first measurement dataset MD1 and the second measurement dataset MD2. In one embodiment, determining the intensity value includes determining the pixel intensity value along the cut line using the first measurement dataset MD1 and the second measurement dataset MD2.

[0067] In one embodiment, determining the intensity value involves: applying a first contour extraction algorithm associated with a first measurement system TS1 to a first measurement dataset MD1; and applying a second contour extraction algorithm associated with a second measurement system TS2 to a second measurement dataset MD2.

[0068] In one embodiment, the process of training the ML model P303 employs a generative adversarial network. In this example, the training process involves training a generator model in conjunction with a discriminator model. The generator model can be trained using a second measurement dataset MD2 as input to generate data similar to a first measurement dataset MD1; and the discriminator model is trained to distinguish the generated data from the first measurement dataset MD1. The generator model and the discriminator model can be trained collaboratively, such that the generator model generates realistic data (e.g., matching the reference data), while the discriminator model classifies such generator model data as potentially unrealistic.

[0069] An exemplary training process for GANs is an iterative process. Each training iteration may include step (i) for randomly selecting training samples from the training data (e.g., MD2 and MD1). Figure 4The process involves several aligned images (IMG1 and IMG2) and step (ii) for training a generator model to transform selected input data from MD2 into transformed data MD2', and training a discriminator model using the transformed data MD2' and reference data MD1. The cost function used to train the discriminator model can be configured to adjust the weights of the discriminator model such that the discriminator model classifies the transformed data MD2' as fake (or false) and the reference data MD1 as real (or true). For example, the discriminator-related cost function can maximize the probability of assigning to real and fake images. The cost function of the generator model can include two parts, for example, configured to minimize the probability that the generator output is classified as fake by the discriminator. The first part can include configured terms that cause the discriminator model to label the transformed data MD2' as real (or true). The second part can include terms such as the intensity difference per pixel between images MD2' and MD1. The second part can be reduced (e.g., minimized) to cause the generator model to generate realistic transformed data MD2'. The above training process (e.g., steps (i) and (ii)) is repeated until the desired result is obtained.

[0070] Figure 5 The illustration shows an exemplary training of an ML model using a Generative Adversarial Network (GAN) according to an embodiment. As an example, the ML model includes a generator model GM and a discriminator model DM. The generator model GM can be configured to receive measurement data as input and generate data similar to reference data. For example, the generator model GM is configured to receive data from a second measurement system (e.g., ...). Figure 3 The measurement data (e.g., SEM image IMG2 or contour) acquired by the first measurement system (e.g., TS2) is used to generate measurement data (e.g., PIMG2 or contour). In one embodiment, the generated measurement data has characteristics (e.g., intensity, feature shape, feature size, etc.) as if the generated measurement data were obtained from a first measurement system (e.g., TS2). Figure 3 The discriminator model DM is configured to receive generated data from the generator model GM and determine whether the generator data is similar to the reference data. For example, the discriminator model DM is configured to receive generated measurement data (e.g., PIMG2) and determine whether the generated measurement data (e.g., PIMG2) is similar to the reference data (e.g., SEM image IMG1). During training, the generator model GM and the discriminator model DM compete with each other, causing the generator model GM to gradually generate realistic images, and the discriminator model to attempt to classify these generated images as fake (fake).

[0071] exist Figure 5In the example shown in (A), the discriminator model DM can use the generated second image PIMG2 as input and the first image IMG1 of the substrate (e.g.) Figure 3 The discriminator model DM is trained using the reference data (W) as input data. In one embodiment, the discriminator model DM classifies the input image into a first category (e.g., labeled as real) or a second category (e.g., labeled as fake). For example, the first category refers to the reference data, and the second category refers to the data generated by the model (e.g., GM). In one embodiment, the weights WTS of the discriminator model DM can be adjusted until the DM model determines that the generated image PIMG2 is real. In one embodiment, real refers to the label assigned to the generated image PIMG2 by the discriminator model DM, where real indicates that the generated image PIMG2 is similar to the first image IMG1. In one embodiment, the adjustment of the weights WTS of the discriminator model DM is guided by a cost function CF. In one embodiment, the cost function CF can be a function of the difference between the first image IMG1 and the generated image PIMG2. In one embodiment, this difference can be characterized by labels (e.g., real vs. fake, intensity differences in the image, etc.).

[0072] exist Figure 5 In the example shown in (B), the generator model GM can use a second image of the substrate, IMG2 (e.g., ...). Figure 3 The first image IMG1 (e.g., W) is used as the input and substrate. Figure 3 The input image (e.g., IMG2) is used as reference data for training. In one embodiment, the generator model GM transforms the input image (e.g., IMG2) into an image PIMG2 with similar properties (e.g., image intensity, intensity slope, or other image properties) to the reference data (e.g., IMG1).

[0073] In one embodiment, the weights WTS' of the generator model GM can be adjusted based on the cost function CF. In one embodiment, the cost function CF can be a function of the difference between the first image IMG1 and the generated image PIMG2 and whether the discriminator model DM classifies the generated image PIMG2 into a second category or a first category. In one embodiment, the cost function guides the adjustment of the weights WTS' of the generator model GM such that the difference between images PIMG2 and IMG1 is reduced (in one embodiment, minimized), and the generated image PIMG2 is classified by the discriminator model DM into the first category (e.g., real).

[0074] In one embodiment, a trained ML model TML can be applied to transform captured measurement data using a measurement tool that trained the ML model. As an example, method 300 may also involve processes P305 and P307. Process P305 involves capturing measurement data 310 of a patterned substrate (e.g., a substrate different from the one used to train the ML model) via a second measurement system TS2. Process P307 involves converting the captured measurement data 310 into transformed measurement data 311 via the trained ML model TML. The transformed measurement data 311 of the patterned substrate has the same characteristics as that captured by the first measurement system TS1.

[0075] Figure 6A The illustration shows a second image 601 acquired by a second measurement system (e.g., TS2) being input into a trained ML model TML (e.g., trained according to method 300) to generate a transformed image 605, according to an embodiment. Figure 6B An exemplary first image 610 acquired by a first measurement system is shown. The first image 610 and the second image 601 are associated with the same substrate. Comparison of the converted image 605 and the first image 610 indicates that these images closely match each other. For example, Figure 7 The illustration shows the signals (e.g., intensity values) drawn along cut lines on features in each of images 601, 605, and 610. A comparison is made from the second image 601 (in... Figure 6A The signal S10 obtained from the first image 610 and the signal S30 obtained from the converted image 605 show a significant difference in intensity values. On the other hand, the signals S20 and S30 obtained from the converted image 605 are closely matched. For example, whether the signals are closely matched can be determined by acquiring the differences between the signals. Therefore, advantageously, the converted image 605 can be used to perform measurements associated with the physical properties of the features in the image 605.

[0076] In one embodiment, the measurement matching scheme can be determined based on first measurement data acquired from a first measurement system TS1. The measurement matching scheme can also be used to transform measurement data obtained from a trained ML model TML to determine measurements of physical properties (e.g., CD, overlap, etc.) of a pattern on a patterned substrate. As an example, method 300 may also involve processes P307 and P311. Process P309 involves determining a measurement matching scheme 315 for a second measurement system TS2 based on a first measurement dataset MD1 and physical property measurements PC1 of the patterned substrate from the first measurement system TS1. For example, measurement matching scheme 315 includes a CD threshold indicating the location where CD measurements are acquired on captured measurement data 311. In one embodiment, determining measurement matching scheme 315 involves: extracting a contour from an image of the first measurement dataset MD1 via a first contour extraction algorithm; and cutting lines (e.g., at locations on the contour) at locations on the contour. Figure 3 CD is measured on line CL1; and based on signals along the cutting line (e.g. Figure 4 The first signal S1 in the CD is used to determine the CD threshold corresponding to the measured CD.

[0077] In one embodiment, measurement data (e.g., 310) of the patterned substrate can be captured using a second measurement system TS2, and the captured measurement data (e.g., 311) can be transformed using a trained ML model TML. Process P311 can be performed on the transformed measurement data 311. Process P311 involves applying a measurement matching scheme 315 to the transformed measurement data 311 to determine a physical property measurement PC2 of the patterned substrate. For example, the physical property measurement PC2 can be a critical dimension (CD) measurement, an overlap measurement, an edge placement error, or other characteristics associated with the patterned substrate.

[0078] Figure 8 This is a block diagram illustrating exemplary operations performed according to embodiments for transforming measurement data and determining measurements using the transformed measurement data. For example, measurement data acquired by a measurement system (e.g., SEM) is transformed to match the measurement behavior of a reference measurement system. In one embodiment, operation 801 involves obtaining training data for training an ML model, wherein the ML model transforms the measurement data to be equivalent to the measurement data acquired by the reference measurement tool. In one embodiment, the training data includes data relative to process P301 (in... Figure 3The measurement dataset is obtained as discussed in the discussion. For example, the measurement dataset includes images IMG1 and IMG2 obtained from a first measurement system TS1 and a second measurement system TS2, respectively. In one embodiment, images IMG1 and IMG2 can be aligned with a design layout to generate aligned images AIMG1 and AIMG2 to be used as training data.

[0079] In operation 803, the training data was used to train the ML model. According to procedure P303 (in... Figure 3 The ML model can be trained. After training, the trained ML model TML is obtained.

[0080] In operation 805, measurement data 810 (e.g., SEM image or contour) of the patterned substrate to be measured can be obtained. For example, measurement data 810 is obtained from a measurement system other than a reference measurement system. Furthermore, measurement data 810 can be input into a trained ML model TML to convert measurement data 810 into transformed data 820.

[0081] In operation 811, reference measurement data (e.g., IMG1) and reference measurement values ​​(e.g., the CD value associated with IMG1) can be obtained from a reference measurement system. Based on this data, a measurement matching scheme R1 can be determined. For example, the measurement matching scheme includes a CD threshold, which is applied to a signal along a cut line on a feature in the image IMG1 to determine the CD value of that feature.

[0082] In operation 813, measurement matching scheme R1 can be applied to the converted measurement data 820 to determine the measured value 830 of the physical properties (e.g., CD) of the patterned substrate. Since the converted measurement data is equivalent to the data acquired by the reference measurement system, and measurement matching scheme R1 also corresponds to the reference measurement system, the measured value 830 will be equivalent to the data acquired by the reference measurement instrument. Therefore, advantageously, the combination of the converted measurement data 820 and measurement matching scheme R1 provides a measured value consistent with the reference measurement system.

[0083] In some embodiments, the inspection or measurement device may be a scanning electron microscope (SEM), which produces images of structures (e.g., some or all of the structure of a device) exposed to or transferred onto a substrate. Figure 9 An embodiment of the SEM tool is described. The primary electron beam EBP emitted from the electron source ESO is focused by the condenser lens CL and then passes through the beam deflectors EBD1, EBD2, and objective lens OL to illuminate the substrate PSub on the substrate stage ST at the focal point.

[0084] When the substrate PSub is irradiated with an electron beam EBP, secondary electrons are generated from the substrate PSub. These secondary electrons are deflected by the ExB deflector EBD2 and detected by the secondary electron detector SED. A two-dimensional electron beam image can be obtained by detecting the electrons generated from the sample in sync with, for example, a two-dimensional scan of the electron beam by the electron beam deflector EBD1, or repeated scans of the electron beam EBP by the electron beam deflector EBD1 in the X or Y direction, and the continuous movement of the substrate stage ST across the substrate PSub in the other direction of the X or Y direction.

[0085] The signal detected by the secondary electronic detector (SED) is converted into a digital signal by an analog-to-digital (A / D) converter (ADC), and this digital signal is sent to the image processing system (IPU). In one embodiment, the image processing system IPU may have a memory (MEM) to store all or part of the digital image for processing by the processing unit (PU). The processing unit (PU) (e.g., specially designed hardware or a combination of hardware and software) is configured to convert or process the digital image into a dataset representing the digital image. Further, the image processing system IPU may have a storage medium (STOR) configured to store the digital image and the corresponding dataset in a reference database. A display device (DIS) may be connected to the image processing system IPU, allowing an operator to perform necessary operations on the device using a graphical user interface.

[0086] As mentioned above, SEM images can be processed to extract contours describing the edges of objects representing a device structure within the image. These contours are then quantized using metrics such as CD. Therefore, images of device structures are typically compared and quantized using simple metrics such as edge-to-edge distance (CD) or simple pixel differences between images. Typical contour models for detecting object edges in an image to measure CD use image gradients. In fact, these models rely on strong image gradients. However, in practice, images are often noisy and have discontinuous boundaries. Techniques such as smoothing, adaptive thresholding, edge detection, erosion, and dilation can be used to process the results of image gradient contour models to address noisy and discontinuous images, but ultimately result in low-resolution quantization of high-resolution images. Therefore, in most instances, the mathematical operations used to reduce noise and automatic edge detection in device structure images lead to a loss of image resolution, resulting in information loss. The result is thus low-resolution quantization, which is equivalent to a simplified representation of a complex high-resolution structure.

[0087] Therefore, it is desirable to have a mathematical representation of a structure (e.g., circuit features, alignment marks, or measurement target portions (e.g., grating features)) generated or anticipated using a patterning process, whether, for example, the structure is in a potential resist image, in a developed resist image, or as a layer transferred to a substrate, for example, by etching, that maintains resolution and also describes the general shape of the structure. In the context of photolithography or other patterning processes, the structure can be a device being manufactured or a portion thereof, and the image can be an SEM image of the structure. In some instances, the structure can be a feature of a semiconductor device (e.g., an integrated circuit). In this case, the structure can be referred to as a pattern or desired pattern comprising multiple features of the semiconductor device. In some instances, the structure can be an alignment mark or a portion thereof (e.g., a grating of alignment marks) used in alignment measurements to determine the alignment of an object (e.g., a substrate) with another object (e.g., a patterning apparatus) or a measurement target or a portion thereof (e.g., a grating of the measurement target), which is used to measure parameters of the patterning process (e.g., overlap, focus, dose, etc.). In one embodiment, the measurement target is used to measure, for example, an overlapping diffraction grating.

[0088] Figure 10 Another embodiment of the inspection apparatus is schematically illustrated. The system is used to inspect a sample 90 (such as a substrate) on a sample stage 88 and includes a charged particle beam generator 81, a condenser lens module 82, a probe forming objective module 83, a charged particle beam deflection module 84, a secondary charged particle detector module 85, and an image forming module 86.

[0089] A charged particle beam generator 81 generates a primary charged particle beam 91. A condenser lens module 82 converges the generated primary charged particle beam 91. A probe forming objective module 83 focuses the converged primary charged particle beam into a charged particle beam probe 92. A charged particle beam deflection module 84 scans the formed charged particle beam probe 92 on the surface of a region of interest on a sample 90 fixed on a sample stage 88. In one embodiment, the charged particle beam generator 81, the condenser lens module 82, and the probe forming objective module 83, or their equivalents, alternatives, or any combination thereof, together form a charged particle beam probe generator that generates the scanning charged particle beam probe 92.

[0090] The secondary charged particle detector module 85 detects secondary charged particles 93 emitted from the sample surface (and possibly other reflected or scattered charged particles from the sample surface) when bombarded by the charged particle beam probe 92, generating a secondary charged particle detection signal 94. An image forming module 86 (e.g., a computing device) is coupled to the secondary charged particle detector module 85 to receive the secondary charged particle detection signal 94 from the secondary charged particle detector module 85 and thus form at least one scan image. In one embodiment, the secondary charged particle detector module 85 and the image forming module 86, or their equivalents, alternatives, or any combination thereof, together form an image forming apparatus that forms a scan image from the detected secondary charged particles emitted from the sample 90 bombarded by the charged particle beam detector 92.

[0091] In one embodiment, the monitoring module 87 is coupled to the image forming module 86 of an image forming apparatus to monitor, control, and / or derive parameters for patterning process design, control, monitoring, etc., using scanned images of the sample 90 received from the image forming module 86. Therefore, in one embodiment, the monitoring module 87 is configured or programmed to cause the execution of the methods described herein. In one embodiment, the monitoring module 87 includes a computing device. In one embodiment, the monitoring module 87 includes a computer program to provide the functionality described herein and is encoded on a computer-readable medium that forms or is disposed within the monitoring module 87.

[0092] In one embodiment, compared with using a probe to inspect the substrate... Figure 9 Similar to electron beam inspection tools, Figure 10 Electronic currents in systems such as Figure 9 The depicted CD SEM is significantly larger, resulting in a sufficiently large probe spot for rapid detection. However, due to the larger probe spot, the resolution may not be as high as that of a CD SEM. In one embodiment, the inspection device discussed above can be a single-beam or multi-beam device, without limiting the scope of this disclosure.

[0093] From, for example Figure 9 and / or Figure 10 The system's SEM images can be processed to extract contours describing the edges of objects representing the device structure in the image. These contours are then typically quantized via user-defined metrics such as CD at the cut lines. Therefore, images of the device structure are typically compared and quantized via metrics such as edge-to-edge distance (CD) measured on the extracted contours or simple pixel differences between images.

[0094] In one embodiment, one or more programs of method 300 may be implemented as instructions (e.g., program code) in a processor of a computer system (e.g., process 104 of computer system 100). In one embodiment, the program may be distributed across multiple processors (e.g., parallel computing) to improve computational efficiency. In one embodiment, the computer program product includes a non-transient computer-readable medium on which instructions are recorded, which, when executed by a computer, implement the methods described herein.

[0095] Figure 11 This is a block diagram illustrating a computer system 100 that may assist in implementing the methods, processes, or apparatus disclosed herein. Computer system 100 includes a bus 102 or other communication mechanism for transmitting information and a processor 104 (or multiple processors 104 and 105) coupled to the bus 102 for processing information. Computer system 100 also includes main memory 106, such as random access memory (RAM) or other dynamic storage device, coupled to the bus 102 for storing information and instructions to be executed by the processor 104. Main memory 106 may also be used to store temporary variables or other intermediate information during the execution of instructions to be executed by the processor 104. Computer system 100 also includes read-only memory (ROM) 108 or other static storage device coupled to the bus 102 for storing static information and instructions for the processor 104. Storage device 110, such as a magnetic disk or optical disk, is provided and coupled to the bus 102 for storing information and instructions.

[0096] Computer system 100 can be coupled to display 112, such as a cathode ray tube (CRT) or flat panel or touchpad display, via bus 102 for displaying information to the computer user. Input device 114, including alphanumeric keys and other keys, is coupled to bus 102 for transmitting information and command selections to processor 104. Another type of user input device is cursor control 116, such as a mouse, trackball, or arrow keys, for transmitting directional information and command selections to processor 104 and controlling cursor movement on display 112. This input device typically has two degrees of freedom on two axes (a first axis, e.g., x, and a second axis, e.g., y), allowing the device to specify a position in a plane. Touchpad (screen) displays can also be used as input devices.

[0097] According to one embodiment, portions of one or more methods described herein can be executed by computer system 100 in response to processor 104 executing one or more sequences of one or more instructions contained in main memory 106. Such instructions may be read into main memory 106 from another computer-readable medium, such as storage device 110. Execution of the sequence of instructions contained in main memory 106 causes processor 104 to perform the process steps described herein. One or more processors in a multiprocessor arrangement may also be employed to execute the sequence of instructions contained in main memory 106. In alternative embodiments, hardwired circuitry may be used in place of or in combination with software instructions. Therefore, the description herein is not limited to any specific combination of hardware circuitry and software.

[0098] As used herein, the term "computer-readable medium" refers to any medium that participates in providing instructions to processor 104 for execution. Such media can take many forms, including but not limited to non-volatile media, volatile media, and transmission media. Non-volatile media include, for example, optical discs or magnetic disks, such as storage device 110. Volatile media include dynamic memory, such as main memory 106. Transmission media include coaxial cables, copper wires, and optical fibers, including wires forming bus 102. Transmission media can also take the form of sound waves or light waves, such as those generated during radio frequency (RF) and infrared (IR) data communications. Common forms of computer-readable media include, for example, floppy disks, retractable disks, hard disks, magnetic tape, any other magnetic media, CD-ROMs, DVDs, any other optical media, punched cards, paper tape, any other physical media with a perforated pattern, RAM, PROMs and EPROMs, flash memory EPROMs, any other memory chips or memory cartridges, the carrier waves described below, or any other media from which a computer can read.

[0099] Various forms of computer-readable media may involve carrying one or more sequences of one or more instructions to processor 104 for execution. For example, the instructions may initially be carried on a disk of a remote computer. The remote computer may load the instructions into its dynamic memory and transmit the instructions over a telephone line using a modem. A modem local to computer system 100 may receive data over a telephone line and convert the data into an infrared signal using an infrared transmitter. An infrared detector coupled to bus 102 may receive the data carried in the infrared signal and place the data on bus 102. Bus 102 carries the data to main memory 106, from which processor 104 retrieves and executes the instructions. Instructions received by main memory 106 may optionally be stored on storage device 110 before or after execution by processor 104.

[0100] Computer system 100 may also include a communication interface 118 coupled to bus 102. Communication interface 118 provides bidirectional data communication coupled to network link 120, which is connected to local network 122. For example, communication interface 118 may be an Integrated Services Digital Network (ISDN) card or a modem to provide data communication connectivity with a corresponding type of telephone line. As another example, communication interface 118 may be a Local Area Network (LAN) card to provide data communication connectivity with a compatible LAN. A wireless link may also be implemented. In any such implementation, communication interface 118 transmits and receives electrical, electromagnetic, or optical signals carrying digital data streams representing various types of information.

[0101] Network link 120 typically provides data communication to other data devices via one or more networks. For example, network link 120 may provide a connection to host computer 124 or a data device operated by Internet Service Provider (ISP) 126 via local network 122. ISP 126, in turn, provides data communication services via a global packet data communication network (now generally referred to as "the Internet" 128). Both local network 122 and Internet 128 use electrical, electromagnetic, or optical signals carrying digital data streams. Signals through various networks, as well as signals on network link 120 and through communication interface 118 (carrying digital data to and from computer system 100), are exemplary forms of carrier waves for transmitting information.

[0102] Computer system 100 can send messages and receive data, including program code, via networks(multiple), network link 120, and communication interface 118. In the Internet example, server 130 can transmit request code for an application via Internet 128, ISP 126, local network 122, and communication interface 118. For example, such a download application can provide all or part of the methods described herein. The received code can be executed by processor 104 upon receipt and / or stored in storage device 110 or other non-volatile storage device for later execution. In this way, computer system 100 can obtain application code in carrier form.

[0103] The embodiments disclosed herein may be further described by the following terms.

[0104] 1. A non-transient computer-readable medium having instructions recorded thereon, which, when executed by a computer, implement a method for converting data associated with a measurement system, the method comprising:

[0105] Access a first SEM dataset acquired by a first scanning electron metrology (SEM) system and a second SEM dataset acquired by a second SEM system, the first and second SEM datasets being associated with a patterned substrate; and

[0106] A machine learning (ML) model is trained using the first and second SEM datasets as training data, such that the trained ML model is configured to convert the measurement dataset acquired by the second SEM system into a transformed dataset with characteristics equivalent to the measurement data acquired by the first SEM system.

[0107] 2. The medium according to Clause 1, wherein training the ML model includes:

[0108] Compare the first SEM dataset and the second SEM dataset; and

[0109] The parameters of the ML model are adjusted based on comparisons to influence the cost function used to train the ML model.

[0110] 3. The medium according to Clause 2, wherein the first SEM dataset and the second SEM dataset comprise: an image set of patterned substrates.

[0111] 4. The medium according to Clause 3, wherein training the ML model includes:

[0112] The first image set acquired by the first SEM system and the second image set acquired by the second SEM system are compared; and

[0113] The parameters of the ML model are adjusted based on comparisons to influence the cost function used to train the ML model, thereby improving the matching between the first image set and the ML-generated images that use the second image set as input to the ML model.

[0114] 5. The medium according to Clause 2, wherein the first SEM dataset and the second SEM dataset comprise:

[0115] The outline of features on a patterned substrate; and / or

[0116] Physical properties associated with patterns on patterned substrates.

[0117] 6. The medium according to Clause 4, wherein the physical characteristics include the critical dimension (CD) of the pattern on the patterned substrate.

[0118] 7. The medium according to Clause 5, wherein training the ML model includes:

[0119] Compare the first CD value of the first SEM dataset with the second CD value of the second SEM dataset; and

[0120] The parameters of the ML model are adjusted based on comparisons to influence the cost function used to train the ML model, thereby improving CD matching between the first and second SEM datasets. This cost function is a function of the first and second CD values.

[0121] 8. The medium according to Clause 3, wherein training the ML model includes:

[0122] Align the first image set or first outline of the first SEM dataset with the design layout image or design outline of the design layout.

[0123] Align the second image set or second outline of the second SEM dataset with the design layout image or design outline of the design layout; and

[0124] The first and second sets of aligned images were used as training data for training the machine learning model.

[0125] 9. The medium according to Clause 6, wherein training the ML model includes:

[0126] Compare the intensity values ​​from the first image set and the second image set; and

[0127] The parameters of the ML model are adjusted based on comparisons to influence the cost function used to train the ML model.

[0128] 10. According to the medium of Clause 7, where training the ML model is an iterative process, each iteration includes:

[0129] (i) Transform the second SEM dataset using the adjusted parameters via an ML model and using the first SEM dataset;

[0130] (ii) Compare the intensity values ​​from the first SEM dataset with those from the transformed dataset;

[0131] (iii) Based on comparisons, further adjust the parameters of the ML model to influence the cost function within the expected threshold;

[0132] (iv) Determine whether the cost function is within the expected threshold; and

[0133] (v) In response to the cost function not being within the expected threshold, repeat steps (i) to (iv).

[0134] 11. The medium according to Clause 8, wherein the cost function is the difference between a first intensity value from a first SEM dataset and a corresponding second intensity value from a second SEM dataset or a transformed dataset.

[0135] 12. The medium according to Clause 7, wherein training the ML model includes:

[0136] Intensity values ​​are determined along cutting lines drawn on the features using the first and second SEM datasets.

[0137] 13. The medium according to Clause 10, wherein determining the intensity value includes: determining the pixel intensity value along the cutting line using a first SEM dataset and a second SEM dataset.

[0138] 14. The medium according to Clause 11, wherein determining the strength value includes:

[0139] The first contour extraction algorithm associated with the first SEM system is applied to the first SEM dataset; and

[0140] The second contour extraction algorithm associated with the second SEM system was applied to the second SEM dataset.

[0141] 15. The medium pursuant to Clause 1 also includes:

[0142] Measurement data of another patterned substrate were captured using a second SEM system; and

[0143] The captured measurement data is converted into transformed measurement data by a trained ML model, and the transformed measurement data of another patterned substrate has the same characteristics as those captured by the first SEM system.

[0144] 16. The medium pursuant to Clause 1 also includes:

[0145] Based on the first SEM dataset and the physical property measurements of the patterned substrate from the first SEM system, the measurement and matching scheme of the second SEM system is determined.

[0146] Measurement data of the patterned substrate were captured using a second SEM system;

[0147] The captured measurement data is transformed using a trained machine learning model; and

[0148] The measurement selection scheme is applied to the converted measurement data to determine another physical characteristic measurement value.

[0149] 17. The medium pursuant to Clause 14, wherein the physical property measurements include at least one of the following: critical dimension (CD) measurements, overlap measurements, and edge placement error.

[0150] 18. The medium according to Clause 15, wherein the measurement option includes a CD threshold, which indicates the location on the captured measurement data where CD measurements are acquired.

[0151] 19. The medium as defined in Clause 16, wherein the determination of the measurement selection scheme includes:

[0152] Contours are extracted from images in the first SEM dataset using a first contour extraction algorithm;

[0153] Draw cutting lines at the locations on the contour to measure CD; and

[0154] Based on the signal along the cutting line, the CD threshold corresponding to the measured CD is determined.

[0155] 20. The medium according to Clause 1, wherein the first SEM system is manufactured by the first manufacturer and the second measurement system is manufactured by the second manufacturer.

[0156] 21. The medium according to Clause 1, wherein the ML model is a convolutional neural network.

[0157] 22. The medium of Clause 1, wherein the ML model is trained using a generative adversarial network architecture, the ML model comprising a generator model and a discriminator model.

[0158] 23. The medium according to Clause 20, wherein training the ML model includes:

[0159] Using the second SEM dataset as input, a generator model is trained in conjunction with a discriminator model to generate data similar to the first SEM dataset; and

[0160] A discriminator model is trained to distinguish the generated data from the first SEM dataset.

[0161] 24. A measurement system, comprising:

[0162] A computer system includes one or more processors on which trained machine learning (ML) models are stored and programmed with computer program instructions that, when executed, cause the computer system to:

[0163] Capture measurement data of patterned substrates; and

[0164] The captured measurement data is transformed into converted measurement data by a trained ML model, which has the same characteristics as that captured by another measurement system.

[0165] 25. A measurement system pursuant to Clause 22, wherein a computer system is used to train an ML model, the training comprising:

[0166] Access a first SEM dataset acquired by a first SEM system and a second SEM dataset acquired by a measurement system, the first and second SEM datasets being associated with a training substrate; and

[0167] A machine learning (ML) model is trained using the first and second SEM datasets as training data, such that the trained ML model is configured to convert the measurement datasets acquired by the measurement system into a transformed dataset with characteristics equivalent to the measurement data acquired by the first SEM system.

[0168] 26. The measurement system according to Clause 23, wherein training the ML model includes:

[0169] Compare the first SEM dataset and the second SEM dataset; and

[0170] The parameters of the ML model are adjusted based on comparisons to influence the cost function used to train the ML model.

[0171] 27. The measurement system according to Clause 24, wherein the first SEM dataset and the second SEM dataset comprise: an image set of the training substrate.

[0172] 28. The measurement system according to Clause 27, wherein training the ML model includes:

[0173] The first image set acquired by the first SEM system and the second image set acquired by the second SEM system are compared; and

[0174] The parameters of the ML model are adjusted based on comparisons to influence the cost function used to train the ML model, thereby improving the matching between the first image set and the ML-generated images that use the second image set as input to the ML model.

[0175] 29. The measurement system according to Clause 25, wherein the first SEM dataset and the second SEM dataset comprise:

[0176] The contours of features on the training substrate; and / or

[0177] Physical properties associated with the pattern on the training substrate.

[0178] 30. A measurement system according to Clause 27, wherein training the ML model includes:

[0179] Align the first image set or first outline of the first SEM dataset with the design layout image or design outline of the design layout.

[0180] Align the second image set or second outline of the second SEM dataset with the design layout image or design outline of the design layout; and

[0181] The first and second sets of aligned images were used as training data for training the machine learning model.

[0182] 31. The measurement system according to Clause 30, wherein training the ML model includes:

[0183] Compare the intensity values ​​from the first image set and the second image set; and

[0184] The parameters of the ML model are adjusted based on comparisons to influence the cost function used to train the ML model.

[0185] 32. The measurement system according to clause 31, wherein training the ML model is an iterative process, each iteration comprising:

[0186] (i) Transform the second SEM dataset using the adjusted parameters via an ML model and using the first SEM dataset;

[0187] (ii) Compare the intensity values ​​from the first SEM dataset with those from the transformed dataset;

[0188] (iii) Based on comparisons, further adjust the parameters of the ML model to influence the cost function within the expected threshold;

[0189] (iv) Determine whether the cost function is within the expected threshold; and

[0190] (v) In response to the cost function not being within the expected threshold, repeat steps (i) to (iv).

[0191] 33. The measurement system according to Clause 32, wherein the cost function is the difference between a first intensity value from a first SEM dataset and a corresponding second intensity value from a second SEM dataset or a transformed dataset.

[0192] 34. The measurement system according to clause 31, wherein training the ML model includes:

[0193] Intensity values ​​are determined along cutting lines drawn on the features using the first and second SEM datasets.

[0194] 35. The measurement system according to Clause 34, wherein determining the intensity value comprises: determining the pixel intensity value along the cut line using a first SEM dataset and a second SEM dataset.

[0195] 36. The measurement system according to Clause 35, wherein determining the strength value includes:

[0196] The first contour extraction algorithm associated with the first SEM system is applied to the first SEM dataset; and

[0197] The second contour extraction algorithm associated with the measurement system was applied to the second SEM dataset.

[0198] 37. The measurement system pursuant to Clause 36 also includes:

[0199] Based on the first SEM dataset and the physical property measurements of the substrate from the first SEM system, the measurement system selection scheme is determined; and

[0200] The measurement selection scheme is applied to the converted measurement data to determine another physical characteristic measurement value.

[0201] 38. A measurement system pursuant to Clause 37, wherein the physical characteristic measurements include at least one of the following: critical dimension (CD) measurements, overlap measurements, and edge placement errors.

[0202] 39. A measurement system according to Clause 38, wherein the measurement scheme includes a CD threshold, which indicates the location where a CD measurement value is acquired on the captured measurement data.

[0203] 40. The measurement system according to Clause 39, wherein the determination of the measurement selection scheme includes:

[0204] Contours are extracted from images in the first SEM dataset using a first contour extraction algorithm;

[0205] Draw cutting lines at the locations on the contour to measure CD; and

[0206] Based on the signal along the cutting line, the CD threshold corresponding to the measured CD is determined.

[0207] 41. The measurement system according to Clause 24, wherein the measurement system is a scanning electron microscope.

[0208] 42. A measurement system according to Clause 24, wherein the trained ML model is a convolutional neural network.

[0209] 43. A measurement system according to Clause 24, wherein the ML model is trained using a generative adversarial network architecture, the ML model comprising a generator model and a discriminator model.

[0210] 44. The measurement system according to clause 43, wherein training the ML model includes:

[0211] The generator model is trained using the second SEM dataset as input, combined with the discriminator model, to generate data similar to the first SEM dataset; and

[0212] A discriminator model is trained to distinguish the generated data from the first SEM dataset.

[0213] 45. A method for converting data associated with a measurement system, the method comprising:

[0214] Access a first SEM dataset acquired by a first scanning electron metrology (SEM) system and a second SEM dataset acquired by a second SEM system, the first and second SEM datasets being associated with a patterned substrate; and

[0215] A machine learning (ML) model is trained using the first and second SEM datasets as training data, such that the trained ML model is configured to convert the measurement dataset acquired by the second SEM system into a transformed dataset with characteristics equivalent to the measurement data acquired by the first SEM system.

[0216] 46. ​​The method according to clause 45, wherein training the ML model includes:

[0217] Compare the first SEM dataset and the second SEM dataset; and

[0218] The parameters of the ML model are adjusted based on comparisons to influence the cost function used to train the ML model.

[0219] 47. The method according to Clause 46, wherein the first SEM dataset and the second SEM dataset comprise: an image set of a patterned substrate.

[0220] 48. The method according to clause 47, wherein training the ML model includes:

[0221] The first image set acquired by the first SEM system and the second image set acquired by the second SEM system are compared; and

[0222] The parameters of the ML model are adjusted based on comparisons to influence the cost function used to train the ML model, thereby improving the matching between the first image set and the ML-generated images that use the second image set as input to the ML model.

[0223] 49. The method according to Clause 42, wherein the first SEM dataset and the second SEM dataset comprise:

[0224] The outline of features on a patterned substrate; and / or

[0225] Physical properties associated with patterns on patterned substrates.

[0226] 50. The method according to Clause 47, wherein the physical properties include the critical dimension (CD) of the pattern on the patterned substrate.

[0227] 51. The method according to clause 50, wherein training the ML model includes:

[0228] Compare the first CD value of the first SEM dataset with the second CD value of the second SEM dataset; and

[0229] The parameters of the ML model are adjusted based on comparisons to influence the cost function used to train the ML model, thereby improving CD matching between the first and second SEM datasets. This cost function is a function of the first and second CD values.

[0230] 52. According to the method of clause 49, training the ML model includes:

[0231] Align the first image set or first outline of the first SEM dataset with the design layout image or design outline of the design layout.

[0232] Align the second image set or second outline of the second SEM dataset with the design layout image or design outline of the design layout; and

[0233] The first and second sets of aligned images were used as training data for training the machine learning model.

[0234] 53. The method according to clause 52, wherein training the ML model includes:

[0235] Compare the intensity values ​​from the first image set and the second image set; and

[0236] The parameters of the ML model are adjusted based on comparisons to influence the cost function used to train the ML model.

[0237] 54. According to the method of Item 53, where training the ML model is an iterative process, each iteration includes:

[0238] (i) Transform the second SEM dataset using the adjusted parameters via an ML model and using the first SEM dataset;

[0239] (ii) Compare the intensity values ​​from the first SEM dataset with those from the transformed dataset;

[0240] (iii) Based on comparisons, further adjust the parameters of the ML model to influence the cost function within the expected threshold;

[0241] (iv) Determine whether the cost function is within the expected threshold; and

[0242] (v) In response to the cost function not being within the expected threshold, repeat steps (i) to (iv).

[0243] 55. The method according to Clause 54, wherein the cost function is the difference between a first intensity value from a first SEM dataset and a corresponding second intensity value from a second SEM dataset or a transformed dataset.

[0244] 56. The method according to clause 53, wherein training the ML model includes:

[0245] Intensity values ​​are determined along cutting lines drawn on the features using the first and second SEM datasets.

[0246] 57. The method according to Clause 56, wherein determining the intensity value comprises: determining the pixel intensity value along the cutting line using a first SEM dataset and a second SEM dataset.

[0247] 58. The method according to Clause 57, wherein determining the strength value includes:

[0248] The first contour extraction algorithm associated with the first SEM system is applied to the first SEM dataset; and

[0249] The second contour extraction algorithm associated with the second SEM system was applied to the second SEM dataset.

[0250] 59. The method pursuant to Clause 45 also includes:

[0251] Measurement data of another patterned substrate were captured using a second SEM system; and

[0252] The captured measurement data is converted into transformed measurement data by a trained ML model, and the transformed measurement data of another patterned substrate has the same characteristics as those captured by the first SEM system.

[0253] 60. The method pursuant to Clause 45 also includes:

[0254] Based on the first SEM dataset and the physical property measurements of the patterned substrate from the first SEM system, the measurement and matching scheme of the second SEM system is determined.

[0255] Measurement data of the patterned substrate were captured using a second SEM system;

[0256] The captured measurement data is transformed using a trained machine learning model; and

[0257] The measurement selection scheme is applied to the converted measurement data to determine another physical characteristic measurement value.

[0258] 61. The method according to Clause 60, wherein the physical property measurements include at least one of the following: critical dimension (CD) measurements, overlap measurements, and edge placement errors.

[0259] 62. The method according to Clause 61, wherein the measurement scheme includes a CD threshold indicating the location where a CD measurement value is acquired on the captured measurement data.

[0260] 63. The method according to Clause 62, wherein determining the measurement selection scheme includes:

[0261] Contours are extracted from images in the first SEM dataset using a first contour extraction algorithm;

[0262] Draw cutting lines at the locations on the contour to measure CD; and

[0263] Based on the signal along the cutting line, the CD threshold corresponding to the measured CD is determined.

[0264] 64. The method according to Clause 45, wherein the first SEM system is manufactured by a first manufacturer and the second measurement system is manufactured by a second manufacturer.

[0265] 65. The method according to clause 45, wherein the ML model is a convolutional neural network.

[0266] 66. The method according to Item 45, wherein the ML model is trained using a generative adversarial network architecture, the ML model comprising a generator model and a discriminator model.

[0267] 67. According to the method of clause 66, training the ML model includes:

[0268] The generator model is trained using the second SEM dataset as input, combined with the discriminator model, to generate data similar to the first SEM dataset; and

[0269] A discriminator model is trained to distinguish the generated data from the first SEM dataset.

[0270] 68. A method for converting SEM images acquired by a scanning electron microscope (SEM) system, the method comprising:

[0271] Access a first SEM image set acquired by a first SEM system and a second SEM image set acquired by a second SEM system, wherein the first and second SEM image sets are associated with a patterned substrate; and

[0272] A machine learning (ML) model is trained using a first SEM image set and a second SEM image set as training data. The trained ML model is configured to convert the SEM image set acquired by the second SEM system into a transformed image set that has characteristics equivalent to those of the SEM images acquired by the first SEM system.

[0273] 69. The method according to clause 68, wherein training the ML model includes:

[0274] Compare the first SEM image set and the second SEM image set; and

[0275] The parameters of the ML model are adjusted based on comparisons to influence the cost function used to train the ML model.

[0276] 70. The method according to clause 68, wherein training the ML model includes:

[0277] Align the first SEM image set with the design layout image;

[0278] Align the second SEM image set with the design layout image; and

[0279] The first and second sets of aligned SEM images were used as training data for training the machine learning model.

[0280] 71. The method according to clause 69, wherein training the ML model includes:

[0281] The intensity values ​​from the first SEM image set and the second SEM image set are compared; and

[0282] The parameters of the ML model are adjusted based on comparison to influence the cost function, thereby reducing the intensity value difference between the first image set and the ML-generated images that use the second SEM image set as input to the ML model.

[0283] 72. According to the method of Item 71, where training the ML model is an iterative process, each iteration includes:

[0284] (i) Transform the second SEM image set using the adjusted parameters via an ML model and using the first SEM image set;

[0285] (ii) Compare the intensity values ​​from the first SEM image set with the transformed image set;

[0286] (iii) Based on comparisons, further adjust the parameters of the ML model to influence the cost function within the expected threshold;

[0287] (iv) Determine whether the cost function is within the expected threshold; and

[0288] (v) In response to the cost function not being within the expected threshold, repeat steps (i) to (iv).

[0289] 73. The method according to Clause 72, wherein the cost function is the difference between a first intensity value from a first SEM image set and a corresponding second intensity value from a second SEM image set or a transformed image set.

[0290] 74. The method pursuant to Clause 68 also includes:

[0291] SEM images of another patterned substrate were captured using a second SEM system; and

[0292] The captured SEM image is converted into a transformed SEM image by a trained ML model. The transformed SEM image of another patterned substrate has properties comparable to those of the image captured by the first SEM system.

[0293] 75. The method pursuant to Clause 68 also includes:

[0294] Based on the first SEM image set and the CD measurement values ​​of the patterned substrate from the first SEM system, the measurement and matching scheme of the second SEM system is determined.

[0295] SEM images of the patterned substrate were captured using a second SEM system;

[0296] Transform captured SEM images using a trained ML model; and

[0297] The measurement selection scheme is applied to the converted SEM image to determine another CD measurement value.

[0298] 76. The method according to Clause 75, wherein the measurement scheme includes a CD threshold indicating the location where CD measurements are acquired on the captured SEM image.

[0299] 77. The method according to Clause 68, wherein the ML model is a convolutional neural network.

[0300] 78. The method according to Item 68, wherein the ML model is trained using a generative adversarial network architecture, the ML model comprising a generator model and a discriminator model.

[0301] Although the concepts disclosed herein can be used for imaging on substrates such as silicon wafers, it should be understood that the disclosed concepts can be used with any type of lithography imaging system, such as those used for imaging on substrates other than silicon wafers.

[0302] The above description is intended to be illustrative and not restrictive. Therefore, it will be apparent to those skilled in the art that modifications may be made as described without departing from the scope of the claims set forth below.

Claims

1. A measurement system, comprising: A computer system comprising one or more processors, on which trained machine learning models are stored and programmed with computer program instructions that, when executed, cause the computer system to: Capture measurement data of patterned substrates; and The captured measurement data is transformed into converted measurement data by the trained machine learning model, and the converted measurement data has the same characteristics as that captured by another measurement system.

2. The measurement system of claim 1, wherein the computer system is instructed to train a machine learning model, the training comprising: Access a first SEM dataset obtained by a first SEM system and a second SEM dataset obtained by the measurement system, wherein the first SEM dataset and the second SEM dataset are associated with a training substrate; and The machine learning model is trained using the first SEM dataset and the second SEM dataset as training data, such that the trained machine learning model is configured to convert the measurement dataset obtained by the measurement system into a transformed dataset with characteristics equivalent to the measurement data obtained by the first SEM system.

3. The measurement system according to claim 2, wherein training the machine learning model comprises: Compare the first SEM dataset and the second SEM dataset; as well as The parameters of the machine learning model are adjusted based on the comparison to influence the cost function used to train the machine learning model.

4. The measurement system according to claim 3, wherein the first SEM dataset and the second SEM dataset comprise: The image set of the training substrate.

5. The measurement system according to claim 4, wherein training the machine learning model comprises: Compare the first image set obtained by the first SEM system with the second image set obtained by the second SEM system; and The parameters of the machine learning model are adjusted based on the comparison to influence the cost function used to train the machine learning model, thereby improving the matching between the first image set and the machine learning-generated images that use the second image set as input to the machine learning model.

6. The measurement system according to claim 2, wherein the first SEM dataset and the second SEM dataset comprise: The outline of the features on the training substrate; and / or Physical properties associated with the pattern on the training substrate.

7. The measurement system according to claim 4, wherein training the machine learning model comprises: Align the first image set or first outline of the first SEM dataset with the design layout image or the design outline of the design layout. Align the second image set or second outline of the second SEM dataset with the design layout image or the design outline of the design layout; and The first and second sets of aligned images are used as training data for training the machine learning model.

8. The measurement system of claim 7, wherein training the machine learning model comprises: Compare the intensity values ​​from the first image set and the second image set; as well as The parameters of the machine learning model are adjusted based on the comparison to influence the cost function used to train the machine learning model.

9. The measurement system of claim 8, wherein training the machine learning model is an iterative process, each iteration comprising: (i) Transform the second SEM dataset using the machine learning model with the adjusted parameters and the first SEM dataset; (ii) Compare the intensity values ​​from the first SEM dataset and the transformed dataset; (iii) Based on the comparison, further adjust the parameters of the machine learning model to influence the cost function within the desired threshold; (iv) Determine whether the cost function is within the desired threshold; and (v) In response to the cost function not being within the expected threshold, repeat steps (i)-(iv).

10. The measurement system of claim 9, wherein the cost function is the difference between a first intensity value from the first SEM dataset and a corresponding second intensity value from the second SEM dataset or the transformed dataset.

11. The measurement system of claim 8, wherein training the machine learning model comprises: Determine the intensity values ​​of the cutting lines drawn across the feature from the first SEM dataset and the second SEM dataset.

12. The measurement system of claim 11, wherein determining the intensity value comprises: Pixel intensity values ​​along the cutting line are determined from the first SEM dataset and the second SEM dataset.

13. The measurement system of claim 12, wherein determining the intensity value comprises: The first contour extraction algorithm associated with the first SEM system is applied to the first SEM dataset; as well as A second contour extraction algorithm associated with the measurement system is applied to the second SEM dataset.

14. The measurement system according to claim 13, further comprising: Based on the first SEM dataset and the physical property measurement results of the patterned substrate from the first SEM system, a measurement configuration scheme for the measurement system is received. and The measurement selection scheme is applied to the converted measurement data to determine the measurement result of another physical characteristic.

15. The measurement system of claim 1, wherein the machine learning model is trained using a generative adversarial network architecture, the machine learning model comprising a generator model and a discriminator model.

Images