Metrology using reference-based synthetic spectra

Abstract

A metrology system can implement a metrology formula by: generating a real training dataset for metrological measurements; generating a synthetic training dataset for the metrological measurements; training a machine learning model using the real training dataset and the synthetic training dataset to generate values ​​for the metrological measurements; and generating metrological measurements for the one or more runtime samples from measurement data associated with one or more runtime samples. The real training dataset can be generated by receiving reference data, performing dimensionality reduction, and identifying one or more relevant principal components of the real training data that satisfy a first correlation threshold with the reference data. The synthetic training dataset can then be generated by extracting the same relevant principal components from the synthetic training data and filtering them to ensure they also satisfy the correlation threshold with the reference data.

Description

[0001] Cross-reference of related applications

[0002] This application claims the right of U.S. Provisional Application No. 63 / 656,107, filed June 5, 2024, entitled “Reference-Based Synthetic Spectroscopic DOE”, by inventor Houssam Chouaib, pursuant to 35 USC § 119(e), the entire contents of which are incorporated herein by reference. Technical Field

[0003] This disclosure generally relates to spectral ellipsometric measurement incorporating machine learning models, and more specifically, to generating synthetic training data for machine learning models used in spectral ellipsometric measurement. Background Technology

[0004] Many current optical metrology applications in integrated circuit manufacturing require complex spectral analysis. Furthermore, the number of critical process steps requiring sensitive metrology continues to increase, while the tolerance windows and accuracy limits of metrology techniques are significantly tightening.

[0005] Some current optical metrology techniques rely on conventional rigorous coupled-wave analysis (RCWA), which has drawbacks such as relatively long solution times and a lack of robustness required to adapt to process variations. RCWA-based solutions further require extensive computational resources and repetitive arrays.

[0006] Some current optical metrology techniques utilize machine learning models to generate metrological measurements based on measured data. However, machine learning models typically require large amounts of training data to provide the desired measurement accuracy. Real training data generated from measurements on manufactured samples with different parameters is usually of high quality but time-consuming. Efforts to supplement real training data with model-based supplementary training data offer only limited success and, in some cases, are counterproductive.

[0007] Therefore, it is necessary to develop systems and methods to address the aforementioned deficiencies. Summary of the Invention

[0008] In an embodiment, the technology described herein relates to a metrology system comprising: a controller including one or more processors configured to execute program instructions that cause the one or more processors to implement a metrology formula by: generating a real training dataset for metrology measurements by: receiving real training data of test features from one or more training samples from a first metrology subsystem; receiving reference data associated with the metrology measurements of the test features from a second metrology subsystem; performing a dimensionality reduction operation on the real training data; identifying one or more principal components of the real training data that satisfy a first correlation threshold with the reference data; and filtering the real training data to include the real training data associated with the one or more training samples. The real training dataset is generated by associating portions of multiple relevant principal components; a synthetic training dataset for the metrology is generated by: generating synthetic training data for multiple simulated test features having known simulated values ​​of the metrology; extracting the one or more relevant principal components from the synthetic training data as dimensionality-reduced synthetic training data; and generating the synthetic training dataset by filtering the dimensionality-reduced synthetic training data to include portions of the dimensionality-reduced synthetic training data that satisfy a second correlation threshold with the reference data; training a machine learning model with the real training dataset and the synthetic training dataset to generate values ​​for the metrology; and generating the metrology for the one or more runtime samples from measurement data associated with the one or more runtime samples.

[0009] In an embodiment, the technology described herein relates to a metrology system in which the dimensionality reduction operation includes principal component analysis.

[0010] In embodiments, the techniques described herein relate to a metrology system in which the metrological measurements include at least one of overlay measurements or critical dimension measurements.

[0011] In an embodiment, the technology described herein relates to a metrology system, wherein the first metrology subsystem comprises optical metrology tools.

[0012] In an embodiment, the technology described herein relates to a metrology system, wherein the first metrology subsystem comprises at least one of a spectral ellipticity measurement tool or a spectral reflectance measurement tool.

[0013] In embodiments, the technology described herein relates to a metrology system, wherein the second metrology subsystem comprises at least one of a particle beam metrology tool or an X-ray metrology tool.

[0014] In embodiments, the techniques described herein relate to a metrology system, wherein the second metrology subsystem comprises at least one of a transmission electron microscope, a transmission small-angle X-ray scattering tool, a scanning electron microscope, a critical-size scanning electron microscope, or an atomic force microscope.

[0015] In an embodiment, the technology described herein relates to a metrology system in which at least one of the first correlation threshold or the second correlation threshold is a goodness-of-fit threshold.

[0016] In an embodiment, the technology described herein relates to a metrology system in which at least one of the first correlation threshold or the second correlation threshold is R. 2 Threshold.

[0017] In an embodiment, the technology described herein relates to a metering system in which the first correlation threshold is equal to the second correlation threshold.

[0018] In an embodiment, the technology described herein relates to a metering system in which the second correlation threshold is greater than the second correlation threshold.

[0019] In embodiments, the technology described herein relates to a metrology system comprising: a first metrology subsystem; a second metrology subsystem; and a controller comprising one or more processors configured to execute program instructions that cause the one or more processors to implement a metrology formula by: generating a real training dataset for metrology measurements by: receiving real training data from the first metrology subsystem of test features on one or more training samples; receiving reference data associated with the metrology measurements of the test features from the second metrology subsystem; performing a dimensionality reduction operation on the real training data; identifying one or more principal components of the real training data that satisfy a first correlation threshold with the reference data; and filtering the real training data to include the real training data. The real training dataset is generated by associating the data with the one or more relevant principal components; a synthetic training dataset for the metrology is generated by: generating synthetic training data for multiple simulated test features having known simulated values ​​of the metrology; extracting the one or more relevant principal components from the synthetic training data as dimensionality-reduced synthetic training data; and generating the synthetic training dataset by filtering the dimensionality-reduced synthetic training data to include the portion of the dimensionality-reduced synthetic training data that satisfies a second correlation threshold with the reference data; training a machine learning model with the real training dataset and the synthetic training dataset to generate values ​​for the metrology; and generating the metrology of the one or more runtime samples from measurement data associated with the one or more runtime samples.

[0020] In embodiments, the techniques described herein relate to a metrology system in which the metrological measurements include at least one of overlay measurements or critical dimension measurements.

[0021] In an embodiment, the technology described herein relates to a metrology system, wherein the first metrology subsystem comprises optical metrology tools.

[0022] In an embodiment, the technology described herein relates to a metrology system, wherein the first metrology subsystem comprises at least one of a spectral ellipticity measurement tool or a spectral reflectance measurement tool.

[0023] In embodiments, the technology described herein relates to a metrology system, wherein the second metrology subsystem comprises at least one of a particle beam metrology tool or an X-ray metrology tool.

[0024] In embodiments, the techniques described herein relate to a metrology system, wherein the second metrology subsystem comprises at least one of a transmission electron microscope, a transmission small-angle X-ray scattering tool, a scanning electron microscope, a critical-size scanning electron microscope, or an atomic force microscope.

[0025] In an embodiment, the technology described herein relates to a metrology system in which at least one of the first correlation threshold or the second correlation threshold is a goodness-of-fit threshold.

[0026] In an embodiment, the technology described herein relates to a metrology system in which at least one of the first correlation threshold or the second correlation threshold is R. 2 Threshold.

[0027] In an embodiment, the technology described herein relates to a metering system in which the first correlation threshold is equal to the second correlation threshold.

[0028] In an embodiment, the technology described herein relates to a metering system in which the second correlation threshold is greater than the second correlation threshold.

[0029] In embodiments, the technology described herein relates to a metrology method comprising: generating a real training dataset for metrological measurements by: receiving real training data of test features from one or more training samples from a first metrology subsystem; receiving reference data associated with the metrological measurements of the test features from a second metrology subsystem; performing a dimensionality reduction operation on the real training data; identifying one or more relevant principal components of the real training data that satisfy a first correlation threshold with the reference data; and generating the real training dataset by filtering the real training data to include portions of the real training data associated with the one or more relevant principal components; and generating the real training dataset by: A synthetic training dataset for the metrology is generated by: generating synthetic training data for multiple simulated test features having known simulated values ​​of the metrology; extracting the one or more relevant principal components from the synthetic training data as dimensionality-reduced synthetic training data; and generating the synthetic training dataset by filtering the dimensionality-reduced synthetic training data to include portions of the dimensionality-reduced synthetic training data that satisfy a second correlation threshold with the reference data; training a machine learning model with the real training dataset and the synthetic training dataset to generate values ​​for the metrology; and generating the metrology of the one or more runtime samples from measurement data associated with the one or more runtime samples.

[0030] It should be understood that the foregoing overview and the following detailed description are for illustrative purposes only and do not necessarily limit the invention. The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and, together with the overview, serve to explain the principles of the invention. Attached Figure Description

[0031] Those skilled in the art can better understand the many advantages of this disclosure by referring to the accompanying drawings.

[0032] Figure 1A A block diagram illustrating a metering system according to one or more embodiments of the present disclosure.

[0033] Figure 1B A simplified schematic diagram illustrating an optical measurement subsystem according to one or more embodiments of the present disclosure.

[0034] Figure 1C A simplified schematic diagram illustrating an x-ray measurement subsystem according to one or more embodiments of the present disclosure.

[0035] Figure 1D A simplified schematic diagram illustrating a particle beam measurement subsystem 102 according to one or more embodiments of the present disclosure.

[0036] Figure 2A The flowcharts illustrate one or more embodiments of the present disclosure, describing the steps performed in the metering method.

[0037] Figure 2B The flowcharts illustrating one or more embodiments of the present disclosure depict sub-steps associated with the step of generating a real training dataset for metrology measurements.

[0038] Figure 2C The flowcharts illustrating one or more embodiments of the present disclosure depict sub-steps associated with step 216, which generates a synthetic training dataset for metrological measurements.

[0039] Figure 3 This describes a plot of the relevant principal components of real training data for test features generated using an online measurement subsystem for spectral ellipsometry according to one or more embodiments of this disclosure, wherein the relevant principal components are plotted as varying based on reference data.

[0040] Figure 4 This describes a plot of the principal components associated with both real training data and synthetic training data according to one or more embodiments of this disclosure.

[0041] Figure 5 The diagram illustrates real training data and filtered synthetic training data according to one or more embodiments of this disclosure.

[0042] Figure 6A This describes a plot of the principal components associated with a series of different test features according to one or more embodiments of this disclosure.

[0043] Figure 6B Description of one or more embodiments of this disclosure Figure 6A The actual training data and the plotted graphs of the filtered and synthesized training data. Detailed Implementation

[0044] Reference will now be made in detail to the disclosed subject matter illustrated in the accompanying drawings. This disclosure has been particularly shown and described with respect to certain embodiments and their specific features. The embodiments set forth herein should be considered illustrative rather than restrictive. It will be readily apparent to those skilled in the art that various changes and modifications in form and detail may be made without departing from the spirit and scope of this disclosure.

[0045] Embodiments of this disclosure relate to systems and methods for providing spectrometric measurements, which incorporate a machine learning model trained on a combination of real training data and synthetic training data, wherein the synthetic training data is generated in a manner that maintains its correlation with reference training data.

[0046] Spectrometry is a powerful metrology technique based on illuminating a sample with broadband light and capturing the spectrum of the reflected light. In a general sense, the reflectance spectrum can be highly sensitive to the physical or optical structure of a sample. Furthermore, different properties of the sample can be measured by controlling the incident illumination and / or the properties of the collected light (e.g., but not limited to polarization, phase, or angle). Various spectrometry techniques, within the spirit and scope of this disclosure, include, but are not limited to, spectral ellipsometric measurements, reflectance measurements, scattering measurements, or Raman metrology. For example, spectrometry can be used to generate measurements representing some or all of the Mueller matrix elements representing the optical response of a sample to incident illumination. As another example, spectrometry can be used to generate specific metrological measurements that can be useful for semiconductor process control, such as, but not limited to, critical dimension (CD) measurements, pattern asymmetry measurements, tilt measurements, or overlay measurements.

[0047] In some embodiments, a machine learning model is developed to generate one or more metrological measurements based on spectral ellipsometric measurements generated with illumination and / or through one or more configurations of light collection. This model can be trained using a combination of real-world training datasets and synthetic training datasets.

[0048] A real training dataset can be generated based on spectral ellipsometric measurements of manufactured samples with different physical parameters (e.g., different CD values, pattern asymmetry, tilt, overlap, or the like) associated with one or more metrologies. This process is often referred to as a design of experiment (DOE). Then, high-resolution reference tools (e.g., scanning electron microscopy (SEM), transmission electron microscopy (TEM), atomic force microscopy (ATM), small-angle X-ray scattering (SAXS), or any variant (e.g., CD-SAXS or the like)) are used to generate reference data containing accurate values ​​of different physical parameters (e.g., one or more metrologies of interest). In this way, a machine learning model can be trained to generate one or more metrologies based on the input spectral ellipsometric measurement data.

[0049] Synthetic training datasets can be generated in a way that ensures relevance to real training data. In some embodiments, candidate synthetic training data is first generated using a simulation model (e.g., a physical model) that simulates optical interactions with simulated samples (e.g., simulated DOEs) that also have different physical parameters. This candidate synthetic training data is then filtered to include only data that meets a relevance threshold to the reference data. As an illustration, dimensionality reduction techniques (e.g., but not limited to principal component analysis (PCA)) can first be applied to the real training data to determine one or more principal components (PCs) that are highly correlated with the reference data for a particular sample design (e.g., a correlation higher than a selected real data relevance threshold). For example, one or more PCs may be linearly correlated with the reference data and have an R-value lower than the real data relevance threshold. 2value.

[0050] Next, the same dimensionality reduction analysis can be applied to the synthetic data, allowing for the extraction of PCs identified as highly correlated with the reference data. Finally, candidate synthetic reference data can be screened to include only data satisfying one or more PCs or combinations thereof at selected synthetic data relevance thresholds. As an illustration, a plot of the PCs of the synthetic data relative to the reference data can show that some candidate synthetic data are not closely related to the reference data, which may be due to various reasons, including but not limited to unrealistic sample geometries. However, candidate synthetic data can be screened to maintain only data that retains the same relevance to the reference data as the real training data. For example, only data in the synthetic training dataset whose R-values ​​remain below the synthetic data relevance threshold can be considered. 2 The value portion, where the correlation threshold of synthetic data may be the same as or different from that of real data.

[0051] With careful consideration, the systems and methods disclosed herein offer numerous advantages over existing synthetic data generation techniques. For example, some synthetic training data generation techniques use only candidate synthetic training data associated with a broad range of DOEs without considering whether the geometry is realistic and / or whether it overlaps with real training data. However, this technique can withstand large amounts of unrealistic or counterproductive training data, which can reduce the robustness of machine learning models. As another example, some synthetic training data generation techniques attempt to filter synthetic training data by comparing different PCs associated with dimensionality reduction techniques, selecting the portion of synthetic training data that overlaps with real training data, and generating additional synthetic training data with similar parameter configurations. However, this technique can be time-intensive and does not actually lead to significantly better results. In some cases, this technique produces worse results than using synthetic training data from a broad range of DOEs indiscriminately. Furthermore, with careful consideration, PC-to-PC comparisons may not be a useful metric for evaluating synthetic training data. Instead, the systems and methods disclosed herein provide synthetic training data associated with reference data.

[0052] For reference Figures 1A to 6B The present disclosure describes a system and method for providing spectral ellipsometric measurements using accurate synthetic training data, according to one or more embodiments.

[0053] Figure 1A A block diagram illustrating a metering system 100 according to one or more embodiments of the present disclosure.

[0054] In some embodiments, the metrology system 100 includes at least one measurement subsystem 102 to generate measurement data associated with test features 104 on the sample 106, and further includes a controller 108 to generate one or more metrological measurements associated with the measurement data. The controller 108 may include one or more processors 110 configured to execute a set of program instructions maintained in memory 112 or a memory device, wherein the program instructions may cause the processors 110 to perform various actions or steps disclosed herein.

[0055] The measurement subsystem 102 may include any component or combination of components adapted to generate measurement data associated with the test feature 104. For example, the measurement subsystem 102 may direct illumination 114 to the test feature 104, capture a light-collecting signal 116 from the test feature 104 in response to illumination 114, and generate measurement data (e.g., using a detector) based on this light-collecting signal 116, wherein the measurement data contains information indicating one or more metrological measurements of interest.

[0056] In some embodiments, the measurement subsystem 102 includes an optical measurement subsystem 102 to generate measurement data based on the interaction between the sample 106 and illumination 114, the illumination 114 containing any suitable wavelength or combination of wavelengths (including, but not limited to, ultraviolet (UV) wavelengths, visible light wavelengths, or infrared (IR) wavelengths). For example, the optical measurement subsystem 102 may include, but is not limited to, a spectroscopic ellipsometer (SE), an SE with multiple illumination angles, an SE (e.g., using a rotation compensator) for measuring Mueller matrix elements, a single-wavelength ellipsometer, a beam profile ellipsometer (angle-resolved ellipsometer), a beam profile reflectometer (angle-resolved reflectometer), a broadband reflectance spectrometer (spectral reflectometer), a single-wavelength reflectometer, an angle-resolved reflectometer, an imaging system, a scatterometer (e.g., a spot analyzer), a Raman metrology tool, a laser-driven spectroscopic reflectance measurement (LDSR) system, or any combination thereof.

[0057] In some embodiments, the measurement subsystem 102 includes an x-ray measurement subsystem 102 to generate measurement data based on the interaction between the sample 106 and the x-ray-containing illumination 114. For example, the measurement subsystem 102 may include, but is not limited to, a small-angle x-ray scatterometer (SAXR), a soft x-ray reflectometer (SXR), or an x-ray photoelectron spectroscopy (XPS) system.

[0058] In some embodiments, the measurement subsystem 102 includes a particle beam measurement subsystem 102 to generate measurement data based on the interaction between the sample 106 and illumination 114, the illumination 114 comprising a particle beam, such as, but not limited to, an electron beam (e-beam), an ion beam, or a neutral particle beam. For illustration, the electron beam measurement subsystem may include a scanning electron microscope (SEM), a critical dimension SEM (CD-SEM), a grayscale SEM (e.g., a model-based SEM similar to a CD-SEM that provides additional information about the measured structure, such as, but not limited to, depth, height, bottom CD, or the like), or a transmission electron microscope (TEM).

[0059] In some embodiments, the metrology system 100 includes at least two different types of measurement subsystems 102 with different optical principles and / or different resolutions. For example, the metrology system 100 may include at least one measurement subsystem 102 configured for online operation (e.g., online measurement subsystem 102) and at least one measurement subsystem 102 configured for reference operation (e.g., reference measurement subsystem 102). In this configuration, the metrology system 100 may implement a trained machine learning model to generate one or more metrological measurements based on measurement data from the online measurement subsystem 102, wherein the reference measurement subsystem 102 is used to generate ground-based ground reference data for training the machine learning model.

[0060] In some applications, the reference measurement subsystem 102 may have relatively higher accuracy than the in-line measurement subsystem 102, but may have reduced processing power. As an illustration, the in-line measurement subsystem 102 may include an optical measurement subsystem 102 configured to capture at least some Mueller matrix elements associated with the test feature 104, while the reference measurement subsystem 102 may include an X-ray or particle beam measurement subsystem 102. As another example, the in-line measurement subsystem 102 may include a particle beam measurement subsystem 102 (e.g., CD-SEM, grayscale SEM, or the like), while the reference measurement subsystem 102 may include a TEM or XPS system. It should be understood that these examples are provided for illustrative purposes only and should not be construed as limiting this disclosure. In fact, the metrology system 100 may include any type of in-line measurement subsystem 102 and any type of reference measurement subsystem 102.

[0061] Multiple measurement subsystems 102 may be provided as a single tool or multiple tools. A single tool providing multiple measurement configurations is generally described in U.S. Patent No. 7,933,026, issued April 26, 2011, the entire contents of which are incorporated herein by reference. Multiple tools and structural analysis are generally described in U.S. Patent No. 7,478,019, issued January 13, 2009, the entire contents of which are incorporated herein by reference.

[0062] Any type of light-collecting signal 116 emitted from test feature 104 in response to illumination 114 (e.g., but not limited to light, X-rays, or particles) can be captured to generate measurement data, regardless of the configuration of measurement subsystem 102.

[0063] Furthermore, the measurement subsystem 102 can be configured to generate metrological measurements based on any number of metrological recipes, wherein the metrological recipes can define various imaging parameters for generating measurement data and / or processing techniques for generating metrological measurements from the measurement data. For example, the metrological recipe may include parameters associated with illumination 114, such as, but not limited to, the number of beams, the angle of incidence (e.g., azimuth and / or polar angle of incidence), polarization, phase characteristics, or wavelength. As another example, the metrological recipe may include parameters associated with the light-collecting signal 116 used to generate measurement data, such as, but not limited to, the light-collecting angle (e.g., for collecting zero-order double diffraction), polarization, phase characteristics, or wavelength. As another example, the metrological recipe may include sampling characteristics, such as, but not limited to, the position on the sample 106 to be measured (e.g., the position of a specific superimposed target or device feature to be characterized) or focusing characteristics.

[0064] The metrology system 100 is adaptable to generate any type of metrological measurement on any type of test feature 104. In addition, the metrology system 100 can generate multiple metrological measurements associated with various sub-features (e.g., key parameters) associated with the test feature 104 and / or metrological measurements associated with multiple test features 104.

[0065] Test feature 104 may be associated with any stage of the manufacturing process (e.g., but not limited to etching, photolithography, or deposition). Non-limiting examples of test feature 104 include, but are not limited to, patterned multilayer structures, patterned monolayer structures (e.g., grating structures or the like), film stacks (e.g., unpatterned film stacks), or combinations thereof. Furthermore, non-limiting examples of metrological measurements include, but are not limited to, CD measurements, height measurements (e.g., the height of a patterned feature, the height of a multilayer feature, or the like), stack-up measurements, tilt measurements, electrical test measurements (e.g., Vt, work function, mobility, Ion / Ioff, or the like), stress or strain measurements, film thickness or material property measurements (e.g., refractive index measurements, spectral measurements, or the like).

[0066] In some embodiments, the metrology system 100 uses a machine learning model to generate one or more metrological measurements based on measurement data provided by the measurement subsystem 102. For example, the machine learning model may be embodied in memory 112 (e.g., as a set of internal weights or the like) such that the processor 110 of the controller 108 can execute program instructions that cause the processor 110 to perform various actions disclosed herein, such as, but not limited to, training the machine learning model, receiving measurement data from the measurement subsystem 102 (e.g., training data, data for online measurements, or any type of data), generating metrological measurements using the machine learning model, generating correctable items or the like based on the metrological measurements.

[0067] The machine learning model can accept measurement data (e.g., raw data) associated with a specific measurement configuration of the measurement subsystem 102. As an illustration, the spectral-based optical measurement subsystem 102 can generate signals associated with 15 Mueller matrix elements, each signal having approximately 670 wavelength pixels to provide approximately 10,000 individual signals for a specific measurement configuration. However, this is for illustrative purposes only and should not be construed as limiting the scope of this disclosure. For example, this system can generate signals associated with any number of Mueller matrix elements (e.g., up to 16 Mueller matrix elements) and provide any number of data points for any number of wavelengths.

[0068] As another example, a machine learning model can be trained to take data derived from one or more sets of measurement data as input and produce one or more metrological measurements as output. For instance, a machine learning model can take dimensionality-reduced measurement data (e.g., one or more PCs) as input and produce one or more metrological measurements as output. Any suitable technique can be used to generate the dimensionality-reduced measurement data, including but not limited to principal component analysis (PCA) (e.g., linear or nonlinear) or fast Fourier transform (FFT) analysis. In a general sense, the principal component set can correspond to aspects of the associated measurement data related to the metrological measurement of interest.

[0069] For reference Figures 2A to 6B The generation of training data for generating a machine learning model for generating metrological measurements is described according to one or more embodiments of this disclosure.

[0070] Machine learning models can incorporate any type or combination of machine learning techniques, such as, but not limited to, supervised, semi-supervised, reinforcement learning, or unsupervised machine learning techniques. For illustration, machine learning models can include, but are not limited to, linear models, neural network models, multinomial models, decision tree models, or random forest models. Furthermore, machine learning models can be trained using training data that can be derived from any source.

[0071] In some embodiments, the machine learning model is trained on a combination of real training data associated with manufactured samples and synthetically generated training data (e.g., simulated training data). With careful consideration, real training data may advantageously include the physical configuration of test feature 104 that could actually occur due to process variations during a typical manufacturing process. However, generating sufficient real training data to provide robust operation of the machine learning model can be difficult and / or impractical. With further careful consideration, synthetic training data may be generated relatively more easily in a quantity useful for training purposes, but may sometimes correspond to unrealistic physical configurations of test feature 104 and therefore have limited utility.

[0072] For example, one technique for generating synthetic training data is to produce synthetic measurements associated with a wide range of manufacturing parameters (e.g., broad DOE) without considering the likelihood that various combinations of manufacturing parameters can actually occur during a typical manufacturing process. This simplification approach can provide a large amount of data, but the potential presence of a large amount of unrealistic training data can adversely affect the robustness of machine learning models.

[0073] As another example, efforts to filter synthetic training data generated by a wide range of DOEs based on comparisons of principal components of real and synthetic training data may not provide many benefits and may result in a lot of synthetic training data being unweighted or excluded.

[0074] In some embodiments, synthetic training data is generated in a manner that maintains relevance to real reference data, which enables robust and accurate performance of the machine learning model.

[0075] Figure 2A The flowcharts illustrating one or more embodiments of this disclosure depict the steps performed in metering method 200. The embodiments and implementation techniques previously described herein within the context of metering system 100 should be interpreted as extensions of method 200. For example, the processor 110 of controller 108 may execute program instructions that cause it to perform one or more steps of method 200, either directly or by generating control signals to direct actions of components internal or external to metering system 100. However, method 200 is not limited to the architecture of metering system 100.

[0076] Method 200 may include step 202 of generating a real training dataset for metrological measurements.

[0077] Figure 2B The flowcharts illustrating one or more embodiments of the present disclosure depict sub-steps associated with step 202, which generates a real training dataset for metrological measurements.

[0078] Step 202, which generates a real training dataset for metrology measurements, may include step 204, which involves creating multiple test features 104 on one or more samples having different values ​​for the metrology measurements. For example, step 204 may involve creating one or more DOE samples in which multiple manufacturing parameters are intentionally varied. In some cases, the manufacturing parameters are varied in a way that reflects the actual process variations of the process tools used in a large-scale manufacturing environment.

[0079] Step 202, which generates a real training dataset for metrological measurements, may include step 206, which generates real training data for multiple test features 104 using a first measurement subsystem 102 (e.g., an online measurement subsystem 102). For example, the first measurement subsystem 102 may correspond to an optical measurement subsystem 102, such as, but not limited to, a spectral ellipticity measurement subsystem 102.

[0080] Step 202, which generates a real training dataset for metrological measurements, may include step 208, which generates reference data for multiple test features 104 using a second measurement subsystem 102 (e.g., a reference measurement subsystem 102). For example, the second measurement subsystem 102 may correspond to a particle beam or X-ray measurement subsystem 102 that provides relatively high-resolution measurements of the multiple test features 104. In this way, the reference data can be used as ground-based data for real training data.

[0081] Step 202, which generates a real training dataset for metrological measurements, may include step 210, which performs a dimensionality reduction operation on the real training data. Any type of dimensionality reduction operation may be used, such as, but not limited to, principal component analysis or Fourier transform analysis. In this way, the real training data can be expressed according to various principal components corresponding to selected portions of the real training data.

[0082] Step 202, which generates a real training dataset for metrological measurements, may include step 212, which identifies one or more principal components of the real training data that satisfy a first correlation threshold with reference data. Step 202, which generates a real training dataset for metrological measurements, may include step 214, which generates the real training dataset by filtering the real training data to include portions of the real measurement data that are associated with one or more principal components.

[0083] In this document, after careful consideration, some principal components may provide a relatively higher correlation with the reference data than other principal components. Therefore, step 212 may involve identifying one or more principal components that satisfy a first correlation threshold, which are then referred to as relevant principal components. Any correlation threshold or metric may be used, such as, but not limited to, R. 2 Value or goodness-of-fit (GoF) threshold. For example, step 212 could involve providing an R value higher than the selected number. 2 One or more related principal components of the value.

[0084] Figure 3 This describes a plot of the relevant principal component (PC9) of the real training data of the DRAM test feature 104 generated using the spectral ellipsometric online measurement subsystem 102 according to one or more embodiments of this disclosure. The PC9 is plotted as varying based on reference data generated using the CD-SEM reference measurement subsystem 102. The observed R of this principal component... 2 The value is ~0.871.

[0085] In this paper, after careful consideration, generating a real training dataset that includes relevant principal components that provide strong correlation with reference data (e.g., ground-based data) can be both efficient and provide robust training.

[0086] For reference Figure 2A and 2C Method 200 may include step 216 of generating a synthetic training dataset for metrological measurements. Figure 2C The flowcharts illustrating one or more embodiments of the present disclosure depict sub-steps associated with step 216, which generates a synthetic training dataset for metrological measurements.

[0087] Step 216, which generates a synthetic training dataset for metrology measurements, may include step 218, which generates synthetic training data for multiple simulated test features 104 with known simulated values ​​of metrology measurements.

[0088] Synthetic training data can be generated using any technique. In some embodiments, synthetic training data is generated using a physics-based measurement model that simulates measurements of test feature 104 (e.g., via the first measurement subsystem 102). For example, in the case where the first measurement subsystem 102 is an optical measurement subsystem 102, the measurement model may include an algorithm-based electromagnetic solver, such as, but not limited to, strictly coupled-wave analysis (RCWA), finite element method (FEM), method of moments, surface integration, volume integration, finite difference time domain (FDTD), or similar techniques.

[0089] Synthetic training data can be generated using a wide range of Design of Elements (DOEs). In some cases, the range of manufacturing parameters simulated to generate synthetic training data is larger than the range of manufacturing parameters used to generate real training data.

[0090] Step 216, which generates a synthetic training dataset for metrological measurements, may include step 220, which extracts one or more relevant principal components from the synthetic training data as dimensionality-reduced synthetic training data.

[0091] continue Figure 3The example described herein can extract the portion of synthetic training data associated with relevant principal components to provide dimensionality-reduced synthetic training data.

[0092] Figure 4 This illustration depicts a plot of the relevant principal components (PC9) associated with both real training data and synthetic training data (combined) according to one or more embodiments of this disclosure. In this manner, Figure 4 yes Figure 3 The extension further includes data from the same principal component (PC9) from the synthetic training data. Figure 4 The R-values ​​observed in the combined real and synthetic training data 2 A value of ~0.253 indicates that some portions of the synthetic measurement training data are inconsistent with the real measurement training data. This can be due to several reasons. For example, portions of the synthetic measurement training data may be inaccurate and may not represent a corresponding correlation with the reference data. As another example, portions of the synthetic measurement training data may cover unrealistic combinations of manufacturing parameters and therefore have little impact on the expected real measurements.

[0093] Step 216, which generates a synthetic training dataset for metrological measurements, may include step 222, which generates the synthetic training dataset by filtering the dimensionality-reduced synthetic training data to include portions of the dimensionality-reduced synthetic training data that satisfy a second correlation threshold with reference data. This second correlation threshold may be the same as (e.g., equal to) or different from the first correlation threshold.

[0094] continue Figures 3 to 4 In an example, step 222 may include filtering the synthetic measurement training data to maintain only the selected R values ​​relative to the reference data. 2 The correlation value (e.g., a second correlation threshold) is used as a portion. In this way, only the portion of the synthetic measurement training data that is relevant to the reference data is retained as part of the synthetic training dataset.

[0095] Figure 5 The diagram illustrates real training data and filtered synthetic training data according to one or more embodiments of this disclosure. Figure 5 Corresponding to Figure 4 The portion of synthetic training data was filtered out to maintain an R-value of 0.87. 2 Values ​​(e.g., the second correlation threshold).

[0096] As previously described in this article, the second relevance threshold may not match the first relevance threshold. Figures 4 to 5 It is displayed in the middle. Figure 6A This illustrates plots of relevant principal components associated with a series of different DRAM test features 104 according to one or more embodiments of this disclosure. For example, Figure 6AIt may include the dimensionality-reduced synthetic training data generated in step 220. Figure 6A In the data, the real training data and the reference data are linearly correlated, with an R-squared of 0.794. 2 The value is 0.048, while the synthetic training data associated with the same principal components has an R-value of 0.048. 2 The value indicates an unfavorable correlation.

[0097] Figure 6B Description of one or more embodiments of this disclosure Figure 6A The actual training data and the plotted graphs of the filtered synthetic training data. For example, Figure 6A It can include data filtered from... Figure 6A The synthetic training dataset is generated by dimensionality reduction and synthetic training data to maintain an R² value of at least 0.66 (the second relevance threshold). For example... Figure 6B As seen in the data, making the second relevance threshold less stringent than the first allows for more synthetic training data within the synthetic training dataset, while still ensuring relevance to the reference data. Therefore, the trained machine learning model can be both accurate and robust.

[0098] It should be understood that, although Figures 3 to 6B The associated descriptions focus on a single relevant principal component, but this is for illustrative purposes only and should not be construed as limiting the scope of this disclosure. In some embodiments, the real training dataset and / or synthetic training dataset contains data associated with multiple relevant principal components.

[0099] In some embodiments, the real training dataset and / or the synthetic training dataset are generated based on mathematical combinations of two or more principal components (e.g., linear combinations, nonlinear combinations, or the like).

[0100] For example, a situation may arise where none of the individual principal components associated with the real training data (e.g., from step 210) satisfy the desired relevance threshold (e.g., the first relevance threshold) with the reference data. However, a situation may arise where a combination of principal components satisfies the desired relevance threshold with the reference data. As an illustration, a weighted linear combination of principal components... It can satisfy the desired relevance threshold with reference data. Any technique can be used to identify the combination of principal components that satisfy the desired relevance threshold, including but not limited to the LINEST technique.

[0101] Therefore, step 212 may involve identifying combinations of principal components that satisfy the desired correlation threshold with the reference data. Similarly, step 220 may involve generating dimensionality-reduced synthetic training data associated with the same combinations of principal components identified in step 212, and step 222 may involve filtering the dimensionality-reduced synthetic training data to maintain a second correlation threshold between the dimensionality-reduced synthetic training data and the reference data.

[0102] Refer again Figure 2A Method 200 may include step 224 of training a machine learning model using a real training dataset and a synthetic training dataset to generate values ​​for a metrology. Method 200 may further include step 226 of generating one or more metrologies for runtime samples from measurement data associated with one or more runtime samples.

[0103] The metrological measurement may correspond to any measurement that can be obtained by analyzing measurement data from the first measurement subsystem 102, including but not limited to measurements of one or more Mueller matrix elements, CD measurements, overlapping measurements, tilt measurements, or asymmetric measurements.

[0104] The metrological measurements can then be used for any purpose. In some embodiments, method 200 includes the step of generating correctable terms for one or more process tools (e.g., but not limited to, lithography tools, etching tools, polishing tools, or the like) based on the metrological measurements. The correctable terms can be used with any type of control technique. For example, a feedback correctable term generated based on metrological measurements of test feature 104 on a test sample can be used to correct deviations in process tools that manufacture similar test feature 104 on subsequent test samples in the same or different batches. As another example, a feedforward correctable term generated based on metrological measurements of test feature 104 on a test sample can be used to control process tools that manufacture additional test features 104 on different sample layers of the same or different test samples.

[0105] For reference Figures 1A to 1D Additional aspects of the metering system 100 will be described in more detail according to one or more embodiments of this disclosure.

[0106] One or more processors 110 of controller 108 may comprise any processor or processing element known in the art. For the purposes of this disclosure, the terms “processor” or “processing element” may be broadly defined to cover any device having one or more processing or logic elements (e.g., one or more microprocessor devices, one or more application-specific integrated circuit (ASIC) devices, one or more field-programmable gate arrays (FPGAs), or one or more digital signal processors (DSPs)). In this sense, one or more processors 110 may comprise any device configured to execute algorithms and / or instructions (e.g., program instructions stored in memory). In some embodiments, one or more processors 110 may embody a desktop computer, a host computer system, a workstation, a graphics computer, a parallel processor, a networked computer, or any other computer system configured to execute a program (which is configured to operate or in conjunction with the measurement subsystem 102), as described throughout this disclosure. Furthermore, different subsystems of the metrology system 100 may include processors or logic elements suitable for implementing at least a portion of the steps described in this disclosure. Therefore, the above description should not be construed as limiting the embodiments of this disclosure but is merely illustrative. Furthermore, the steps described throughout this disclosure can be implemented by a single controller or alternatively by multiple controllers. Additionally, controller 108 may comprise one or more controllers housed within a common housing or multiple housings. In this manner, any controller or combination of controllers can be individually packaged as a module suitable for integration into the metering system 100.

[0107] Memory 112 may comprise any storage medium known in the art suitable for storing program instructions executable by one or more associated processors 110. For example, memory 112 may comprise a non-transitory memory medium. By another example, memory 112 may comprise, but is not limited to, read-only memory (ROM), random access memory (RAM), magnetic or optical storage devices (e.g., magnetic disks), magnetic tape, solid-state drives, and the like. It should be further noted that memory 112 may be housed together with one or more processors 110 within a common controller housing. In some embodiments, memory 112 may be remotely located relative to the physical location of one or more processors 110 and controller 108. For example, one or more processors 110 of controller 108 may access remote memory (e.g., a server) accessible via a network (e.g., the Internet, an intranet, and the like).

[0108] Figures 1B to 1D Variations of the measurement subsystem 102 according to one or more embodiments of the present disclosure are described.

[0109] Figure 1BA simplified schematic diagram of an optical measurement subsystem 102 according to one or more embodiments of the present disclosure is provided. For example, the measurement subsystem 102 may include, but is not limited to, a spectroscopic ellipsometer (SE), an SE having multiple illumination angles, an SE (e.g., using a rotation compensator) for measuring Mueller matrix elements, a single-wavelength ellipsometer, a beam profile ellipsometer (angle-resolved ellipsometer), a beam profile reflectometer (angle-resolved reflectometer), a broadband reflectance spectrometer (spectral reflectometer), a single-wavelength reflectometer, an angle-resolved reflectometer, an imaging system, a scatterometer (e.g., a spot analyzer), or any combination thereof.

[0110] In some embodiments, the measurement subsystem 102 includes an illumination source 118 configured to generate illumination 114 in the form of at least one illumination beam. Illumination 114 from illumination source 118 may include light of one or more selected wavelengths, including, but not limited to, ultraviolet (UV) radiation, visible light radiation, or infrared (IR) radiation. Furthermore, the spatial profile of illumination 114 on sample 106 may be controlled by a field plane aperture to have any selected spatial profile.

[0111] The illumination source 118 may comprise any type of illumination source suitable for providing illumination 114 formed by light. In some embodiments, the illumination source 118 is a laser source. For example, the illumination source 118 may comprise, but is not limited to, one or more narrowband laser sources, broadband laser sources, supercontinuum laser sources, white light laser sources, or the like. In some embodiments, the illumination source 118 comprises a laser sustained plasma (LSP) source. For example, the illumination source 118 may comprise, but is not limited to, an LSP lamp, LSP bulb, or LSP chamber suitable for housing one or more elements capable of emitting broadband illumination when excited into a plasma state by a laser source. In some embodiments, the illumination source 118 comprises a lamp source. In some embodiments, the illumination source 118 may comprise, but is not limited to, an arc lamp, a discharge lamp, an electrodeless lamp, or the like.

[0112] Lighting source 118 may use free space technology and / or optical fiber to provide lighting 114.

[0113] In some embodiments, the measurement subsystem 102 directs illumination 114 to the sample 106 via illumination path 122 through at least one illumination lens 120 (e.g., an objective lens). Illumination path 122 may include one or more optical components adapted to modify and / or adjust illumination 114 and direct illumination 114 to the sample 106. In some embodiments, illumination path 122 includes one or more illumination path optics 124 to shape or otherwise control illumination 114. For example, illumination path optics 124 may include, but are not limited to, one or more field stops, one or more pupil stops, one or more polarizers, one or more filters, one or more beam splitters, one or more diffusers, one or more homogenizers, one or more apodizers, one or more beam shapers, or one or more mirrors (e.g., static mirrors, translation mirrors, scanning mirrors, or the like).

[0114] The measurement subsystem 102 can use any suitable technique to position the sample 106 for measurement. In some embodiments, such as Figure 1B As described, the measurement subsystem 102 includes a sample stage 126, which includes one or more actuators (e.g., linear actuators, tilt / flip actuators, rotation actuators, or the like) to position the sample 106 relative to the illumination beam. In some embodiments, although not explicitly shown, the measurement subsystem 102 includes beam scanning optics (e.g., galvanometer mirrors, scanning prisms, or the like) to adjust the position and / or scan one or more beams of illumination 114.

[0115] In some embodiments, the measurement subsystem 102 includes at least one light-collecting lens 128 to capture a light-collecting signal 116 (e.g., light) and directs this light-collecting signal 116 to one or more detectors 130 via a light-collecting path 132. The light-collecting path 132 may include one or more optical elements adapted to modify and / or adjust the light-collecting signal 116 from the sample 106. In some embodiments, the light-collecting path 132 includes one or more light-collecting path optics 134 to shape or otherwise control the light-collecting signal 116. For example, the light-collecting path optics 134 may include, but are not limited to, one or more field stops, one or more pupil stops, one or more polarizers, one or more filters, one or more beam splitters, one or more diffusers, one or more homogenizers, one or more apodizers, one or more beam shapers, or one or more mirrors (e.g., static mirrors, translation mirrors, scanning mirrors, or the like).

[0116] Measurement subsystem 102 may typically include any number or type of detectors 130. For example, measurement subsystem 102 may include at least one single-pixel detector 130, such as, but not limited to, a photodiode, a burst photodiode, or a single-photon detector. As another example, measurement subsystem 102 may include at least one multi-pixel detector 130, such as, but not limited to, a charge-coupled device (CCD) or complementary metal-oxide-semiconductor (CMOS) device, a line detector, or a time-delay integration (TDI) detector.

[0117] Detector 130 can be positioned at any selected location within the light-collecting path 132. In some embodiments, the measurement subsystem 102 includes detector 130 at a field plane (e.g., a plane conjugate to sample 106) to generate an image of sample 106. In some embodiments, the measurement subsystem 102 includes detector 130 at a pupil plane (e.g., a diffraction plane) to generate a pupil image. In this respect, the pupil image may correspond to the angular distribution of light from sample 106 at detector 130. For example, the diffraction order associated with the diffraction of illumination 114 from sample 106 may be imaged in the pupil plane or otherwise observed. In a general sense, detector 130 may capture any combination of reflected (or transmitted), scattered, or diffracted light from sample 106.

[0118] The illumination path 122 and light-collecting path 132 of the measurement subsystem 102 can be oriented into a wide range of configurations. For example, such as Figure 1B As explained, illumination path 122 and light-collecting path 132 may contain non-overlapping optical paths. In some embodiments, although not explicitly shown, measurement subsystem 102 may include a beam splitter oriented such that a common objective lens can simultaneously direct illumination 114 to sample 106 and capture light-collecting signal 116.

[0119] Figure 1CA simplified schematic diagram of an x-ray measurement subsystem 102 according to one or more embodiments of the present disclosure is shown. This measurement subsystem 102 may be configured as, but is not limited to, a SAXR, SXR, or XPS system. The X-ray characterization system and associated measurement techniques are broadly described in the following: U.S. Patent No. 7,929,667, issued April 19, 2011; U.S. Patent No. 9,885,962, issued February 6, 2018; U.S. Patent No. 10,013,518, issued July 3, 2018; U.S. Patent No. 10,324,050, issued June 18, 2019; U.S. Patent No. 10,352,695, issued July 16, 2019; U.S. Patent No. 10,775,323, issued September 15, 2020; and Lemaillet et al., “Intercomparison between optical and x-ray scatterometry measurements of FinFET structures,” Proceedings of SPIE (The International Society for Optical Engineering). Optical Engineering, April 2013; Kline et al., “X-ray scattering critical dimensional metrology using a compact x-ray source for next generation semiconductor devices,” Journal of Micro / Nanolithography, MEMS, and MOEMS 16.1 (2017); U.S. Patent No. 11,333,621, published May 17, 2022; and U.S. Patent Application No. 2021 / 0207956, published July 8, 2021; the entire contents of all the aforementioned U.S. patents are incorporated herein by reference.

[0120] In some embodiments, the illumination source 118 is an X-ray source configured to produce X-ray illumination 114 with any particle energy (e.g., soft X-rays, hard X-rays, or the like). The measurement subsystem 102 may then include any combination of components suitable for capturing the associated light-collecting signal 116 (which may include, but is not limited to, X-ray emission, optical emission, or particle emission).

[0121] For example, measurement subsystem 102 may include at least one x-ray illumination lens 120 and / or illumination path optics 124 adapted to collimate or focus x-ray illumination 114. Although not shown, measurement subsystem 102 may further include at least one x-ray collecting path lens and / or collecting path optics adapted to collect, collimate, and / or focus the light-collecting signal 116 from sample 106. Furthermore, measurement subsystem 102 may include various illumination path optics 124 and / or collecting path optics 134, such as, but not limited to, x-ray collimators, spectral x-ray optics (e.g., grazing incidence ellipsoidal mirrors), multi-capillary optics (e.g., hollow capillary x-ray waveguides), multilayer optics or systems, or any combination thereof. In an embodiment, the measurement subsystem 102 includes an x-ray detector 130, such as, but not limited to, an x-ray monochromator (e.g., a crystal monochromator, such as a Loxley-Tanner-Bowen monochromator or the like), an x-ray aperture, an x-ray beam stop, or a diffraction optics (e.g., a zone plate).

[0122] Figure 1D A simplified schematic diagram of a particle beam measurement subsystem 102 according to one or more embodiments of the present disclosure is shown. This measurement subsystem 102 may be configured as, but is not limited to, SEM, CD-SEM, grayscale SEM, or TEM.

[0123] In some embodiments, illumination source 118 includes a particle source (e.g., an electron beam source, an ion beam source, or the like) such that illumination 114 includes a particle beam (e.g., an electron beam, a particle beam, or the like). Illumination source 118 may include any particle source known in the art suitable for generating particle illumination 114. For example, illumination source 118 may include, but is not limited to, an electron gun or an ion gun. In some embodiments, illumination source 118 is configured to provide a particle beam with tunable energy. For example, illumination source 118 including an electron source may, but is not limited to, provide an accelerating voltage in the range of 0.1 kV to 30 kV. As another example, illumination source 118 including an ion source may, but does not need to, provide an ion beam with energy in the range of 1 to 50 keV.

[0124] In some embodiments, the measurement subsystem 102 includes one or more particle focusing elements. For example, the one or more particle focusing elements may include, but are not limited to, a single particle focusing element or one or more particle focusing elements forming a composite system. In some embodiments, the one or more particle focusing elements include an illumination lens 120 configured to direct a particle illumination beam to the sample 106. Furthermore, the one or more particle focusing elements may include any type of electronic lens known in the art, including, but not limited to, electrostatic, magnetic, single-potential, or dual-potential lenses.

[0125] In some embodiments, the measurement subsystem 102 includes one or more particle detectors 130 for imaging or otherwise detecting particles emitted from the sample 106. For example, detector 130 may include an electron collector (e.g., a secondary electron collector, a backscattered electron detector, or the like). As another example, detector 130 may include a photon detector (e.g., a light detector, an X-ray detector, a scintillation element coupled to a photomultiplier tube (PMT) detector, or the like) for detecting electrons and / or photons from the sample surface.

[0126] For general reference Figures 1A to 6B The systems and methods disclosed herein offer numerous advantages and applications, such as, but not limited to, identifying process spaces by correlation with a true measured PC to a real reference, identifying accurate synthetic spectra using the same correlation, and filtering inaccurate synthetic spectra using the relationship between the true PC and the real reference. The systems and methods can further provide options for selecting a different set of synthetic spectra for each key parameter, training each key parameter with different synthetic spectra while maintaining accurate and process-consistent synthetic spectra, setting criteria using measured spectra and a real reference to maintain or exclude synthetic spectra, establishing linear or nonlinear relationships between combinations of measured PC and measured reference, and setting criteria to maintain synthetic spectra and exclude non-compliant synthetic spectra.

[0127] Reference-based synthesis of DOEs generates low-sensitivity features applicable to any type of low-reference sampling use case or to measure low-sensitivity characteristics, such as semiconductor process CDs and film thicknesses, as well as real-time process monitoring of changes in material size (thickness), composition, or other properties, such as atomic layer growth in semiconductor processes, chemical or biological reactions that lead to changes in material quantity or properties. Target semiconductor metrology applications and benefits include, but are not limited to, etched SCDs of logic devices at nodes of 2 nm and above 14 nm, GAA nanosheets and nanowire devices forming FEOLs, and interconnect MEOL processes with many features ranging from several nanometers to even as low as 1 nm. This approach is expected to improve robustness by 2 to 3 times and enable capable metrology for these advanced node processes, thereby increasing the time required for formulation re-invention from 3 weeks to 3 months and providing formulations with better robustness, accuracy, precision, and stability.

[0128] Furthermore, the systems and methods disclosed herein can be applied to the etching SCD of DRAM devices at nodes of 10 nm or below, where many small features and dimensions need to be measured and controlled, but sufficient reference sampling is lacking. The methods will amplify accurate synthetic spectra and references for these small features, thereby enabling the development of formulations and libraries with better accuracy, stability, and robustness to process changes. For logic / wafer foundry GAA devices, the systems and methods disclosed herein can facilitate the monitoring and control of high dielectric constants and metal gate processes at GAA nodes of 2 nm and above, where thicknesses become increasingly thinner and more composite layers need to be measured, including dipole-doped layers of only one or two monoatom layers. It also improves the measurement performance and capability of individual layers in logic / wafer foundry GAA devices with Si / SiGe superlattices, having 7 to 12 or more layers requiring individual layer thickness and composition measurements. This reduces resolution time and increases blind test pass rates. Finally, it is expected to reduce formulation development time for DRAM die in-place pairs (IDOs) and enable more robust formulation quality metrics (QM) for tracking measurement robustness.

[0129] The objects described herein sometimes refer to different components contained within or connected to other components. It should be understood that such depicted architectures are illustrative only, and many other architectures can in fact be implemented to achieve the same functionality. Conceptually, any arrangement of components used to achieve the same functionality is effectively “associated” to achieve the desired functionality. Therefore, any two components combined herein to achieve a particular functionality can be considered “associated” with each other to achieve the desired functionality, regardless of the architecture or intermediate components. Similarly, any two such associated components can also be considered “connected” or “coupled” to each other to achieve the desired functionality, and any two components that can be suchly associated can also be considered “coupleable” to each other to achieve the desired functionality. Specific examples of coupleability include, but are not limited to, physically interactive and / or physically interactive components and / or wirelessly interactive and / or logically interactive components.

[0130] It is believed that this disclosure and its many accompanying advantages will be understood from the foregoing description, and it will be appreciated that various changes can be made to the form, construction, and arrangement of the components without departing from the subject matter of the disclosure or sacrificing all its significant advantages. The forms described are for illustrative purposes only, and the appended claims are intended to cover and encompass such changes. Furthermore, it should be understood that the invention is defined by the appended claims.

Claims

1. A measurement system comprising: A controller comprising one or more processors configured to execute program instructions, thereby causing the one or more processors to implement a metering formulation by: The following methods are used to generate a real training dataset for metrological measurements: Receive real training data of test features from one or more training samples from the first measurement subsystem; Receive reference data associated with the metrological measurements of the test feature from the second metrology subsystem; Perform dimensionality reduction on the real training data; Identify one or more principal components of the real training data that satisfy a first correlation threshold with the reference data; and The real training dataset is generated by filtering the real training data to include the portions of the real training data that are associated with the one or more relevant principal components. The synthetic training dataset for the metrological measurements is generated through the following: Generate synthetic training data for multiple simulated test features with known simulated values ​​of the metrological measurements; Extract one or more relevant principal components from the synthetic training data as dimensionality-reduced synthetic training data; and The synthetic training dataset is generated by filtering the dimensionality-reduced synthetic training data to include the portion of the dimensionality-reduced synthetic training data that satisfies a second correlation threshold with the reference data. The machine learning model is trained using the real training dataset and the synthetic training dataset to generate the values ​​of the measurement. and The metrological measurements of the one or more runtime samples are generated from measurement data associated with one or more runtime samples.

2. The metering system according to claim 1, wherein the dimensionality reduction operation includes principal component analysis.

3. The metrology system according to claim 1, wherein the metrology measurement includes at least one of stacking measurement or critical dimension measurement.

4. The metrology system according to claim 1, wherein the first metrology subsystem includes an optical metrology tool.

5. The metrology system according to claim 1, wherein the first metrology subsystem comprises at least one of a spectral ellipticity measuring tool or a spectral reflectance measuring tool.

6. The metrology system of claim 1, wherein the second metrology subsystem comprises at least one of a particle beam metrology tool or an X-ray metrology tool.

7. The metrology system according to claim 1, wherein the second metrology subsystem comprises at least one of a transmission electron microscope, a transmission small-angle X-ray scattering instrument, a scanning electron microscope, a critical-size scanning electron microscope, or an atomic force microscope.

8. The metrology system according to claim 1, wherein at least one of the first correlation threshold or the second correlation threshold is a goodness-of-fit threshold.

9. The metering system according to claim 1, wherein at least one of the first correlation threshold or the second correlation threshold is R. 2 Threshold.

10. The metering system according to claim 1, wherein the first correlation threshold is equal to the second correlation threshold.

11. The metering system according to claim 1, wherein the second correlation threshold is greater than the second correlation threshold.

12. A measurement system comprising: First metrology subsystem; Second metering subsystem; and A controller comprising one or more processors configured to execute program instructions, thereby causing the one or more processors to implement a metering formulation by: The following methods are used to generate a real training dataset for metrological measurements: Receive real training data of test features from one or more training samples from the first measurement subsystem; Receive reference data associated with the metrological measurement of the test feature from the second metrology subsystem; Perform dimensionality reduction on the real training data; Identify one or more principal components of the real training data that satisfy a first correlation threshold with the reference data; and The real training dataset is generated by filtering the real training data to include the portions of the real training data that are associated with the one or more relevant principal components. The synthetic training dataset for the metrological measurements is generated through the following: Generate synthetic training data for multiple simulated test features with known simulated values ​​of the metrological measurements; Extract one or more relevant principal components from the synthetic training data as dimensionality-reduced synthetic training data; and The synthetic training dataset is generated by filtering the dimensionality-reduced synthetic training data to include the portion of the dimensionality-reduced synthetic training data that satisfies a second correlation threshold with the reference data. and The machine learning model is trained using the real training dataset and the synthetic training dataset to generate the values ​​of the measurement. and The metrological measurements of the one or more runtime samples are generated from measurement data associated with one or more runtime samples.

13. The metrology system of claim 12, wherein the metrology measurement includes at least one of stacking measurement or critical dimension measurement.

14. The metrology system of claim 12, wherein the first metrology subsystem comprises an optical metrology tool.

15. The metrology system of claim 12, wherein the first metrology subsystem comprises at least one of a spectral ellipticity measuring tool or a spectral reflectance measuring tool.

16. The metrology system of claim 12, wherein the second metrology subsystem comprises at least one of a particle beam metrology tool or an X-ray metrology tool.

17. The metrology system of claim 12, wherein the second metrology subsystem comprises at least one of a transmission electron microscope, a transmission small-angle X-ray scattering instrument, a scanning electron microscope, a critical-size scanning electron microscope, or an atomic force microscope.

18. The metrology system of claim 12, wherein at least one of the first correlation threshold or the second correlation threshold is a goodness-of-fit threshold.

19. The metering system of claim 12, wherein at least one of the first correlation threshold or the second correlation threshold is R. 2 Threshold.

20. The metering system of claim 12, wherein the first correlation threshold is equal to the second correlation threshold.

21. The metering system according to claim 12, wherein the second correlation threshold is greater than the second correlation threshold.

22. A measurement method, comprising: The following methods are used to generate a real training dataset for metrological measurements: Receive real training data of test features from one or more training samples from the first measurement subsystem; Receive reference data associated with the metrological measurements of the test feature from the second metrology subsystem; Perform dimensionality reduction on the real training data; Identify one or more principal components of the real training data that satisfy a first correlation threshold with the reference data; and The real training dataset is generated by filtering the real training data to include the portions of the real training data that are associated with the one or more relevant principal components. The synthetic training dataset for the metrological measurements is generated through the following: Generate synthetic training data for multiple simulated test features with known simulated values ​​of the metrological measurements; Extract one or more relevant principal components from the synthetic training data as dimensionality-reduced synthetic training data; and The synthetic training dataset is generated by filtering the dimensionality-reduced synthetic training data to include the portion of the dimensionality-reduced synthetic training data that satisfies a second correlation threshold with the reference data. The machine learning model is trained using the real training dataset and the synthetic training dataset to generate the values ​​of the measurement. and The metrological measurements of the one or more runtime samples are generated from measurement data associated with one or more runtime samples.

Images