Measurement of structures in the presence of signal contamination

Abstract

A hybrid modeling or machine learning model based on local gradients can be used to reduce or eliminate the contamination of metrology data by non-target signals caused by measurement spots incident on target and non-target regions. The hybrid modeling method based on local gradients uses a hybrid model based on a model of the target region and a local gradient term of the measured signal, which can be determined from the scan data. The machine learning method obtains hybrid metrology data from a plurality of positions relative to the target. The hybrid metrology data is a mixture of target signals from the target and non-target signals from non-target regions, and is used by a trained machine learning model to determine target metrology data comprising target signals from the target without non-target signals, and to determine parameters of interest of the target.

Description

Cross-references to related applications

[0001] This application claims priority to U.S. non-provisional application No. 18 / 495,247, filed October 26, 2023, entitled “MEASUREMENTS OF STRUCTURES INPRESENCE OF SIGNAL CONTAMINATIONS”, which has been assigned to the assignee of this application and is incorporated herein by reference in its entirety. Technical Field

[0002] The subject described in this article generally relates to metrology, and more specifically to the measurement of structures using metrological equipment with a measurement spot size larger than the target area. Background Technology

[0003] The semiconductor and other similar industries typically use optical metrology equipment to provide non-contact evaluation of substrates during processing. Optical metrology techniques, such as ellipsometrics and reflectometry, generally operate by irradiating a sample with a probe beam of electromagnetic radiation and then detecting and analyzing the energy reflected and / or transmitted. The probe beam can be polarized or unpolarized radiation and can include radiation of one or more wavelengths. Ellipsometry typically measures the change in the polarization state of the reflected beam after interaction with the sample, while reflectometry measures the magnitude change in the intensity of the reflected beam.

[0004] For example, in optical metrology, it is desirable for the measurement spot generated by the optical metrology equipment to be completely incident on the target area of ​​the sample under test in order to improve the accuracy and precision of the measurement. If the size of the measurement spot is larger than the size of the target area, the resulting optical signal will be contaminated by signals from areas outside the target area, which can make it difficult to accurately measure the desired structure. The size of the measurement spot is conventionally determined by the optical system of the optical metrology equipment, but redesigning the optical system to reduce the size of the measurement spot is an expensive and time-consuming task, and may ultimately be insufficient to accurately measure increasingly smaller geometries. Therefore, techniques other than optical system redesign are desired to compensate for signal contamination. Summary of the Invention

[0005] Measurement spots larger than or misaligned with the target area can obtain metrological data from both the target and non-target areas, resulting in contamination of the obtained metrological data by signals from the non-target areas. Local gradient-based mixture models or machine learning models can be used to reduce or eliminate this contamination. A local gradient-based mixture model includes a model of the target area and local gradient terms of the measured signal, which can be determined from the scan data. Metrological data from the target area can be fitted to the local gradient-based mixture model by fitting parameters in the model of the target area and fitting at least one coefficient of the local gradient. In another approach, a machine learning model is trained based on the variation in the pattern of metrological data obtained from different locations of a reference target. During measurement, target mixed metrological data is obtained from multiple different locations relative to the target and provided as input data to a trained machine learning model. This target mixed metrological data is a mixture of target signals from the target and non-target signals from non-target areas. The machine learning model determines target metrological data based on the input data, which includes target signals but no non-target signals. The target metrological data can then be used to determine parameters of interest in the target area, for example, by fitting the target metrological data to the modeling data of the target.

[0006] In one specific implementation, a method for measuring parameters of interest (POIs) of a target on a sample includes: acquiring metrological data of the target using a metrological device, the metrological data being a mixture of target signals from the target and non-target signals from non-target regions; fitting the metrological data to a hybrid model incorporating the target and the local gradients of the measured signals; and determining the POIs of the target based on the fitting of the metrological data to the hybrid model.

[0007] In one embodiment, a metrology device configured to measure parameters of interest of a target on a sample includes: a source configured to generate radiation to be incident on the target on the sample; at least one detector configured to detect radiation generated from the target in response to the incident radiation; and at least one processor coupled to the at least one detector. The at least one processor is configured to acquire metrological data of the target using the metrology device. The metrological data is a mixture of target signals from the target and non-target signals from non-target regions. The at least one processor is further configured to fit the metrological data to a hybrid model comprising a model of the target and local gradients of the measured signals. The at least one processor is further configured to determine the parameters of interest of the target based on the fitting of the metrological data to the hybrid model.

[0008] In one specific implementation, a method for measuring a parameter of interest (MOI) of a target on a sample includes: acquiring mixed metrological data at multiple different locations relative to the target using a metrological device, wherein the mixed metrological data from each different location is a mixture of a target signal from the target and a non-target signal from a non-target region. The method further includes: determining target metrological data based on the mixed metrological data of the target using a trained machine learning model, wherein the target metrological data includes the target signal from the target but not the non-target signal from the non-target region; and determining the MOI of the target based on the target metrological data.

[0009] In one embodiment, a metrology device configured to measure a parameter of interest (MOI) of a target on a sample includes: a source configured to generate radiation to be incident on the target on the sample; at least one detector configured to detect radiation generated from the target in response to the incident radiation; and at least one processor coupled to the at least one detector. The at least one processor is configured to acquire mixed metrology data at multiple different locations relative to the target using the metrology device, wherein the mixed metrology data from each different location is a mixture of a target signal from the target and a non-target signal from a non-target region. The at least one processor is further configured to determine target metrology data based on the mixed metrology data using a trained machine learning model, wherein the target metrology data includes the target signal from the target but no non-target signal from the non-target region; and to determine the MOI of the target based on the target metrology data. Attached Figure Description

[0010] Figure 1 A schematic diagram of a metering device configured to reduce or eliminate contamination from signals originating from non-target areas is shown.

[0011] Figure 2 A schematic diagram of a metering device configured to reduce or eliminate contamination from signals originating from non-target areas is shown.

[0012] Figure 3A and Figure 3B The measurement spots on the surface of the sample produced by normal incident and oblique incident irradiation are illustrated respectively.

[0013] Figure 3C and Figure 3D Top views of the measurement spot generated by the metrology equipment and graphs of the measured spectrum relative to the true spectrum are shown respectively.

[0014] Figure 4A The image shows top views of several targets of different sizes and measurement spots.

[0015] Figure 4BThis is an example for... Figure 4A The graph shows the mean square error of the fit of the signal obtained from the measurement spot on the target area.

[0016] Figure 5A and Figure 5B Top views are shown, illustrating line scan measurements of a target area by moving the measurement spot above the target area along the X and Y axes, respectively.

[0017] Figure 6 This is an example in, for example Figure 5A The graph illustrates examples of different spectral signals collected when a measurement spot is linearly scanned over a target region.

[0018] Figure 7 An exemplary flowchart is shown, depicting an example method for measuring the parameters of interest of a target on a sample using a local gradient-based hybrid model, according to some specific implementations.

[0019] Figure 8A A top view is shown illustrating a line scan measurement of a target area by moving the measuring spot above the target area.

[0020] Figure 8B A top view of the target area is shown, where the measurement spot is positioned at different locations and covers different percentages of the target area.

[0021] Figure 9 An example is shown of using a trained machine learning model to measure the parameters of interest in a target region.

[0022] Figure 10 An exemplary flowchart is shown, depicting an example method for measuring parameters of interest of a target on a sample using a machine learning model, according to some specific implementations. Detailed Implementation

[0023] During the manufacture of semiconductors and similar devices, it is often necessary to monitor the manufacturing process through non-destructive measurements of these devices. One type of metrology that can be used to perform non-destructive measurements of samples during processing is optical metrology, which uses radiation incident normally or obliquely, and can use a single wavelength or multiple wavelengths. Examples of optical metrology include ellipsometrics, reflectance measurement, Fourier transform infrared spectroscopy (FTIR), etc. Other types of metrology can also be used, including X-ray metrology, photoacoustic metrology, electron beam (E-beam) metrology, etc.

[0024] Metrological instruments used for measurements such as thin film metrology and optical critical dimension (OCD) metrology, as well as other types of measurements, illuminate a sample with a probe beam of electromagnetic radiation and detect and analyze reflected or transmitted signals. The measurement spot is a region on the sample surface from which light is reflected (or transmitted through) and subsequently received by the detector of the optical metrology instrument. The probe beam typically does not produce a measurement spot with sharp edges or boundaries; instead, the probe beam has an intensity distribution such that the total beam power is confined to a small area (i.e., the measurement spot).

[0025] The target area measured by optical metrology equipment can be a specially designed target, such as within a scribe line placed on a substrate between processed dies, or it can be within the equipment, such as a specific area within the die. The target area can be determined by the structure or characteristics of the sample produced during processing. The target area can be a defined region, such as a physical frame or area on the sample sometimes referred to as a device pad, or it can simply be an undefined area on the sample to be measured.

[0026] For accuracy and precision, the incident probe beam should produce a measurement spot on the sample that is completely confined to the target area. If the measurement spot is larger than the target area, the resulting optical signal will be contaminated by signals from areas outside the target area, making accurate measurements difficult to obtain. As the geometry of devices in semiconductor and similar industries, along with their corresponding target areas, continues to shrink, it becomes increasingly difficult to generate a measurement spot that is spatially confined to the target area. The problems associated with measurement spot sizes larger than the target area pose a significant challenge, especially for measurements using advanced equipment.

[0027] One possible solution is to reduce the measurement spot size to be smaller than the target area, so that the obtained signal originates only from the target area. However, reducing the actual measurement spot size requires a redesign of the current tool's optical system, which is an expensive and time-consuming task. Furthermore, physics imposes limitations on reducing the measurement spot size, which may make a redesign of the optical system insufficient for some device sizes.

[0028] Another solution is to use an incoherent mixing model, which models these signals based on the assumption that the mixed signals from the target region and the surrounding region are incoherently mixed in the intensity domain. However, the assumption of incoherent signal mixing is not necessarily valid, as coherent signal mixing can also occur. Furthermore, incoherent mixing models rely on structural information from the surrounding region, which is often unknown and irregular (e.g., aperiodic, with arbitrary defects), and therefore difficult to model, if not impossible.

[0029] As discussed in this paper, signal contamination from non-target regions can be reduced or eliminated to generate accurate measurements. This contamination can be eliminated or reduced based on either modeling or machine learning methods. Modeling methods use a hybrid model based on a pure target model and the local gradient term of the measured signal, which can be used to absorb signal contamination from surrounding non-target regions. Machine learning methods use variations in patterns across multiple spectra to predict signals from the target region without contamination from surrounding non-target regions. Neither method requires a redesign of the metrology system and therefore can utilize data collected from current tools. Furthermore, the methods discussed in this paper are not limited to large measurement spots and small target regions but can be applied to any situation involving an invariant target region and varying, unknown non-target regions.

[0030] Both hybrid model methods based on local gradients and machine learning methods use local signals as direct inputs, thus they can be applied to any type of data, such as elliptic polarization measurement data, such as Mueller matrix (MM), Jones matrix, psi and delta data; reflectance measurement data, such as reflectance collected at different polarizer angles; interferometric measurement data; Fourier transform infrared spectroscopy (FTIR) data, etc.

[0031] Figure 1 A schematic diagram of a metering device 100 configured to reduce or eliminate contamination from signals originating from non-target areas is shown. The metering device includes a head 102 coupled to a computing system 160, as discussed herein. Figure 1 The metrology device 100 illustrated is a normal incidence system and can be, for example, a reflectometer, spectroreflectometer, FTIR, or any other normal incidence metrology device. Multiple heads (i.e., different metrology devices) can be combined in the metrology device 100 if desired. The computing system 160 can be configured to analyze data obtained from the sample 130 by the metrology device 100 and to control the movement of the stage 120 holding the sample 130 via actuator 121 and / or head 102. The stage 120 can be horizontally moved in Cartesian (i.e., X and Y) coordinates (as indicated by arrows 123 and 124) or polar (i.e., R and θ) coordinates, or some combination of both. The stage 120 and / or head 102 can also be vertically moved, for example, for focusing.

[0032] The head 102 may include an optical system 104 comprising a detector 116 (such as a spectrometer including wavelength dispersive elements and a camera) and a light source 106 (such as a xenon arc lamp and / or a deuterium lamp) that generates light incident on a target. The detector detects light generated from the target in response to the incident light. In operation, the light generated by the light source 106 may be directed along an optical axis 108 (e.g., via a beam splitter 110) toward a sample 130 including a target region 132. The target region 132 may be a specially designed target (e.g., a device pad) or may be part of the device itself. An objective lens 112 focuses light onto the target region 132 and receives light reflected from the target region 132. The reflected light may pass through the beam splitter 110 and be focused onto the detector 116 by a lens 114. The detector 116 provides data signals to a computing system 160. The objective lens 112, beam splitter 110, lens 114, and detector 116 are merely examples of typical optical elements that can be used. Additional optical elements, such as polarizers and / or analyzers, may be used if desired. Furthermore, additional optical elements such as field stops and lenses may typically be present in the optical system 104.

[0033] Optical system 104 generates a measurement spot on the surface of sample 130. The measurement spot has a spot size constrained by the components of optical system 104. As discussed, it is desirable that the measurement spot size be smaller than the size of target region 132, such that the reflected light received by optical system 104 originates only from target region 132 and does not include light reflected from the target neighborhood (i.e., the area on sample 130 outside and around target region 132). As the geometry of devices in semiconductor and similar industries continues to shrink, the size of the target similarly decreases, making it more difficult to generate a measurement spot size smaller than the target. Therefore, in some cases, for example, if the measurement spot size is larger than the size of target region 132 or if the measurement spot is not properly aligned with target region 132, the reflected light received by optical system 104 may be reflected not only from target region 132 but also from outside the target region.

[0034] The computing system 160 is coupled to the head 102 and receives signals that may be contaminated by non-target regions from the detector 116, and may reduce or eliminate signal contamination based on modeling or machine learning methods, as discussed herein, to perform measurements of at least one parameter of the sample 130. For example, the computing system 160 may be a workstation, personal computer, central processing unit, or other suitable computer system, or multiple systems. It should be understood that the computing system 160 may be a single computer system or multiple separate or linked computer systems, which are interchangeably referred to as computing system 160, at least one computing system 160, or one or more computing systems 160. The computing system 160 may be included in, connected to, or otherwise associated with the metrology device 100. Different subsystems of the metrology device 100 may each include a computing system configured to perform steps associated with the associated subsystem. The computing system 160 may be communicatively coupled to the detector 116 in any manner known in the art. For example, the computing system 160 may be coupled to a separate computing system associated with the detector 116. The computing system 160 may be configured to receive and / or acquire metering data or information from one or more subsystems of the metering device 100 (e.g., detector 116) via a transmission medium that may include wired and / or wireless portions. Therefore, the transmission medium may serve as a data link between the computing system 160 and other subsystems of the metering device 100.

[0035] The computing system 160 includes at least one processor 162, a memory 164, and a user interface (UI) 168 communicatively coupled via a bus 161. The memory 164 or other non-transitory computer-available storage medium includes computer-readable program code 166 embodied therein, and can be used by the computing system 160 to cause one or more computing systems 160 to control the metering device 100 and perform the functions discussed herein, including eliminating or reducing signal contamination based on modeling or machine learning methods. For example, as illustrated, the memory 164 may include computer-readable program code 166 or instructions to cause the processor 162 to perform modeling 166m or machine learning (ML) 166ml, as discussed herein. It should be understood that the processor 162 may be configured to perform only one of modeling 166m and machine learning 166ml, or may be configured to perform both modeling 166m and machine learning 166ml.

[0036] The computing system 160 may be further configured to analyze the acquired data and determine one or more parameters of the sample 130 under test. The results of the data analysis (e.g., parameters characterizing the structure of the device under test) may be reported, for example, stored in a memory 164 associated with the sample 130 and / or indicated to a user via a UI 168, an alarm, or other output device. Furthermore, the results from the analysis may be reported and fed forward or back to processing equipment to adjust appropriate manufacturing steps, thereby compensating for any detected discrepancies during the manufacturing process. For example, the computing system 160 may include a communication port 169, which may be any type of communication connection, such as a connection to the Internet or any other computer network. The communication port 169 may be used to receive instructions for programming the computing system 160 to perform any one or more of the functions described herein and / or, during feedforward or feedback, to output signals, for example, containing measurement results and / or instructions, to another system, such as an external process tool, to adjust process parameters associated with the manufacturing process steps of the sample based on the measurement results.

[0037] In view of this disclosure, those skilled in the art can implement the data structures and software code described in this detailed description for automatically performing one or more actions, and these data structures and software code are stored on, for example, a computer-usable storage medium (e.g., memory 164), which can be any device or medium capable of storing code and / or data for use by computing system 160. Computer-usable storage media can be, but is not limited to, read-only memory, random access memory, magnetic and optical storage devices such as disk drives, magnetic tape, etc. Furthermore, the functions described herein can be wholly or partially embodied within a circuit system of application-specific integrated circuits (ASICs) or programmable logic devices (PLDs), and these functions can be embodied in a computer-understandable descriptor language that can be used to create ASICs or PLDs that operate as described herein.

[0038] Figure 2 A schematic diagram of another metering device 200 configured to reduce or eliminate contamination from signals from non-target areas is shown. This other metering device may be coupled to a computing system 160, as discussed herein. Figure 2 The metrology device 200 illustrated herein can be an oblique incident system, and can be, for example, an ellipsometer, a spectral ellipsometer, a Mueller matrix ellipsometer, etc. Multiple heads (i.e., different metrology devices) can be combined in the same metrology device 200 if desired. The operation of the calculation system 160 to reduce or eliminate signal contamination from non-target areas can be similar to that described above. Figure 1The operation discussed herein, but when coupled to the metrology device 200, the computing system 160 is also configured to analyze the data acquired by the metrology device 200 (e.g., elliptic polarization measurement) and determine one or more parameters of the sample to be tested.

[0039] The metrology apparatus 200 is illustrated as including a broadband light source 202, a polarization state generator 203 having a polarizer 204 and a rotation compensator 205, and a lens system 206 that focuses illumination light 211 onto a measurement spot positioned on the surface of a sample 230 on a stage 208. Due to the polarizer 204 and the rotation compensator 205, the incident illumination light 211 has a known polarization state. The polarization state of the light reflected from the sample 230 is analyzed by a polarization state analyzer 215, for example, by passing the reflected light 213 through another polarizer (typically referred to as analyzer 212) after passing through another lens system 210. After passing through analyzer 212, the reflected light 213 is focused by lens system 214 onto a detector 216 (e.g., a spectrometer) coupled to a computing system 160. In use, the sample to be measured will change the polarization state of the incident light, which will change the amplitude and phase of the resulting signal from detector 216. By using changes in intensity and phase, the material properties of sample 230 can be determined. This is the essence of ellipsometric measurement and is well known in the art.

[0040] The optical system of the spectrometer 200 generates a measurement spot on the surface of the sample 230, including the target region 232. Similarly, the size of the measurement spot is constrained by the components of the system of the meter 200. In some cases, for example, the measurement spot size may be larger than the target region 232, or the measurement spot may not be properly aligned with the target region 232, and therefore, reflected light may be reflected not only from the target region 232 but also from outside the target region. A computational system 160, coupled to receive a signal (which may be contaminated by non-target regions) from a detector 216, can be configured to reduce or eliminate signal contamination based on modeling or machine learning methods, as discussed herein, to perform a measurement of at least one parameter of the sample 230.

[0041] It should be understood that while spectroreflectometers and spectrorellometric ellipsometers are specifically discussed in this paper, the processes for reducing or eliminating contamination of signals from non-target regions are not limited to these. The reduction or elimination of contamination of signals from non-target regions as discussed herein can be applied to any desired optical metrology equipment.

[0042] Figure 3A An example is a measurement spot 310 on the surface of a sample 301 generated by normal incident irradiation 302, such as by... Figure 1The measuring spot 310 is generated by the metrology device 100. The measuring spot 310 is illustrated as being generated using a refractive lens 304, but a reflective lens may also be used. As illustrated, the measuring spot 310 is incident on a target region 306, which, for clarity, is shown with crosshairs and an exaggerated size relative to the sample 301. The target region 306 may be determined by the structure or features of the sample 301 and may be a physically designated area on the sample (e.g., a square or framed device pad manufactured on the surface of the sample), or it may simply be an undefined area on the sample to be measured. The size of the measuring spot 310 is determined by the optical system of the metrology device (e.g., represented by lens 304). Figure 3A As can be seen, the size of the measuring spot 310 is exemplified as being larger than the target region 306. Therefore, the signal received by the metrology device will include a mixture of light reflected from the target region 306 and light reflected from the non-target region 307 (i.e., the surrounding area outside the target region 306). However, the contamination of the signal by the non-target region can be reduced or eliminated by using a modeling method or by using a machine learning method. This modeling method uses a mixture model based on the target model and local spectral gradient terms, while the machine learning method uses variations in the patterns of multiple spectra to predict the signal from the target region.

[0043] Figure 3B Similar to Figure 3A This illustrates a measurement spot 312 that can be produced by tilted illumination 303, such as by... Figure 2 The measurement spot 312 is generated by the metrology device 200. The measurement spot 312 is illustrated as being generated using a reflecting lens 314, but a refractive lens may also be used if desired. The size of the measurement spot 312 is determined, for example, by the lens 314, and is illustrated as being larger than the target region 306. Additionally, as illustrated, oblique illumination 303 is obliquely incident on the surface of the sample 301, thereby producing an elliptical magnified measurement spot 312. Contamination of the resulting signal from non-target regions (i.e., outside the target region 306) can be reduced or eliminated by using modeling methods or by using machine learning methods. The modeling method uses a hybrid model based on the target model and local spectral gradient terms, while the machine learning method uses variations in patterns of multiple spectra to predict the signal from the target region.

[0044] Figure 3C Examples are given by metering equipment and in Figure 3A A top view of the measurement spot 310 as illustrated in the image. Figure 3C The diagram illustrates a measuring spot 310, the center of which (illustrated as a cross pattern 320) is aligned with the center of the target region 306. The measuring spot 310 is illustrated as being larger than the target region 306, such that the measuring spot 310 at least partially covers the target location, for example, the non-target region 307.

[0045] Figure 3D This is a graph illustrating the measured spectrum 322 Y(λ) generated by the measurement spot 310. The measurement spot 310 is at least partially outside the target, i.e., it includes the non-target region 307, and therefore, the measured spectrum 322 is generated by mixing the signal from the target region 306 with the signal generated from the non-target region 307. For comparison, in... Figure 3D The diagram also illustrates a "true" target spectrum 324, which is the spectrum that would be generated if the measuring spot 310 illuminates only the target region 306 (i.e., only the location on the target). The measured spectrum 322, generated by the mixed signals from the target region 306 and the surrounding non-target regions 307, differs from the true target spectrum 324, and therefore, the parameters of the target region 306 determined using the measured spectrum 322 will be inaccurate.

[0046] During optical metrology, the measured data (e.g., the measured spectrum 322) is typically compared with modeling data of the structure under test. Modeling data can be calculated using physics-based techniques such as tightly coupled-wave analysis (RCWA), finite-difference time-domain (FDTD), or the finite element method (FEM), which require a detailed understanding of the structure under test. For example, modeling requires that preliminary structural and material information about the sample be known in order to generate an accurate representative model of the sample, which may include one or more variable parameters. Preliminary structural and material information about the sample may include the structure type and a physical description of the sample, with nominal values ​​for various parameters such as layer thickness, linewidth, spacing width, sidewall angles, etc., and the range in which these parameters can vary. The sample may also include one or more immutable sample parameters, i.e., parameters that are not expected to change significantly during manufacturing. Typically, a library of modeling data with variations of multiple parameters in the model can be pre-generated to increase measurement throughput; however, in some instances, modeling data can be calculated in real time. For example, in nonlinear regression, the measured data is compared with the modeling data for each parameter variation until a good fit is achieved between the modeling data and the measured data. This can be determined, for example, based on the mean squared error (MSE). When a good fit is achieved between the measured data and the modeling data, the model parameters corresponding to the modeling data can be considered an accurate representation of the parameters of the structure under test.

[0047] Representative models of the structure under test typically do not include structural or material information from non-target regions. For example, the structural information of non-target regions is often unknown and may be irregular (e.g., aperiodic, with arbitrary defects), and theoretically may be difficult or impossible to model. Therefore, when the measured data is contaminated by signals from non-target regions (e.g., such as...), the structure may be difficult or impossible to model. Figure 3DWhen the measured spectrum (illustrated in 322) is used to determine the parameters of the target region, the goodness of fit may be affected, resulting in a decrease in the accuracy and precision of the measurement.

[0048] Figure 4A A top view of multiple targets, labeled PAD_A, PAD_B, PAD_C, PAD_D, and PAD_E, is illustrated. Each target has a different size, for example, 40 µm × 40 µm, 35 µm × 35 µm, 30 µm × 30 µm, 25 µm × 25 µm, and 20 µm × 20 µm, respectively. A measuring spot 410 is illustrated as having a center (illustrated as a cross pattern 412) at the center of each of the multiple targets. The measuring spot 410 may, for example, have a spot size diameter of 35 µm, and thus fits perfectly within targets PAD_A and PAD_B when properly centered, but is larger than targets PAD_C, PAD_D, and PAD_E, and therefore includes non-target areas.

[0049] Figure 4B This is a graph illustrating the MSE achieved by fitting the measured data obtained from the measurement spot 410 with the modeling data of each of the targets PAD_A, PAD_B, PAD_C, PAD_D, and PAD_E. The signals obtained from targets PAD_A and PAD_B are purely from the targets and therefore can be well fitted by the target model (i.e., low MSE). For targets PAD_C, PAD_D, and PAD_E, which are smaller than the measurement spot 410, the obtained signals are contaminated by signals from non-target regions, and therefore, the signals cannot be well fitted to the target-only model. Furthermore, as can be seen, the fitting quality (e.g., MSE) is poor for targets relatively smaller than the measurement spot 410.

[0050] This can reduce or eliminate signal contamination from non-target areas (e.g., caused by measurement spot size larger than the target area and / or by misalignment of the measurement spot with the target area), thereby improving measurement accuracy and precision. In one approach, a hybrid model method can be used to eliminate or reduce signal contamination. This hybrid model method models the target area based on a pure target model (i.e., a model of only the target area) and a local signal gradient (LG) term from the target area. The signal can be a spectral signal; that is, the local signal gradient can be a local spectral gradient, including Mueller (MM), Jones (MM), psi and delta data, reflectance measurement data, interferometric measurement data, Fourier transform infrared spectroscopy (FTIR) data, etc.; or any other measured signal, including X-ray metrology, photoacoustic metrology, electron beam (E-beam) metrology, etc.

[0051] In the simple mixing model method based on linear incoherent mixing of signals, the mixed signal (S) obtained from the target region is the signal (S0) from the target region. tgt ) and signals from surrounding non-target areas (S srd The linear incoherent mixture of signals (S) can be written as follows: Equation 1 Where f is the fraction of the target signal.

[0052] The incoherent mixture model based on Equation 1 can only be used to generate modeling data if information about both the target region and the surrounding non-target regions is known. However, information about the surrounding non-target regions is often unknown and often irregular (e.g., non-periodic, arbitrary defects, etc.). Therefore, generating an accurate incoherent mixture model is impractical.

[0053] The local gradient-based mixture model can be generated based on reformatting Equation 1, such that it contains a pure target signal (S). tgt ) and the signal difference (S) between the target area and the surrounding non-target area. srd –S tgt ) items, as follows: Equation 2 The second term in Equation 2 (i.e., the signal difference (S) between the target area and the surrounding non-target area) srd –S tgt The local signal gradient can be estimated by the correct coefficient c, which is determined based on a set of measured signals (e.g., based on collected scan data) containing different fractions of the target region signal and the non-target signal, as follows.

[0054] Equation 3 Therefore, the hybrid model based on local gradients can be written as follows: Equation 4 In some implementations, additional or different coefficients of the local gradient, such as higher-order coefficients, can be used. During the analysis (e.g., comparison of the measured signal and the calculated signal determined based on Equation 4), the coefficient c floats along with the target structural parameters (i.e., is a variable parameter) to absorb contamination from the surrounding non-target regions. The local gradient (LG) term is determined directly from the measured signal from the reference target, and therefore, the hybrid model based on the local gradient does not require structural information from the surrounding non-target regions.

[0055] Figure 5A and Figure 5BAn example is illustrated by moving the measurement spot 510 above the target area 502 while collecting data (each by...). Figure 5A and Figure 5B (See arrows 504 and 514 in the image) for a top view of line scan measurement of target area 502. Figure 5A An example of a scan along the x-axis is shown, while Figure 5B A linear scan along the y-axis is illustrated. As an example, the target region 502 can be a 30µm × 30µm target pad, and the measurement spot size can be 35µm. Data can be collected as the measurement spot moves into (or out of) the target region, with data collected during the scan in steps, for example, every 1µm. During in-line scan measurements, as the measurement spot 510 moves across the target region 502, the collected data will contain signals from both the target region 502 and the surrounding non-target region 503, with their fractions varying as the measurement spot 510 moves. Alternatively, scan data collection can be in several other styles, including a checkerboard or grid pattern, where local gradients can be determined by the difference between their average values ​​or the difference between any two locations, or fitted to multiple measurement data that are a mixture of target and non-target signals with different levels.

[0056] Figure 6 This illustrates a linear scanning measurement spot over the target area (e.g., as shown in the image). Figure 5A Graph 600 illustrates examples of different spectral signals collected during the process (as illustrated). For example, graph 600 illustrates the measurement of the absolute Mueller matrix (MM) of the Mueller matrix m12.

[0057] The local gradient of the measured signal can be determined as the derivative or rate of change of the signal measured from different locations along a spot scan, which is capable of capturing spectral variations. The local gradient of the measured signal can be constant or wavelength-dependent. The local gradient of the measured signal can be determined from two different locations (e.g., adjacent or neighboring locations, e.g., S). n S n+1 , where n is the difference between the collected spectra (where n is the scan sequence number).

[0058] Equation 5 The local gradient of the measured signal can be generated from a left scan of the target region, a right scan of the target, or both. The local gradient of the measured signal can be determined between two locations or from multiple locations. Two or more locations are close to each other to define the "locality" of the local gradient, and, for example, can, but do not have to, be adjacent during the scan. For example, a single local gradient of the measured signal can be determined based on the difference between two locations, or multiple local gradients of the measured signal can be determined based on the differences between multiple locations, and multiple local gradients of the measured signal can be used together in a hybrid model, or can be combined (e.g., averaged) to produce a single local gradient of the measured signal for use in a hybrid model. Furthermore, to better represent the signal difference between the target region and the surrounding non-target regions, the gradient can be determined based on multiple scan directions (e.g., the X direction used to generate LG_X, such as...). Figure 5A (as illustrated) and the Y direction used to generate LG_Y (e.g.) Figure 5B The local gradient of the measured signal is determined by scanning data (as illustrated) or in any other direction. A local gradient-based mixture model may include one or more local gradients of the measured signal. When the local gradient of the measured signal is combined with a pure target model in a local gradient-based mixture model, the local gradient of the measured signal can absorb signals from surrounding non-target regions, thus allowing the measurement to approximate the true value, for example, without signal combinations from non-target regions.

[0059] In one implementation of a hybrid model based on local gradients, local gradient determination can be performed for each target region or for each die. A drawback of measuring the local gradient for each target region or die is that it requires measuring signals from multiple locations, which reduces throughput. If the local gradients exhibit good similarity over all target regions, the local gradient from a single reference target region can be used to represent the local gradients of other target regions. Similarly, if the local gradients exhibit good similarity over all dies, the local gradients from multiple target regions within a reference die can be used to represent the local gradients of corresponding multiple target regions in other dies. For example, a good match in batch results indicates the feasibility of using a reference target region or a reference die.

[0060] The reference target region can be a specially designed reference pad or the target region to be measured. For example, a first target region can be scanned to measure signals from multiple locations, which are used to determine the local gradient of the measured signal in the first target region. In some implementations, to determine one or more parameters of the first target region, measured signals from the first target region can be selected from the scan and used to fit a hybrid model based on local gradients (which includes a model of the target region and the local gradient of the measured signal determined for the first target region). During measurements in other target regions, the first target region can serve as a reference target region; that is, the local gradient of the measured signal determined for the first target region is used for the other target regions.

[0061] Figure 7 An exemplary flowchart is shown depicting an example method 700 for measuring parameters of interest of a target on a sample using a local gradient-based mixture model, according to some specific implementations. In some specific implementations, example method 700 may be executed by a metrology device (such as metrology device 100 or 200) using a computing system 160 having one or more processors (e.g., processor 162) configured to perform the functions discussed herein, including eliminating or reducing signal contamination based on a local gradient-based mixture model approach.

[0062] At point 702, measurement data of the target is obtained using a measurement device. This measurement data is a mixture of target signals from the target and non-target signals from non-target areas, for example, as referenced. Figure 1 , Figure 2 and Figures 3A to 3D The discussion focuses on the use of radiation incident on the sample above the measurement area. For example, the metrology device utilizes radiation incident on the sample above the measurement area and obtains metrological data of the target from the location of the measurement area, which includes both the target and non-target regions. The measurement area may be larger than the target in at least one dimension. In some embodiments, the metrological data of the target may be spectral data. The components used to obtain the metrological data of the target may be, for example, metrology device 100 or 200, which includes a computing system 160 having a processor 162 configured to perform modeling, such as... Figure 1 and Figure 2 As shown.

[0063] At point 704, the measurement data is fitted to a hybrid model that includes the target model and the local gradients of the measured signal, for example, as referenced. Figure 5A , Figure 5B and Figure 6As discussed in Figure 8. A hybrid model incorporating the local gradients of the target model and the measured signal is, for example, a hybrid model based on local gradients, as discussed herein. For example, fitting metrological data to a hybrid model incorporating the local gradients of the target model and the measured signal may include: floating the parameter of interest in the target model and floating at least one coefficient of the local gradient of the measured signal. The hybrid model may contain one or more local gradients. Local gradients may be constant or wavelength-dependent. Components for fitting metrological data to a hybrid model incorporating the local gradients of the target model and the measured signal may be, for example, metrological device 100 or 200, which includes a computing system 160 having a processor 162 configured to perform modeling, such as... Figure 1 and Figure 2 As shown.

[0064] At point 706, the parameters of interest for the target are determined based on the fitting of the econometric data to the mixture model, for example, as referenced. Figure 5A , Figure 5B and Figure 6 As discussed in Figure 8. The component used to determine the parameters of interest for the target based on fitting the measurement data using a hybrid model can be, for example, a measurement device 100 or 200, which includes a computing system 160 with a processor 162 configured to perform modeling, such as... Figure 1 and Figure 2 As shown.

[0065] In some specific implementations, the method may further include: obtaining a local gradient econometric dataset by acquiring signals from multiple measurements at different locations, wherein the signals from each different location are different mixtures of target and non-target signals, for example, as referenced. Figure 5A , Figure 5B and Figure 6 As discussed in Figure 8, the components for obtaining a local gradient metrology dataset by acquiring signals from multiple measurements at different locations (where the signals from each measurement at different locations are different mixtures of target and non-target signals) are, for example, metrology devices 100 or 200, which include a computing system 160 with a processor 162 configured to perform modeling, such as... Figure 1 and Figure 2 As shown. The method may further include: determining the local gradient of the measured signal based on the differences in signals measured from different locations within a local gradient econometric dataset, for example, as referenced... Figure 5A , Figure 5B and Figure 6As discussed in Figure 8. The component used to determine the local gradient of a measured signal based on the differences in signals from different locations within a local gradient metrology dataset can be, for example, metrology device 100 or 200, which includes a computing system 160 with a processor 162 configured to perform modeling, such as... Figure 1 and Figure 2 As shown.

[0066] In some specific implementations, a local gradient econometric dataset can be obtained from the target object. For example, the econometric data for the target can be selected from a local gradient econometric dataset.

[0067] In some specific implementations, a local gradient econometric dataset can be obtained from a reference target. For example, the reference target may have the same nominal values ​​of the parameters of interest as the target. For example, the reference target may be on a sample that has the target. For example, the reference target may be on a sample that is different from the target. For example, a different sample may be fabricated under the same conditions as the target.

[0068] In some specific implementations, a local gradient measurement dataset of a target can be obtained by acquiring signals from multiple measurements at different locations along at least one scan in at least one dimension.

[0069] In another approach, machine learning methods can be used to eliminate or reduce signal contamination. This machine learning uses variations in the patterns of multiple signals to predict signals from the target region, without signal contamination from surrounding non-target regions.

[0070] As the measuring spot moves from a position where it is partially incident on the target region to a position where it is fully incident on the target region, a pattern of changes in multiple signals is generated. When the measuring spot is scanned over the target region, the collected signals will display a well-formed pattern controlled by the movement of the measuring spot from partial to full incident on the target region, which can be controlled experimentally. Therefore, a machine learning model can be built to identify the changing pattern of signals based on the movement of the measuring spot. The machine learning model can be trained using signals (e.g., spectral or other measurement data) collected at multiple positions relative to the target region while the measuring spot is scanned over the target region. For example, signals can be collected as the measuring spot moves at constant step sizes. The trained machine learning model can be used to predict the signal for the next (or any other desired) step size of movement of the measuring spot. The final predicted signal will represent an accurate signal as if it were collected when the measuring spot was fully incident on the target.

[0071] Figure 8AA top view is shown illustrating a line scan measurement of the target area 802 by moving the measurement spot 810 (illustrated by arrow 804 and dashed line) above the target area 802 while collecting data. Figure 8A The measurement spot 810 is illustrated as an ellipse, for example, generated by oblique incident light from metrology device 200, but could be circular if generated by normal incident light from metrology device 100. As illustrated, the measurement spot 810 may be larger than the target region 802 in at least one dimension, which would result in a signal contaminated by signals from the surrounding non-target region 803 (i.e., the measured spectrum). For example, the target region 802 may be a 30µm × 30µm target pad, and the measurement spot size along one axis may be 40µm. The measurement spot 810 can be scanned across the target region 802 in consistent steps (e.g., 1µm). Figure 8A A further example is a graph 820 showing the spectral signal collected from the measurement spot 810 when it is scanned over the target region 802 (e.g., as shown in the figure). Figure 8A (As illustrated). The collected signal is a mixture of the target signal from the target region 1002 and the non-target signal from the non-target region 803. Arrow 824 illustrates the change in the spectral signal in graph 820 when the measurement spot 810 is scanned across the target region 802.

[0072] A machine learning model can be trained based on mixed signals collected over multiple locations, where the trained model predicts the signal for the next (or other desired) step size of the measured spot based on previously measured locations. For example, the machine learning model can be trained by performing regression to minimize the difference between the two sides of the following equation: f(Mi, Mj) = Mk Equation 6 Where i, j, and k are indices representing different positions of the scan along the measurement spot that spans the target region, M represents the measured signal, and f(·) represents the machine learning model.

[0073] Figure 8B Several examples of top-view views of the target region 802 are shown, where different percentages of the measuring spot 810 cover the target region 802, such as 60%, 70%, 80%, 90%, and 100%. It should be understood that the relative size of the measuring spot 810 relative to the target region 802 is exaggerated, and the measuring spot 810 may be larger than the target region 802, such as... Figure 8B As shown. It should be further understood that the percentage of the target area 802 covered by the measurement spot 810 is equivalent to the position of the measurement spot 810 relative to the target area 802.

[0074] As an example, based on Equation 6 and referring to Figure 8BA machine learning model can be trained by performing regression to minimize the difference between the two sides of the following equation: f(M60%, M70%) = M80% Equation 7 f(M70%, M80%) = M90% Equation 8 Once trained, the machine learning model can predict target signals generated only from the target region 802, that is, as if the measured signal were obtained only from the target, such that the target signal does not include non-target signals from the surrounding non-target regions (M100%), as follows: f(M80%, M90%) = M100%. Equation 9 Therefore, even if the measuring spot 810 is larger than the target area 802, the machine learning model can be trained to predict the accurate signal (M100%) of the target area 802, for example, generated by the measuring spot 810 without any contamination signal from the surrounding non-target areas.

[0075] Machine learning models can also be used for dynamic noise reduction. For example, while machine learning models can be trained based on Equations 7 and 8 to improve accuracy, when the dynamic precision signal on the right side of the equations is available, the data can be further processed by taking the average spectrum of multiple cycles (cy1, cy2… cyk), and then used as training labels for training machine learning models, as illustrated below: M80%_mean = (M80%_cy1 + M80%_cy2 + ... + M80%_cyk) / k. Equation 10 A machine learning model can be trained by performing regression to minimize the difference between the two sides of all the following equations.

[0076] f(M60%_cy1, M70%_cy1) = M80%_mean, f(M60%_cy2, M70%_cy2) = M80%_mean,…, f(M60%_cyk, M70%_cyk) = M80%_mean Equation 11 M80%_mean = M80%_cy1 + M80%_cy2 + … + M80%_cyk) / k Equation 12 f(M70%_cy1, M80%_cy1) = M90%_mean, f(M70%_cy2, M80%_cy2) = M90%_mean,…, f(M70%_cyk, M80%_cyk) = M90%_mean Equation 13 M90%_mean = M90%_cy1 + M90%_cy2 + … + M90%_cyk) / k Equation 14 A well-trained machine learning model can predict signals that are free from signal contamination from surrounding non-target areas and have reduced dynamic noise as follows: f(M80%_cy1, M90%_cy1) = M100%_cy1, f(M80%_cy2, M90%_cy2) = M100%_cy2, …, f(M80%_cyk, M90%_cyk) = M100%_cyk Equation 15 M100%_mean = M100%_cy1 + M100%_cy2 + … + M100%_cyk) / k Equation 16 In some implementations, multiple measurement targets used for signal collection can be used to train a machine learning model. Signals collected from multiple measurement targets can be used to generate training data for the machine learning model, but may not require additional reference metrics. For example, multiple measurement targets can cover typical process conditions of the equipment. Generally, the more signals collected and the more measurement targets used, the better the yield of the machine learning model. Signals can be collected from each measurement target using scanning, for example, as... Figure 8A As illustrated, the movement of the measurement spot is known and, in some specific implementations, a constant step distance, such as 1 µm. Once trained, the machine learning model can be deployed on the metrology equipment.

[0077] Figure 9 An example is illustrated using a trained machine learning model 920 to measure parameters of interest in a target region 902 to eliminate or reduce contamination of the signal from surrounding non-target regions 903. Multiple measured mixed signals are obtained from different locations using a measurement spot 910, which may have a spot size larger than the target region 902 in at least one dimension and / or may be misaligned with the target region 902. Each mixed signal is a mixture of a target signal from the target region 902 and a non-target signal from the non-target region 903. The measurement spot 910 is illustrated with dashed lines, where the center of the measurement spot (illustrated as a crosshair) is located at different locations to illustrate the signals M measured from multiple different locations. Figure 9In this method, different locations are identified based on the percentage of the measurement spot 910 covering the target area 902, and the corresponding measured signal M obtained from each corresponding location is labeled accordingly, such as M90%, M80%, M70%, and M60%. The measured signal can be collected from the target area by moving the measurement spot a known and, in some specific implementations, a constant step distance (e.g., 1 µm), which can be the same distance used to generate training data for a machine learning model. One of the locations from which the measured signal is obtained should be the location on the target area 902 that provides the most accurate signal (e.g., M90%). The location from which the most accurate signal can be obtained can be determined from the line scan signal obtained during training. If desired, for example, to increase throughput, the measured signal can be obtained from only two locations.

[0078] At least two signals (which may include the most accurate signal) from the measured mixed signal are provided as input data to a trained machine learning model 920. The machine learning model 920 receives the measured mixed signal (e.g., M80% and M90%) and outputs an improved signal M100%, which reduces or eliminates non-target signals from the non-target region 903. Therefore, the machine learning model 920 determines the target signal, for example, as if the measurement spot 910 is only incident on the target region 902 and not on the surrounding non-target region 903, thereby reducing or eliminating signal contamination from the non-target region. Functionally, the improved signal M100% is equivalent to the signal measured from the smaller measurement spot 930 that is only incident on the target region 902, such as... Figure 9 As illustrated. After determining the improved signal, for example, the improved signal can be used to determine one or more parameters of the target region, for example, by fitting the improved signal to the modeling data of the target region until a good fit is achieved.

[0079] Figure 10 An exemplary flowchart is shown depicting an example method 1000 for measuring parameters of interest of a target on a sample using a machine learning model, according to some specific implementations. In some specific implementations, example method 1000 may be executed by a metrology device (such as metrology device 100 or 200) using a computing system 160 having one or more processors (e.g., processor 162) configured to perform the functions discussed herein, including machine learning model-based methods for eliminating or reducing signal contamination.

[0080] As illustrated in the figure, at position 1002, mixed measurement data is obtained at multiple different locations relative to the target using a measuring device. The mixed measurement data from each different location is a mixture of target signals from the target and non-target signals from non-target areas. For example, as shown in the reference... Figure 9The discussion focuses on the use of radiation incident on the sample above the measurement area, where the measurement area is larger than the target, for example, as... Figure 8A , Figure 8B and Figure 9 As illustrated. The metrological data of the target can be, for example, spectral data. The component for obtaining mixed metrological data at multiple different locations relative to the target using a metrological device (where the mixed metrological data from each different location is a mixture of target signals from the target and non-target signals from non-target regions) can be, for example, metrological device 100 or 200, which includes a computing system 160 with a processor 162 configured to perform machine learning modeling, such as... Figure 1 and Figure 2 As shown.

[0081] At point 1004, a trained machine learning model is used to determine the target measurement data based on mixed measurement data, wherein the target measurement data includes target signals from the target region but no non-target signals from non-target regions, for example, as referenced. Figure 9 The component used to determine target measurement data (where the target measurement data includes target signals from the target area but no non-target signals from the non-target area) based on mixed measurement data using a trained machine learning model can be, for example, a measurement device 100 or 200, which includes a computing system 160 having a processor 162 configured to perform actions such as... Figure 1 and Figure 2 The machine learning modeling and reference shown in the figure Figure 8A , Figure 8B and Figure 9 The machine learning models discussed.

[0082] At position 1006, the target's parameters of interest are determined based on the target's target metrics, for example, as referenced. Figure 9 The components used to determine the parameters of interest of a target based on the target's measurement data can be, for example, measurement devices 100 or 200, which include a computing system 160 having a processor 162 configured to perform machine learning modeling, such as... Figure 1 and Figure 2 As shown.

[0083] The above description is intended to be illustrative and not restrictive. For example, the examples (or one or more aspects thereof) described above can be used in combination with each other. Other specific embodiments can be used, as will be apparent to those skilled in the art after reading the above description. Furthermore, various features can be grouped together, and fewer than all features of a particular disclosed specific embodiment can be used. Therefore, the following aspects are thus incorporated into the above description as examples or specific embodiments, each aspect being an independent, separate specific embodiment, and it is contemplated that such specific embodiments can be combined or arranged in various ways with each other. Therefore, the substance and scope of the appended claims should not be limited to the foregoing description.

Claims

1. A method for measuring a parameter of interest of a target on a sample, the method comprising: Measurement data of the target is obtained using a metering device. The measurement data is a mixture of target signals from the target and non-target signals from non-target areas. The measurement data is fitted to a hybrid model, which includes the model of the target and the local gradients of the measured signal; as well as The parameters of interest for the target are determined based on the fitting of the measurement data to the hybrid model.

2. The method of claim 1, wherein the metrology device uses radiation incident on the sample above the measurement area, and the metrology data of the target is obtained using the metrology device from a location in the measurement area that includes both the target and the non-target area.

3. The method of claim 2, wherein the measurement area is larger than the target in at least one dimension.

4. The method according to claim 1, further comprising: A local gradient measurement dataset is obtained by acquiring signals from multiple measurements at different locations, wherein the signals from each different location are different mixtures of target signals and non-target signals; as well as The local gradient of the measured signal is determined based on the difference between the measured signals from the different locations in the local gradient econometric dataset.

5. The method of claim 4, wherein the local gradient econometric dataset is obtained from the target.

6. The method of claim 5, wherein the measurement data of the target is selected from the local gradient measurement dataset.

7. The method of claim 4, wherein the local gradient econometric dataset is obtained from a reference target.

8. The method of claim 7, wherein the reference target is on the sample together with the target.

9. The method of claim 7, wherein the reference target is on a different sample compared to the target.

10. The method of claim 4, wherein the local gradient measurement dataset of the target is obtained by acquiring signals of the plurality of measurements from different locations along at least one scan in at least one dimension.

11. The method of claim 1, wherein fitting the metrological data to the hybrid model comprising the model of the target and the local gradient of the measured signal comprises floating the parameter of interest in the model of the target and floating at least one coefficient of the local gradient of the measured signal.

12. The method of claim 1, wherein the hybrid model comprises one or more local gradients of the measured signal.

13. The method of claim 1, wherein the local gradient of the measured signal is constant or wavelength-dependent.

14. The method of claim 1, wherein the metrological data of the target includes spectral data.

15. A metrological apparatus configured to measure a parameter of interest of a target on a sample, the metrological apparatus comprising: A source, configured to generate radiation to be incident on the target on the sample; At least one detector, the at least one detector being configured to detect radiation from the target generated in response to the radiation incident on the target; as well as At least one processor, coupled to the at least one detector, wherein the at least one processor is configured to: Measurement data of the target is obtained using a metering device. The measurement data is a mixture of target signals from the target and non-target signals from non-target areas. The measurement data is fitted to a hybrid model, which includes the model of the target and the local gradients of the measured signal; as well as The parameters of interest for the target are determined based on the fitting of the measurement data to the hybrid model.

16. The metrology apparatus of claim 15, wherein the radiation is incident on the apparatus on the sample above the measurement area, and the target metrology data of the target is obtained from a location in the measurement area that includes both the target and the non-target area.

17. The measuring device of claim 16, wherein the measuring area is larger than the target in at least one dimension.

18. The metering device according to claim 15, wherein the at least one processor is further configured to: A local gradient econometric dataset is obtained by acquiring signals from multiple measurements at different locations, wherein the signals from each different location are different mixtures of target and non-target signals; and The local gradient of the measured signal is determined based on the difference between the measured signals from the different locations in the local gradient econometric dataset.

19. The metering device of claim 18, wherein the local gradient metering dataset is obtained from the target.

20. The metering device of claim 19, wherein the metering data of the target is selected from the local gradient metering dataset.

21. The metering device of claim 18, wherein the local gradient metering dataset is obtained from a reference target.

22. The measuring device of claim 21, wherein the reference target is on the sample together with the target.

23. The measuring device of claim 21, wherein the reference target is on a different sample compared to the target.

24. The metering device of claim 15, wherein the local gradient metering dataset of the target is obtained by acquiring the metering data from different locations by scanning along at least one line in at least one dimension.

25. The metrology apparatus of claim 15, wherein the at least one processor is further configured to fit the target metrology data to the hybrid model comprising the model of the target and the local gradient of the measured signal by being configured to float the parameter of interest in the model of the target and to float at least one coefficient of the local gradient of the measured signal.

26. The metering device of claim 15, wherein the hybrid model comprises one or more local gradients of the measured signal.

27. The measuring device of claim 15, wherein the local gradient of the measured signal is constant or wavelength-dependent.

28. The metering device of claim 15, wherein the metering data of the target includes spectral data.

29. A method for measuring a parameter of interest of a target on a sample, the method comprising: Mixed measurement data are obtained at multiple different locations relative to the target using a metering device, wherein the mixed measurement data from each different location is a mixture of target signals from the target and non-target signals from non-target areas; The target measurement data is determined using a trained machine learning model based on the mixed measurement data, wherein the target measurement data includes the target signal from the target, but not the non-target signal from the non-target region; as well as The parameter of interest for the target is determined based on the target measurement data.

Images