Whole wafer measurement based on trained whole wafer measurement model

Abstract

Methods and systems are described herein for measuring semiconductor structures based on a trained full wafer measurement model that is effective for all possible measurement locations on a wafer. The full wafer measurement model is trained based on experimental design (DOE) measurement data collected across an entire wafer or set of wafers subjected to the same set of process steps. By employing DOE measurement data across an entire wafer or a set of wafers, information about process behavior across the entire wafer is implicitly incorporated into the trained model at all locations across the wafer under test. The model training process facilitates physical process behavior that reduces the degree of freedom of the base model, breaks the correlation between parameters, and reduces the dimensionality of the solution space. Therefore, measurement performance and robustness are improved.

Description

Technical Field

[0001] The described embodiments relate to metrology systems and methods, and more particularly, to methods and systems for improved measurement of semiconductor structures. Background Art

[0002] Semiconductor devices, such as logic and memory devices, are typically manufactured using a sequence of processing steps applied to a sample. These processing steps form the various features and structural levels of the semiconductor device. For example, photolithography is one semiconductor manufacturing process that involves creating patterns on a semiconductor wafer. Additional examples of semiconductor manufacturing processes include, but are not limited to, chemical mechanical polishing, etching, deposition, and ion implantation. Multiple semiconductor devices can be manufactured on a single semiconductor wafer and then separated into individual semiconductor devices.

[0003] Metrology processes are used at various steps during the semiconductor manufacturing process to detect defects on wafers to facilitate higher yields. Optical and X-ray-based metrology techniques offer high throughput potential without the risk of sample destruction. Many metrology-based techniques, including scatterometry, reflectometry, and ellipsometry implementations and associated analysis algorithms, are commonly used to characterize critical dimensions, film thickness, composition, overlay, and other parameters of nanoscale structures.

[0004] Many metrology techniques are indirect methods of measuring the physical properties of the sample under test. In most cases, the raw measurement signal cannot be used to directly determine the physical properties of the sample. Instead, a measurement model is used to estimate the value of one or more parameters of interest based on the raw measurement signal. For example, ellipsometry is an indirect method of measuring the physical properties of the sample under test. Generally, a physics-based measurement model or a machine learning-based measurement model is needed to estimate the value of one or more parameters of interest based on the raw measurement signal (e.g., α meas and β meas ) to determine the physical properties of the sample.

[0005] In some examples, a physics-based measurement model is generated that attempts to predict the original measurement signal (e.g., α meas and β meas As described in equations (1) and (2), the measurement model includes parameters associated with the metrology tool itself (e.g., machine parameters (P machine )) and parameters associated with the sample being tested. When solving for the parameters of interest, some sample parameters are considered fixed values ​​(P spec-fixed ) and other sample parameters of interest are floating (P spec-float ), i.e., based on the original measurement signal.

[0006] α model =f(P machine ,P spec-fixed ,Pspec-float ) (1)

[0007] β model =g(P machine ,P spec-fixed ,P spec-float ) (2)

[0008] Machine parameters are parameters used to characterize a metrology tool (e.g., ellipsometer 101). Exemplary machine parameters include angle of incidence (AOI), analyzer angle (A0), polarizer angle (P0), illumination wavelength, numerical aperture (NA), compensator or waveplate (if present), and the like. Sample parameters are parameters used to characterize a sample (e.g., material and geometric parameters that characterize the structure under test). For thin film samples, exemplary sample parameters include refractive index, dielectric function tensor, nominal layer thicknesses of all layers, layer sequence, and the like. For CD samples, exemplary sample parameters include geometric parameter values ​​associated with different layers, refractive indices associated with different layers, and the like. For measurement purposes, machine parameters and many sample parameters are considered known, fixed-value parameters. However, the values ​​of one or more sample parameters are considered unknown, floating parameters of interest.

[0009] In some examples, the value of the unknown floating parameter of interest is determined by an iterative process (e.g., regression) that produces the best fit between theoretical predictions and experimental data. The value of the unknown floating parameter of interest is varied and the model output value (e.g., α model and β model ) are iteratively calculated and compared with the original measured data until the model output value is determined to be consistent with the experimental measurement value (e.g. α meas and β meas ). In some other examples, floating parameters are resolved by searching a library of pre-computed solutions to find the closest match.

[0010] In some other examples, a measurement model based on trained machine learning is used to directly estimate the value of the parameter of interest based on the raw measurement data. In these examples, the measurement model based on machine learning considers the raw measurement signal as the model input and produces the value of the parameter of interest as the model output.

[0011] Machine learning-based measurement models must be trained to produce useful estimates of parameters of interest for specific measurement applications. Generally speaking, model training is based on raw measurement signals collected from samples with known values ​​of the parameters of interest (i.e., design of experiments (DOE) data).

[0012] The machine learning-based measurement model is parameterized by several weight parameters. The machine learning-based measurement model is typically trained by a regression process (e.g., ordinary least squares regression) that minimizes the total output error. The values ​​of the weight parameters are iteratively adjusted to minimize the difference between the known reference value of the parameter of interest and the value of the parameter of interest estimated by the machine learning-based measurement model based on the measured raw measurement signal.

[0013] Physics-based or machine learning-based measurement models are typically used to independently estimate the value of one or more parameters of interest at each measurement site. By performing independent measurements at each measurement site, measurement information from neighboring measurement sites is not utilized. Furthermore, wafer-level process information is also not utilized. This limits measurement accuracy and increases measurement costs.

[0014] Current state-of-the-art measurement applications suffer from low measurement sensitivity for parameters of interest and high correlation levels between parameters that characterize the structure under test. This increases measurement complexity. Measurements are typically performed over long periods of time and at multiple angles of incidence to break down correlations and increase sensitivity. This increases measurement time and overall measurement cost.

[0015] In one example, overlay is measured at a single measurement site without information from neighboring sites or wafer-level process information. Therefore, at the time of measurement at a particular measurement site, no a priori information is available about the overlay parameter and other parameters that may be correlated with overlay. In fact, many parameters are correlated with overlay, and rather than improving the overlay estimate, these parameters tend to reduce the robustness of the overlay measurement itself.

[0016] Future metrology applications face metrology challenges due to increasingly smaller resolution requirements, multi-parameter dependencies, increasingly complex geometries, and the increasing use of opaque materials. Therefore, methods and systems for improving measurement model training and parameter reasoning that incorporate wafer-level process information are desirable. Summary of the Invention

[0017] Described herein are methods and systems for measuring semiconductor structures based on trained full-wafer measurement models. The full-wafer measurement model is trained based on design of experiment (DOE) measurement data collected across an entire wafer or wafer group undergoing the same set of process steps. By training the model using DOE measurement data across the entire wafer or wafer group, information about process behavior across the entire wafer is implicitly incorporated into the trained model at all locations across the wafer under test. The model training process promotes a smooth mapping of physical process behavior, i.e., the values ​​of one or more parameters of interest, across the wafer. This reduces the degrees of freedom of the underlying model, breaks down dependencies between parameters, and reduces the dimensionality of the solution space. Consequently, measurement performance and robustness are improved.

[0018] In one aspect, a full-wafer measurement model is trained in parallel based on DOE measurement data at multiple locations across one or more wafers and corresponding values ​​of one or more parameters of interest. The data required to train the full-wafer measurement model includes a full-wafer DOE training dataset (S DOE ) and corresponding trusted values ​​of one or more parameters of interest at various measurement locations across one or more wafers (POI DOE ).

[0019] In some embodiments, the full-wafer DOE training dataset includes actual measurement data collected from structures fabricated according to trusted values ​​of one or more parameters of interest across one or more wafers.

[0020] In some other embodiments, the DOE set of values ​​for the one or more parameters of interest is a set of known programmed parameter values, and the corresponding full-wafer training data set (S DOE ) is generated through metrological simulation.

[0021] In some embodiments, the DOE set of values ​​for the one or more parameters of interest is derived from a parameterized model having independent variables describing different locations on the wafer. In this way, the parameterized model generates values ​​for the parameters of interest at any location on the wafer for a given set of values ​​for the model parameters. The corresponding full-wafer training data set (S DOE ) is generated through metrological simulation.

[0022] In some of these embodiments, coefficient values ​​for the parameterized model are generated randomly, and for each set of coefficient values, values ​​of one or more parameters of interest are sampled across the wafer (ie, based on prescribed or randomly selected locations).

[0023] In other embodiments, the coefficient values ​​of the parameterized model are generated by a fitting or training process based on measured or assumed values ​​of the parameter of interest. The measured values ​​include values ​​of the parameter of interest measured by a trusted metrology system. The assumed values ​​include values ​​of the parameter of interest measured by a process simulator (e.g., etc.) or the value of the attention parameter generated based on user experience.

[0024] In another aspect, a full-wafer metrology model is trained based on DOE measurement data and corresponding values ​​of one or more parameters of interest at specified locations across one or more wafers. Additional information about process variations across the DOE wafer is implicitly incorporated into the model by training based on both DOE measurement data and site locations. In this way, the trained full-wafer metrology model captures the physical process behavior across the wafer and is further constrained to derive model results towards a mapping family of values ​​for the one or more parameters of interest, i.e., parameter values ​​that vary as a function of location on the wafer.

[0025] In another aspect, a trained full-wafer measurement model is used to estimate the values ​​of one or more parameters of interest at measurement locations across the entire wafer under test. The trained full-wafer measurement model estimates the values ​​of the parameters of interest at each specified measurement location based on the measurement data collected at the specified location. However, the trained full-wafer measurement model is valid for all possible measurement locations on the wafer.

[0026] In yet another aspect, the full-wafer measurement model estimates coefficient values ​​that characterize the parameterized wafer map, and the estimated coefficient values ​​are mapped to values ​​of the parameters of interest to characterize the structure under test at each measurement site using the trained wafer map model.

[0027] In some embodiments, a trained wafer map model is used to synthetically generate the DOE datasets described above. In these embodiments, a full-wafer DOE set of values ​​for one or more parameters of interest is generated based on the parameterized model, and the corresponding full-wafer training dataset (S) of measured data is generated. DOE ) is determined through metrological simulation.

[0028] In yet another aspect, the coefficients of the parameterized wafer map are trained to accurately map a DOE set of values ​​of one or more auxiliary parameters to characterize the structure under test across a wafer or wafer group. The auxiliary parameters are required to accurately simulate the measurement of the structure. In addition, the trained wafer map model is used to synthesize a DOE data set of the auxiliary parameters based on the parameterized model. The corresponding full-wafer training data set (S) of the measured data DOE ) The DOE dataset of the parameter of interest and the DOE dataset of the auxiliary parameters of the structure are determined by metrology simulation. In this way, the metrology simulation does not have to rely on assumed values ​​of the auxiliary parameters.

[0029] In some embodiments, the full-wafer measurement model is a machine learning-based measurement model. In other embodiments, the full-wafer measurement model is a physics-based measurement model.

[0030] The foregoing is a summary and therefore necessarily contains simplifications, generalizations, and omissions of detail; therefore, those skilled in the art will appreciate that the summary is illustrative only and is in no way limiting. Other aspects, inventive features, and advantages of the devices and / or processes described herein will become apparent in the non-limiting detailed description set forth herein. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 An illustration of a metrology system 100 for measuring characteristics of a semiconductor structure according to exemplary methods presented herein is depicted.

[0032] Figure 2 is a diagram illustrating a full-wafer metrology model training engine 200 in one embodiment.

[0033] Figure 3 is a diagram illustrating the full-wafer metrology model training engine 215 in another embodiment.

[0034] Figure 4 is a diagram illustrating a trained full-wafer measurement model inference engine 220 in one embodiment.

[0035] Figure 5 is a diagram illustrating a trained full-wafer measurement model inference engine 230 in another embodiment.

[0036] Figure 6 Instructions as reference Figure 2 Plot 120 of the tracking performance of the full-wafer measurement model trained as described.

[0037] Figure 7 Instructions as reference Figure 3 Plot 122 of the tracking performance of the full-wafer measurement model trained as described.

[0038] Figure 8 is with Figure 6 A wafer error map 125 is associated with the difference between the trusted value and the predicted value at each measurement location illustrated in FIG.

[0039] Figure 9 is with Figure 7 FIG. 126 shows a wafer error graph 126 associated with the difference between the credible and predicted values ​​at each measurement location illustrated in FIG.

[0040] Figure 10 is a diagram illustrating a trained full-wafer measurement model inference engine 240 in another embodiment.

[0041] Figure 11 Graph 250 depicting DOE values ​​for a parameter of interest at various measurement site locations.

[0042] Figure 12 Depicts the estimated θ from a trained 5th order polynomial wafer map model in one embodiment. Figure 11 Graph 251 depicting the values ​​of the parameters of interest.

[0043] Figure 13 Describes another embodiment of the estimated θ by a trained 10th order polynomial wafer map model. Figure 11 Graph 252 depicting the values ​​of the parameters of interest.

[0044] Figure 14 Describes another embodiment of the estimated by the trained neural network wafer map model Figure 11 253 shows the values ​​of the parameters of interest depicted in FIG.

[0045] Figure 15 Describes another embodiment of the estimated 5th order polynomial wafer map model Figure 11 261 of the values ​​of the parameters of interest depicted in FIG.

[0046] Figure 16 Describes another embodiment of the estimated θ by a trained 10th order polynomial wafer map model. Figure 11 Graph 262 depicting the values ​​of the parameters of interest.

[0047] Figure 17 Describes another embodiment of the estimated by the trained neural network wafer map model Figure 11 263 shows the values ​​of the parameters of interest depicted in FIG.

[0048] Figure 18 FIG. 2 is a diagram illustrating a full-wafer metrology model training engine 270 in another embodiment.

[0049] Figure 19 is a diagram illustrating a trained full-wafer measurement model inference engine 290 in another embodiment.

[0050] Figure 20 A flow chart illustrating a method 300 for training a full-wafer measurement model for estimating values ​​of parameters of interest in one example. DETAILED DESCRIPTION

[0051] Reference will now be made in detail to background examples and some embodiments of the invention, examples of which are illustrated in the accompanying drawings.

[0052] Methods and systems for measuring semiconductor structures based on a trained full-wafer measurement model are described herein. The full-wafer measurement model is trained based on design of experiment (DOE) measurement data collected across an entire wafer or group of wafers undergoing the same set of process steps. Furthermore, the trained full-wafer measurement model is used to estimate the values ​​of one or more parameters of interest at measurement sites across the entire wafer under test.

[0053] By utilizing DOE measurement data across an entire wafer or wafer set, information about process behavior across the entire wafer is implicitly incorporated into the trained model at all locations across the measured wafer. Consequently, the trained full-wafer measurement model estimates the values ​​of parameters of interest across the measured wafer without interruption. The model training process promotes a smooth mapping of physical process behavior, i.e., the values ​​of one or more parameters of interest, across the wafer. This reduces the degrees of freedom of the underlying model, breaks down dependencies between parameters, and reduces the dimensionality of the solution space. Consequently, measurement performance and robustness are improved.

[0054] Figure 1A system 100 for measuring a characteristic of a sample according to the exemplary methods presented herein is illustrated. Figure 1 As shown in FIG. 1 , the system 100 may be used to perform Figure 1 1 . In this regard, system 100 may include a spectroscopic ellipsometer equipped with an illuminator 102 and a spectrometer 104. The illuminator 102 of system 100 is configured to generate illumination of a selected wavelength range (e.g., 100 nm to 2500 nm) and direct it to a structure on the surface of a sample disposed at a measurement point 110. The spectrometer 104 is, in turn, configured to receive illumination reflected from the structure 101. It should be further noted that a polarization state generator 107 is used to polarize light emitted from the illuminator 102 to produce a polarized illumination beam 106. Radiation reflected by the structure 101 passes through a polarization state analyzer 109 to the spectrometer 104. The radiation received by the spectrometer 104 in a collection beam 108 is analyzed with respect to polarization state to allow the spectrometer to perform spectral analysis of the radiation passed by the analyzer. These spectra 111 are passed to a computing system 130 for analysis of the structures described herein.

[0055] like Figure 1 , system 100 includes a single measurement technique (i.e., SE). However, in general, system 100 may include any number of different measurement techniques. By way of non-limiting example, system 100 may be configured as a spectroscopic ellipsometer (including Mueller matrix ellipsometry), a spectroscopic reflectometer, a spectroscopic scatterometer, an overlay scatterometer, an angle-resolved beam profile reflectometer, a polarization-resolved beam profile reflectometer, a beam profile reflectometer, a beam profile ellipsometer, any single-wavelength or multi-wavelength ellipsometer, or any combination thereof. Furthermore, measurement data collected by different measurement techniques and analyzed according to the methods described herein may be collected from multiple tools, a single tool integrating one measurement technique, a single tool integrating multiple techniques, or a combination thereof, including, by way of non-limiting example, soft X-ray reflectometry, small-angle X-ray scatterometry, an imaging-based metrology system, a hyperspectral imaging-based metrology system, a scatterometry overlay metrology system, and the like.

[0056] In a further embodiment, the system 100 may include one or more computing systems 130 for performing measurements of a structure based on a developed measurement model according to the methods described herein. The one or more computing systems 130 may be communicatively coupled to the spectrometer 104. In one aspect, the one or more computing systems 130 are configured to receive measurement data 111 associated with measurements of a structure under test (e.g., structure 101).

[0057] In some embodiments, computing system 130 is configured to develop and train a full-wafer measurement model and execute the trained full-wafer measurement model to estimate values ​​of one or more parameters of interest, as described herein.

[0058] In one aspect, a full-wafer metrology model is trained based on DOE measurement data and corresponding values ​​of one or more parameters of interest at multiple locations across one or more parallel wafers. In this way, the trained full-wafer metrology model captures the physical process behavior across the wafer.

[0059] Full-wafer DOE training dataset of measured data (S DOE ) and the corresponding credible values ​​of one or more parameters of interest (POI DOE ) are generated at various measurement locations across one or more wafers.

[0060] In some embodiments, the full-wafer DOE training data set includes actual measurement data collected from structures fabricated according to trusted values ​​for one or more parameters of interest across one or more wafers. In some embodiments, the trusted values ​​are measured by a trusted reference metrology system (e.g., SEM, TEM, etc.). In some embodiments, the trusted values ​​for the one or more parameters of interest associated with each of the measured structures are known programmed values ​​used to fabricate the measured structures. In some embodiments, the trusted values ​​for the one or more parameters of interest associated with each of the measured structures are assumed values ​​used to fabricate the measured structures.

[0061] In some other embodiments, the DOE set of values ​​for one or more parameters of interest is a corresponding full-wafer training data set (S) of known programmed parameter values ​​and measured data. DOE ) is generated by metrology simulation. In these examples, the metrology simulation tool simulates a training data set (S DOE ) generated by a metrology tool in response to measuring a structure having a known programmed shape characterized by a DOE of parameters of interest. In some embodiments, the simulated metrology tool is used to ultimately measure a structure having unknown values ​​of one or more parameters of interest (POIs). DOE ) of the same measuring tool as the structure of the .

[0062] In some of these embodiments, a DOE set of values ​​for one or more parameters of interest is derived from a parameterized model. In the example illustrated by Equation (3), the parameterized model is a second-order polynomial model with two independent variables, x and y, corresponding to rectangular coordinate values ​​representing different locations on the wafer. In this way, the parameterized model generates a value for the parameter of interest at any location {x, y} on the wafer for a given set of values ​​{C1...C6} for the polynomial coefficients.

[0063] POI(x,y)=C1x 2 +C2xy+C3y 2 +C4x+C5y+C6 (3)

[0064] In some embodiments, the DOE dataset is composed of a full-wafer DOE dataset that determines the values ​​of one or more parameters of interest based on a parameterized model and a corresponding full-wafer training dataset (S) that determines the measurement data via metrology simulation. DOE ) to synthesize.

[0065] In some of these embodiments, coefficient values ​​or parameterized models are generated randomly, and for each set of coefficient values ​​(e.g., {C1 . . . C6}), values ​​of one or more parameters of interest are sampled across the wafer (i.e., based on prescribed or randomly selected {x, y} coordinate values).

[0066] A second-order polynomial model using a rectangular basis is used to characterize the set of DOEs for the values ​​of the parameter of interest, as illustrated by Equation (3). However, in general, any order of polynomial model and any suitable basis may be considered within the scope of this patent profile. In some examples, for example, principal component analysis of measured or assumed values ​​of the parameter of interest across the wafer may be analyzed to arrive at a suitable basis for the parameterized model.

[0067] In the example illustrated by equation (3), the parameterized model includes two independent variables x and y, which correspond to rectangular coordinate values ​​representing different locations on the wafer. However, in some other examples, the parameterized model includes four independent variables: two independent variables x and y corresponding to rectangular coordinate values ​​representing different locations on the wafer and two additional independent variables field x and field y corresponding to rectangular coordinate values ​​representing different locations within a field on the wafer, i.e., POI(x, y, field x, field y).

[0068] In some other embodiments, the parameterized model of the DOE set that characterizes the values ​​of the parameter of interest may be a machine learning-based model, such as a neural network-based model. In these examples, the coefficients of the neural network model may be trained based on measured or assumed values ​​of the parameter of interest across the wafer.

[0069] The coefficients of the parameterized model of the values ​​of the parameter of interest across the wafer may be selected randomly or by a fitting or training process based on measured or assumed values ​​of the parameter of interest. Measured values ​​include values ​​of the parameter of interest measured by a trusted metrology system. Assumed values ​​include values ​​obtained by a process simulator (e.g., etc.) or the value of the attention parameter generated based on user experience.

[0070] Corresponding full-wafer DOE measurement data (S DOE) are generated through metrology simulation based on the determined values ​​of one or more parameters of interest at each sampled wafer location and each set of coefficient values. These synthetically generated DOE datasets are then used to train a full-wafer measurement model. Synthetically generated full-wafer DOE measurement data more accurately represents true measurements than random sampling because the true wafer profile and process variations are accounted for in the simulation. Therefore, using synthetically generated DOE datasets generally results in improved robustness.

[0071] Figure 2 is a diagram illustrating a full-wafer measurement model training engine 200 in one embodiment. In some embodiments, the computing system 130 is configured as the full-wafer measurement model training engine 200 described herein. Figure 2 As depicted in FIG, the full-wafer measurement model training engine 200 includes a machine learning module 206, an error evaluation module 208, and a control module 210. 1..M DOE 202 together with the corresponding credible value of the parameter of interest 1…N POI 1…M DOE 205 is provided as input to the machine learning module 206.

[0072] The training dataset S of full-wafer measurement data 1..M DOE 202 contains measurement data associated with measurements at M different measurement sites across a wafer or wafer set, where M is any non-negative integer value. Corresponding trusted values ​​of the parameters of interest 1… N POI 1…M DOE 205 includes the values ​​of each of N parameters of interest at each of M different measurement sites, where N is any non-negative integer value.

[0073] In some examples, the full-wafer measurement model is a neural network model. Figure 2 As depicted, the machine learning module 206 evaluates the dataset S 1…M DOE The output of the neural network model is the estimated value of each of the parameters of interest at each measurement location. 1…N POI 1…M * 207, which is passed to the error evaluation module 208. The error evaluation module 208 compares the estimated values ​​of the parameters of interest determined by the neural network model 1…N POI 1…M * 207 Corresponding credible values ​​of the concerned parameters 1… N POI 1…MDOE 205. Error assessment module 208 updates neural network weights 212 to minimize a function of the difference between the determined value and the trusted value of the characterization parameter of interest (e.g., a quadratic error function, a linear error function, or any other suitable difference function). The updated neural network weights 212 are transmitted to machine learning module 206. Machine learning module 206 uses the updated neural network weights to update the neural network model for the next iteration of the training process. Iterations continue until the function of the difference between the determined value and the known value of the characterization parameter of interest is minimized. The resulting trained full-wafer measurement model 214 is transmitted to a memory (e.g., memory 132).

[0074] like Figure 2 As depicted, the full wafer measurement model training engine 200 receives the values ​​of the parameters of interest from a reference source 201. 1… N POI 1…M DOE 205. The reference source 201 is a trusted metrology system, a simulator, or both for generating the aforementioned DOE set of measurement data and corresponding DOE parameter values.

[0075] In another aspect, the full-wafer measurement model is trained by dynamically controlling weights associated with one or more measurement performance metrics, which are used to regularize the optimization of the measurement model training process. By way of non-limiting example, key performance metrics include R-squared (R 2 ), slope, gauge repeatability and reproducibility (GRR), etc. At each training iteration, the measurement model training engine 200 verifies the model performance against each performance metric. This information is provided as input to the dynamic controller, which adjusts the weight of each different performance target at each iteration.

[0076] The measurement model training engine 200 trains a measurement model based on an optimization function regularized by one or more measurement performance metrics while dynamically controlling the weights associated with each regularization term of the optimization function. In some examples, the measurement model is a neural network model. Figure 2 As depicted, the machine learning module 206 evaluates the dataset S 1…M DOE 202 neural network model h( . ).

[0077] At each iteration of the training process, the control module 210 determines each regularization weighting term γ associated with each measurement target k Each updated value is determined based on the achieved value of the measurement target and the desired value of each measurement target. Figure 2, the control module 210 receives an indication 209 of a value for each achieved measurement objective and a desired value 213 for each measurement objective. At each iteration, the control module 210 compares the achieved value associated with each measurement objective with the desired value and determines an updated value for each regularization weighting term 211. The updated value for each regularization weighting term 211 is communicated to the loss assessment module 208. The loss assessment module 208 uses the updated value 211 to evaluate the optimization function at the next iteration.

[0078] By continuously adjusting the weights of each measurement target during the training process, the neural network is trained to achieve the desired specifications for each measurement target with less computational effort.

[0079] The control module 210 employs a controller optimized for multiple measurement objectives. By way of non-limiting example, the controller is any of a linear quadratic regularizer (LQR) based controller, a proportional integral derivative (PID) controller, an optimal controller, an adaptive controller, a model predictive controller, and the like.

[0080] In some embodiments, the parameters of the controller are optimized for robust performance through a search algorithm (eg, genetic algorithm, simulated annealing algorithm, gradient descent algorithm, etc.).

[0081] In some examples, each measurement performance metric is represented as a separate distribution. In one example, the distribution of measurement accuracy is an inverted gamma distribution. Equation (1) describes the probability density function p for a measurement accuracy data set x, where Γ(.) represents the gamma function, the constant a represents the shape parameter, and the constant b represents the scale parameter.

[0082]

[0083] In another example, the distribution of the mean values ​​of the examples of the measured structures on the wafer is described by a normal distribution. Equation (2) describes the probability density function m of the measured wafer average data set x, where μ represents a particular mean value and σ represents a particular variance associated with the distribution.

[0084]

[0085] In yet another aspect, statistical information characterizing actual measurement data collected from the structure (e.g., known distributions associated with important measurement performance metrics such as measurement accuracy, wafer average, etc.) is specifically used for regularization-driven optimization of measurement model training.

[0086] At each iteration, the optimization function drives the change of the weight value W and the bias value b of the neural network model that minimizes the optimization function. When the optimization function reaches a sufficiently low value, the measurement model is considered trained, and the trained measurement model 214 is stored in a memory (e.g., memory 132).

[0087] In some examples, the multiple metrics that characterize the tracking performance are stable and converge quickly to a final value. In one example, the multiple metrics include R 2 The invention provides a method for determining the performance of a DOE structure by measuring a critical dimension of the DOE structure, wherein the weighted values ​​associated with the terms of the objective function associated with each of the plurality of metrics converge quickly and stably to a small number when achieving the desired value of the performance goal.

[0088] On the other hand, a full-wafer metrology model is trained based on DOE measurement data and corresponding values ​​of one or more parameters of interest at specified locations across one or more wafers. In this way, the trained full-wafer metrology model captures the physical process behavior across the wafer and is further constrained to drive model results toward a mapping family of values ​​for one or more parameters of interest (i.e., parameter values ​​that vary as a function of location on the wafer).

[0089] Figure 3 is a diagram illustrating the full-wafer metrology model training engine 215 in one embodiment. Figure 3 The component symbols described in Figure 2 The same element symbols as described in Figure 3 As depicted in the figure, the training dataset S of full-wafer measurement data 1..M DOE 202 and each corresponding chip position LOC 1…M DOE 203 is provided as input to the machine learning module 206.

[0090] like Figure 3 As depicted in FIG. 1 , the machine learning module 206 evaluates the measurement data S associated with each position of the set of M measurement sites. 1…M DOE The output of the neural network model is the estimated value of each parameter of interest at each measurement position. 1…N POI 1…M * 207, which is transmitted to the error evaluation module 208. The error evaluation module 208 compares the error at each specified location LOC determined by the neural network model. 1…M DOE The estimated value of the parameter of interest determined at 203 1…N POI 1…M * 207 Corresponding credible values ​​of the concerned parameters 1…N POI 1…M DOE205. Error assessment module 208 updates neural network weights 212 to minimize a function of the difference between the determined value and the trusted value of the characterization parameter of interest (e.g., a quadratic error function, a linear error function, or any other suitable difference function). The updated neural network weights 212 are transmitted to machine learning module 206. Machine learning module 206 uses the updated neural network weights to update the neural network model for the next iteration of the training process. Iterations continue until the function of the difference between the determined value and the known value of the characterization parameter of interest is minimized. The resulting trained full-wafer measurement model 214 is transmitted to a memory (e.g., memory 132).

[0091] Using both DOE measurement data (e.g., spectra) and the site locations associated with each measurement as input for training the full-wafer metrology model implicitly introduces additional information about process variations across the DOE wafer. This improves the robustness of the trained full-wafer metrology model.

[0092] Generally, position-adjusted full-wafer measurements are trained based on position information and the resulting corresponding DOE measurement dataset to introduce variations similar to those introduced across the wafer. In some examples, multiple random and smoothed wafer maps are simulated to provide the aforementioned position and DOE measurement data. DOE measurement samples are collected across a large number of different measurement locations across different wafers and wafer batches to learn the process behaviors that are critical to achieving accurate full-wafer measurements.

[0093] In yet another aspect, the trained full-wafer measurement model is used to predict values ​​of parameters of interest across the wafer based on actual measurement signals (e.g., spectra) collected across the wafer by a measurement system (e.g., metrology system 100). In some embodiments, the measurement system is the same measurement system used to collect the DOE measurement data. In other embodiments, the measurement system is a system simulated to synthetically generate the DOE measurement data. In one example, the actual measurement data includes measurement spectra 111 collected by metrology system 100 from one or more metrology targets having unknown values ​​for one or more parameters of interest.

[0094] Figure 4 is a diagram illustrating a trained full-wafer measurement model inference engine 220 in one embodiment. Figure 4 As depicted in FIG, the trained full-wafer measurement model inference engine 220 includes a trained full-wafer measurement module 221. Figure 4 In the depicted embodiment, the measurement data S collected by the metrology system or a combination of metrology systems at M measurement sites 1…M MEAS 222 is provided as input to the trained full-wafer measurement module 221. The trained full-wafer measurement module 221 employs the trained full-wafer measurement model to determine the values ​​of the N parameters of interest corresponding to the M measurement sites measured across the wafer.1…N POI 1…M MEAS 224. In these embodiments, the measurement of the value of the parameter of interest is performed in parallel at all measurement sites across the wafer, i.e., all sites together.

[0095] Figure 5 FIG. 2 is a diagram illustrating a trained full-wafer measurement model inference engine 230 in another embodiment. Figure 5 As depicted in FIG. 2 , the trained full-wafer measurement model inference engine 230 includes a reference Figure 3 The trained full-wafer measurement module 231 is trained in the manner described. Figure 5 In the depicted embodiment, the measurement data S collected by the metering system or a combination of metering systems at a specific measurement site i MEAS 232 and a set of coordinates LOC describing the location of the measurement site on the wafer i MEAS 233 is provided as input to the full wafer measurement module 231. The trained full wafer measurement module 231 uses the trained full wafer measurement model to determine the input value LOC corresponding to the measurement location. i MEAS The value of one or more parameters of interest at position 233 1…N POI i MEAS 224. In these embodiments, the measurement of the value of the parameter of interest is performed sequentially at each measurement site across the wafer, ie, each measurement is adjusted according to the location of the measurement site on the wafer.

[0096] The trained full-wafer metrology model estimates the value of the parameter of interest at each specified measurement location based on the measurement data collected at that location.However, the trained full-wafer metrology model is valid for all possible measurement locations on the wafer.

[0097] Figure 6 A plot 120 indicating tracking performance is shown. Figure 6 As illustrated in FIG. 1 , the x position of each data point on the plot 120 indicates a trusted reference value (e.g., a DOE reference value) for the parameter of interest at a particular measurement site, and the y position of each data point indicates the value of the parameter of interest using the reference value. Figure 2 The predicted values ​​of the parameter of interest at the same measurement site of the trained full-wafer measurement model trained as described above. Ideal tracking performance is indicated by the dashed line 121. If all predicted values ​​perfectly matched the corresponding known trustworthy values, then all data points would lie on line 121. However, in reality, tracking performance is not perfect. Figure 6 As depicted, the correlation between known and predicted values ​​is shown by an R of 0.79. 2 Value characterization.

[0098] Figure 7 A plot 122 indicating tracking performance is shown. Figure 7 As illustrated in FIG. 1 , the x position of each data point on the plot 122 indicates a trusted reference value (e.g., a DOE reference value) for the parameter of interest at each measurement site, and the y position of each data point indicates the value of the parameter of interest using the reference value. Figure 3 The predicted values ​​of the parameter of interest at the same measurement site of the trained full-wafer measurement model trained as described above. Ideal tracking performance is indicated by the dashed line 123. If all predicted values ​​perfectly matched the corresponding known trustworthy values, then all data points would lie on line 123. However, in reality, tracking performance is not perfect. Figure 7 As depicted, the correlation between known and predicted values ​​is represented by an R of 0.88. 2 Value characterization.

[0099] As Figure 6 and 7 As illustrated, a full-wafer metrology model trained with measurement location input improves the correlation between the credible and predicted values ​​of a parameter of interest.

[0100] Figure 8 is with Figure 6 A wafer error map 125 is associated with the difference between the trusted value and the predicted value at each measurement location illustrated in FIG.

[0101] Figure 9 is with Figure 7 A wafer error graph 126 is associated with the difference between the trusted value and the predicted value at each measurement location as described in FIG. Figure 8 and 9 As described in , a full-wafer metrology model trained with measurement position input reduces the error across the wafer and smoothes the error map, ie, reduces the error gradient across the wafer.

[0102] On the other hand, the full-wafer measurement model estimates values ​​of coefficients that characterize the parameterized wafer map, and the estimated coefficient values ​​are mapped to values ​​of parameters of interest that characterize the structure under test at each measurement site.

[0103] In one aspect, the coefficients of a parameterized wafer map are trained to accurately map a set of DOEs that characterize the values ​​of one or more parameters of interest for structures under test across a wafer or wafer set.

[0104] Figure 11 A graph 250 depicts DOE values ​​for a parameter of interest (eg, overlay) at various measurement site locations within each field of the wafer.

[0105] Figure 12 Plots the estimated θ from the trained 5th order polynomial wafer map model at various measurement site locations across the wafer. Figure 11 Graph 251 depicting the values ​​of the parameters of interest.

[0106] Figure 13 Plots the estimated θ from the trained 10th order polynomial wafer map model at various measurement site locations across the wafer. Figure 11 Graph 252 depicting the values ​​of the parameters of interest.

[0107] Figure 14 Plotting the estimated θ by the trained neural network wafer map model at various measurement site locations across the wafer. Figure 11 253 shows the values ​​of the parameters of interest depicted in FIG.

[0108] exist Figures 12 to 14 In the illustrated example, the wafer map model is parameterized by two independent variables x and y corresponding to rectangular coordinate values ​​representing different wafer locations.

[0109] Figure 15 Plots the estimated θ at various measurement site locations within each field of the wafer by the trained 5th order polynomial wafer map model. Figure 11 261 of the values ​​of the parameters of interest depicted in FIG.

[0110] Figure 16 Plots the estimated θ at various measurement site locations within each field of the wafer by the trained 10th order polynomial wafer map model. Figure 11 Graph 262 depicting the values ​​of the parameters of interest.

[0111] Figure 17 Plots the estimated θ at various measurement site locations within each field of the wafer by the trained neural network wafer map model. Figure 11 263 shows the values ​​of the parameters of interest depicted in FIG.

[0112] exist Figures 15 to 17 In the illustrated example, the wafer map model is parameterized by four independent variables: two independent variables x and y corresponding to rectangular coordinate values ​​representing different wafer locations and two additional independent variables field x and field y corresponding to rectangular coordinate values ​​representing different locations within any field on the wafer. The same number of measurement sites and measurement site locations are used to achieve Figures 12 to 17 The differences are a result of three different modeling approaches (5th order polynomial, 10th order polynomial, and trained neural network) and different numbers of independent variables.

[0113] like Figures 12 to 14 and Figures 15 to 17 As depicted in Figure 2, the trained neural network model predicts more accurately than the 10th-order polynomial model. Figure 11 The DOE values ​​of the parameters of interest depicted in the figure are more accurately predicted by the 10th-order polynomial model than by the 5th-order polynomial model. Figure 11 Furthermore, the model parameterized by both x and y chip positions and x and y field positions more accurately predicts the DOE values ​​of the parameters of interest depicted in FIG. Figure 11 DOE values ​​for the parameters of interest depicted in .

[0114] like Figures 12 to 14 and Figures 15 to 17 As depicted in

[0015] , a higher-order model (e.g., a higher-order polynomial model or a neural network model with more nodes) is used to more accurately capture high-frequency local variations. Additionally, high-frequency local variations are more accurately captured when the model is parameterized by both x and y wafer positions and x and y field positions.

[0115] After the coefficients of the parameterized wafer map are trained to accurately map the DOE set of values ​​for one or more parameters of interest, a full-wafer measurement model is trained at multiple locations across one or more parallel wafers based on DOE measurement data corresponding to the determined coefficients of the parameterized wafer map. 1…N POI 1…M DOE 205 DOE value WMP of the coefficients of the trained wafer map model 1…U DOE Replace and focus on the estimated values ​​of the parameters 1…N POI 1…M * 207 is estimated from the coefficients of the trained wafer map model WMP 1…U * 207 replacement, the training process is similar to the reference Figure 2 The training process described. In this way, the trained full-wafer measurement model captures the physical process behavior across the wafer.

[0116] Figure 10 FIG. 2 is a diagram illustrating a trained full-wafer measurement model inference engine 240 in another embodiment. Figure 10 As depicted in FIG, the trained full-wafer measurement model inference engine 240 includes a trained full-wafer measurement module 241 and a trained wafer map module 242. Figure 10 In the depicted embodiment, the measurement data S collected by the metrology system or a combination of metrology systems at M measurement sites 1…M MEAS 243 is provided as input to the trained full-wafer metrology module 241. The trained full-wafer metrology module 241 employs the trained full-wafer metrology model to determine the values ​​WMP for the U coefficients defined by the trained wafer map. 1…U MEAS244. The determined coefficient values ​​are communicated to the trained wafer map module 242. The trained wafer map module 242 maps the determined coefficient values ​​to values ​​corresponding to the parameter of interest at each of the M measurement sites measured across the wafer based on the trained wafer map. 1…M MEAS 245. In these embodiments, the measurement of the value of the parameter of interest is performed in parallel at all measurement sites across the wafer, i.e., all sites together.

[0117] In general, any parameterized mathematical function can be used as a parameterized wafer map, for example, a principal component model, a neural network model, a polynomial model, a discrete cosine transform model, a wavelet model, etc. The coefficients of the parameterized mathematical function are selected to best fit the DOE map that characterizes the values ​​of each parameter of interest of the structure under test.

[0118] In another aspect, the trained wafer map model is used to synthetically generate a DOE dataset, as previously described. In these embodiments, a full-wafer DOE set of values ​​for one or more parameters of interest is generated based on the parameterized model, and a corresponding full-wafer training dataset S of measured data is generated. DOE Determined by metrological simulation.

[0119] In yet another aspect, the coefficients of a parameterized wafer map are trained to accurately map a set of DOEs to the values ​​of one or more auxiliary parameters that characterize the structure under test across a wafer or wafer set. The auxiliary parameters are required to accurately simulate measurements of the structure. The values ​​of these parameters (e.g., underlying parameters) are typically assumed, and their variation across the wafer is not modeled for metrology simulations.

[0120] Furthermore, the trained wafer map model is used to synthesize a DOE dataset that generates auxiliary parameters based on the parameterized model. DOE The DOE data set of the structure-based parameter of interest and the DOE data set of the auxiliary parameters are determined by metrological simulation. In this way, the metrological simulation does not have to rely on the assumed values ​​of the auxiliary parameters.

[0121] In some embodiments, the DOE dataset is generated by a process simulator that generates a wafer map based on a specific process setup. The trained wafer map is then used to generate synthetic spectra and reference values ​​for parameters of interest to train the metrology model, as described above.

[0122] In some embodiments, the full-wafer measurement model is a physics-based measurement model. Measurements of the parameter of interest are measured in parallel across multiple measurement sites on the wafer by regression of the measurement data, for example, by spectral fitting across all measurement sites together using a trained physics-based measurement model.

[0123] Figure 18is a diagram illustrating a full-wafer measurement model training engine 270 in another embodiment. In some embodiments, the computing system 130 is configured as the full-wafer measurement model training engine 270 described herein. Figure 18 As depicted in FIG, the full-wafer measurement model training engine 270 includes a physics-based model module 274 and an error evaluation module 280. The training dataset S of the full-wafer measurement data 1..M DOE 273 together with the corresponding credible values ​​of the parameters of interest 1…N POI 1…M DOE 272 is provided as input to the physics-based model module 274 .

[0124] like Figure 18 As depicted, the full wafer measurement model training engine 270 receives the values ​​of the DOE parameters of interest from a reference source 271. 1…N POI 1…M DOE 272. Reference source 271 is a trusted metrology system, simulator, or both used to generate the aforementioned DOE parameter values.

[0125] The training dataset S of full-wafer measurement data 1..M DOE 273 contains measurement data associated with measurements at M different measurement sites across a wafer or wafer set, where M is any non-negative integer value. Corresponding trusted values ​​of the parameters of interest 1… N POI 1…M DOE 272 includes the values ​​of each of N parameters of interest at each of M different measurement sites, where N is any non-negative integer value.

[0126] like Figure 18 As depicted, the physics-based model module 274 is based on the credible values ​​of the parameters of interest. 1…N POI 1…M DOE 272 to evaluate the physics-based model. The output of the physics-based model is the estimated full-wafer measurement data S at M measurement sites associated with the credible values ​​of the parameters of interest. 1..M * 275. Estimated full-wafer measurement data S 1..M * 275 and DOE full wafer measurement data S 1..M DOE 273 comparison. Error evaluation module 280 updates the value of floating variable 277 of the physics-based model to minimize the characterization estimated full wafer measurement data S 1..M * 275 and DOE full wafer measurement data S1..M DOE 273. During the training phase, floating parameters are typically model parameters that are not parameters of interest, such as material parameters and geometric parameters, but rather parameters that need to be tuned to ensure the accuracy of the physics-based model.

[0127] The updated values ​​of the floating variables 277 are communicated to the physics-based model module 274. The physics-based model module 274 uses the updated values ​​of the floating variables 277 to update the physics-based model for the next iteration of the training process. Iterations continue until the function that characterizes the difference between the determined values ​​of the full-wafer measurement data and the DOE values ​​is minimized. The resulting trained full-wafer measurement model 279 is communicated to a memory (e.g., memory 132).

[0128] Figure 19 FIG2 is a diagram illustrating another embodiment of a trained full-wafer measurement model inference engine 290. A trained physics-based full-wafer measurement module 291 employs a trained physics-based full-wafer measurement model to determine values ​​for N parameters of interest corresponding to M measurement sites measured across the wafer. In these embodiments, the measurement of the values ​​of the parameters of interest is performed in parallel at all measurement sites across the wafer, i.e., all sites together.

[0129] like Figure 19 As depicted in FIG, the full-wafer measurement model inference engine 290 includes a trained physics-based full-wafer measurement module 292. Figure 19 In the depicted embodiment, the physical full-wafer measurement module 292 estimates the seed value of the parameter of interest. 1…N POI 1…M SEED 299 associated full-wafer measurement data S at M measurement sites 1..M * 293. Estimated full-wafer measurement data S 1..M * 293 and actual full-wafer measurement data S collected at M measurement sites by a metrology system or a combination of metrology systems 1..M MEAS 291 comparison. Error evaluation module 295 determines the updated value of the parameter of interest 1…N POI 1…M * 296, which minimizes the characterization estimate of the full-wafer measurement data S 1..M * 293 and actual full-wafer measurement data S 1..M MEAS 291 is the difference between the two.

[0130] Pay attention to the updated value of the parameter 1…N POI1…M * 296 is transmitted to the trained physics-based full-wafer measurement module 292. The trained physics-based full-wafer measurement module 292 updates the physics-based full-wafer measurement model with the updated values ​​of the parameters of interest for the next iteration of the inference process. Iterations continue until the objective function is minimized. The resulting estimated values ​​of the parameters of interest are transmitted to a memory (e.g., memory 132).

[0131] The optimization performed by the error evaluation module 295 is optionally regularized by one or more regularization terms REG 297. In some embodiments, the regularization term 297 includes an expected wafer map associated with each of the N parameters of interest. In these embodiments, the solution is driven toward estimating the fit of the full-wafer measurement data to the actual full-wafer measurement data at the M measurement sites and the fit of the values ​​of the parameters of interest to the corresponding expected wafer maps used as regularization terms.

[0132] In general, different forms of regularization are considered within the scope of this patent profile, such as terms that promote smooth variations in the value of a parameter of interest across a wafer, terms that penalize discontinuities in the value of a parameter of interest across a wafer, and so on.

[0133] In some embodiments, a physics-based full-wafer model is trained and used to infer values ​​of a parameterized wafer map corresponding to each of the parameters of interest. Estimates of the parameters of interest across the measured wafer are derived directly from the values ​​of the parameterized wafer map based on the location of each measurement site.

[0134] In general, the parameters of interest determined based on the trained full-wafer metrology model described herein include, but are not limited to, geometric parameters characterizing the measurement structure, dispersion parameters characterizing the measurement structure, process parameters characterizing the process used to fabricate the measurement structure, electrical properties of the measurement structure, etc. Exemplary geometric parameters include critical dimension (CD), overlay, etc. Exemplary process parameters include lithography focus, lithography dose, etch time, etc.

[0135] In some examples, DOE measurement data associated with measurements of instances of one or more DOE metrology targets at multiple locations across one or more wafers by a metrology system is simulated. The simulated data is generated from a parameterized model of measurements of each of the one or more DOE metrology structures by the metrology system.

[0136] In some other examples, DOE measurement data associated with measurements of one or more instances of a DOE metrology target at multiple locations across one or more wafers by a metrology system is actual measurement data collected by the metrology system or multiple instances of the metrology system. In some embodiments, the same metrology system or multiple instances of the metrology system are used to collect actual measurement data from instances of the metrology target having unknown values ​​for one or more parameters of interest. In some embodiments, different instances of the metrology system or multiple different instances of the metrology system are used to collect actual measurement data from instances of the metrology target having unknown values ​​for one or more parameters of interest.

[0137] In some embodiments, the values ​​of the parameters of interest used to train the full-wafer measurement model are derived from measurements of DOE wafers using a reference metrology system. The reference metrology system is a trusted measurement system that produces sufficiently accurate measurements. In some instances, the reference metrology system is too slow to be used for measuring wafers inline as part of a wafer manufacturing process flow, but is suitable for offline use, such as for model training. By way of non-limiting example, the reference metrology system may include a stand-alone optical metrology system, such as a spectroscopic ellipsometer (SE), a SE with multiple illumination angles, a SE that measures Mueller matrix elements, a single wavelength ellipsometer, a beam profile ellipsometer, a beam profile reflectometer, a broadband reflectance spectrometer, a single wavelength reflectometer, an angle-resolved reflectometer, an imaging system, a scatterometer (e.g., a spot analyzer), an X-ray based metrology system (e.g., a small angle X-ray scatterometer (SAXS) operating in transmission or grazing incidence mode), an X-ray diffraction (XRD) system, an X-ray fluorescence (XRF) system, an X-ray photoelectron spectroscopy (XPS) system, an X-ray reflectometry (XRR) system, a Raman spectroscopy system, an atomic force microscope (AFM) system, a transmission electron microscope system, a scanning electron microscope system, a soft X-ray reflectometry system, an imaging-based metrology system, a hyperspectral imaging-based metrology system, a scatterometry overlay metrology system, or other technology capable of determining device geometry.

[0138] In some embodiments, the measurement model trained as described herein is implemented as a neural network model. In other examples, the measurement model can be implemented as a linear model, a nonlinear model, a polynomial model, a response surface model, a support vector machine model, a decision tree model, a random forest model, a kernel regression model, a deep network model, a convolutional network model, or other types of models.

[0139] In yet another aspect, the measurement results described herein can be used to provide active feedback to process tools (e.g., lithography tools, etching tools, deposition tools, etc.). For example, the values ​​of measurement parameters determined based on the measurement methods described herein can be transmitted to an etching tool to adjust the etching time for achieving a desired etching depth. In a similar manner, etching parameters (e.g., etching time, diffusion rate, etc.) or deposition parameters (e.g., time, concentration, etc.) can be included in the measurement model to provide active feedback to the etching tool or deposition tool, respectively. In some examples, corrections to process parameters determined based on measurement device parameter values ​​determined using a trained full-wafer measurement model can be transmitted to the process tool. In one embodiment, the computing system 130 determines the values ​​of one or more parameters of interest based on the measurement signal 111 received from the measurement system during the process. In addition, the computing system 130 transmits control commands to a process controller (not shown) based on the determined values ​​of the one or more parameters of interest. The control commands cause the process controller to change the state of the process (e.g., stop the etching process, change the diffusion rate, change the lithography focus, change the lithography dose, etc.).

[0140] In some embodiments, the methods and systems described herein for metrology of semiconductor devices are applied to measuring memory structures. These embodiments enable optical critical dimension (CD), film, and composition metrology of periodic and planar structures.

[0141] In some examples, the measurement model is implemented as a KLA-Tencor® available from KLA-Tencor Corporation (Milpitas, CA). Elements of an optical critical dimension metrology system. In this way, a model is generated and ready to be used in real time after spectra are collected by the system.

[0142] In some other examples, the measurement model is implemented by a CT-1000, available from Ketian Corporation (Milpitas, CA, USA). The resulting trained model can be incorporated into a database accessible by the metrology system performing the measurement. Components of the library.

[0143] Figure 20 A method 300 for training a full-wafer measurement model illustrating at least one novel aspect is provided. The method 300 is suitable for use by a metrology system (e.g., the present invention). Figure 1 100). In one aspect, it should be appreciated that the data processing blocks of method 300 may be implemented via pre-programmed algorithms executed by one or more processors of computing system 130 or any other general-purpose computing system. It should be appreciated that the specific structural aspects of metrology system 100 are not intended to be limiting, but rather should be interpreted as illustrative only.

[0144] In block 301, a certain amount of measurement data is collected from each of a plurality of measurement sites across a wafer. Each measurement site includes one or more instances of one or more structures disposed on the wafer.

[0145] In block 302, an estimated value of a parameter of interest characterizing each instance of one or more structures at each of a plurality of measurement sites across a wafer is determined based on a certain amount of measurement data using a trained full-wafer measurement model. The trained full-wafer measurement model is valid across the wafer and is evaluated based on a certain amount of measurement data at each of the plurality of measurement sites.

[0146] In a further embodiment, system 100 includes one or more computing systems 130 for performing measurements of semiconductor structures based on a trained full-wafer measurement model according to the methods described herein. The one or more computing systems 130 can be communicatively coupled to one or more spectrometers, active optical elements, process controllers, and the like. In one aspect, the one or more computing systems 130 are configured to receive measurement data associated with spectroscopic measurements of structures on wafer 101.

[0147] It should be appreciated that one or more steps described throughout this disclosure may be performed by a single computer system 130 or, alternatively, multiple computer systems 130. Furthermore, the various subsystems of the system 100 may include computer systems suitable for performing at least a portion of the steps described herein. Therefore, the above description should not be construed as limiting the present invention, but is provided for illustration only.

[0148] Additionally, computer system 130 may be communicatively coupled to the spectrometer in any manner known in the art. For example, one or more computer systems 130 may be coupled to a computer system associated with the spectrometer. In another example, the spectrometer may be directly controlled by a single computer system coupled to computer system 130.

[0149] The computer system 130 of the system 100 may be configured to receive and / or obtain data or information from subsystems of the system (e.g., spectrometers and the like) via a transmission medium that may include wired and / or wireless portions. In this manner, the transmission medium may serve as a data link between the computer system 130 and the other subsystems of the system 100.

[0150] The computer system 130 of system 100 can be configured to receive and / or obtain data or information (e.g., measurement results, modeling inputs, modeling results, reference measurement results, etc.) from other systems via a transmission medium that can include wired and / or wireless components. In this manner, the transmission medium can serve as a data link between the computer system 130 and other systems (e.g., memory onboard system 100, external memory, or other external systems). For example, the computer system 130 can be configured to receive measurement data from a storage medium (i.e., memory 132 or external memory) via a data link. For example, spectral results obtained using the spectrometer described herein can be stored in a permanent or semi-permanent memory device (e.g., memory 132 or external memory). In this regard, spectral results can be input from onboard memory or an external memory system. Furthermore, the computer system 130 can send data to other systems via the transmission medium. For example, measurement models or estimated parameter values ​​determined by the computer system 130 can be transmitted to and stored in external memory. In this regard, measurement results can be output to another system.

[0151] Computing system 130 may include, but is not limited to, a personal computer system, a mainframe computer system, a workstation, a graphics computer, a parallel processor, or any other device known in the art. In general, the term "computing system" may be broadly defined to encompass any device having one or more processors that execute instructions from a memory medium.

[0152] Program instructions 134 implementing a method, such as the method described herein, may be transmitted via a transmission medium such as a wire, cable, or wireless transmission link. Figure 1 As illustrated in FIG, program instructions 134 stored in memory 132 are transferred to processor 131 via bus 133. Program instructions 134 are stored in a computer-readable medium, such as memory 132. Exemplary computer-readable media include read-only memory, random access memory, magnetic or optical disks, or tape.

[0153] As used herein, the term "critical dimension" includes any critical dimension of a structure (e.g., bottom critical dimension, middle critical dimension, top critical dimension, sidewall angle, grating height, etc.), any critical dimension between two or more structures (e.g., distance between two structures), and displacement between two or more structures (e.g., overlay displacement between overlay grating structures, etc.). Structures may include three-dimensional structures, patterned structures, overlay structures, and the like.

[0154] As described herein, the term "critical dimension application" or "critical dimension measurement application" includes any critical dimension measurement.

[0155] As described herein, the term "metrology system" encompasses any system used, at least in part, to characterize any aspect of a sample, including measurement applications such as critical dimension metrology, overlay metrology, focus / dose metrology, and composition metrology. However, such technical terms do not limit the scope of the term "metrology system" as described herein. Additionally, system 100 can be configured to measure patterned and / or unpatterned wafers. The metrology system can be configured as an LED inspection tool, an edge inspection tool, a backside inspection tool, a macro inspection tool, or a multi-mode inspection tool (involving data from one or more platforms simultaneously), as well as any other metrology or inspection tool that benefits from the techniques described herein.

[0156] Various embodiments of semiconductor measurement systems are described herein that can be used to measure samples within any semiconductor processing tool, such as an inspection system or a lithography system. The term "sample" is used herein to refer to a wafer, a reticle, or any other sample that can be processed (e.g., printed or inspected for defects) by means known in the art.

[0157] As used herein, the term "wafer" generally refers to a substrate formed of semiconductor or non-semiconductor materials. Examples include, but are not limited to, single crystal silicon, gallium arsenide, and indium phosphide. Such substrates are commonly found and / or processed in semiconductor manufacturing facilities. In some cases, a wafer may comprise only a substrate (i.e., a bare wafer). Alternatively, a wafer may comprise one or more layers of different materials formed on a substrate. The one or more layers formed on a wafer may be "patterned" or "unpatterned." For example, a wafer may comprise multiple dies having repeatable pattern features.

[0158] A "reticle" can be a reticle at any stage of the reticle manufacturing process or a finished reticle that may or may not be released for use in a semiconductor manufacturing facility. A reticle or "mask" is generally defined as a substantially transparent substrate having substantially opaque regions formed thereon and configured in a pattern. The substrate may comprise, for example, a glass material such as amorphous SiO2. The reticle may be placed over a photoresist blanket wafer during the exposure step of the photolithography process so that the pattern on the reticle can be transferred to the photoresist.

[0159] The one or more layers formed on a wafer may be patterned or unpatterned. For example, a wafer may include multiple dies, each having repeatable pattern features. The formation and processing of such material layers may ultimately result in a completed device. Many different types of devices may be formed on a wafer, and the term "wafer" as used herein is intended to encompass a wafer having any type of device fabricated thereon known in the art.

[0160] In one or more exemplary embodiments, the functions described may be implemented in hardware, software, firmware, or any combination thereof. If implemented in software, the functions may be stored on or transmitted through a computer-readable medium as one or more instructions or codes. Computer-readable media include both computer storage media and communication media, including any media that facilitates the transfer of a computer program from one location to another. The storage medium may be any available media that can be accessed by a general-purpose or special-purpose computer. By way of example and not limitation, such computer-readable media may include RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other media that can be used to carry or store desired program code components in the form of instructions or data structures and that can be accessed by a general-purpose or special-purpose computer or a general-purpose or special-purpose processor. In addition, any connection is appropriately referred to as a computer-readable medium. For example, if the software is transmitted from a website, server, or other remote source using a coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies (such as infrared, radio, and microwave), then the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies (such as infrared, radio, and microwave) are included in the definition of medium. As used herein, disk and optical disc include compact disc (CD), laser disc, optical disc, digital versatile disc (DVD), floppy disk, and Blu-ray disc, where disks typically reproduce data magnetically, while optical discs use lasers to reproduce data optically. Combinations of the above should also be included within the scope of computer-readable media.

[0161] Although certain specific embodiments have been described above for instructional purposes, the teachings of this patent file have general applicability and are not limited to the specific embodiments described above. Therefore, various modifications, adaptations, and combinations of the various features of the described embodiments may be practiced without departing from the scope of the invention as set forth in the claims.

Claims

1. A system comprising: a metrology system comprising an illumination source and a detector, the metrology system configured to collect an amount of measurement data from each of a plurality of measurement sites across a wafer, each measurement site comprising one or more instances of one or more structures disposed on the wafer; and A computing system configured to: receiving the amount of measurement data from each of the plurality of measurement sites on the wafer; and An estimate of a parameter of interest characterizing each instance of the one or more structures at each of the plurality of measurement sites across the wafer is determined based on the amount of measurement data using a trained full-wafer metrology model valid across the wafer, wherein the trained full-wafer metrology model is evaluated based on the amount of measurement data at each of the plurality of measurement sites.

2. The system of claim 1 , wherein the computing system is further configured to: receiving a quantity of design of experiments (DOE) measurement data associated with measurements of one or more DOE instances of the one or more structures at each of a plurality of DOE measurement sites; receiving reference values ​​of one or more parameters of interest characterizing the one or more DOE instances of the one or more structures at each of the plurality of DOE measurement sites; and The full-wafer measurement model is iteratively trained in parallel based on the certain amount of DOE measurement data at the plurality of DOE measurement sites and the corresponding reference values.

3. The system of claim 2, wherein the computing system is further configured to: An indication of a location of each of the plurality of DOE measurement sites is received, wherein the training of the full-wafer measurement model is also based on the location of each of the plurality of DOE measurement sites, and wherein the determining of the estimated value of the parameter of interest characterizing each instance of the one or more structures disposed on the wafer at each of the plurality of measurement sites is also based on the location of each of the plurality of measurement sites.

4. The system of claim 1 , wherein the determining of the estimated value of the parameter of interest characterizing each instance of the one or more structures at each of the plurality of measurement sites across the wafer involves: estimating values ​​of coefficients of a function of a parameterized wafer map that characterizes the value of the parameter of interest at any location across the wafer; and The value of the parameter of interest characterizing each instance of the one or more structures at each of the plurality of measurement sites across the wafer is determined based on the estimate of the coefficient value.

5. The system of claim 2, wherein the computing system is further configured to: determining the reference values ​​of the one or more parameters of interest for the one or more DOE instances characterizing the one or more structures at each of the plurality of DOE measurement sites based on a function of a DOE parameterized wafer map characterizing reference values ​​of the parameters of interest at any location across the wafer; and The amount of design of experiments (DOE) measurement data associated with measurements of the one or more DOE instances of the one or more structures at each of the plurality of DOE measurement sites is determined based on simulations of the metrology system including the reference values ​​for the one or more parameters of interest.

6. The system of claim 5, wherein the computing system is further configured to: Values ​​of coefficients of a function of the DOE parameterized wafer map characterizing a reference value of the parameter of interest at any location across the wafer are estimated based on measured or assumed values ​​of the reference value of the parameter of interest.

7. The system of claim 6, wherein the computing system is further configured to: and estimating values ​​of coefficients of a function of a DOE parameterized wafer map that characterizes reference values ​​of one or more auxiliary parameters that characterize one or more structures under test at any location across the wafer, wherein the simulation of the metrology system also includes the reference values ​​of the one or more auxiliary parameters.

8. The system of claim 2, wherein the reference values ​​of one or more parameters of interest for the one or more DOE instances characterizing the one or more structures at each of the plurality of DOE measurement sites are generated by a process simulator.

9. The system of claim 2, wherein the reference values ​​of one or more parameters of interest characterizing the one or more DOE instances of the one or more structures at each of the plurality of DOE measurement sites are measured by a trusted reference metrology system.

10. The system of claim 1, wherein the trained full-wafer measurement model is based on machine learning.

11. The system of claim 1 , wherein the trained full-wafer metrology model is physics-based.

12. The system of claim 1, wherein the amount of measurement data comprises measurements of the one or more structures by at least one optically based metrology system, at least one X-ray based metrology system, or any combination of these.

13. A method comprising: collecting an amount of measurement data from each of a plurality of measurement sites across a wafer, each measurement site comprising one or more instances of one or more structures disposed on the wafer; and An estimate of a parameter of interest characterizing each instance of the one or more structures at each of the plurality of measurement sites across the wafer is determined based on the amount of measurement data using a trained full-wafer metrology model valid across the wafer, wherein the trained full-wafer metrology model is evaluated based on the amount of measurement data at each of the plurality of measurement sites.

14. The method according to claim 13, further comprising: receiving a quantity of design of experiments (DOE) measurement data associated with measurements of one or more DOE instances of the one or more structures at each of a plurality of DOE measurement sites; receiving reference values ​​of one or more parameters of interest characterizing the one or more DOE instances of the one or more structures at each of the plurality of DOE measurement sites; and The full-wafer measurement model is iteratively trained in parallel based on the certain amount of DOE measurement data at the plurality of DOE measurement sites and the corresponding reference values.

15. The method according to claim 14, further comprising: An indication of a location of each of the plurality of DOE measurement sites is received, wherein the training of the full-wafer measurement model is also based on the location of each of the plurality of DOE measurement sites, and wherein the determining of the estimated value of the parameter of interest characterizing each instance of the one or more structures disposed on the wafer at each of the plurality of measurement sites is also based on the location of each of the plurality of measurement sites.

16. The method of claim 13 , wherein the determining of the estimated value of the parameter of interest characterizing each instance of the one or more structures at each of the plurality of measurement sites across the wafer involves: estimating values ​​of coefficients of a function of a parameterized wafer map that characterizes the value of the parameter of interest at any location across the wafer; and The value of the parameter of interest characterizing each instance of the one or more structures at each of the plurality of measurement sites across the wafer is determined based on the estimate of the coefficient value.

17. The method of claim 14, further comprising: determining the reference values ​​of the one or more parameters of interest for the one or more DOE instances characterizing the one or more structures at each of the plurality of DOE measurement sites based on a function of a DOE parameterized wafer map characterizing reference values ​​of the parameters of interest at any location across the wafer; and The amount of design of experiments (DOE) measurement data associated with measurements of the one or more DOE instances of the one or more structures at each of the plurality of DOE measurement sites is determined based on simulations of the metrology system including the reference values ​​for the one or more parameters of interest.

18. The method of claim 13, wherein the trained full-wafer metrology model is physics-based or machine learning-based.

19. A system comprising: a metrology system comprising an illumination source and a detector, the metrology system configured to collect an amount of measurement data from each of a plurality of measurement sites across a wafer, each measurement site comprising one or more instances of one or more structures disposed on the wafer; and A non-transitory computer-readable medium comprising instructions that, when executed by one or more processors of a computing system, cause the computing system to: receiving the amount of measurement data from each of the plurality of measurement sites on the wafer; and An estimate of a parameter of interest characterizing each instance of the one or more structures at each of the plurality of measurement sites across the wafer is determined based on the amount of measurement data using a trained full-wafer metrology model valid across the wafer, wherein the trained full-wafer metrology model is evaluated based on the amount of measurement data at each of the plurality of measurement sites.

20. The system of claim 19, the non-transitory computer-readable medium further comprising instructions that, when executed by one or more processors of the computing system, cause the computing system to: receiving a quantity of design of experiments (DOE) measurement data associated with measurements of one or more DOE instances of the one or more structures at each of a plurality of DOE measurement sites; receiving reference values ​​of one or more parameters of interest characterizing the one or more DOE instances of the one or more structures at each of the plurality of DOE measurement sites; and The full-wafer measurement model is iteratively trained in parallel based on the certain amount of DOE measurement data at the plurality of DOE measurement sites and the corresponding reference values.

Images