Method and system for determining the quality of semiconductor measurements - Patents.com
A quality indicator model trained on actual and simulated data enhances semiconductor measurement accuracy by normalizing metrics across diverse scenarios, addressing indirect measurement challenges and improving process control.
Patent Information
- Application Number
- JP2023519029
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- Priority Date
- 2020-09-24
- Filing Date
- 2021-09-01
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2041-09-01
AI Technical Summary
Existing metrology techniques struggle to accurately measure semiconductor parameters due to indirect measurement methods, limited applicability across different measurement scenarios, and high computational costs, leading to inaccurate estimation of measurement quality and uncertainty.
A quality indicator model is trained using actual and simulated measurement data to estimate the quality of semiconductor measurements, with normalized metrics ensuring applicability across various measurement scenarios, and providing feedback for process adjustments.
The quality indicator model effectively identifies measurement uncertainties and outliers, enabling precise process control and improving the accuracy of semiconductor manufacturing by providing real-time feedback for process adjustments.
Smart Images

Figure 0007595756000006 
Figure 0007595756000007 
Figure 0007595756000008
Abstract
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 technology]
[0002] Semiconductor devices, such as logic devices and memory devices, are typically fabricated by a series of processing steps applied to a specimen. The various features and multiple structural hierarchies of the semiconductor device are formed by these processing steps. For example, lithography, among others, is one semiconductor manufacturing process that involves generating patterns on a semiconductor wafer. Further examples of semiconductor manufacturing processes include, but are not limited to, chemical mechanical polishing, etching, deposition, and ion implantation. Multiple semiconductor devices may be fabricated 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 the wafer to improve yield. Optical and X-ray based metrology techniques offer the potential for high throughput without the risk of destroying the sample. Many metrology-based techniques, including scatterometry, reflectometry, and ellipsometry implementations and associated analysis algorithms, are commonly used to characterize the 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 specimen being measured. In most cases, the raw measurement signal cannot be used to directly determine the physical properties of the specimen. Instead, a measurement model is employed to estimate the values 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 specimen being measured. Generally, the raw measurement signal (e.g., α meas and β meas) to determine the physical properties of a test specimen, a physics-based or machine learning-based measurement model is required.
[0005] In some examples, the raw measurement signal (e.g., α meas and β meas A physics-based metrology model is created that attempts to predict the parameters related to the metrology tool itself, such as the machine parameters (P machine ) and parameters related to the test piece being measured. When determining the value of the parameter of interest, some test piece parameters are set to fixed values (P spec-fixed ) and other specimen parameters of interest are treated as floating values (P spec-float ), i.e., evaluated on the basis of the raw measurement signal.
[0006] α model =f(P machine ,P spec-fixed ,P spec-float ) (1) β model =g(P machine ,P spec-fixed ,P spec-float ) (2)
[0007] Machine parameters are parameters used to characterize the metrology tool (e.g., the ellipsometer 101). Exemplary machine parameters include the angle of incidence (AOI), analyzer angle (A0), polarizer angle (P0), illumination wavelength, numerical aperture (NA), compensator or waveplate (if present), etc. Specimen parameters are parameters used to characterize the specimen (e.g., material parameters and geometric parameters that characterize the structure under measurement). For a thin film specimen, exemplary specimen parameters include the refractive index, dielectric function tensor, nominal layer thicknesses of all layers, layer order, etc. For a CD specimen, exemplary specimen parameters include geometric parameter values associated with different layers, refractive indices associated with different layers, etc. For measurement purposes, the machine parameters and many specimen parameters are treated as known, fixed-value parameters. However, the values of one or more of the specimen parameters are treated as unknown, floating parameters of interest.
[0008] In some instances, values for the floating parameters of interest are determined by an iterative process (e.g., regression) that produces the best fit between theoretical predictions and experimental data. meas and β meas The values of the unknown floating parameters of interest are varied and the model output values (e.g., α model and β model ) is calculated and compared to the raw measurement data. In some other examples, the floating parameters are determined by searching a library of pre-calculated solutions to find the closest match.
[0009] In some other examples, trained machine learning based measurement models are employed to directly estimate values of the parameters of interest based on raw measurement data. In these examples, the machine learning based measurement models take raw measurement signals as model inputs and generate values of the parameters of interest as model outputs.
[0010] To generate useful estimates of the parameters of interest for a particular measurement application, both physics-based and machine learning-based measurement models must be trained. Typically, model training is based on raw measurement signals collected from test specimens with known values of the parameters of interest (Design of Experiments (DOE) data).
[0011] The machine learning-based measurement model is parameterized by several weight parameters. Conventionally, the machine learning-based measurement model is trained by a regression process (e.g., ordinary least squares regression). The values of the weight parameters are iteratively adjusted to minimize the difference between the known reference value of the target parameter and the value of the target parameter estimated by the machine learning-based measurement model based on the measured raw measurement signal.
[0012] As mentioned above, the physics-based metrology model is parameterized by several machine and specimen parameters. Traditionally, physics-based metrology models are also trained by a regression process (e.g., ordinary least squares regression). One or more of the machine and specimen parameters are iteratively adjusted to minimize the difference between the raw measurement data and the modeled measurement data. At each iteration, the value of the particular specimen parameter of interest is maintained at a known DOE value.
[0013] Both trained machine learning-based measurement models and physics-based models estimate the values of the parameters of interest based on measurement data. The accuracy of the estimation depends on the measurement data. For example, measurement data outside the domain of the measurement data observed during the model training phase can cause large prediction errors due to extrapolation.
[0014] Previously, several techniques have been considered to quantify the output uncertainty associated with each measurement site, including Bayesian regression (e.g., linear process regression, Gaussian process regression, etc.) using variational inference and Markov Chain Monte Carlo (MCMC) methods, Monte Carlo dropout as a Bayesian neural network, bootstrap sampling, output variation from different hyperparameter combinations, and estimating the distance of the sample to the center of the training set.
[0015] Unfortunately, these conventional techniques have limited practical success. Each of the aforementioned techniques represents an ensemble average prediction that is not associated with a trained measurement model. The ensemble average prediction is not the same as the prediction of an optimally trained measurement model. Therefore, the ensemble average prediction cannot accurately estimate the measurement uncertainty and cannot directly measure the quality of the measurement model. Furthermore, Bayesian regression and bootstrap sampling are computationally expensive and are practically unsuitable for implementation as part of a large-scale semiconductor manufacturing facility. Finally, the aforementioned techniques are not normalized. Without normalization, the results cannot be consistently interpreted across different measurement recipes or structures during measurement. [Prior art documents] [Patent documents]
[0016] [Patent Document 1] US Patent Application Publication No. 2016 / 0154319 Summary of the Invention [Problem to be solved by the invention]
[0017] Increasing demand for fine resolution, multi-parameter correlation, increasing geometric complexity, and increasing use of opaque materials pose metrology challenges in future metrology applications. Therefore, methods and systems for improving the assessment of measurement quality are desired. [Means for solving the problem]
[0018] A method and system are presented herein for estimating a quality index value indicative of one or more performance characteristics of a semiconductor measurement. The quality index is determined from a quality index model trained based on actual measurement data, simulated measurement data, or both, and reference measurement data. Once trained, the quality index model allows for the calculation of a quantitative quality index indicative of the measurement quality of a particular measurement recipe based on measurement data collected from an in-line wafer at each measurement site without the use of a reference metrology. The quality index value is normalized to ensure applicability across a wide range of measurement scenarios.
[0019] In some embodiments, a value of the quality index is determined for each measurement sample during measurement inference. In some embodiments, the trained measurement model is a defect classification model trained to detect and classify defects from measurement data.
[0020] In general, the quality index may be used for a variety of other purposes. The quality index may be used to characterize uncertainties associated with measurement parameters of interest, defect classification, etc. In some other examples, the quality index is employed as a filter to detect outliers in measurements. In some other examples, the quality index is employed as a trigger to adjust a semiconductor process. Similarly, in some examples, the quality index value is employed to detect wafers that are not within manufacturing specifications. In some of these examples, the quality index value is employed to detect when a measurement tool bias has deviated from specifications. In other examples, the quality index value is employed to detect when a measurement error bias has deviated from specifications.
[0021] In one embodiment, the quality index model is trained by training parameters of a data reconstruction model, which reconstructs each feature of the measurement dataset based on other features of the measurement dataset and values of one or more target parameters estimated by the trained measurement model based on the measurement dataset.
[0022] In another aspect, a trained quality index model is employed to determine a value of a quality index associated with a particular measurement based on an amount of measurement data.
[0023] In a further aspect, the raw quality index values are normalized to broaden the applicability of the quality index values to multiple recipes, tools, layers, use cases, etc. In this manner, the normalized quality index is an indication of the measurement performance for a wide range of measurement scenarios.
[0024] In another aspect, the training data and measurement data employed to train and use the quality indicator models described herein are simulated and / or collected from actual measurements from multiple targets located in close proximity on a wafer, in some embodiments the measurement targets during the measurements described herein may be actual device structures rather than dedicated metrology targets.
[0025] In another embodiment, the training data and measurement data used to train and use the quality indicator models described herein are collected from one or more targets by multiple different measurement systems.
[0026] In some embodiments, the measurement system employed to collect the measurement data and estimate the target parameter is different from the measurement system employed to collect the measurement data and estimate the value of the quality index.
[0027] In another further aspect, a plurality of different parameters of interest are measured and a corresponding quality index is evaluated for each of the plurality of different parameters of interest, hi some embodiments, the quality indexes associated with the plurality of different parameters of interest are employed to more clearly identify measurement uncertainties associated with a particular measurement recipe, to more clearly identify when changes in process parameters are required, or both.
[0028] In another further aspect, values of the parameters of interest, corresponding quality index values, or both, associated with structures in close proximity to the measurement target are employed to more clearly identify when changes in process parameters are required.
[0029] In yet another aspect, the measurements described herein can be used to provide active feedback to a process tool (eg, a lithography tool, an etch tool, a deposition tool, etc.).
[0030] The above is a summary and thus contains, by necessity, simplifications, generalizations, and omissions of detail; thus, those skilled in the art will appreciate that the summary is for illustrative purposes 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] FIG. 1 is a diagram of a wafer metrology system 100 for monitoring measurement quality of measurements performed on a wafer according to example methods presented herein. [Diagram 2] FIG. 1 illustrates an exemplary measurement quality model training engine 150 in one embodiment. [Diagram 3] FIG. 1 illustrates an exemplary measurement quality engine 160 in another embodiment. [Figure 4A]1 is a plot showing the correlation between the absolute error in the value of a parameter of interest determined by a measurement model and the corresponding quality index value associated with each measurement. [Figure 4B] 1 is a plot showing the correlation between the absolute error in the value of a parameter of interest determined by a measurement model and the corresponding quality index value associated with each measurement. [Figure 4C] 1 is a plot showing the correlation between the absolute error in the value of a parameter of interest determined by a measurement model and the corresponding quality index value associated with each measurement. [Figure 4D] 1 is a plot showing the correlation between the absolute error in the value of a parameter of interest determined by a measurement model and the corresponding quality index value associated with each measurement. [Figure 4E] 1 is a plot showing the correlation between the absolute error in the value of a parameter of interest determined by a measurement model and the corresponding quality index value associated with each measurement. [Figure 4F] 1 is a plot showing the correlation between the absolute error in the value of a parameter of interest determined by a measurement model and the corresponding quality index value associated with each measurement. [Diagram 5] 1 is a plot showing a distribution of quality index values calculated for several measurement sites on each of several wafers. [Figure 6] 3 is a flow diagram of a method 300 for estimating a value of a quality index indicative of one or more performance characteristics of a semiconductor measurement. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0032] Reference will now be made in detail to example background and certain embodiments of the invention, examples of which are illustrated in the accompanying drawings.
[0033] A method and system are presented herein for estimating a quality index value indicative of one or more performance characteristics of a semiconductor measurement. The quality index is determined from a quality index model trained based on actual measurement data, simulated measurement data, or both, and reference measurement data. Once trained, the quality index model allows for the calculation of a quantitative quality index indicative of the measurement quality of a particular measurement recipe based on measurement data collected from an in-line wafer at each measurement site without the use of a reference metrology. The quality index value is normalized to ensure applicability across a wide range of measurement scenarios.
[0034] In general, the quality index values indicate measurement uncertainty, anomalies in the measurement data compared to the training data, confidence in the values of the target parameters estimated using the trained measurement model, and the need for adjustments to process variables at the wafer level or wafer lot level.
[0035] In some embodiments, a value of the quality index is determined for each measurement sample during measurement inference. In some of these embodiments, the semiconductor measurement includes inferring values of one or more parameters of interest from the measurement data using a trained measurement model. In some of these embodiments, the trained measurement model is physics-based. In some other embodiments, the trained measurement model is a machine learning based measurement model (e.g., a linear model, a neural network model, a convolutional network model, etc.). If the quality index estimate indicates that the measurement quality is insufficient, the measurement is flagged. In some examples, repeated instances of insufficient measurement quality indicate that the measurement model is stale and requires retraining.
[0036] In some embodiments, the trained measurement model is a defect classification model trained to detect and classify defects from the measurement data. In these embodiments, the trained quality index model is employed to determine the uncertainty of the defect classification. In some of these embodiments, the measurement data is image data and the trained quality index model is employed to determine the uncertainty of the defect detection from the image data.
[0037] In some embodiments, the trained measurement model is a metrology model trained to estimate a geometric parameter of interest (e.g., critical dimension, overlay, etc.), a dispersion parameter of interest (e.g., refractive index, etc.), a process parameter of interest (e.g., lithography focus, dose, etc.), an electrical property of interest (e.g., bandgap, etc.), or any combination thereof. In these embodiments, the trained quality indicator model is employed to determine the uncertainty of the estimated parameter of interest.
[0038] Quality indices may be employed to characterize uncertainties associated with measurement parameters of interest, defect classification, and the like, among other things. In general, however, quality indices may be used for a variety of other purposes. In some examples, quality indices are employed as filters to detect outliers in measurements. In one example, the value of a quality index associated with a particular measurement is compared to a predefined threshold. If the quality index value exceeds the predefined threshold, the measurement is considered an outlier. In some examples, quality indices are employed as triggers to adjust a semiconductor process. For example, if an average value of several quality indices each associated with different instances of the same measurement exceeds a predefined threshold, the process is adjusted to bring the average value of the quality index back into a desired range.
[0039] Similarly, in some examples, the quality index value is employed to detect wafers that are not within manufacturing specifications. In some of these examples, the quality index value is employed to detect when a measurement tool bias has deviated from specifications. In other examples, the quality index value is employed to detect when a measurement error bias has deviated from specifications.
[0040] FIG. 1 illustrates a system 100 for measuring properties of a specimen and evaluating quality indicators associated with each measurement, according to an exemplary method presented herein. As illustrated in FIG. 1, the system 100 may be used to perform spectroscopic ellipsometric measurements of a structure 101. In this embodiment, the system 100 may include a spectroscopic ellipsometer with an illuminator 102 and a spectrometer 104. The illuminator 102 of the system 100 is configured to generate illumination in a selected wavelength range (e.g., 100-2500 nm) and direct it to a structure disposed on a surface of the specimen on which the structure 101 is fabricated. The spectrometer 104 is then configured to receive illumination reflected from the structure 101. It is further noted that the light originating from the illuminator 102 is polarized using a polarization state generator 107 to generate a polarized illumination beam 106. The radiation reflected by the structure 101 passes through a polarization state analyzer 109 and is sent to the spectrometer 104. The radiation received by the spectrometer 104 in the collected beam 108 is analyzed for polarization state, which allows the spectrometer to perform spectral analysis of the radiation that passes through the analyzer. These spectra 111 are passed to a computing system 130 for analysis of the structures described herein.
[0041] As shown in FIG. 1, system 100 includes a single measurement technique (i.e., SE). In general, however, system 100 may include any number of different measurement techniques. As non-limiting examples, 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 or multiple wavelength ellipsometer, or any combination thereof. Furthermore, in general, the measurement data collected by the different measurement techniques and analyzed according to the methods described herein may be collected from multiple tools, a single tool integrating multiple techniques, or a combination thereof.
[0042] In further embodiments, system 100 may include one or more computing systems 130 employed to perform measurements of the structure and determine quality index values according to methods described herein. The one or more computing systems 130 may be communicatively coupled to spectrometer 104. In one aspect, the one or more computing systems 130 are configured to receive measurement data 111 related to measurements of the structure (e.g., structure 101) being measured.
[0043] In one aspect, the computing system 130 is configured as a measurement quality model training engine for training a quality indicator model associated with a measurement of the structure being measured. Figure 2 illustrates an exemplary measurement quality model training engine 150 in one embodiment. As shown in Figure 2, the measurement quality model training engine 150 includes an optional data transformation module 151, a trained measurement model module 153, and a data reconstruction model training module 155.
[0044] The quality index model is trained based on a quantity of training data and reference data. The training data includes actual measurement data, simulated measurement data, or both. In some embodiments, the training data includes simulated measurement data from a simulation of measurement of a particular target with a particular measurement tool. In some embodiments, the training data includes actual measurement data from measurement of a particular target with a particular measurement tool. In some of these embodiments, the measurement data includes actual measurements of targets on a nominal wafer. In some of these embodiments, the measurement data includes actual measurements of targets on a design of experiments (DOE) wafer having known programmed values of parameters of interest.
[0045] The training data may be simulated, collected, or both, for any of a variety of metrology purposes, which in some embodiments are measurement accuracy, measurement precision tracking, tool-to-tool matching, wafer-to-wafer variation, etc.
[0046] As mentioned above, training data is required to train a measurement model. The training data employed to train a quality indicator model may be the same training dataset as the dataset employed to train the corresponding measurement model, a subset of the training dataset employed to train the corresponding measurement model, or a dataset different from the training dataset employed to train the corresponding measurement model. In a preferred embodiment, the training data employed to train a quality indicator model is the same dataset or a subset of the dataset employed to train the corresponding measurement model.
[0047] The reference data employed to train the quality index model includes reliable values of one or more parameters of interest associated with the measurement. In some embodiments, the reference data is obtained from a reference metrology system, i.e., a metrology system that is trusted to provide good measurements of the parameters of interest. In some embodiments, the reference data is provided empirically by a user of the metrology system. In some embodiments, the reference data is provided as known programmed values of the parameters of interest associated with the measured DOE wafer. In some embodiments, the reference data is derived from a statistical analysis of measurements of the parameters of interest by a number of different measurement techniques. In some embodiments, the reference data is derived from known physical constraints.
[0048] Generally, the reference data represents the values of one or more parameters of interest in various terms, such as precision, accuracy (e.g., bias), tracking (e.g., correlation to a reference value), tool-to-tool matching, within-wafer variation, wafer average, wafer signature, and wafer-to-wafer variation.
[0049] In some embodiments, the features of the training dataset are directly adopted as input data for training the quality index model. However, in some other embodiments, the training dataset is transformed into a reduced dimension, and the features of the reduced training dataset are directly adopted to train the quality index model. In many practical situations, it is preferable to reduce the dimension of the training dataset to reduce the computational load. In one example, a set of measurement spectra adopted as training data may include several thousand features (e.g., 15,000 measurement signals). However, using data reduction, the dimension of the training dataset may be reduced by one order of magnitude or more (e.g., 200 signals).
[0050] In some embodiments, dimensionality reduction of the data features is achieved by principal component analysis, where the measured data set is decomposed into a smaller set of principal components. In general, any suitable data reduction technique may be employed, such as Fourier analysis, wavelet analysis, discrete cosine transform analysis, etc.
[0051] 2, a training data set 157 is received by a data transformation module 151. In one example, the data transformation module 151 transforms the features of the training data set 157 into a set of principal components 152 having n components as shown in equation (1), where P T is the vector of principal components of the training data set 152 that is communicated to the trained measurement model module 153 and the data reconstruction model training module 155.
number
[0052] In one embodiment, the quality index model is trained by training parameters of a data reconstruction model, which reconstructs each feature of the measurement dataset based on other features of the measurement dataset and values of one or more target parameters estimated by the trained measurement model based on the measurement dataset.
[0053] In one example, the data reconstruction model is shown in Equation (2), where PC i is the i-th feature of the measurement dataset, and PC i * is PC i is the reconstructed value of , y is the measurement parameter of interest, and g i is the reconstruction function associated with the i-th feature of the measurement dataset, and P i is the reconstruction model constant associated with the i-th reconstruction function.
[0054] PC1 * =g1(PC2,PC3,…,PC n ,y,y 2 ,y3 ,y 4 ,P1)]) PC2 * =g2(PC1,PC3,…,PC n ,y,y 2 ,y 3 ,y 4 ,P2)]) (2) … PC n * =g n (PC1, PC2, …, PC n-1 ,y,y 2 ,y 3 ,y 4 ,P n )]) In the example shown in Equation (2), each feature of the measurement data set is a principal component. In general, the reconstruction function g may be any suitable parameterized model, such as a linear model, a neural network model, etc. In the example shown in Equation (2), each principal component reconstruction is a function of the measurement parameters of interest up to a fourth order polynomial basis, the principal components of the measurement data set excluding the principal component being reconstructed, and a constant parameter of each reconstruction function g.
[0055] The data reconstruction model is trained based on a training dataset (with or without data reduction) and the values of the parameters of interest, y, both of which are either known from reference measurements or calculated from the training dataset using a trained measurement model. The result of training is a set of adjusted model parameters, e.g., P i As shown in FIG. 2, the trained data reconstruction model 156 is stored in the memory 132.
[0056] As shown in FIG. 2, the set of principal components 152 is communicated to a trained measurement model module 153. The trained measurement model module 153 determines values of one or more object parameters y based on the set of principal components 152 using the trained measurement model. The calculated values 154 of the object parameters are communicated to a data reconstruction model training module 155. In this example, the value of the object parameter y is calculated from the training data set using the trained measurement model. In an alternative example, the value of the object parameter 154 is communicated directly to the data reconstruction model training module 155.
[0057] In some embodiments, the data reconstruction model is trained by minimizing the sum of the differences between each feature of the training data set and each corresponding reconstructed feature, which is typically expressed as a least-squares minimization. Equation (3) illustrates, as a non-limiting example, the training of the data reconstruction model.
number
[0058] In some other embodiments, the data reconstruction model is trained based on specific domain knowledge related to the training data. The domain knowledge is represented as one or more probability distributions, and the one or more probability distributions are used to regularize the optimization process employed during training. In this way, the optimization process is physically regularized by one or more representations of physics-based probability distributions. As a non-limiting example, probability distributions related to measurement accuracy, tool-to-tool matching, tracking, within-wafer variation, etc. are employed to physically regularize the optimization process. In this way, domain knowledge obtained from experience, measurement data, and physics is directly represented in the objective function that drives the optimization of the data reconstruction model. As a result, the domain knowledge is fully utilized in the development process of the quality index model.
[0059] In another aspect, a trained quality index model is employed to determine a value of a quality index associated with a particular measurement based on an amount of measurement data.
[0060] The measurement data may include actual measurement data, simulated measurement data, or both, In some embodiments, the measurement data may include simulated measurement data from a simulation of measurement of one or more measurement targets that are structurally different from the measurement targets employed as part of the training data set.
[0061] In some embodiments, the metrology data includes actual metrology data from a measurement of a particular target by a particular metrology tool. Typically, the metrology data includes actual measurements of a target on a nominal wafer. In some embodiments, metrology data is collected from multiple wafers to estimate wafer-to-wafer variation. In some embodiments, metrology data is collected from DOE wafers to estimate measurement accuracy, precision tracking, or both. In some embodiments, metrology data is collected by multiple metrology systems to estimate tool-to-tool matching.
[0062] In some embodiments, the features of the measurement dataset are directly adopted as input data to the trained quality indicator model. However, in some other embodiments, the measurement dataset is transformed into a reduced dimension, and the features of the reduced measurement dataset are directly adopted as input data to the trained quality indicator model. In many practical situations, it is preferable to reduce the dimension of the measurement dataset to reduce the computational load. In general, the same data feature dimension reduction techniques are adopted to reduce the measurement dataset as those adopted to reduce the training dataset.
[0063] In some embodiments, dimensionality reduction of the data features is achieved by principal component analysis, where the measured data set is decomposed into a smaller set of principal components. In general, any suitable data reduction technique may be employed, such as Fourier analysis, wavelet analysis, discrete cosine transform analysis, etc.
[0064] 3, the measurement dataset 173 is received by the data transformation module 161. In one example, the data transformation module 161 transforms the features of the training dataset 173 into a set of principal components 162 having n components as shown in Equation (4), where PC is a vector of principal components of the measurement dataset 162 that is communicated to the trained measurement model module 163 and the trained data reconstruction module 165.
[0065] PC = [PC1, PC2, ..., PC n ] (4)
[0066] 3, the set of principal components 162 is communicated to a trained measurement model module 163. The trained measurement model module 163 uses the trained measurement model to determine values y of one or more parameters of interest based on the set of principal components 162. The calculated values 164 of the parameters of interest are communicated to a trained data reconstruction module 165. Exemplary parameters of interest include geometric parameters, dispersion parameters, process parameters, or electrical parameters that characterize the measured structure.
[0067] Alternatively, the measurement data set 173 is processed by a trained measurement model to estimate values of one or more parameters of interest, e.g., y, that characterize the measured structure.
[0068] The data reconstruction model reconstructs each feature of the measurement dataset (or the reduced measurement dataset) based on other features of the measurement dataset and the values of one or more target parameters estimated by the trained measurement model based on the measurement dataset. In one example, the data reconstruction model is shown in Equation (2). In this example, PC i is the i-th feature of the measurement dataset, and PC i * is PC i is the reconstructed value of , y is the measurement parameter of interest, and g iis the reconstruction function associated with the i-th feature of the measurement dataset, and P i is the vector of reconstruction model constants associated with the i-th reconstruction function.
[0069] 3, the set of principal components 162 and the calculated values 164 of the parameters of interest are communicated to a trained data reconstruction module 165. The reconstruction function and reconstruction model constants are known. The trained data reconstruction module 165 uses the trained data reconstruction model to determine a reconstructed measurement data set, e.g., values of the reconstructed principal components 166, as shown in equation (2), by way of non-limiting example.
[0070] In general, when the measurement data under evaluation is close to the training space, the features of each data, e.g., PC i are the features of the corresponding reconstructed data, PC i * It will be the same as:
[0071] As shown in FIG. 3, a trained measurement model module 167 receives the reconstructed measurement data set 166 and uses a trained measurement model to derive values of one or more parameters of interest 168 that characterize the measured structure, e.g., y * Estimate.
[0072] 3, a quality index module 169 receives values of one or more target parameters 168 determined based on the reconstructed measurement data set and values of one or more target parameters 164 determined based on the measurement data set. A raw quality index value 170 is calculated by dividing the measured value of the target parameter, e.g., y, by the value of the target parameter estimated using the reconstructed measurement data, e.g., y, as shown by equation (5). * where QM raw are the raw quality index values. The difference indicates the uncertainty in the values of the parameters of interest estimated using the measurement model.
[0073] QM raw=|yy * | (5)
[0074] 3, a normalization module 171 normalizes the raw quality index values 170 to generate normalized quality index values 172, which are stored in memory 132. The scale of the raw quality index values varies by layer, tool, measurement recipe, use case, etc. In a further aspect, the raw quality index values are normalized to broaden the applicability of the quality index values to multiple recipes, tools, layers, use cases, etc. In this manner, the normalized quality index is an indication of the measurement performance for a wide range of measurement scenarios.
[0075] In some examples, the normalized quality index values are bounded between 0 and 1. In this way, the normalized quality index values are standardized and kept consistent: a smaller value indicates a more reliable measurement and vice versa.
[0076] In some embodiments, normalization of the raw quality index values is accomplished iteratively. First, the raw quality index values and an initial value of the normalization factor α 0 As a non-limiting example, equation (6) shows a first iteration of normalizing the raw quality index values, where QM norm is the normalized quality index value, and QM raw is the raw quality index value, and α 0 is the initial value of the normalization coefficient α, and QM critical is the raw quality index value QM raw As shown in equation (6), QM norm The value of is bounded by the values 0 and 1.
number
number
number
[0077] 4A-4F are plots showing the correlation between the absolute error in the value of the parameter of interest determined by the measurement model and the corresponding quality index value associated with each measurement.
[0078] In these examples, the absolute error in the measurement of the parameter of interest, Error abs is the known reference value y of the parameter of interest ref and the value of the parameter of interest, y, estimated by the measurement model from each simulated measurement spectrum.
[0079] Error abs =|yy ref | (9)
[0080] The raw quality index value QM associated with each measurement rawis expressed as the measured value of the parameter of interest, e.g., y, and the value of the parameter of interest estimated using the reconstructed measured data, e.g., y * It is determined as the absolute difference between
[0081] QM raw =|yy * | (10)
[0082] 4A-4F show the correlation between the absolute error and the corresponding quality index values for different focus values, different process conditions, and different measurement conditions, respectively.
[0083] FIG. 4A shows a plot 210 illustrating the correlation between absolute error and corresponding quality index values for 100 different measurements, including a 5% variation range of underlayer thickness and a range of focus values in the training data set employed to train the measurement model and the quality index model. As shown in FIG. 4A, plot line 211 shows the absolute error and plot line 212 shows the corresponding quality index values. As shown in FIG. 4A, for all test conditions, the absolute error is very low, as are the corresponding values of the quality index. Both the machine learning-based measurement model and the quality index model are trained based on training data similar to the measurements shown in FIG. 4A, and therefore the small values of absolute error and the corresponding quality index values indicate that the quality index methodology described herein is effective within the domain of the training data.
[0084] FIG. 4B shows a plot 220 illustrating the correlation between absolute error and corresponding quality index values for 100 different measurements, including a 10% variation range of underlayer thickness and a range of focus values within and outside the training data set employed to train the measurement model and the quality index model. As shown in FIG. 4B, plot line 221 shows the absolute error, and plot line 222 shows the corresponding quality index values. As shown in FIG. 4B, the absolute error varies depending on the test condition, as do the corresponding quality index values. Furthermore, for all test conditions, the quality index values are well correlated with the absolute error.
[0085] FIG. 4C shows a plot 230 illustrating the correlation between absolute error and corresponding quality index values for 100 different measurements, including a 5% variation range of underlayer thickness, a 10% mean shift of one layer, and a range of focus values within and outside the training data set employed to train the measurement model and the quality index model. As shown in FIG. 4C, plot line 231 shows the absolute error, and plot line 232 shows the corresponding quality index values. As shown in FIG. 4C, the absolute error varies with the test condition, as do the corresponding quality index values. Again, for all test conditions, the quality index values correlate well with the absolute error.
[0086] FIG. 4D shows a plot 240 illustrating the correlation between absolute error and corresponding quality index values for 100 different measurements, including a 5% variation range of underlayer thickness and a range of focus values within and outside the training data set employed to train the measurement model and the quality index model. As shown in FIG. 4D, plot line 241 shows the absolute error and plot line 242 shows the corresponding quality index values. As shown in FIG. 4D, the absolute error varies depending on the test condition, as do the corresponding quality index values. Again, for all test conditions, the quality index values correlate well with the absolute error.
[0087] FIG. 4E shows a plot 250 illustrating the correlation between absolute error and corresponding quality index values for 100 different measurements, including a 5% variation range in underlayer thickness, a 0.1 degree shift in incidence angle, and a range of focus values within and outside the training data set employed to train the measurement model and the quality index model. As shown in FIG. 4E, plot line 251 shows the absolute error, and plot line 252 shows the corresponding quality index values. As shown in FIG. 4E, the absolute error varies with the test condition, as do the corresponding quality index values. Again, for all test conditions, the quality index values correlate well with the absolute error.
[0088] FIG. 4F shows a plot 260 illustrating the correlation between absolute error and corresponding quality index values for 100 different measurements, including a 5% variation range in underlayer thickness, a 3-fold increase in measurement system noise, and a range of focus values within and outside the training data set employed to train the measurement model and the quality index model. As shown in FIG. 4F, plot line 261 shows the absolute error, and plot line 262 shows the corresponding quality index values. As shown in FIG. 4F, the absolute error varies depending on the test condition, as do the corresponding quality index values. Again, for all test conditions, the quality index values correlate well with the absolute error.
[0089] Both the machine learning-based measurement model and the quality index model are trained on training data with a smaller domain than the measurements shown in Figures 4B-4F. Nevertheless, the absolute error values and the corresponding quality index values are well correlated. This indicates that the quality index methodology described herein is effective outside the domain of the training data. Thus, the quality index methodology described herein can be used to detect uncertainty in the measurements of the parameters of interest.
[0090] 5 is a plot 270 showing a distribution of quality index values calculated for a number of measurement sites on each of a number of wafers. As shown in FIG. 5, each vertical bar represents a 3 sigma value of the distribution of quality index values calculated for a number of measurement sites on each particular wafer. As shown in FIG. 5, the 3 sigma value of each distribution is very small, indicating a relatively tight distribution. As a result, measurements having a relatively large quality index value, for example, greater than 0.8, may be outliers.
[0091] In another aspect, the training data and measurement data employed to train and use the quality indicator models described herein are simulated and / or collected from multiple targets located in close proximity on a wafer.
[0092] In some embodiments, the measurement target during the measurements described herein may be an actual device structure rather than a dedicated metrology target.
[0093] In another aspect, the training data and measurement data used to train and use the quality indicator models described herein are collected from one or more targets by a number of different metrology systems. In one example, measurement data from a first target may be collected by a spectroscopic ellipsometer, measurement data from the same target or a different target may be collected by a small-angle x-ray scatterometry (SAXS) system, and the same target or a different target may be collected by an imaging-based metrology system.
[0094] In some embodiments, the measurement system employed to collect the measurement data and estimate the target parameter is different from the measurement system employed to collect the measurement data and estimate the value of the quality index.
[0095] In another further aspect, a plurality of different parameters of interest are measured and a corresponding quality index is evaluated for each of the plurality of different parameters of interest, hi some embodiments, the quality indexes associated with the plurality of different parameters of interest are employed to more clearly identify measurement uncertainties associated with a particular measurement recipe, to more clearly identify when changes in process parameters are required, or both.
[0096] In another further aspect, values of the parameters of interest, corresponding quality index values, or both, associated with structures in close proximity to the measurement target are employed to more clearly identify when changes in process parameters are required.
[0097] In general, the training data and measurement data employed to train and use the quality indicator models described herein may be collected from any suitable semiconductor metrology system, including, by way of non-limiting example, spectroscopic ellipsometers, spectroscopic reflectometers, soft x-ray based metrology systems, small angle x-ray scattering metrology systems, imaging systems, hyperspectral imaging systems, and the like.
[0098] In yet another aspect, the measurements described herein can be used to provide active feedback to a process tool (e.g., a lithography tool, an etch tool, a deposition tool, etc.). For example, values of the measured parameters determined based on the measurement methods described herein can be communicated to an etch tool to adjust the etch time to achieve a desired etch depth. Similarly, an etch parameter (e.g., etch time, diffusivity, etc.) or a deposition parameter (e.g., time, concentration, etc.) may be included in the measurement model to provide active feedback to an etch tool or a deposition tool, respectively. In some examples, modifications to the process parameters determined based on the measured device parameter values may be communicated to the process tool. In one embodiment, the computing system 130 determines values of one or more target parameters. Additionally, the computing system 130 communicates a control command to a process controller based on the determined values of the one or more target parameters. The control command causes the process controller to change a state of the process (e.g., stop the etch process, change the diffusivity, etc.). In one example, the control command causes the process controller to adjust the focus of the lithography system, the dose of the lithography system, or both. In another example, the control command may cause the process controller to vary the etch rate to improve the wafer uniformity of a measured CD parameter.
[0099] In some examples, the metrology model is implemented as an element of a SpectraShape® optical critical dimension metrology system available from KLA-Tencor Corporation, Milpitas, Calif., U.S.A. In this manner, the model is created and ready to use immediately after a spectrum is collected by the system.
[0100] In some other examples, the measurement model is implemented offline, for example by a computing system implementing AcuShape® software available from KLA-Tencor Corporation, Milpitas, California, USA. The resulting trained model may be incorporated as an element of an AcuShape® library accessible by the metrology system that performs the measurements.
[0101] 6 illustrates a method 300 for estimating a value of a quality index indicative of one or more performance characteristics of a semiconductor measurement in at least one novel aspect. The method 300 is suitable for implementation by a metrology system, such as the metrology system 100 illustrated in FIG. 1 of the present invention. It should be understood that in one aspect, the data processing blocks of the method 300 may be implemented via pre-programmed algorithms executed by one or more processors of the computing system 130 or any other general-purpose computing system. It should be understood that the specific structural aspects of the metrology system 100 are not intended to represent limitations herein, but should be construed as merely illustrative.
[0102] In block 301, an amount of illumination radiation is directed to a first measurement location on a surface of a semiconductor wafer.
[0103] At block 302, an amount of radiation collected from a first measurement location in response to the amount of illumination radiation is detected.
[0104] At block 303, a first plurality of features characterizing the amount of radiation detected from the first measurement location is generated.
[0105] At block 304, values of one or more target parameters are estimated, the one or more parameters characterizing the measurement site from values of the first plurality of features of the detected radiation based on the trained measurement model.
[0106] A reconstructed value for each of the first plurality of features is determined at block 305. The reconstructed value for a feature of the first plurality of features is determined from one or more target parameters based on the trained data reconstruction model and the values of each of the first plurality of features excluding the value of the feature.
[0107] In block 306, reconstructed values of one or more object parameters characterizing the measurement site are estimated from the values of the plurality of reconstructed features based on the trained measurement model.
[0108] In block 307, a quality index value is determined based on a difference between values of the one or more target parameters determined from values of the first plurality of features and values of the one or more target parameters determined from reconstructed values of the first plurality of features.
[0109] In a further embodiment, the system 100 includes one or more computing systems 130 employed to perform measurements of the semiconductor structures based on the spectroscopic measurement data collected according to the methods described herein. The one or more computing systems 130 may be communicatively coupled to one or more spectrometers, active optical elements, process controllers, etc. In one aspect, the one or more computing systems 130 are configured to receive measurement data related to spectral measurements of the structures of the wafer 104.
[0110] It should be appreciated that one or more steps described throughout this disclosure may be performed by a single computer system 130, or alternatively, by multiple computer systems 130. Additionally, different subsystems of system 100 may include computer systems suitable for performing at least some of the steps described herein. Thus, the foregoing description should not be construed as a limitation on the present invention, but is merely illustrative.
[0111] Additionally, computer system 130 may be communicatively coupled to the spectrometer in any manner known in the art. For example, one or more of computing systems 130 may be coupled to a computing system associated with the spectrometer. In another example, the spectrometer may be directly controlled by a single computer system coupled to computer system 130.
[0112] The computer system 130 of the system 100 may be configured to receive and / or acquire data or information from subsystems of the system (e.g., spectrometer, etc.) by way of a transmission medium, which may include wired and / or wireless portions. In this manner, the transmission medium may act as a data link between the computer system 130 and other subsystems of the system 100.
[0113] The computer system 130 of the system 100 may be configured to receive and / or acquire data or information (e.g., measurement results, modeling inputs, modeling results, reference measurement results, etc.) from other systems by way of a transmission medium, which 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 system (e.g., memory on-board the system 100, an external memory, or another external system). For example, the computing system 130 may be configured to receive measurement data from a storage medium (i.e., memory 132 or an external memory) via a data link. For example, spectral results obtained using the spectrometer described herein may be stored in a persistent or semi-persistent memory device (e.g., memory 132 or an external memory). In this regard, the spectral results may be imported from the on-board memory or from an external memory system. Additionally, the computer system 130 may transmit data to other systems via the transmission medium. For example, a measurement model, or estimated parameter values determined by the computer system 130 may be communicated and stored in an external memory. In this regard, the measurement results may be exported to another system.
[0114] 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.
[0115] Program instructions 134 implementing methods such as those described herein may be transmitted over a transmission medium such as a wire, cable, or wireless transmission link. For example, as shown in Figure 1, program instructions 134 stored in memory 132 are transmitted to processor 131 via bus 133. Program instructions 134 are stored in a computer-readable medium (e.g., memory 132). Exemplary computer-readable media include read-only memory, random-access memory, magnetic or optical disks, or magnetic tape.
[0116] The term "critical dimension" as used herein includes any critical dimension of a structure (e.g., bottom critical dimension, middle critical dimension, top critical dimension, sidewall angle, grating height, etc.), the critical dimension between any two or more structures (e.g., the distance between two structures), and the 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, etc.
[0117] As used herein, the terms "critical dimension application" or "critical dimension measurement application" include any critical dimension measurement.
[0118] The term "metrology system" as used herein includes any system employed at least in part to characterize a specimen in any manner, including metrology applications such as critical dimension metrology, overlay metrology, focus / dose metrology, and composition metrology. However, such terminology does not limit the scope of the term "metrology system" as used herein. Additionally, the system 100 may be configured for measurements of patterned and / or unpatterned wafers. The metrology system may 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 (including data from one or more platforms simultaneously), and any other metrology or inspection tool that benefits from calibration of system parameters based on critical dimension data.
[0119] Described herein are various embodiments of a semiconductor metrology system that may be used to measure specimens within any semiconductor processing tool (e.g., an inspection system or a lithography system). The term "specimen" is used herein to refer to a wafer, a reticle, or any other sample that may be processed (e.g., printed or inspected for defects) by means known in the art.
[0120] The term "wafer" as used herein generally refers to a substrate formed of a semiconductor or non-semiconductor material. Examples include, but are not limited to, monocrystalline silicon, gallium arsenide, and indium phosphide. Such substrates may be commonly found and / or processed in semiconductor manufacturing facilities. In some cases, a wafer may include only a substrate (i.e., a bare wafer). Alternatively, a wafer may include 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 include multiple dies having repeatable pattern features.
[0121] A "reticle" may be a reticle at any stage of a reticle manufacturing process or a finished reticle that may or may not have been 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 include, for example, a glass material such as amorphous SiO2. During an exposure step of a lithography process, the reticle may be placed over a resist-covered wafer such that the pattern on the reticle is transferred to the resist.
[0122] 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 on which any type of device known in the art is manufactured.
[0123] 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 or transmitted as one or more instructions or code on a computer-readable medium. Computer-readable media includes both computer storage media and communication media, including any medium that facilitates transfer of a computer program from one place to another. Storage media may be any available medium 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 medium that can be used to hold or store desired program code means 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 the strict sense, any connection is also referred to as a computer-readable medium. For example, if the software is transmitted from a website, server, or other remote source using coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared, radio, microwave, etc., the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, microwave, etc., are included within the definition of media. As used herein, disk and disc include compact discs (CDs), laser discs, optical discs, digital versatile discs (DVDs), floppy disks, and Blu-ray discs, where disks typically reproduce data magnetically and discs reproduce data optically using lasers. Combinations of the above should also be included within the scope of computer-readable media.
[0124] Although certain specific embodiments are described above for purposes of illustration, the teachings of this patent document have general applicability and are not limited to the specific embodiments described above. Thus, various modifications, adaptations, and combinations of the various features of the described embodiments can be made without departing from the scope of the invention as set forth in the claims.
Claims
1. an illumination source configured to provide an amount of illumination radiation toward a first measurement site on a surface of the semiconductor wafer; a detector configured to detect an amount of radiation collected from the first measurement site in response to the amount of illumination radiation. a metrology subsystem comprising: generating a first plurality of characteristic values characterizing the amount of radiation detected from the first measurement site; estimating values of one or more object parameters characterizing the first measurement location from the generated values of the first plurality of features of the detected radiation based on a trained measurement model; determining a reconstructed value for each of the first plurality of features, where the reconstructed value for a feature of the first plurality of features is determined from the estimated values of the one or more subject parameters and the generated values for each of the first plurality of features excluding the generated value for the feature of the first plurality of features based on a trained data reconstruction model; estimating reconstructed values of the one or more object parameters characterizing the first measurement site from the reconstructed values of the first plurality of features based on the trained measurement model; and determining a value of a quality indicator based on a difference between estimated values of the one or more target parameters determined from the generated values of the first plurality of features and reconstructed values of the one or more target parameters determined from the reconstructed values of the first plurality of features; A computing system configured to: A system comprising:
2. The system of claim 1 , wherein the computing system is further configured to normalize the value of the quality index.
3. 2. The system of claim 1, wherein the computing system comprises: The system is further configured to train one or more parameters of the data reconstruction model by minimizing a difference between values of a second plurality of features of measurements of a second measurement site different from the first measurement site and reconstructed values of the second plurality of features, wherein the reconstructed value of a feature of the second plurality of features is determined from values of the one or more target parameters characterizing the second measurement site and the values of the second plurality of features excluding the feature.
4. The system of claim 3 , wherein the second plurality of features of the measurements of the second measurement site are simulated.
5. 4. The system of claim 3, wherein the values of the one or more target parameters characterizing the second measurement site are estimated from values of the second plurality of features based on the trained measurement model.
6. The system of claim 3 , wherein the values of the one or more target parameters characterizing the second measurement site are known reference values.
7. The system according to claim 1 , wherein the first plurality of characteristic quantities are a plurality of detector signals measured in response to the amount of radiation detected from the first measurement site.
8. 2. The system of claim 1 , wherein generating the first plurality of feature values characterizing the amount of radiation detected from the first measurement site includes reducing a dimension of a plurality of detector signals measured in response to the amount of radiation detected from the first measurement site.
9. 9. The system of claim 8, wherein the reduction of the dimensionality of the plurality of detector signals comprises any of principal component analysis, Fourier analysis, wavelet analysis, and discrete cosine transform analysis.
10. 10. The system of claim 1, wherein the illumination source and the detector are configured as part of one of a spectroscopic ellipsometer, a spectroscopic reflectometer, a soft x-ray reflectometer, a small angle x-ray scatterometer, an imaging system, and a hyperspectral imaging system.
11. 2. The system of claim 1, wherein the trained measurement model is one of a trained defect classification model and a trained metrology model.
12. 2. The system of claim 1, wherein the one or more parameters of interest include any of a geometric parameter of interest, a process parameter of interest, an electrical parameter of interest, and a dispersion parameter of interest.
13. 2. The system of claim 1, wherein the one or more parameters of interest include any of an overlay dimension, a critical dimension, a lithography focus, and a lithography dose.
14. providing an amount of illumination radiation toward a first measurement location on a surface of the semiconductor wafer; detecting an amount of radiation collected from the first measurement site in response to the amount of illumination radiation; generating a first plurality of characteristic values characterizing the amount of radiation detected from the first measurement site; estimating values of one or more object parameters characterizing the first measurement location from the generated values of the first plurality of features of the detected radiation based on a trained measurement model; determining a reconstructed value for each of the first plurality of features, wherein a reconstructed value for a feature of the first plurality of features is determined from the estimated values of the one or more target parameters and the generated values for each of the first plurality of features excluding the generated value for the feature of the first plurality of features based on a trained data reconstruction model; estimating reconstructed values of the one or more object parameters characterizing the first measurement site from the reconstructed values of the first plurality of features based on the trained measurement model; determining a value of a quality indicator based on a difference between the estimated values of the one or more target parameters determined from the generated values of the first plurality of features and the reconstructed values of the one or more target parameters determined from the reconstructed values of the first plurality of features; The method includes:
15. 15. The method of claim 14, The method further comprising normalizing the value of the quality index.
16. 15. The method of claim 14, The method further includes a step of training one or more parameters of the data reconstruction model by minimizing a difference between values of a second plurality of features of measurements of a second measurement site different from the first measurement site and reconstructed values of the second plurality of features, wherein the reconstructed value of a feature of the second plurality of features is determined from values of the one or more target parameters characterizing the second measurement site and the values of the second plurality of features excluding the feature.
17. 15. The method of claim 14, wherein the step of generating a first plurality of feature values characterizing the amount of radiation detected from the first measurement site comprises reducing a dimension of a plurality of detector signals measured in response to the amount of radiation detected from the first measurement site.
18. 20. The method of claim 17, wherein the reduction of the dimensionality of the plurality of detector signals comprises any of principal component analysis, Fourier analysis, wavelet analysis, and discrete cosine transform analysis.
19. an illumination source configured to provide an amount of illumination radiation toward a first measurement site on a surface of the semiconductor wafer; a detector configured to detect an amount of radiation collected from the first measurement site in response to the amount of illumination radiation; When executed by one or more processors, the one or more processors are generating a first plurality of characteristic values characterizing the amount of radiation detected from the first measurement site; estimating values of one or more object parameters characterizing the first measurement location from the generated values of the first plurality of features of the detected radiation based on a trained measurement model; determining a reconstructed value for each of the first plurality of features, where the reconstructed value for a feature of the first plurality of features is determined from the estimated values of the one or more subject parameters and the generated values for each of the first plurality of features excluding the generated value for the feature of the first plurality of features based on a trained data reconstruction model; estimating reconstructed values of the one or more object parameters characterizing the first measurement site from the reconstructed values of the first plurality of features based on the trained measurement model; and determining a value of a quality indicator based on a difference between the estimated values of the one or more target parameters determined from the generated values of the first plurality of features and the reconstructed values of the one or more target parameters determined from the reconstructed values of the first plurality of features; normalizing the value of the quality index; A non-transitory computer-readable medium storing instructions for causing A system comprising:
20. 20. The system of claim 19, wherein the non-transitory computer readable medium, when executed by the one or more processors, causes the one or more processors to: training one or more parameters of the data reconstruction model by minimizing a difference between values of a second plurality of features of measurements of a second measurement site different from the first measurement site and reconstructed values of the second plurality of features, wherein the reconstructed value of a feature of the second plurality of features is determined from values of the one or more target parameters characterizing the second measurement site and the values of the second plurality of features excluding the feature. The system further stores instructions to cause the
Citation Information
Patent Citations
Method for evaluating control strategies in a semicondcutor manufacturing process
EP3547031A1
Method and system for providing quality measurement values for improved process control
JP2014512101A
Systems and methods incorporating neural networks and forward physical models in semiconductor applications
JP2019525450A
Systems and methods for fabricating semiconductor device structures
US20150033201A1
Method and Inspection Apparatus and Computer Program Product for Assessing a Quality of Reconstruction of a Value of a Parameter of Interest of a Structure
US20160154319A1
