System and method for library construction and use for measurement of patterned structures

Abstract

A computer system is presented that is configured and operable as a library builder for extracting one or more parameters of a patterned structure, the one or more parameters being extracted from real-time measurement data acquired on the structure. The system comprises: a data input function component for receiving input data comprising preliminary measurement data acquired from at least a portion of a structure and comprising data indicative of a user-defined quality of measurement (QOR); and a data processor.

Description

Technical Field

[0001] The present invention belongs to the field of model-based automatic measurement of parameters of patterned structures and concerns the construction and use of libraries for interpreting measurement data, which is particularly useful in the semiconductor industry, for example, for controlling manufacturing processes. Background Art

[0002] Characterizing the properties of semiconductor structures is a common goal at all stages in the semiconductor industry. For this reason, and in view of the decreasing size of semiconductor devices based on such structures, highly sensitive metrology tools and precise data analysis are required to monitor the properties of semiconductor structures.

[0003] It is known that optical critical dimension (OCD) measurement techniques (also known as scatterometry) are effective for measuring parameters of patterned (periodic) structures. The measurement data is typically optical data that can be analyzed to derive information about the geometric parameters of the pattern, including thickness, critical dimension (CD), line spacing, line width, wall depth, wall profile, etc., as well as the optical constants of the materials included in the sample. The optical metrology tools used for such measurements are typically tools based on ellipsometry and / or reflectometry. Tools based on reflectometry typically measure changes in the amplitude of radiation returned from / transmitted through a sample, while tools based on ellipsometry typically measure changes in the polarization state of the radiation after interacting with the sample. In addition, or as an alternative to these techniques, angular analysis of light returned (reflected and / or scattered) from a patterned (periodic) structure can be used to measure parameters that define / characterize the structure.

[0004] The data analysis is usually performed using a fitting procedure based on data based on theoretical models and / or reference data, and the structural parameters are derived from the theoretical data that satisfies a "best fit" condition with the measured data. More specifically, this method of measurement data interpretation usually includes a comparison between theoretical data and measured data. The theoretical data are based on one or more optical models, each model being based on various combinations of multiple parameters. The parameters considered in the model are usually of two types, one associated with the structure and the other associated with the measurement technique. If the comparison stage does not provide the desired results, the model parameters of the theoretical data are changed, thereby changing the theoretical input data, and the comparison is repeated until the desired fit is obtained (e.g., convergence to a minimum of the merit function). Summary of the invention

[0005] There is a need in the art for a new approach for measurement data interpretation that allows optimization of the data analysis to meet the specific (customized) requirements of process control applications.

[0006] As mentioned above, theoretical data (including various models or multiple parameter sets for a specific model) are usually provided for a specific type of structure (with certain geometric and material properties). Such theoretical data are usually generated offline, i.e., before and independently of actual measurements on the specific structure, and present a collection (library) of theoretical signals (signatures), each signal (signature) corresponding to data (i.e., parameter values) that can be measured from a specific type of structure under specific measurement conditions. In the case of spectrometry-based OCD measurements in patterned structures (e.g., semiconductor wafers), these can be spectral signatures.

[0007] Most semiconductor process control applications operate at the batch level, where wafers in the batch are measured sequentially between process steps. As geometries shrink and performance and chip density continue to increase, control of various manufacturing processes requires finer levels of control and monitoring, such as wafer-to-wafer (W2W), within-wafer (WIW), and wafer-to-wafer variations of one or more critical parameters (e.g., pattern features).

[0008] Process control in high-end applications poses new challenges. Specifically, manufacturing process stabilization and yield optimization require strict process control and high-precision OCD metrology. Important process parameters that need to be strictly controlled are: repeatability, correlation with reference, tool-to-tool (T2T) variation in measurement, radial trend, decorrelation of parameters, site / layer to site / layer matching, DOE, etc. On the one hand, library search, one of the most commonly used methods to solve the inverse problem in optical scattering measurement, is highly dependent on the grid spacing selected for each parameter in the library for its final measurement accuracy. When the grid spacing decreases, the time cost of parameter extraction increases dramatically.

[0009] The present invention provides a new technique for creating an optimized library, for example, for theoretical data based on scattering measurements. This optimized library on the one hand meets the basic requirement that for any measured spectrum it returns parameter values ​​and a theoretical spectrum, so that the best fit can be used as one of the criteria for determining the result, and on the other hand it has internal degrees of freedom (is flexible) so that the result quality of the available measured data can be used as a supplement (or alternative) to the best fit to obtain the best interpretation solution. This library optimization procedure is performed off-line, and after the optimized library has been created (after optimization of the available data), it can be used to interpret new measured spectra.

[0010] The present technology eliminates the need for manual trial-and-error optimization (eg, library matching), which is critical to the very tight timeframes during the production process, and enables a novel automatic and robust interpretation process to help achieve an accurate solution.

[0011] According to a broad aspect of the present invention, there is provided a computer system configured and operable as a library builder for extracting one or more parameters of a patterned structure, the one or more parameters being extracted from real-time measurement data acquired on the structure, the computer system comprising:

[0012] a data input function component for receiving input data, the input data comprising preliminary measurement data acquired from at least a portion of the structure and including data indicative of a user-defined quality of measurement result (QOR); and

[0013] A data processor is configured and operable to process and analyze input data and predetermined theoretical modeling data corresponding to the measurement data to modify the theoretical modeling data and define optimized theoretical data, so that one or more parameters of the structure that satisfy a first condition of a best fit criterion between the optimized theoretical data and the preliminary measurement data and a second condition of the QOR can be extracted from the optimized theoretical data in response to preliminary measurement data, thereby enabling the library to be further used to interpret real-time measurement data to extract one or more parameters of the structure being measured.

[0014] In some embodiments, the data processor includes a library optimizer functional component that is configured and operable to provide theoretical modeling data that meets a first condition, and to iteratively train and test the theoretical modeling data on preliminary measured data according to one or more indicators determined by data indicating a user-defined QOR, thereby applying a modification procedure to the theoretical modeling data until the one or more indicators meet a predetermined ranking, thereby obtaining optimized theoretical data.

[0015] In some embodiments, the library optimizer utilizes previously prepared and stored initial theoretical modeling data that satisfies the first condition. In some other embodiments, the library optimizer functional component can be configured to generate theoretical modeling data that satisfies the first condition. To this end, the library optimizer is operable to apply an iterative training, testing, and validation procedure to at least one predetermined model using training and testing parameters of the structure.

[0016] The library optimizer functional components may include the following functional components / modules:

[0017] a library estimator with a library convergence criterion on the model data;

[0018] an interpretation engine associated with the library estimator and configured and operable to interpret the input measurement data and operate with the library estimator to perform iterative data interpretation to provide interpretation results that enable identification of theoretical modeling data that matches the input measurement data, wherein the interpretation engine has internal degrees of freedom to modify the interpretation results by interpreting preliminary measurement data and QORs to enable determination of a theoretical data set formed by the matched theoretical data and corresponding theoretical values ​​of the set of structural parameters;

[0019] a metric evaluator functional component configured and operable to analyze a quality of result represented by the set of parameters regarding the QOR and generate data indicative of a corresponding at least one metric;

[0020] A ranking function component is configured and operable to rank the data indicating at least one indicator, and upon identifying that the ranking does not satisfy the second condition, initiate the operation of the interpretation engine and the library estimator using a modified library convergence criterion to perform an iterative data interpretation procedure by modifying the matching theoretical data until the matching theoretical data satisfies the second condition.

[0021] Considering measurements of structures being patterned on a production line (e.g., a semiconductor wafer), data indicative of a user-defined QOR may include one or more of: repeatability of measurements of at least one parameter; correlation of at least one parameter of the structure with a reference, tool-to-tool (T2T) variations in measurements, radial trends, decorrelation of parameters, site / layer to site / layer matching, design of experiments (DOE).

[0022] The library optimizer functional component is configured to generate theoretical modeling data satisfying a first condition by applying an iterative training, testing and validation procedure to at least one predetermined model using training and testing parameters of the structure.

[0023] In some embodiments, the library optimizer functional component includes:

[0024] a library estimator with a library convergence criterion on the model data;

[0025] an interpretation engine associated with the library estimator and configured and operable to interpret the input measurement data and operate with the library estimator to perform iterative data interpretation to provide interpretation results that enable identification of theoretical modeling data that matches the input measurement data, wherein the interpretation engine has internal degrees of freedom to modify the interpretation results by interpreting preliminary measurement data and QORs to enable determination of a theoretical data set formed by the matched theoretical data and corresponding theoretical values ​​of the set of structural parameters;

[0026] a metric evaluator functional component configured and operable to analyze a quality of result represented by the set of parameters regarding the QOR and generate data indicative of a corresponding at least one metric;

[0027] A ranking function component is configured and operable to rank the data indicating at least one indicator, and upon identifying that the ranking does not satisfy the second condition, initiate the operation of the interpretation engine and the library estimator using a modified library convergence criterion to perform an iterative data interpretation procedure by modifying the matching theoretical data until the matching theoretical data satisfies the second condition.

[0028] In some embodiments, the measurement data is optical data. For example, the measurement data includes spectral data.

[0029] In some examples, the data indicative of the user-defined QOR includes, for different theoretical spectra, a degree to which the theoretical modeling data predicts at least one of a geometric and material-related parameter of the structure.

[0030] In some embodiments, the data indicative of the user-defined QOR includes a smoothness of at least one of a geometric and a material-related parameter across the same structure or within several structures.

[0031] According to another broad aspect of the present invention, there is provided a data processing method for extracting one or more parameters of a patterned structure, the one or more parameters being extracted from real-time measurement data acquired on the structure, the method comprising:

[0032] receiving input data comprising preliminary measurement data acquired from at least a portion of a structure and data indicative of a user-defined quality of measurement result (QOR); and

[0033] Process and analyze input data and predetermined theoretical modeling data corresponding to the measurement data to modify the theoretical modeling data and define optimized theoretical data, so that one or more parameters of the structure that satisfy a first condition of the best fit criterion between the optimized theoretical data and the preliminary measurement data and a second condition of the QOR can be extracted from the optimized theoretical data in response to preliminary measurement data, thereby enabling the library to be further used to interpret real-time measurement data to extract one or more parameters of the structure being measured. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to better understand the subject matter disclosed herein and to illustrate how it may be implemented in practice, embodiments will now be described, by way of non-limiting examples only, with reference to the accompanying drawings, in which:

[0035] Figures 1A to 1E The library construction and use techniques known in the art are schematically shown, wherein: Figure 1ALibrary training, testing, and validation phases are shown; Figure 1B The operation of the library and its associated interpretation engine to interpret newly measured data by fitting is schematically illustrated; Figure 1C The principles of library construction using neural networks are illustrated; Figure 1D The principles of regression-based library construction are illustrated; and Figure 1E An example of self-convergence results of spectral fitting during library training / testing according to a conventional library construction method is shown;

[0036] Figure 2 is a block diagram illustrating the main principles of the present invention for the configuration and operation of a library builder;

[0037] Figure 3 is a flow chart illustrating a method performed by the library builder system of the present invention;

[0038] FIG. 4A to FIG. 4C The main operation steps of the library construction according to the present invention are shown, wherein: Figure 4A illustrates the operation of a library estimator and its associated optimized interpretation engine configured in accordance with the present invention to utilize additional data related to user-defined measurement quality and construct an optimized library; Figure 4B The operation of the optimization library for interpreting real-time measurement data and extracting structural parameters is shown; and Figure 4C Some examples of result quality conditions that are typically required in the semiconductor industry are shown;

[0039] Figure 5 Schematically illustrates a solution flow that requires incorporating additional user input requirements while using a library configured according to a conventional approach;

[0040] Figure 6 is a top-level flow chart of the method of the present invention for optimizing library creation and use;

[0041] Figure 7 The best self-convergence of a library achievable using conventional methods (two solutions) is illustrated as a comparison to that achievable with the optimized library according to the present invention which provides a better fitting model for a wider set of constraints. DETAILED DESCRIPTION

[0042] In order to better understand the principle of the present invention, first refer to Figures 1A to 1E , Figures 1A to 1E The known process of constructing and using a library (data based on a theoretical model, e.g., spectral data) to interpret measured data (e.g., spectral data) to estimate structural parameters is schematically illustrated.

[0043] like Figure 1AAs shown, each of one or more predetermined models (eg, RCWA) is trained and tested, and the resulting theoretical spectra are input into a library builder to be further validated to form library data.

[0044] Each model is configured to describe the detectable optical response of a structure having predetermined general characteristics (material and geometry related parameters) under predetermined illumination and detection conditions (measurement conditions). A model (e.g., a rigorous coupled wave analysis (RCWA) model) describes the relationship between the detectable optical response and various parameters of the structure, each parameter varying within a corresponding parameter space / range.

[0045] During the library construction phase, the training parameter set Input into the model and obtain the corresponding theoretical spectrum set Therefore, these theoretical spectra represent the library data (multiple spectra) for a given model. Training parameter set Included are parameters that are relatively highly dependent on changes in one or more other parameters of the measurement conditions and / or structure.

[0046] The library construction also includes applying the same model to a second set of parameters randomly selected from the range of parameters used during training. And determine the corresponding spectral test set The test spectral set is then used in the library validation phase That is, testing whether the theoretical data correctly responds to new input data not used in training

[0047] The library responds to new input parameter sets with corresponding theoretical spectra The validation phase uses a fitting procedure (iterative procedure) which is executed until the best fitting theoretical spectrum is obtained. That is, when the specified error standard is met, the test parameter set provided by the library data Theoretical spectra provided The corresponding set of theoretical spectra determined directly from the model (RCWA) The closest possible match is between: It defines the library convergence criteria. The library so constructed presents theoretical data simulating a given model (eg, RCWA).

[0048] This library and its corresponding interpretation function components (for a given model) can then be used to interpret new measured data (e.g. spectra) to determine the parameters of the structure under investigation. Figure 1B As shown, the measured data Enter the library and interpret the functional components to find the theoretical spectrum with the library data and measurement spectrum The best fitting condition between the theoretical spectra corresponding to, This can be achieved using an iterative algorithm / interpretation engine (e.g., Levenberg-Marquard) and the parameters corresponding to the best fit theoretical spectrum are The value of is determined as a parameter of the structure being measured.

[0049] Typically, a library builder can be any of a neural network, a grid estimator, a simple regression, etc. Figure 1C and Figure 1D The principles of library construction based on neural networks and simple regression are illustrated in a self-explanatory manner, respectively. Before training, the user (or software functional component) defines which model parameters are to be floated and chooses their initial values, and which parameters are to remain fixed during the library construction process and sets nominal values ​​for them. Furthermore, the spectral settings / channels (wavelength selection) are fixed. The purpose of the training process is to find appropriate internal parameters w of the neural network or function. i , the appropriate internal parameters will be each training parameter vector P1, ..., P n optimally transforms into the corresponding spectra S(λ1),…,S(λ n Once training is started, the library builder sets the internal parameters / coefficients (e.g., neuron activation weights) w i . For weight w i Training is performed to give the best library that converges towards the closest model (mimicking RCWA).

[0050] like Figure 1E As shown, it adapts (self-converges) to the minimum error criterion (condition is Model spectrum is not unique, since there are other minima and they represent other "close" models. Therefore, additional information is needed in order to select the best fitting minimum.

[0051] Reference now Figure 2 , Figure 2 The main principles of the present invention for the configuration and operation of the library builder 100 of the present invention are shown in a block diagram. The library builder 100 is configured and operable to build a new library 130. As described above, such a library 130 meets the basic requirements of any measured spectrum because it identifies the best fitting theoretical spectrum and returns the corresponding parameter values. In other words, the best fit can be one of the criteria for determining the result, however, as will be further described below, the library estimator of the present invention provides a single solution for the best fit condition. In addition, the library 130 is also configured with internal degrees of freedom (flexibility) to allow the best interpretation solution to be obtained using a predefined degree of result quality QOR of the measured data.

[0052] Such flexibility is needed in practical process control applications, which usually require that the theoretical data (library) automatically self-adjusts to additional external data (user input) related to the predetermined conditions of the specific application (i.e., the result quality of the structural inspection / measurement). Such specific conditions include one or more of the following: repeatability, correlation with reference, tool-to-tool (T2T) variation in measurement, radial trend, decorrelation of parameters, site / layer to site / layer matching, DOE, etc.

[0053] The library builder 100 is configured as a computer system, and in particular includes functional components such as a data input component 102 and an output component 104, a memory 106, and a data processor 107. The data processor 107 includes a modeling component 108 and a training and testing component 110, which are configured as described above to train and test each of the models and the structural parameters using one or more predetermined models. and Model data (theoretical data) generated in association and

[0054] A novel aspect of the present invention is to replace the traditional library builder with a new library optimizer functional component 103. The library optimizer is configured and can be operated to utilize user input, and this user input includes the data of the requirements / conditions of the indication user to the quality of result (QOR). The library optimizer 103 includes the optimized data interpretation scheme realized by the interpretation engine 114, the index evaluator 116 and the index ranking functional component 118, which operate together to assess whether the best fit theoretical data also meets the predetermined standard of the other index defined by the quality of result (QOR) condition, and if it does not meet, then further adjust the model and re-estimate the library. For this reason, the library optimizer also includes a model adjustment functional component 120. The result of this scheme is to create an optimized library 130 formed by the library estimator 112 and the interpretation engine 114.

[0055] Thus, for a given one or more user-defined result quality conditions, the library 130 is built once offline and can then be used in an online mode to interpret real-time measurement data to determine structural parameters.

[0056] Now refer to Figure 2 and Figure 3 To describe the operation of the library builder 100, wherein, Figure 3 An example is given by Figure 2 Flowchart 200 of operational steps of a method performed by system 100.

[0057] First, input data is provided (step 202). The input data includes at least one model suitable for describing the measurement of a specific structure by a specific measurement scheme / condition (e.g., RCWA), as well as user input related to the quality of the result QOR, and some preliminary measurement data, including N spectra, i.e.

[0058] In this specific non-limiting example, scatterometry measurements are considered and a model of the RCWA type is used. However, it should be understood that the present invention is not limited to any specific type of measurement and any specific model describing a collection of measurement data (ie any measurement scheme).

[0059] The input data is typically stored in memory 106 - step 204. In a well-known manner as described above, the functional components 108 and 110 use the model data to compare the structural parameters. and Generate theoretical data in a correlated manner and - Step 206. This enables preliminary library construction to be performed by operation of the library estimator 112 and the interpreter 114 using the initial library convergence criteria to achieve the best fit criteria with respect to the selected model, as described above (step 206).

[0060] Furthermore, according to the present invention, the optimized library estimator 103 and the interpretation engine 114 are configured to further participate in the library construction phase to enable the library to meet the QOR criteria - step 210. Once constructed, the library 130 can be used to interpret newly measured data (spectra).

[0061] More specifically, the interpretation engine 114 is configured such that it has internal degrees of freedom that allow the interpretation engine 114 to be used by interpreting the measurement data including the QOR (user input) and the preliminary provided measurement data. To this end, the interpretation engine 114 operates with the library estimator 112 to determine the Corresponding / matching theoretical spectrum And thus determine the corresponding theoretical parameter values ​​and Thus forming the corresponding data set ——Step 212.

[0062] The parameter set is evaluated by the indicator evaluator 116. Quality of results represented - step 214. The metric is evaluated with respect to one or more QOR conditions. As indicated above, such a QOR may be associated with one or more of the following conditions to be met / achieved by the measurement (i.e., by the interpretation of the measurement data): repeatability, correlation with a reference, tool-to-tool (T2T) variation in the measurement, radial trend, decorrelation of parameters, site / layer to site / layer matching, DOE, etc. Considering, for example, the repeatability condition, the measurement data includes a plurality of measurement data blocks corresponding to a plurality of measurement sites, and the variation of the results of the values ​​of the structural parameters should be within specified metrics.

[0063] Therefore, the metric is calculated (as further exemplified below) and analyzed by the metric ranking function 118 using a predetermined ranking function (e.g., metric summation) - step 216. If the ranking meets the requirement of a predetermined threshold (usually a minimum value), the corresponding theoretical spectrum determined by the library data (preliminary library estimator function) is The construction of the optimized library 130 is completed with the best fit criteria and the QOR criteria being met, and it can be saved for further use in online measurements - step 218. However, if the metric ranking functional component 118 identifies that the ranking does not meet the threshold requirement / condition (e.g., is above the threshold), the model adjustment functional component 120 is activated to modify the model - step 220. In this regard, it should be understood that the model is generally configured to allow for extrapolation of model data, and this is achieved by the modeling functional component 108 to provide a modified library convergence criterion (step 222) and possibly modified model data (step 224), and the entire process is repeated using the modified data until the above two main criteria are achieved, and the corresponding optimized library 130 is completed and replaces the preliminary library - step 226.

[0064] refer to FIG. 4A to FIG. 4C The technology of the present invention as described in more detail above is summarized. Figure 4A As shown, the library estimator 112 and its associated interpretation engine 114 (after training and testing in the conventional manner to prepare a preliminary / initial library) undergo an optimization procedure together. To this end, the preliminary measurement data And after being interpreted through an iterative fitting process, the measured parameter results are generated Then, the metric evaluator 116 analyzes the data, taking into account the user input (i.e., the QOR conditions defining the desired metrics), and ranks the calculated metrics (e.g., by determining the sum of the metrics), and repeats the iterative procedure by modifying the library convergence criteria until a minimum ranking is obtained. The inventors have shown that this optimization procedure is very fast, i.e., runs for a few minutes each time.

[0065] When the optimization library 130 and its interpretation engine are completed, they can be used in an online automatic measurement / inspection process of a structure to process / interpret real-time measurement data. Figure 4B As shown, the library interpretation engine of the optimization library 130 receives the measurement data (Although the same instructions are used, the measured data are real data rather than preliminary data), and the library is obtained by measuring the parameter results In response, the measured parameter results are obtained by performing a typical fitting procedure when applied to the optimized library data. As such, two criteria are met, namely, the best fit criterion and the user-defined result quality criterion. In addition, as will be further illustrated below, the ambiguity of the solution is improved. Figure 4C Examples of such result quality conditions that are typically required in the semiconductor industry are shown.

[0066] In order to demonstrate the advantages of the present invention, the following should be considered. Figures 1A to 1E The described conventional library creation method does not allow automatic and robust adaptation of the measurement data interpretation to additional user requirements regarding the quality of the results while providing the required accuracy. These additional requirements on the part of the user are dictated by the need for strict process control and high-precision OCD metrology in order to achieve the requirements of manufacturing process stability and yield optimization.

[0067] refer to Figure 5 , Figure 5 The schematic diagram shows a complex, inefficient and therefore impractical solution flow for incorporating additional user input requirements into library adaptation using conventional methods. Figure 1A to Figure 1B Similar to the process scheme described in , using the test parameter set During online use of the training library, the experimental spectra are Input to the training library interpretation engine to obtain the same spectrum as the measured spectrum The corresponding parameter set To account for additional user requirements associated with the desired quality of results, the solution (i.e., parameters) must be checked / validated against various additional metrics, including, for example, global wafer metrics (e.g., tool-to-tool, etc.) and local wafer metrics (e.g., spectral fit, match to reference, etc.) listed in Table 1. ).

[0068] Table 1

[0069]

[0070] By e.g. the sum of The solution ranking is calculated by the sum of all marked (floating) and unmarked (fixed) external references and is a measure of the goodness of the solution. If the solution ranking meets the predetermined specification (usually a minimum value), the output parameter is accepted. as a solution. If the solution ranks outside the specification, the model needs to be revised and the library estimator needs to be rebuilt. However, this has to be done manually, i.e., the goal of this manual approach is to optimize the solution. However, considering the very tight time frame during the manufacturing process of current high-end applications, this trial-and-error library optimization process cannot keep up with the pace.

[0071] The inventors recognized this need in library optimization and developed the novel method described above for performing automatic library optimization offline, which is performed offline and once for a specific model or set of models and specific user input regarding one or more result quality criteria. The result quality is evaluated using user-defined metrics, which are combined into an overall ranking. The optimized library can then be used to interpret new spectra that were not included in the original data used to build the optimized library.

[0072] The above method for optimizing library creation of the present invention is Figure 6 , shown in a self-explanatory manner as a top-level flow chart. As shown, after the training and testing steps, the library estimator is optimized using interpretations of predetermined measurement data prepared according to user input. Such measurement data is shown in the figure as:

[0073]

[0074] Among them, characterizations corresponding to the full wafer map (FWM), correlation with reference (Ref), repeatability (Rep) and tool-to-tool matching (T2T) are provided in the library construction stage. The process is as described above with reference to Figure 2 and Figure 3 The described procedure is continued until an optimized library is created that meets the following two criteria: Best Fit Criteria and the result quality criteria, i.e., the minimum ranking of the required indicators When both criteria (model fit and minimum rank) are achieved, the iterative process is stopped and the optimization library can be used online to interpret new measurement data (spectra).

[0075] It should be noted that the optimization library of the present invention is still a model-based library, and therefore it is not allowed to obtain a bad spectral fit but achieve the condition of minimum ranking, that is, the convergence conditions of spectral fitting can be modified and appropriately relaxed to achieve the minimum ranking, but the optimization library is still a model-based library, and spectral fitting is always required.

[0076] As already mentioned above, building a library using conventional methods often results in several minima, which reflects the fact that several spectra (i.e. several parameter sets) may match the same convergence condition. The present invention also provides improvements in this regard. In this regard, reference is made to Figure 7 As shown in the figure, the best fit criterion result (best self-convergence of the library) provides two minima, i.e., two solutions, and thus there is ambiguity. As described above, further optimization of the library according to the present invention, by considering the overall ranking of the desired indicators, i.e., appropriately modifying the library convergence criteria, shifts the library training to another local minimum, which may represent a better fitting model for a wider set of constraints.

[0077] As already described above, reference Figure 2 and 3 , the optimization library of the present invention uses an optimizable interpretation engine 114, that is, it has internal degrees of freedom that allow the results of its interpretation to be changed. For any block of measurement data (spectrum), it returns the value of the corresponding parameter and the theoretical data (spectrum), and the best fit between the measurement data and the theoretical data is one of the criteria for determining the result (i.e., parameter). However, due to its internal degrees of freedom, in addition to (e.g., during the interpretation of new measurement data) spectral fitting, additional input data related to the required result quality of the available measurement data (via indicators and combined rankings) can be used to obtain the best interpretation solution. After the available data is optimized, such an interpretation engine 114 can be used as part of a library management system 130 to operate with the library estimator 112 to interpret new measurement spectra.

[0078] Refer to above Figure 4A The novel optimizable interpret engine of the present invention is described. A specific example of the operation of the optimizable interpret engine 114 follows.

[0079] The optimized interpretation engine 114 utilizes the theoretical spectrum set This set of theoretical spectra is calculated from the model at random points in parameter space. The corresponding parameter values ​​are estimated as follows:

[0080]

[0081] in, and The theoretical spectrum and the measured spectrum are The corresponding parameters, and is a vector of weights that minimize the index between the selected set of theoretical spectra and the measured spectra.

[0082] Explanation (i.e. weight The selection of N spectra is performed by interpolating the N theoretical spectra (e.g., using equation (1) above). The internal degree of freedom of the interpretation algorithm is the selection of subsets from the original N theoretical spectra used for interpolation, where different subsets of the N spectra can give different interpretation results.

[0083] The above specific examples of the inventive technique of finding a solution, ie a theoretical spectrum and parameters that give the best fit to the measured spectrum, can be formulated in a general (mathematical) form.

[0084] The optimizable interpretation engine of the present invention provides a function for selecting a theoretical spectrum from a family of parameterized functions (e.g., the sum in equation (1) above) They typically use only part of the theoretical spectrum in and ), so that the selected function best predicts the values ​​of the geometrical and / or material parameters of the structure (pattern on the wafer) for a given measured spectrum.

[0085] For example, let us consider h θ (S) is a selected function of the theoretical spectrum. This function can be represented by the weighted sum of N theoretical spectra (selected subsets):

[0086]

[0087] where θ is the selected subset, and According to the condition of formula (1) The selected weights. The mathematical description of the interpretation method is:

[0088] Find θ, from which h θ (S) = P gives the best parameter prediction.

[0089] An appropriate subset of N spectra and the corresponding function h is performed while optimizing the result quality of the available measurement data. θ The selection of (S), as evaluated by a ranking / combination of rankings calculated from parameter results (for all indicators, where the number of indicators can be 1 or more according to user input). The interpretation method can involve minimizing the ranking (loss function) formed by the indicators over the parameter space of the function family.

[0090] The mathematical description can be as follows: By making the loss function L = ∑ i M i (h θ ) is minimized to find the optimal θ, where M i is an indicator (i≥1) based on the results of the user input and is therefore included in the ranking calculation (e.g., the results appearing in Table 1). In the following, some non-limiting examples of indicators are described:

[0091] The first indicator can be the function h θ The degree to which the geometrical / material parameters are correctly predicted given multiple representative theoretical spectra, which are generated from the model by solving Maxwell's equations. Different theoretical spectra have different weights on the indicator, and the weights belong to the set of parameters optimized by the interpretation method. These weights can be continuous or discrete (e.g., 0 and 1). The mathematical description is as follows: The first indicator can be:

[0092]

[0093] Among them, S t is related to the parameter P t The corresponding representative theoretical spectrum set. Weight ρ(S t ) can be continuous or discrete (or even binary) and is optimized together with the parameter θ.

[0094] The second indicator can be the degree of agreement between the predicted and measured geometric / material parameters (reference) for a specific set of measured spectra. The reference parameters can be obtained according to different metrological methods. The mathematical description is as follows: The second indicator can be

[0095]

[0096] Among them, S ref is the same as the available reference parameter P ref The corresponding measured spectrum.

[0097] A third metric may be the smoothness of the predicted geometric / material parameters across one or several wafers.

[0098] A fourth indicator may be a mathematical formula that describes the extent to which the wafer-level signature (spectral signature) of one or several geometric / material parameters has radial characteristics.

[0099] A fifth metric may be the degree to which the repeatability spectra predict similar geometric / material parameter values.

[0100] A sixth metric may be the extent to which spectra from the same structure measured on different tools predict similar geometrical / material parameter values.

[0101] A seventh metric may be how well the predictions of the geometric / material parameters agree with the customer-generated Design of Experiments (DOE).

[0102] An eighth metric may be how stable the predictions of the geometry / material parameters are to process variations, as reflected in measurements on additional wafers.

[0103] A ninth metric may be the degree to which the geometry / material parameter predictions are insensitive to model errors (eg, fixed parameter noise).

[0104] Techniques for selecting functions from a family of parameterized functions can utilize arbitrary choices and any number of indicators with arbitrary sets of weights, depending on the available data.

[0105] In some examples, the optimizable interpretation engine of the present invention can be configured as a linear interpretation engine. The engine uses a set of theoretical spectra calculated from the model at random points in parameter space. And the interpretation is performed by linear interpolation in parameter space and spectral space: Among them, the coefficient w i Depending on the interpreted measured spectrum and theoretical spectrum, for example, the condition of formula (1) is satisfied.

[0106] As described above, once the interpretation engine has been optimized for a specific application (ie, specific user input / QOR defining a metric) based on selected / preliminary measurement data, the interpretation engine can be used as part of a library to interpret new measurement data.

Claims

1. A computer system configured and operable as a library builder for extracting one or more parameters of a patterned structure, the one or more parameters being extracted from real-time measurement data acquired on the structure, the computer system comprising: a data input function component for receiving input data, the input data comprising preliminary measurement data acquired from at least a portion of the structure and including data indicative of a user-defined quality of measurement result (QOR); as well as A data processor is configured and operable to process and analyze the input data and predetermined theoretical modeling data corresponding to the measurement data to modify the theoretical modeling data and define optimized theoretical data, so that one or more parameters of the structure that satisfy a first condition of a best fit criterion between the optimized theoretical data and the preliminary measurement data and a second condition of the QOR can be extracted from the optimized theoretical data in response to the preliminary measurement data, thereby enabling the library to be further used to interpret the real-time measurement data to extract one or more parameters of the structure being measured.

2. The computer system according to claim 1, wherein: The data processor includes a library optimizer functional component, which is configured and operable to provide the theoretical modeling data that meets the first condition, and iteratively train and test the theoretical modeling data on the preliminary measurement data according to one or more indicators determined by the data indicating the user-defined QOR, thereby applying a modification procedure to the theoretical modeling data until the one or more indicators meet a predetermined ranking, thereby obtaining the optimized theoretical data.

3. The computer system according to claim 2, wherein: The library optimizer functional component is configured to generate the theoretical modeling data satisfying the first condition by applying an iterative training, testing and validation procedure to at least one predetermined model using training and testing parameters of the structure.

4. The computer system according to claim 2, wherein: The library optimizer functional components include: a library estimator with a library convergence criterion on the model data; an interpretation engine associated with the library estimator and configured and operable to interpret the input measurement data and operate with the library estimator to perform iterative data interpretation to provide interpretation results, the interpretation results enabling identification of theoretical modeling data that match the input measurement data, wherein the interpretation engine has internal degrees of freedom to modify the interpretation results by interpreting the preliminary measurement data and the QOR, enabling determination of a theoretical data set formed by the matched theoretical data and corresponding theoretical values ​​of a set of structural parameters; a metric evaluator functional component configured and operable to analyze a quality of result represented by said set of parameters regarding said QOR and generate data indicative of a corresponding at least one metric; A ranking function component is configured and operable to rank the data indicating the at least one indicator, and upon identifying that the ranking does not satisfy the second condition, initiate the operation of the interpretation engine and the library estimator using a modified library convergence criterion to perform an iterative data interpretation procedure by modifying the matching theoretical data until the matching theoretical data satisfies the second condition.

5. The computer system according to claim 1, wherein: The data indicative of the user-defined QOR includes at least one of: repeatability of measurement of at least one parameter; correlation of at least one parameter of the structure with a reference, tool-to-tool (T2T) variations in measurements, radial trends, decorrelation of parameters, site / layer to site / layer matching, design of experiments (DOE).

6. The computer system according to claim 1, wherein: The measured data are optical data.

7. The computer system according to claim 6, wherein: The measurement data includes spectral data.

8. The computer system according to claim 7, wherein: The data indicative of the user-defined QOR includes, for different theoretical spectra, an extent to which the theoretical modeling data predicts at least one of a geometric and material-related parameter of the structure.

9. The computer system according to claim 1, wherein: The data indicative of the user-defined QOR comprises a smoothness of at least one of a geometric and a material-related parameter across the same structure or within several structures.

10. A computer system configured and operable as a library builder for extracting one or more parameters of a patterned structure, the one or more parameters being extracted from real-time measurement data acquired on the structure, the computer system comprising: a data input function component for receiving input data, the input data comprising preliminary measurement data acquired from at least a portion of the structure and including data indicative of a user-defined quality of measurement result (QOR); as well as A data processor configured and operable to process and analyze the input data and predetermined theoretical modeling data and define optimized theoretical data, so that one or more parameters of the structure satisfying a first condition of a best fit criterion between the optimized theoretical data and the preliminary measured data and a second condition of the QOR can be extracted from the optimized theoretical data in response to the preliminary measured data, thereby enabling the library to be further used to interpret the real-time measured data to extract one or more parameters of the structure being measured, the data processor comprising: a library optimizer functional component configured and operable to provide the theoretical modeling data satisfying the first condition and apply a modification procedure to the theoretical modeling data, and obtain the optimized theoretical data; a metric evaluator functional component configured and operable to analyze a quality of result represented by said set of parameters regarding said QOR and generate data indicative of a corresponding at least one metric; and A ranking function component is configured and operable to rank the data indicating the at least one indicator and selectively initiate an iterative data interpretation procedure by modifying the matching theoretical data until the matching theoretical data satisfies the second condition.

11. A computer system configured and operable as a library builder for extracting one or more parameters of a patterned structure, the one or more parameters being extracted from real-time measurement data acquired on the structure, the computer system comprising: a data input function component for receiving input data, the input data comprising: preliminary measurement data acquired from at least a portion of a structure; and comprising data indicative of a user-defined quality of measurement result (QOR), the data comprising the degree to which predetermined theoretical model data predicts at least one of a geometric and material-related parameter of the structure for different theoretical spectra; and A data processor is configured and operable to process and analyze the input data and predetermined theoretical modeling data corresponding to the measurement data to modify the theoretical modeling data and define optimized theoretical data, so that one or more parameters of the structure that satisfy a first condition of a best fit criterion between the optimized theoretical data and the preliminary measurement data and a second condition of the QOR can be extracted from the optimized theoretical data in response to the preliminary measurement data, thereby enabling the library to be further used to interpret the real-time measurement data to extract one or more parameters of the structure being measured.

12. A data processing method for extracting one or more parameters of a patterned structure, the one or more parameters being extracted from real-time measurement data acquired on the structure, the method comprising: receiving input data comprising preliminary measurement data acquired from at least a portion of a structure and data indicative of a user-defined quality of measurement result (QOR); as well as Processing and analyzing the input data and predetermined theoretical modeling data corresponding to the measurement data to modify the theoretical modeling data and define optimized theoretical data, so that one or more parameters of the structure that satisfy a first condition of a best fit criterion between the optimized theoretical data and the preliminary measurement data and a second condition of the QOR can be extracted from the optimized theoretical data in response to the preliminary measurement data, thereby enabling the library to be further used to interpret the real-time measurement data to extract one or more parameters of the structure being measured.

13. The method according to claim 12, wherein: The processing includes: providing the theoretical modeling data that satisfies the first condition; and iteratively training and testing the theoretical modeling data on the preliminary measurement data according to one or more indicators determined by the data indicating the user-defined QOR, thereby applying a modification procedure to the theoretical modeling data until the one or more indicators satisfy a predetermined ranking, thereby obtaining the optimized theoretical data.

14. The method according to claim 13, wherein: The providing the theoretical modeling data satisfying the first condition comprises generating the theoretical modeling data satisfying the first condition by applying an iterative training, testing and validation procedure to at least one predetermined model using training and testing parameters of the structure.

15. The method according to claim 13, wherein: Providing the theoretical modeling data satisfying the first condition includes: providing a library estimator having a library convergence criterion with respect to model data; and an interpretation engine associated with the library estimator, the interpretation engine being configured and operable to interpret input measurement data and to operate with the library estimator to perform iterative data interpretation, thereby providing interpretation results, the interpretation results enabling identification of the theoretical modeling data that matches the input measurement data, wherein the interpretation engine has internal degrees of freedom to modify the interpretation results by interpreting the preliminary measurement data and the QOR, enabling determination of a theoretical data set formed by the matched theoretical data and corresponding theoretical values ​​of a set of structural parameters; analyzing a quality of result represented by said set of parameters for said QOR and generating data indicative of a corresponding at least one metric; The data indicating the at least one indicator is ranked, and upon identification that the ranking does not satisfy the second condition, operation of the interpretation engine and the library estimator is initiated using a modified library convergence criterion to perform an iterative data interpretation procedure by modifying the matching theoretical data until the matching theoretical data satisfies the second condition.

16. The method according to claim 12, wherein: The data indicative of the user-defined QOR includes at least one of: repeatability of measurement of at least one parameter; correlation of at least one parameter of the structure with a reference, tool-to-tool (T2T) variations in measurements, radial trends, decorrelation of parameters, site / layer to site / layer matching, design of experiments (DOE).

17. The method according to claim 12, wherein: The measured data are optical data.

18. The method according to claim 17, wherein: The measurement data includes spectral data.

19. The method according to claim 18, wherein: The data indicative of the user-defined QOR includes, for different theoretical spectra, an extent to which the theoretical modeling data predicts at least one of a geometric and material-related parameter of the structure.

20. The method according to claim 12, wherein: The data indicative of the user-defined QOR comprises a smoothness of at least one of a geometric and a material-related parameter across the same structure or within several structures.

Images