Method and device for training semiconductor structure prediction model and method and device for measuring optical critical dimension

Abstract

One or more embodiments of the invention provide a training method and device of a semiconductor structure prediction model and an optical critical dimension measurement method and device. The training method comprises the steps that an original training sample is acquired, the original training sample comprises an original spectral value and multiple sets of target parameters corresponding to a semiconductor structure, and the following steps are executed circularly until conditions are met: the original training sample is adopted as a training sample used for first-round training to train a structure prediction model; after training is completed, whether the precision of the structure prediction model reaches the standard or not is judged; under the condition that the target parameters do not reach the standard, the prediction effect of the current round of each group of target parameters is determined; for each group of target parameters, under the condition that the prediction effect of the current round is superior to the prediction effect of the previous round, updating model weights corresponding to the target parameters; determining an incremental training sample used in the next round of training; and determining the original training sample and the incremental training sample as training samples used in the next round of training to continue to train the structure prediction model.

Description

Technical Field

[0001] One or more embodiments of this specification relate to the field of semiconductor manufacturing, and more particularly, to a method and device for training a semiconductor structure prediction model and a method and device for measuring optical critical dimensions. Background Art

[0002] Optical Critical Dimension (OCD) is a technology used to measure and control the dimensions of microstructures, widely used in fields such as semiconductor manufacturing. OCD can predict target parameters of a sample (such as the semiconductor structure under test) by measuring its spectral data.

[0003] The target parameters of the structure of a sample to be tested (e.g., a semiconductor to be tested) typically encompass multiple dimensions. The structural prediction model training methods offered by related technologies struggle to simultaneously address the accuracy of target parameters across these various dimensions. To achieve appropriate accuracy for each target parameter dimension, it is typically necessary to train these parameters separately. This involves constructing multiple structural prediction models, one for each target parameter dimension, and then training each separately. However, this approach often consumes significantly more training time and computing resources. Summary of the Invention

[0004] In view of this, one or more embodiments of this specification provide the following technical solutions: According to a first aspect of one or more embodiments of this specification, a method for training a semiconductor structure prediction model is provided. The structure prediction model is used to predict target parameters of multiple groups of non-identical semiconductor structures, each group of target parameters corresponding to a group of model weights. The method includes: Obtaining original training samples, the original training samples including original spectral values ​​and multiple sets of target parameters corresponding to semiconductor structures, wherein the original spectral values ​​are sample features, and the multiple sets of target parameters are corresponding multiple sets of non-completely identical sample labels; Loop through the following steps until the condition is met: Using the original training samples as training samples for the first round of training to train the structure prediction model; After the training is completed, determining whether the accuracy of the structure prediction model meets the requirements; When the accuracy of the structural prediction model does not meet the requirements, determining the prediction effect of each group of target parameters in this round respectively; For each set of target parameters, if the prediction results of this round are better than those of the previous round, the model weights corresponding to the target parameters are updated; Determining incremental training samples for the next round of training, wherein the incremental training samples include incremental spectral values ​​and corresponding sets of target parameters; The original training samples and the incremental training samples are determined as training samples for the next round of training to continue training the structure prediction model.

[0005] Optionally, also include: When the prediction results of the current round are not better than those of the previous round, the model weights corresponding to the target parameters are retained.

[0006] Optionally, the process of determining whether the accuracy of the structure prediction model meets the requirements includes: Calculating a training score for the structure prediction model; When the training score reaches a training score threshold, testing the structure prediction model using a test sample to obtain a test score; When the test score reaches a test score threshold, determining that the accuracy of the structure prediction model meets the standard; When the test score does not reach the test score threshold, it is determined that the accuracy of the structure prediction model does not meet the standard.

[0007] Optionally, the training score or the test score is obtained based on the weighted prediction effect of each group of target parameters.

[0008] Optionally, the condition includes: the accuracy of the structure prediction model meets the standard or reaches a preset training round.

[0009] According to a second aspect of one or more embodiments of this specification, a method for measuring an optical critical dimension is provided, the method comprising: Obtaining a measured spectrum value of the semiconductor to be measured; Inputting the measured spectrum value into a structure prediction model with satisfactory accuracy to obtain a target parameter prediction result of the semiconductor structure to be measured output by the structure prediction model; Wherein, the structure prediction model is trained using the aforementioned method.

[0010] According to a third aspect of one or more embodiments of this specification, a training device for a semiconductor structure prediction model is provided. The structure prediction model is used to predict target parameters of multiple groups of non-identical semiconductor structures, each group of target parameters corresponding to a set of model weights. The device includes: A sample acquisition unit is configured to acquire an original training sample, wherein the original training sample includes an original spectral value and multiple sets of target parameters corresponding to the semiconductor structure, wherein the original spectral value is a sample feature, and the multiple sets of target parameters are corresponding multiple sets of non-identical sample labels; The model training unit loops through the following steps until the conditions are met: Using the original training samples as training samples for the first round of training to train the structure prediction model; After the training is completed, determining whether the accuracy of the structure prediction model meets the requirements; When the accuracy of the structural prediction model does not meet the requirements, determining the prediction effect of each group of target parameters in this round respectively; For each set of target parameters, if the prediction results of this round are better than those of the previous round, the model weights corresponding to the target parameters are updated; Determining incremental training samples for the next round of training, wherein the incremental training samples include incremental spectral values ​​and corresponding sets of target parameters; The original training samples and the incremental training samples are determined as training samples for the next round of training to continue training the structure prediction model.

[0011] According to a fourth aspect of one or more embodiments of this specification, an optical critical dimension measurement device is provided, the device comprising: A spectrum value acquisition unit, which acquires a measured spectrum value of the semiconductor to be measured; a structure prediction unit, inputting the measured spectrum value into a structure prediction model with satisfactory accuracy, and obtaining a target parameter prediction result of the semiconductor structure to be measured outputted by the structure prediction model; Wherein, the structure prediction model is trained using the aforementioned method.

[0012] According to a fifth aspect of one or more embodiments of this specification, an electronic device is proposed, comprising: a processor; a memory for storing processor-executable instructions; wherein the processor implements the steps of the aforementioned method by running the executable instructions.

[0013] According to a sixth aspect of one or more embodiments of this specification, a computer-readable storage medium is provided, on which computer instructions are stored. When the instructions are executed by a processor, the steps of the aforementioned method are implemented.

[0014] According to a seventh aspect of one or more embodiments of this specification, a computer program product is proposed, comprising a computer program / instruction, which implements the steps of the aforementioned method when executed by a processor.

[0015] As can be seen from the above embodiments, using the above implementation method, after a round of training of the structural prediction model, based on the prediction results of the target parameters of different dimensions, the corresponding model weights can be updated if the current prediction results are better than the previous prediction results. If the original prediction results are not better than the previous prediction results, the corresponding model weights can be retained, that is, the corresponding model weights are not updated, thereby achieving simultaneous optimization of the prediction accuracy of multi-dimensional target parameters. The purpose of separate training of multi-dimensional parameters can be achieved by training a single structural prediction model, and the training time remains almost unchanged. Compared with the method of building multiple models and training them separately, training time and computing resources are greatly saved. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a schematic diagram of a grating structure provided by an exemplary embodiment.

[0017] Figure 2 This is a schematic diagram of another grating structure provided by an exemplary embodiment.

[0018] Figure 3 This is a schematic diagram of another semiconductor structure to be tested provided by an exemplary embodiment.

[0019] Figure 4 This is a flowchart of a method for training a semiconductor structure prediction model provided by an exemplary embodiment.

[0020] Figure 5 This is a flowchart of another method for training a semiconductor structure prediction model provided by an exemplary embodiment.

[0021] Figure 6 It is a structural diagram of a device provided by an exemplary embodiment.

[0022] Figure 7 It is a block diagram of a training device for a semiconductor structure prediction model provided by an exemplary embodiment.

[0023] Figure 8 1 is a block diagram of an optical critical dimension measuring apparatus provided by an exemplary embodiment. DETAILED DESCRIPTION

[0024] Optical Critical Dimension (OCD) is a technology used to measure and control the dimensions of microstructures, widely used in fields such as semiconductor manufacturing. The general principle of OCD measurement technology can be described as follows: first, a theoretical spectral database corresponding to the sample's topographic model is established. The scattering signal (measured spectrum) of the periodic structure of the specific measured area of ​​the sample is then matched with the parameters in the theoretical spectral database to estimate the sample's specific topographic parameters.

[0025] When establishing a theoretical spectral database, the algorithm used to calculate the theoretical spectral data is generally rigorous coupled wave analysis (RCWA). The RCWA algorithm substitutes the dielectric function of the sample and the Fourier series expansion of the electromagnetic field in the sample region into Maxwell's equations. Using the continuity condition of the electromagnetic field, the electric field in the incident region is solved to obtain the sample's reflection coefficient. The RCWA algorithm is then used to calculate the sample's reflection coefficient at each wavelength, yielding the sample's theoretical spectral data.

[0026] Modern OCD systems often incorporate machine learning and / or deep learning methods, training models with large amounts of real-world data to improve the efficiency and accuracy of predicting target parameters of sample structures based on measured spectral data. However, this approach is only suitable for use with a large number of training samples. With fewer training samples, overfitting is prone to occur, reducing the accuracy of target parameter predictions.

[0027] Based on this, an embodiment of the present invention provides a technology for training a model for predicting semiconductor structure target parameters based on measured spectral data, which can solve the problem of inaccurate prediction when the structural parameter prediction model is trained through machine learning technology in the traditional way when there are few training samples.

[0028] Specifically, the OCD measurement technology involved in the embodiments of the present invention can be roughly divided into the following aspects: 1) Establish a theoretical spectral database: Based on process parameters (such as materials and exposure conditions), a morphological model of the semiconductor structure to be measured is constructed. This morphological model includes all possible morphologies within the corresponding range of structural parameters. Theoretical spectral data corresponding to all possible morphologies are then fitted based on the model. A theoretical spectral database is then established by mapping the "possible morphologies" to the "theoretical spectra." The theoretical spectral database is constructed as follows: a grating model is pre-constructed based on the process parameters, and several sets of structural parameter values ​​are determined. Several structural points are then constructed through permutations and combinations, each consisting of one or more structural parameters. The corresponding spectral values ​​are then calculated for each structural point, and a theoretical spectral database is constructed. The quality of the theoretical spectral database construction is a key factor affecting the accuracy of OCD measurements. Therefore, when constructing the theoretical spectral database, it is necessary to accurately construct the structural model of the grating structure and accurately calculate the theoretical spectral values ​​corresponding to each structural point to ensure the reliability and accuracy of the measurement results.

[0029] 2) Get the measured spectrum value: The optical signal scattered or diffracted from the sample surface is obtained by the OCD measurement device and converted into spectral information, which is the measured spectral value of the semiconductor structure to be measured.

[0030] Specifically, OCD measurement systems typically use a broadband light source, such as a halogen lamp or a supercontinuum laser, with a wavelength range covering the ultraviolet to near-infrared (e.g., 200-1000 nm), to provide rich spectral information. During OCD measurement, light emitted by the OCD system's light source passes through a polarizer and is divided into two polarization modes: TE and TM. The electric field direction of the TE mode is perpendicular to the plane of incidence, while the electric field direction of the TM mode is parallel to the plane of incidence. When polarized light strikes the surface of the sample to be measured, the sample's three-dimensional structure (e.g., lines and holes) behaves like a periodic grating, causing the incident light to produce diffraction orders (e.g., 0th order, ±1st order, etc.) at specific angles. The light propagation path at the sample's edges and surface varies due to structural parameters (e.g., height, line width, etc.), thereby affecting the light's phase and intensity distribution. During the measurement process, a detector collects the light signal scattered or diffracted from the sample surface and converts it into spectral information, which is the measured spectral value. In some embodiments, polarized light is also provided to the sample surface at multiple rotation angles for testing to obtain more comprehensive spectral information.

[0031] 3) Perform spectral matching to obtain the target parameters of the actual structure: The measured spectrum is compared and matched with the theoretical spectrum in the theoretical spectrum database. The theoretical spectrum with the highest similarity to the measured spectrum is determined through a matching algorithm (such as a genetic algorithm or a neural network), and its corresponding target parameters are used as the structural parameters of the grating to be measured.

[0032] When performing OCD measurement, the size of the grating structure to be measured on the wafer is usually measured. Figure 1 As shown, Figure 1 This is a schematic diagram of a grating structure provided in this specification. The grating structure is composed of multiple periodically arranged semiconductor structures. In one or more embodiments of this specification, the grating structure of the grating refers to the cross-sectional shape of the periodically arranged semiconductor structures constituting the grating, that is, Figure 1 The cross section of each semiconductor structure is a trapezoid. Of course, in addition to this, the cross section of the semiconductor structure can also be composed of multiple trapezoids, such as Figure 2 As shown, Figure 2 A schematic diagram of another grating structure provided in this specification. Figure 1The figure shows some parameters used to describe the cross-sectional characteristics of the semiconductor structure in the grating, such as top width, bottom width, sidewall angle, height, and depth. These parameters describe the semiconductor structure that forms the periodic grating arrangement. Generally, each semiconductor structure requires a target parameter consisting of a set of permutations and combinations of at least two structural parameters or / and combinatorial operations (e.g., addition, subtraction, multiplication, and division) to describe or determine. In rare cases, all possible sets of a single structural parameter are used as the target parameter to describe or determine a semiconductor structure. These structural parameters are the optical critical dimensions to be measured during the OCD measurement process. The specific type and number of structural parameters to be measured depends on the types of structural parameters contained in the structural points stored in the theoretical spectral database used, and can be selected based on actual needs.

[0033] For example, in one embodiment, a grating structure includes three structural parameters, namely, top width, height, and sidewall angle. If the top width of a specific grating structure is 3nm, the height is 4nm, and the sidewall angle is 53°, the structural parameters of a specific structural point are expressed as [3nm, 4nm, 53°]. Then, based on the determined structural points, the spectral values ​​corresponding to each structural point are calculated, and then the corresponding spectrum is determined based on the determined spectral values, wherein the spectral value is a multidimensional characteristic value used to characterize the corresponding spectral characteristics, that is, the number of spectral dimensions of the spectral value is often very high, for example [0.1533 0.12360.1766 ... 0.4650 0.5579 0.4802]. The multidimensional numerical value as a whole represents a spectral value, which includes 100 spectral dimensions, and each numerical value represents a spectral dimension. The spectrum corresponding to the structural point can be determined based on the multidimensional numerical value. Finally, a theoretical spectral database is constructed based on each structural point and the corresponding spectrum.

[0034] For further information, please refer to Figure 3 , Figure 3 The example of a semiconductor structure to be tested with a minimum period in a grating structure is shown, which includes two trapezoidal columns, wherein the upper trapezoidal column includes structural parameters TCD1, BCD1, HT1, etc., and the lower trapezoidal column includes structural parameters BCD1, BCD2, HT2, etc. Among them, the target parameter Reference A can be Figure 1 The set of permutations and combinations of the specific structural parameters (such as BCD1, BCD2, HT2, etc.) included in the trapezoidal column below is similar to the set of permutations and combinations of {BCD1, BCD2, HT2}, or / and BCD1, BCD2, HT2 combined operations (such as addition, subtraction, multiplication, division, etc.). The target parameter Reference B can be Figure 1The set of permutations and combinations of the specific structural parameters (such as TCD1, BCD1, HT1, etc.) included in the upper trapezoidal column, similar to {TCD1, BCD1, HT1}, or / and the set of combined operations (for example, addition, subtraction, multiplication, division, etc.) of the specific structural parameters TCD1, BCD1, HT1 included in the upper trapezoidal column, etc., can be specifically set based on business needs, and this manual does not impose any special restrictions on this.

[0035] The target parameters of the semiconductor being tested typically encompass multiple dimensions. The structural prediction model training methods offered by related technologies struggle to simultaneously address the accuracy of target parameters across these dimensions. To achieve appropriate accuracy for each dimension, it is typically necessary to train these parameters separately. This involves constructing multiple structural prediction models, one for each dimension, and then training each separately. However, this approach often consumes significantly more training time and computing resources.

[0036] The target parameter dimensions typically refer to combinations or permutations of different structural parameters. For example, they can be different combinations of at least two different structural parameters within a single structural point. For example, if a specific grating structure has a top width of 3nm, a height of 4nm, and a sidewall angle of 53°, the target parameters may include dimensions such as [3nm, 4nm, 53°], [3nm, 53°], [4nm, 53°], and [3nm, 4nm].

[0037] This specification provides a training method for a structure prediction model, which can achieve the purpose of separate training of multi-dimensional target parameters by training a structure prediction model. The training time remains almost unchanged, which greatly saves training time and computing resources compared to the method of building multiple models and training them separately.

[0038] In this specification, the training method of the structure prediction model can be applied in servers, and can also be applied in terminal devices such as PCs, mobile phones, tablet devices, and laptops. This specification does not impose any special restrictions on the application subjects of the training method of the semiconductor structure prediction model.

[0039] In an exemplary embodiment, the structure prediction model can be used to predict multiple sets of non-identical target parameters of the semiconductor structure under test, that is, the structure prediction model can be used to predict target parameters of multiple dimensions of the semiconductor structure, wherein each set of target parameters can correspond to a target parameter of a dimension and a set of model weights, please refer to Figure 4 , the training method of the structure prediction model may include the following steps: Step 402: Obtain original training samples, where the original training samples include original spectral values ​​and multiple sets of target parameters corresponding to semiconductor structures, wherein the original spectral values ​​are sample features, and the multiple sets of target parameters are corresponding multiple sets of non-identical sample labels.

[0040] In this embodiment, the original training sample may include original spectral values ​​and multiple sets of target parameters of the corresponding semiconductor structures, that is, the original training sample includes labels with multiple dimensions.

[0041] For example, the original training sample includes two dimensions of target parameters. The target parameters of these two dimensions can be respectively recorded as Ref1 and Ref2.

[0042] Step 404, looping through the following steps until a condition is met: using the original training samples as the training samples for the first round of training to train the structure prediction model; after the training is completed, determining whether the accuracy of the structure prediction model meets the standard; if the accuracy of the structure prediction model does not meet the standard, determining the prediction effect of this round for each group of target parameters; for each group of target parameters, if the prediction effect of this round is better than the prediction effect of the previous round, updating the model weight corresponding to the target parameter; determining the incremental training samples for the next round of training, the incremental training samples including incremental spectral values ​​and corresponding multiple groups of target parameters; determining the original training samples and the incremental training samples as the training samples for the next round of training to continue training the structure prediction model.

[0043] Based on the aforementioned step 402, after obtaining the original training samples, the initial structure prediction model can be trained using the original training samples. After the training is completed, it can be determined whether the accuracy of the structure prediction model meets the standard, and if the accuracy of the structure prediction model does not meet the standard, the incremental training samples used for the next round of training can be determined. Similar to the original training samples, the incremental training samples include incremental spectral values ​​and corresponding multiple sets of target parameters. Then, the original training samples and the incremental training samples are determined as the training samples for the next round of training to continue training the structure prediction model. After the training is completed, it can be continued to determine whether the model accuracy meets the standard.

[0044] Among them, when judging whether the model accuracy meets the standard or not, it is also possible to first judge whether the training score of the model reaches the training score threshold, and when the training score reaches the training score threshold, the original test sample can be used to test the structural prediction model to obtain the test score, and it is possible to continue to judge whether the evaluation score reaches the test score threshold, etc. The processing and implementation of this part will be described in detail in subsequent embodiments.

[0045] In this embodiment, if the accuracy of the structural prediction model does not meet the requirements, the prediction effect of each set of target parameters in this round can be determined separately. For example, the prediction effect of target parameters Ref1 and Ref2 can be determined separately. The method for evaluating the prediction effect will be described in detail in subsequent embodiments.

[0046] In this embodiment, for a certain set of target parameters, if the prediction effect of this round is better than the prediction effect of the previous round, the model weight corresponding to the target parameter can be updated. For example, assuming that the prediction effect of the target parameter Ref1 in this round is better than the prediction effect of the previous round, the model weight corresponding to the target parameter Ref1 can be updated (for example, the weight corresponding to Ref1 can be changed from Updated to ).

[0047] Among them, the comparison method of the prediction effect of this round and the prediction effect of the previous round can be judged by the test score. If the test score of the structural parameter Ref1 of this round is higher than the test score of the structural parameter Ref1 of the previous round, it can be determined that the prediction effect of this round is better than the prediction effect of the previous round.

[0048] In this embodiment, for a certain set of target parameters, if the prediction effect of the current round is not better than the prediction effect of the previous round, the model weight corresponding to the target parameter is retained, that is, the model weight corresponding to the target parameter is not updated. For example, assuming that the prediction effect of the target parameter Ref2 in the current round is not better than the prediction effect of the previous round, the model weight corresponding to the target parameter is retained (for example, the weight corresponding to Ref2 is retained). ).

[0049] The test score can also be used to compare the prediction results of the current round with those of the previous round. If the test score of the structural parameter Ref2 of the current round is lower than or equal to the test score of the structural parameter Ref2 of the previous round, it can be determined that the prediction results of the current round are not better than those of the previous round.

[0050] In this embodiment, the training score or test score of the structure prediction model can be obtained based on the weighted prediction effect of each dimensional target parameter. The weights used in the weighting can be pre-set. For example, the same weight can be set for the prediction effect of each dimensional target parameter. For another example, different weights can be set for the prediction effect of different dimensional target parameters. Specifically, relatively large weights can be set for important target parameters, and relatively small weights can be set for other target parameters. This specification does not impose any special restrictions on this.

[0051] As can be seen from the above implementation, after a round of training of the structural prediction model, the corresponding model weights can be updated based on the prediction results of the target parameters of different dimensions if the current prediction results are better than the previous prediction results. If the prediction results are not better than the previous prediction results, the corresponding model weights can be retained, that is, the corresponding model weights are not updated, thereby achieving simultaneous optimization of the prediction accuracy of multi-dimensional target parameters. By training a single structural prediction model, the purpose of separate training of multi-dimensional target parameters can be achieved, and the training time is almost unchanged. Compared with the method of building multiple models and training them separately, training time and computing resources are greatly saved.

[0052] The following is a complete introduction to the training method of the semiconductor structure prediction model from two aspects: original sample preprocessing and the overall training process of the structure prediction model.

[0053] 1. Raw sample preprocessing In one exemplary embodiment, before training a semiconductor structure prediction model, raw samples may be obtained. The raw samples include spectral values ​​and corresponding target parameters of the semiconductor structure. The spectral values ​​may be sample features, and the target parameters may be corresponding sample labels.

[0054] After the original samples are obtained, they can be divided into training samples (referred to as original training samples) and testing samples (referred to as original testing samples). The original training samples and the original testing samples can both include original spectral values ​​and corresponding target parameters.

[0055] In one example, the original samples may be divided by clustering the spectral values ​​of the original samples.

[0056] Specifically, the number of first-type cluster centers may be determined first, and the original samples may be clustered using the spectral values ​​of the original samples as clustering objects to obtain the first-type cluster centers.

[0057] The number of the first type of class centers may be pre-set based on the number of original samples.

[0058] Optionally, due to the high dimensionality of the spectral values, before clustering them, the spectral values ​​in the original sample can be reduced in dimensionality. The reduced spectral values ​​can then be clustered to obtain the first type of cluster centers. In this example, dimensionality reduction effectively reduces the spectral values' dimensionality, thereby reducing the cost of subsequent calculations and improving computational efficiency.

[0059] Next, the original sample with the shortest distance to the center of the first type class among the original samples is determined as the original training sample.

[0060] For each first-type class center, the original sample with the shortest distance to the first-type class center among the original samples may be determined as the original training sample.

[0061] For example, taking clustering to obtain 9 first-type class centers as an example, for each first-type class center, an original sample with the shortest distance to the first-type class center can be found in the original samples, and then the original sample with the shortest distance is determined as the original training sample, obtaining 9 original training samples.

[0062] Then, the remaining original samples are determined as original test samples.

[0063] As can be seen from the above description, this embodiment uses a clustering algorithm to cluster the spectral values ​​in the original samples and selects the original samples closest to the center of the first type class as the original training samples. This can make the original training samples obtained by division more representative and more dispersed, ensuring that the original training samples are evenly distributed among the original samples. This eliminates the need for manual screening of representative training data and improves the efficiency of sample data division. In addition, the remaining original samples are designated as original test samples, which can also improve the testing effect of the test set and effectively ensure the accuracy of the subsequent structure prediction model.

[0064] In another example, the original samples may be divided by clustering the target parameters of the original samples.

[0065] Specifically, the number of the second type class centers may be determined first, and the original samples may be clustered using the target parameters of the original samples as clustering objects to obtain the second type class centers.

[0066] The number of the second type of class centers may also be pre-set based on the number of original samples.

[0067] Next, the original sample with the shortest distance to the center of the second type class among the original samples is determined as the original training sample.

[0068] For each second-type class center, the original sample with the shortest distance to the second-type class center among the original samples may be determined as the original training sample.

[0069] For example, taking clustering to obtain 9 second-type class centers as an example, for each second-type class center, an original sample with the shortest distance to the second-type class center can be found in the original samples, and then the original sample with the shortest distance is determined as the original training sample, obtaining 9 original training samples.

[0070] Then, the remaining original samples are determined as original test samples.

[0071] When the original sample corresponds to multiple groups of semiconductor structural parameters, clustering can be performed using one group of structural parameters as a clustering object, and this specification does not impose any special restrictions on this.

[0072] As can be seen from the above description, this embodiment uses a clustering algorithm to cluster the structural parameters in the original samples and selects the original samples closest to the center of the second type class as the original training samples. This can make the divided original training samples representative and also make the original training samples relatively dispersed, ensuring that the original training samples are evenly distributed among the original samples. This eliminates the need for manual screening of representative training data and improves the efficiency of sample data partitioning. In addition, determining the remaining original samples as original test samples can also improve the testing effect of the test set and effectively ensure the accuracy of the subsequent structural prediction model.

[0073] In an exemplary embodiment, before using the original samples to train the structure prediction model, the original samples may be subjected to optimization preprocessing to optimize the internal training time of the subsequent model.

[0074] Taking the original training samples as an example, the process of preprocessing the original training samples to optimize subsequent model internal training time can first determine the sample size of the original training samples and the characteristic dimension of the original spectral values. Then, the relationship between the sample size and the characteristic dimension is compared. If the sample size is greater than the characteristic dimension, the original spectral values ​​can be solved using the characteristic root method to obtain the model input corresponding to the original spectral values.

[0075] If the number of samples is greater than the number of eigendimensionality, indicating a large number of original training samples, the eigenvalue matrix composed of the original sample data can be solved using the eigenroot method to obtain the corresponding eigenvalues ​​and eigenvectors. Each row in the original spectral matrix represents the original spectral value corresponding to an original training sample, and each column represents the original spectral value of a dimension.

[0076] In this embodiment, the score matrix obtained by the characteristic root method can be used as the model input corresponding to the original training sample, so as to replace the original spectral values ​​in the original training sample to input the structure prediction model in the subsequent model training process, so as to train the structure prediction model and realize feature dimensionality reduction, thereby optimizing the training time of the structure prediction model and improving the training efficiency.

[0077] When the number of samples is less than or equal to the feature dimension, a matrix decomposition method may be used to decompose the original spectral values ​​into low-dimensional features as model inputs corresponding to the original spectral values.

[0078] When the number of samples is less than or equal to the feature dimension, it means that the number of original training samples is small. The matrix decomposition method can be used to reduce the dimension of the original training samples, thereby optimizing the training time of the structure prediction model and improving the training efficiency.

[0079] For example, the singular value decomposition (SVD) method can be used to decompose the original spectral value matrix into the product of two low-rank matrices, thereby obtaining a score matrix, which is used to replace the original spectral values ​​in the original training samples and input into the structure prediction model to train the structure prediction model. When using the singular value decomposition method, the left orthogonal matrix can be omitted to further save computing time and memory. In this example, the amount of computation required for the low-dimensional matrix is ​​much smaller than that for the original high-dimensional matrix (i.e., the original spectral value matrix), thereby reducing computational complexity.

[0080] 2. Overall training process of the structure prediction model In an exemplary embodiment, please refer to Figure 5 , the training process of the semiconductor structure prediction model may include the following steps: Step 502: Use the original training samples as training samples for the first round of training to train the structure prediction model.

[0081] In this embodiment, the structure prediction model may be a neural network model, etc. This specification does not impose any special restrictions on the structure of the structure prediction model.

[0082] Step 504: After the training is completed, the training score of the structure prediction model is calculated.

[0083] During the first round of training, if the structure prediction model is trained using the original training samples and the structure prediction model has converged, the training can be confirmed to be complete. Next, the training score of the structure prediction model can be calculated to evaluate the effect of the first round of model training.

[0084] During a non-initial training round, if the structure prediction model is trained using the original training samples and the corresponding incremental training samples, and the structure prediction model has converged, training can be considered complete. A training score for the structure prediction model can then be calculated to evaluate the effectiveness of this round of model training.

[0085] In one example, the training samples (including the original training samples and the incremental training samples) used in the training can be input into a trained structure prediction model to obtain a prediction result (i.e., a predicted target parameter) output by the structure prediction model. The deviation between the predicted target parameter and the sample label (i.e., the target parameter corresponding to the training sample) can then be calculated. The number of first samples whose deviation does not exceed a preset deviation threshold can then be counted, and a training score for the model can be determined based on the first number of samples. The training score and the first number of samples are positively correlated, i.e., the greater the number of first samples whose deviation does not exceed the deviation threshold, the higher the training score of the model, and the smaller the number of first samples whose deviation does not exceed the deviation threshold, the lower the training score of the model.

[0086] If only this method of calculating the bias is used to calculate the training score, it can be directly determined that the training score reaches the training score threshold when the number of first samples reaches more than 80% of the number of training samples, and it can be directly determined that the training score does not reach the training score threshold when the number of first samples is less than 80% of the number of training samples.

[0087] In another example, a Design of Experiments (DoE) approach can be used to analyze the differences between different wafers to determine the training score of the model.

[0088] Specifically, the mean values ​​of the target parameters for each wafer can be sorted based on the sample labels to obtain a label sequence. The sample labels are the structural parameters of the training samples. The label sequence and the predicted sequence can then be compared, item by item. If the two sequences are exactly the same, it indicates that the ranking of the model prediction values ​​is exactly the same as the true values, the model prediction performance is good, and the training score can be determined as a value that reaches the training score threshold. If the two sequences are not exactly the same, it indicates that the ranking of the model prediction values ​​is different from the true values, the model prediction performance is poor, and the training score can be determined as a value that does not reach the training score threshold.

[0089] In another example, the model training score can be calculated by comparing the difference between the maximum and minimum values ​​of the target parameter for each wafer and the floating range between the model prediction value and the sample label.

[0090] Specifically, the label range of the target parameter can be calculated based on the sample label, and a preset floating parameter can be obtained. The corresponding prediction range can be determined based on the label range and the floating parameter. The floating parameter can be a percentage or a fixed value. Then, for the training samples used in the training, it can be determined whether the predicted parameters predicted by the structural prediction model are within the aforementioned prediction range. If the predicted structural parameters are within the prediction range, the prediction of the sample is considered to meet the standard.

[0091] Then, the number of samples with the prediction parameter within the prediction range can be counted as a second sample size, where the second sample size reflects the prediction performance of the structural prediction model. Furthermore, the training score or the test score is determined based on the second sample size, and the training score or the test score is positively correlated with the second sample size. That is, a larger second sample size indicates a higher training score.

[0092] In another example, the training score can also be calculated based on the linear correlation between the sample label and the prediction parameter. Generally speaking, the closer the linear correlation is to 1, the higher the training score is.

[0093] In another example, the training score may also be calculated based on the slope of the linear regression line between the sample label and the prediction parameter. Generally speaking, the closer the slope of the linear regression line is to 1, the higher the training score is.

[0094] In another example, a training score calculation method based on Gauge Repeatability and Reproducibility (GRR) can be used. If 3×Sigma (standard deviation) does not exceed the preset threshold, the model's repeatability and stability are good, and the training score meets the standard. If 3×Sigma (standard deviation) exceeds the preset threshold, the model's variability is large, the repeatability and stability are poor, and the training score fails to meet the standard.

[0095] It should be noted that, in practical applications, one or more of the aforementioned calculation methods may be used to calculate the training score of the structure prediction model. If multiple calculation methods are used, the training scores obtained by each calculation method may be weighted to determine the final training score. Of course, other methods may also be used to determine the training score, and this specification does not impose any particular limitation thereto.

[0096] Step 506 : When the training score reaches the training score threshold, the structure prediction model is tested using the original test sample to obtain a test score.

[0097] Based on the training score calculation result of the aforementioned step 504, it can be determined whether the training score reaches a preset training score threshold. If the training score does not reach the training score threshold, step 512 can be executed. If the training score reaches the training score threshold, it indicates that the training effect of the structure prediction model is good. The structure prediction model can be tested using the original test sample and a test score can be calculated.

[0098] The test score calculation process can refer to the training score calculation process described above. For example, the test score can be calculated using a bias calculation method, a design of experiment (DoE) method, etc., or a combination of multiple calculation methods. This manual will not elaborate on each of these methods here.

[0099] Step 508: If the test score does not reach the test score threshold, determine an incremental training sample.

[0100] Based on the test score calculation result of the aforementioned step 506, it can be determined whether the test score reaches a preset test score threshold. If the test score does not reach the test score threshold, it can indicate that the structure prediction model performs poorly in the test set, and incremental training samples can be determined to continue training the structure prediction model.

[0101] Optionally, a threshold for the number of incremental training iterations can be set. When the threshold is reached, if the test score still does not reach the threshold, training can be terminated and the model with the best training effect from the multiple training runs can be selected as the output. For example, the model with the highest test score from the multiple training runs can be selected as the trained structure prediction model output.

[0102] For example, if the test score does not reach the test score threshold, a determination may be made as to whether the number of incremental training runs has reached the incremental number threshold. If not, the step of determining incremental training samples may be performed. If the incremental number threshold has been reached, training may be terminated, and the model with the highest test score obtained during training may be selected as the trained prediction model and output to the user.

[0103] In step 510 , the original training samples and the incremental training samples are determined as training samples for the next round of training to continue training the structure prediction model. After the training is completed, the process returns to step 504 to calculate the training score.

[0104] In this embodiment, when the test score of the structure prediction model does not reach the test score threshold, it means that the accuracy of the structure prediction model does not meet the standard. The original training samples and the determined incremental training samples can be used to continue training the structure prediction model, and after the training is completed, the process returns to step 504 to calculate the model training score to evaluate the training effect of the model.

[0105] The process of determining the incremental training samples will be described in detail later.

[0106] Step 512 , when the training score does not reach the training score threshold, adjust the regularization factor to continue training the structure prediction model, and after the training is completed, return to step 504 to calculate the training score.

[0107] In this embodiment, if the training score of the structure prediction model does not reach the training score threshold, the regularization factor may be adjusted to continue training the structure prediction model. For example, the regularization factor may be increased when the structure prediction model is overfitting, and the regularization factor may be decreased when the structure prediction model is underfitting.

[0108] In this embodiment, after the structure prediction model is retrained, step 504 may be executed to calculate the training score. If the number of regularization factor adjustments reaches the adjustment threshold and the training score still does not reach the training score threshold, the model training may be terminated.

[0109] Step 514: When the test score reaches the test score threshold, it is determined that the model accuracy meets the standard and the model training ends.

[0110] In this embodiment, when the test score of the structure prediction model reaches the test score threshold, it can be determined that the accuracy of the structure prediction model meets the standard, and then the model training can be ended. Subsequently, the structure prediction model that meets the accuracy standard can be used to predict semiconductor target parameters.

[0111] Specifically, a spectrum value of the semiconductor to be measured can be first obtained, referred to as a measured spectrum value. The measured spectrum value is then input into a structure prediction model with satisfactory accuracy to obtain a target parameter prediction result for the semiconductor structure to be measured, output by the structure prediction model. After obtaining the target parameter prediction result, the target parameter prediction result can be corrected using a determined correction function to obtain the target parameter of the semiconductor to be measured.

[0112] As can be seen from the above description, using the structure prediction model training method provided in this specification, a training score of the model can be calculated after model training is completed. After the training score meets the standard, the model can be tested using the original test samples and a test score can be calculated. If the test score also meets the standard, it can be determined that the accuracy of the structure prediction model meets the standard. Using this automated method to determine whether the model accuracy meets the standard can improve the efficiency of model accuracy determination.

[0113] At the same time, joint training and test scoring can ensure that the structure prediction model not only performs well on the training data, but also maintains good performance on unseen test data. This dual evaluation mechanism can improve the generalization ability of the structure prediction model and make it more reliable in practical applications.

[0114] In addition, when calculating model training scores and test scores, various calculation methods such as deviation calculation and designed experiments are used to comprehensively evaluate the performance of the structural prediction model and improve the accuracy of the model performance evaluation results.

[0115] In one exemplary embodiment, a regression method may be used to search an optical critical dimension model library for standard parameters corresponding to the raw spectral values ​​in the original training samples. Each set of raw spectral values ​​may correspond to a set of standard parameters. The optical critical dimension model library may be pre-established, storing spectral values ​​and their corresponding structural parameters. For ease of distinction, the spectral values ​​stored in the optical critical dimension model library may be referred to as standard spectral values ​​and standard structural parameters.

[0116] Then, parameter boundaries may be determined based on the standard parameters.

[0117] After finding the standard parameters, the parameter boundary of the structure can be determined based on the standard parameters. Generally speaking, the semiconductor structure corresponding to the parameters within the parameter boundary is similar to the semiconductor structure to be tested and can be used to construct incremental training samples.

[0118] In actual situations, the range of the parameter boundary of the structure is related to the floating range of the structural parameter. The floating range of the structural parameter is generally closely related to the specific process parameters and material parameters. Specifically, the critical value of the structural parameter can be determined based on expert experience, the critical value of historical data, the design value of the semiconductor structure, the specific semiconductor process parameters, etc., thereby determining the floating value range of the structural parameter. In most cases, it is less than or equal to the floating range of the structural parameter. In a few cases, the actual process and actual measurement result in the data of a small number of samples exceeding the floating range of the structural parameter. In this embodiment, the range of the structural parameter boundary is constrained to be less than or equal to the floating range of the structural parameter.

[0119] Then, several parameters may be determined as incremental parameters within a range that does not exceed the parameter boundaries.

[0120] For example, since the target parameters in the original sample data are derived from real measurement data and are typically accurate structural parameters, the optical critical dimension model library uses theoretical calculations or model inference to obtain structural parameters and spectral values, such as those constructed using the RCWA algorithm, grid points, or trained models (e.g., neural network models). Therefore, the corresponding standard parameters found in the optical critical dimension model library using the original spectral values ​​may contain errors. A parameter correction function can then be determined based on the target parameters and the corresponding standard parameters. The randomly determined parameters can then be corrected based on the correction function to obtain more accurate incremental parameters.

[0121] Specifically, for ease of distinction, several parameters determined within the parameter boundary range can be called candidate parameters, and then the correction function is used to correct each candidate parameter separately, and the correction result is called incremental parameter, which is used to construct incremental training samples.

[0122] This embodiment can determine a correction function based on target parameters and standard parameters, and subsequently use the correction function to correct the candidate parameters to obtain incremental parameters, which can improve the accuracy and reliability of the incremental parameters and thus improve the accuracy of the subsequent structure prediction model.

[0123] Furthermore, after the incremental parameters are determined, the spectral values ​​corresponding to the incremental structural parameters can be determined as corresponding incremental spectral values, thereby constructing incremental training samples.

[0124] Specifically, the following method can be used to determine the incremental spectrum value corresponding to the incremental parameter: In this embodiment, the theoretical spectrum value corresponding to the incremental parameter may be determined first.

[0125] In one example, the optical critical dimension model library can be searched for spectral values ​​of several parameters adjacent to the incremental parameter (referred to as library spectral values). By searching for adjacent parameters, library spectral values ​​close to the incremental parameter can be obtained. Next, the incremental parameter can be interpolated based on these library spectral values ​​to obtain its corresponding theoretical spectral value. In this example, by searching for library spectral values ​​of adjacent parameters and performing interpolation calculations, a theoretical spectral value for the new parameter is constructed. This effectively utilizes existing spectral data, reduces reliance on experimental measurements, and improves the efficiency and accuracy of spectral calculations.

[0126] In another example, the incremental parameters can be used as model inputs, and a trained spectral value prediction model can be used to predict theoretical spectral values ​​corresponding to the incremental parameters. The spectral value prediction model can be a model structure such as a neural network model, which can be trained based on the parameters and their corresponding spectral values ​​to predict spectral values ​​based on the parameters.

[0127] In another example, the Rigorous Coupled-Wave Analysis (RCWA) algorithm can also be used to calculate the theoretical spectral values ​​corresponding to the incremental parameters. Specifically, a geometric model of the periodic structure can be established based on the incremental parameters, and the theoretical spectral values ​​corresponding to the incremental parameters can be obtained by solving partial differential equations.

[0128] Then, interpolation may be performed between the wavelength points corresponding to the original spectrum values ​​according to the theoretical spectrum values ​​to obtain incremental spectrum values.

[0129] Specifically, interpolation calculation can be performed between the wavelength points corresponding to the original spectrum values ​​to obtain incremental spectrum values. The interpolation is performed in the wavelength dimension and is used to convert the theoretical spectrum values ​​into incremental spectrum values ​​with the same wavelength points as the original spectrum values.

[0130] Among them, when the original sample corresponds to multiple groups of target parameters of the semiconductor structure, when determining the incremental training sample, the incremental training sample can be constructed for each group of target parameters respectively, and this specification does not impose any special restrictions on this.

[0131] It can be seen from this that this embodiment can use the original training samples to construct incremental training samples without the need for experimental measurement, which can improve the efficiency and accuracy of incremental training sample construction.

[0132] Figure 6 This is a schematic structural diagram of a device provided by an exemplary embodiment. Figure 6 At the hardware level, the device includes a processor 602, an internal bus 604, a network interface 606, a memory 608, and a non-volatile memory 610. Of course, it may also include hardware required for other functions. One or more embodiments of this specification can be implemented based on software, such as the processor 602 reading the corresponding computer program from the non-volatile memory 610 into the memory 608 and then running it. Of course, in addition to software implementation, one or more embodiments of this specification do not exclude other implementation methods, such as logic devices or a combination of software and hardware, etc., that is, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0133] Please refer to Figure 7 The semiconductor structure prediction model training device 700 can be applied to Figure 6The device shown in FIG. 1 is used to implement the technical solution of this specification. The semiconductor structure prediction model training device 700 may include: A sample acquisition unit 701 acquires original training samples, wherein the original training samples include original spectral values ​​and multiple sets of target parameters corresponding to semiconductor structures, wherein the original spectral values ​​are sample features, and the multiple sets of target parameters are corresponding multiple sets of non-identical sample labels; The model training unit 702 executes the following steps in a loop until the following conditions are met: Using the original training samples as training samples for the first round of training to train the structure prediction model; After the training is completed, determining whether the accuracy of the structure prediction model meets the requirements; When the accuracy of the structural prediction model does not meet the requirements, determining the prediction effect of each group of target parameters in this round respectively; For each set of target parameters, if the prediction results of this round are better than those of the previous round, the model weights corresponding to the target parameters are updated; Determining incremental training samples for the next round of training, wherein the incremental training samples include incremental spectral values ​​and corresponding multiple sets of incremental structural parameters; The original training samples and the incremental training samples are determined as training samples for the next round of training to continue training the structure prediction model.

[0134] Optionally, when the prediction result of the current round is not better than the prediction result of the previous round, the model training unit 702 retains the model weight corresponding to the target parameter.

[0135] Optionally, the process of determining whether the accuracy of the structure prediction model meets the requirements includes: Calculating a training score for the structure prediction model; When the training score reaches a training score threshold, testing the structure prediction model using a test sample to obtain a test score; When the test score reaches a test score threshold, determining that the accuracy of the structure prediction model meets the standard; When the test score does not reach the test score threshold, it is determined that the accuracy of the structure prediction model does not meet the standard.

[0136] Optionally, the training score or the test score is obtained based on the weighted prediction effect of each group of target parameters.

[0137] Optionally, the condition includes: the accuracy of the structure prediction model meets the standard or reaches a preset training round.

[0138] Please refer to Figure 8The optical critical dimension measuring device 800 can also be used for Figure 5 The optical critical dimension measuring device 800 may include: The spectrum value acquisition unit 801 acquires the measured spectrum value of the semiconductor to be measured; A structure prediction unit 802 inputs the measured spectrum value into a structure prediction model with satisfactory accuracy, and obtains a target parameter prediction result of the semiconductor structure to be measured output by the structure prediction model; The structure prediction model is trained using the aforementioned method in this specification.

[0139] Based on the same concept as the above method, this specification also provides an electronic device, including: a processor; a memory for storing processor-executable instructions; wherein the processor implements the steps of the method described in any of the above embodiments by running the executable instructions.

[0140] Based on the same concept as the above method, this specification also provides a computer-readable storage medium on which computer instructions are stored. When the instructions are executed by a processor, the steps of the method described in any of the above embodiments are implemented.

[0141] Based on the same concept as the above method, this specification also provides a computer program product, including a computer program / instruction, which implements the steps of the method described in any of the above embodiments when executed by a processor.

Claims

1. A method for training a semiconductor structure prediction model, characterized in that: The structure prediction model is used to predict target parameters of multiple groups of semiconductor structures that are not completely identical, each group of target parameters corresponding to a group of model weights, and the method includes: Obtaining original training samples, the original training samples including original spectral values ​​and multiple sets of target parameters corresponding to semiconductor structures, wherein the original spectral values ​​are sample features, and the multiple sets of target parameters are corresponding multiple sets of non-completely identical sample labels; Loop through the following steps until the condition is met: Using the original training samples as training samples for the first round of training to train the structure prediction model; After the training is completed, determining whether the accuracy of the structure prediction model meets the requirements; When the accuracy of the structural prediction model does not meet the requirements, determining the prediction effect of each group of target parameters in this round respectively; For each set of target parameters, if the prediction results of this round are better than those of the previous round, the model weights corresponding to the target parameters are updated; Determining incremental training samples for the next round of training, wherein the incremental training samples include incremental spectral values ​​and corresponding sets of target parameters; The original training samples and the incremental training samples are determined as training samples for the next round of training to continue training the structure prediction model.

2. The method according to claim 1, characterized in that Also includes: When the prediction results of the current round are not better than those of the previous round, the model weights corresponding to the target parameters are retained.

3. The method according to claim 1, characterized in that The process of determining whether the accuracy of the structure prediction model meets the requirements includes: Calculating a training score for the structure prediction model; When the training score reaches a training score threshold, testing the structure prediction model using a test sample to obtain a test score; When the test score reaches a test score threshold, determining that the accuracy of the structure prediction model meets the standard; When the test score does not reach the test score threshold, it is determined that the accuracy of the structure prediction model does not meet the standard.

4. The method according to claim 3, characterized in that The training score or the test score is obtained based on the weighted prediction effect of each group of target parameters.

5. The method according to claim 1, wherein The conditions include: the accuracy of the structure prediction model meets the standard or reaches a preset training round.

6. An optical critical dimension measurement method, characterized in that: The method comprises: Obtaining a measured spectrum value of the semiconductor to be measured; Inputting the measured spectrum value into a structure prediction model with satisfactory accuracy to obtain a target parameter prediction result of the semiconductor structure to be measured output by the structure prediction model; Wherein, the structure prediction model is trained using the method described in any one of claims 1 to 5.

7. A training device for a semiconductor structure prediction model, characterized in that: The structure prediction model is used to predict target parameters of multiple groups of semiconductor structures that are not completely identical, each group of target parameters corresponding to a group of model weights, and the device includes: A sample acquisition unit is configured to acquire an original training sample, wherein the original training sample includes an original spectral value and multiple sets of target parameters corresponding to the semiconductor structure, wherein the original spectral value is a sample feature, and the multiple sets of target parameters are corresponding multiple sets of non-identical sample labels; The model training unit loops through the following steps until the conditions are met: Using the original training samples as training samples for the first round of training to train the structure prediction model; After the training is completed, determining whether the accuracy of the structure prediction model meets the requirements; When the accuracy of the structural prediction model does not meet the requirements, determining the prediction effect of each group of target parameters in this round respectively; For each set of target parameters, if the prediction results of this round are better than those of the previous round, the model weights corresponding to the target parameters are updated; Determining incremental training samples for the next round of training, wherein the incremental training samples include incremental spectral values ​​and corresponding sets of target parameters; The original training samples and the incremental training samples are determined as training samples for the next round of training to continue training the structure prediction model.

8. An optical critical dimension measuring device, characterized in that: The device comprises: A spectrum value acquisition unit, which acquires a measured spectrum value of the semiconductor to be measured; a structure prediction unit, inputting the measured spectrum value into a structure prediction model with satisfactory accuracy, and obtaining a target parameter prediction result of the semiconductor structure to be measured outputted by the structure prediction model; Wherein, the structure prediction model is trained using the method described in any one of claims 1 to 5.

9. An electronic device, characterized in that: include: processor; A memory for storing processor-executable instructions; wherein the processor implements the steps of the method according to any one of claims 1 to 6 by executing the executable instructions.

10. A computer-readable storage medium, characterized in that Computer instructions are stored thereon, and when the instructions are executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

11. A computer program product, characterized in that The method comprises a computer program / instruction, which implements the steps of the method according to any one of claims 1 to 6 when the computer program / instruction is executed by a processor.

Images