Method for analyzing and optimizing focusing sensor optical parameters and silicon wafer film layer process parameters
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- INST OF MICROELECTRONICS CHINESE ACAD OF SCI LTD
- Filing Date
- 2024-12-24
- Publication Date
- 2026-06-26
AI Technical Summary
In existing technologies, the optical models for calculating height-related errors are too complex, making it difficult to perform sensitivity analysis and parameter optimization design using analytical methods. This results in the difficulty in accurately controlling the height measurement error of the focusing sensor under a multi-layer structure.
An error proxy model is constructed using machine learning methods. The model is trained using a training dataset. Combined with a parameter optimization algorithm, the optical parameters of the focusing sensor and the silicon wafer film layer process parameters are optimized to achieve global optimization and obtain the optimal parameter set.
It improves the assessment accuracy and calculation speed of highly process-dependent errors, simplifies the parameter optimization process, and enables efficient adjustment of the focusing sensor optical system and the multi-layer structure of the silicon wafer.
Smart Images
Figure CN119989627B_ABST
Abstract
Description
Technical Field
[0001] This disclosure relates to the field of semiconductor manufacturing technology, and in particular to a method, apparatus, storage medium, and electronic device for analyzing and optimizing optical parameters of a focusing sensor and silicon wafer film layer process parameters. Background Technology
[0002] Focusing sensors, used in semiconductor manufacturing and inspection equipment, are primarily used for measuring the surface height and morphology of silicon wafers. These sensors can obtain the surface height of the silicon wafer online, calculate the surface morphology map, and through further data processing, convert it into adjustments to the vertical direction of the wafer stage. The wafer stage then makes real-time vertical adjustments to ensure that the silicon wafer remains within the depth of focus range and near the optimal focal plane of the exposure or inspection optical system, thereby achieving the most ideal exposure or inspection results.
[0003] The incident light beam from the focusing sensor is reflected off the silicon wafer surface. The focusing sensor receives the reflected beam, performs photoelectric conversion, and calculates the silicon wafer surface height online. Due to the multi-layer structure of the silicon wafer during the photolithography process, the incident beam undergoes multiple reflections and refractions at each layer. This results in an overall phase shift in the reflected beam, directly causing an error in the silicon wafer surface height measurement, known as the height process-dependent error. The factors influencing this error mainly include two aspects: the optical parameters of the focusing sensor and the silicon wafer layer process parameters. The traditional method for calculating this error is based on the multi-layer electromagnetic wave propagation theory. It calculates the characteristic matrices of the incident and reflected light at each layer boundary, ultimately obtaining the overall reflectivity and average phase shift of the multiple layer structures on the silicon wafer surface. Then, complex partial derivative and multiple integral operations are used to obtain the height process-dependent error. However, the optical model for this error is too complex, making subsequent sensitivity analysis and parameter optimization difficult using analytical methods. Summary of the Invention
[0004] The purpose of this disclosure is to provide a method for analyzing and optimizing the optical parameters of a focusing sensor and the process parameters of a silicon wafer film layer, so as to solve the above-mentioned problems existing in the prior art.
[0005] The embodiments of this disclosure adopt the following technical solution: a method for analyzing and optimizing optical parameters of a focusing sensor and process parameters of a silicon wafer film layer, comprising: constructing an error proxy model with optical parameters of a focusing sensor and process parameters of a silicon wafer film layer as input and highly process-related errors as output; constructing a training dataset to train the error proxy model until the model converges; and, according to a preset error target and parameter optimization range, using the error proxy model to globally optimize the optical parameters of the focusing sensor and the process parameters of the silicon wafer film layer to obtain an optimal parameter set that meets the preset error target.
[0006] This disclosure also provides an analysis and optimization apparatus for focusing sensor optical parameters and silicon wafer film layer process parameters, comprising: a surrogate model construction module for constructing an error surrogate model that takes focusing sensor optical parameters and silicon wafer film layer process parameters as input and highly process-related errors as output; a training module for constructing a training dataset to train the error surrogate model until the model converges; and a parameter optimization module for globally optimizing the focusing sensor optical parameters and silicon wafer film layer process parameters using the error surrogate model according to a preset error target and parameter optimization range, to obtain an optimal parameter set that meets the preset error target.
[0007] This disclosure also provides a storage medium storing a computer program, which, when executed by a processor, implements the steps of the analysis and optimization method for the optical parameters of the focusing sensor and the silicon wafer film layer process parameters described in the first embodiment of this disclosure.
[0008] This disclosure also provides an electronic device, including at least a memory and a processor. The memory stores a computer program, and when the processor executes the computer program in the memory, it implements the steps of the method for analyzing and optimizing the optical parameters of the focusing sensor and the silicon wafer film layer process parameters described in the first embodiment of this disclosure.
[0009] The beneficial effects of this disclosure are as follows: an error proxy model based on machine learning with high process-related errors is established, which in principle can approximate the physical model of the focusing sensor and the multi-layer structure of the silicon wafer with high precision. The training and optimization of the error proxy model through a training dataset formed by training with a large amount of simulation or actual test data can improve the accuracy of the error proxy model in evaluating the real situation. At the same time, the optimal parameter design is achieved by combining parameter optimization, without relying on the analytical calculation of the physical model, making the parameter analysis and optimization process more feasible and faster in terms of calculation speed. Attached Figure Description
[0010] To more clearly illustrate the technical solutions in one or more embodiments of this specification or in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0011] Figure 1 This is a flowchart of the method for analyzing and optimizing the optical parameters of the focusing sensor and the silicon wafer film process parameters in the first embodiment of this disclosure;
[0012] Figure 2 This is a schematic diagram of highly process-dependent errors in the first embodiment of this disclosure;
[0013] Figure 3 This is a schematic diagram of the optical design and parameters of the focusing sensor in the first embodiment of this disclosure;
[0014] Figure 4 This is a schematic diagram of the error proxy model in the first embodiment of this disclosure;
[0015] Figure 5 This is a schematic diagram illustrating the variation of highly process-dependent errors with photoresist thickness in the first embodiment of this disclosure;
[0016] Figure 6 This is a schematic diagram illustrating the variation of highly process-dependent error with anti-reflective layer thickness in the first embodiment of this disclosure;
[0017] Figure 7 This is a schematic diagram illustrating the variation of highly process-dependent error with hard mask thickness in the first embodiment of this disclosure;
[0018] Figure 8 This is a schematic diagram illustrating the variation of highly process-dependent error with incident angle in the first embodiment of this disclosure;
[0019] Figure 9 This is a schematic diagram illustrating the variation of highly process-dependent error with photoresist thickness when using different anti-reflection layer thicknesses and hard mask thicknesses in the first embodiment of this disclosure.
[0020] Figure 10 This is a schematic diagram illustrating the variation of highly process-dependent error with photoresist thickness when different optical parameters of the focusing sensor are used in the first embodiment of this disclosure.
[0021] Figure 11 The results of the global sensitivity analysis performed using the variance-based method in the first embodiment of this disclosure;
[0022] Figure 12 This is a schematic diagram of the Pareto front for multi-objective optimization of the optical parameters of the focusing sensor and the silicon wafer film process parameters using a genetic algorithm in the first embodiment of this disclosure.
[0023] Figure 13 This is a schematic diagram of the structure of the device for analyzing and optimizing the optical parameters of the focusing sensor and the process parameters of the silicon wafer film layer in the second embodiment of this disclosure. Detailed Implementation
[0024] To enable those skilled in the art to better understand the technical solutions in one or more embodiments of this specification, the technical solutions in one or more embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this specification, and not all of the embodiments. Based on one or more embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of this document.
[0025] Focusing sensors, used in semiconductor manufacturing and inspection equipment, are primarily used for measuring the surface height and morphology of silicon wafers. These sensors can obtain the surface height of the silicon wafer online, calculate the surface morphology map, and through further data processing, convert it into adjustments to the vertical direction of the wafer stage. The wafer stage then makes real-time vertical adjustments to ensure that the silicon wafer remains within the depth of focus range and near the optimal focal plane of the exposure or inspection optical system, thereby achieving the most ideal exposure or inspection results.
[0026] The incident light beam from the focusing sensor is reflected off the silicon wafer surface. The focusing sensor receives the reflected beam, performs photoelectric conversion, and calculates the silicon wafer surface height online. Due to the multi-layer structure of the silicon wafer during the photolithography process, the incident beam undergoes multiple reflections and refractions at each layer. This results in an overall phase shift in the reflected beam, directly causing an error in the silicon wafer surface height measurement, known as the height process-dependent error. The factors influencing this error mainly include two aspects: the optical parameters of the focusing sensor and the silicon wafer layer process parameters. The traditional method for calculating this error is based on the multi-layer electromagnetic wave propagation theory. It calculates the characteristic matrices of the incident and reflected light at each layer boundary, ultimately obtaining the overall reflectivity and average phase shift of the multiple layer structures on the silicon wafer surface. Then, complex partial derivative and multiple integral operations are used to obtain the height process-dependent error. However, the optical model for this error is too complex, making subsequent sensitivity analysis and parameter optimization difficult using analytical methods.
[0027] To address the aforementioned issues, the first embodiment of this disclosure provides a method for analyzing and optimizing the optical parameters of a focusing sensor and the process parameters of silicon wafer films, which serves as a guide for the design of the optical system of a focusing sensor and the design of multi-film parameters of silicon wafers during semiconductor fabrication. Figure 1 A flowchart of the analysis and optimization method of this embodiment is shown, such as... Figure 1 As shown, the method mainly includes steps S10 to S30:
[0028] S10, construct an error surrogate model that takes the optical parameters of the focusing sensor and the silicon wafer film layer process parameters as inputs and the highly process-dependent error as the output.
[0029] High process-dependent errors arise because the multi-layered structure of the silicon wafer causes multiple reflections and refractions between different layers in the incident and reflected light beams of the focusing sensor. This results in an overall phase shift in the reflected beam, ultimately causing a discrepancy between the measured height of the silicon wafer and the actual height. Figure 2 As shown, the conventional evaluation model for highly process-dependent errors is a complex optical model of the optical parameters of the focus sensor and the silicon wafer film process parameters. Due to the complexity of the physical principles, it requires approximation methods for modeling, which reduces the accuracy of the evaluation model. Furthermore, the optical calculation formula for highly process-dependent errors is complex and cannot explicitly include the optical parameters of the focus sensor and the silicon wafer film process parameters, making it difficult to perform sensitivity analysis and optimization design of the design parameters using analytical methods.
[0030] To address the aforementioned issues, this embodiment establishes an error proxy model for highly process-dependent errors. This error proxy model takes the optical parameters of the focusing sensor and the silicon wafer film layer process parameters as inputs and the highly process-dependent errors as outputs. Based on machine learning, it approximates the optical models of the focusing sensor and the multi-film structure of the silicon wafer with high precision, avoiding the complex analysis of highly process-dependent errors through optical models, solving the problems of wasted computing power and time consumption, and optimizing the calculation time of highly process-dependent errors.
[0031] In this embodiment, the optical module of the focus sensor generates an incident light beam, which is reflected off the surface of the silicon wafer under test. The focus sensor receives the reflected light beam and then calculates the height of the silicon wafer surface based on the intensity of the reflected light beam. The schematic diagram is shown below. Figure 3 As shown. The optical parameters of the focusing sensor mainly include at least one of the following parameters: numerical aperture NA, incident light angle, incident light center wavelength, incident light wavelength bandwidth, incident light spectral distribution, and incident light polarization coefficient; wherein, numerical aperture NA is the product of the refractive index n of the incident light in the transmission medium and the sine of the aperture angle θ. Since the optical path of the focusing sensor's detection light is in air, NA = sin(θ); the incident light angle is the angle between the incident light and the normal to the silicon wafer surface; the focusing sensor uses a broadband light source, and the incident light center wavelength is the median value of the spectral wavelength; the incident light bandwidth is the effective width of the incident light in the above spectrum; the incident light spectral distribution is the distribution function of the incident light intensity with wavelength; the incident light has two polarization states, s-ray and p-ray, and the incident light polarization coefficient is the intensity ratio of these two polarization states in the total light intensity, ranging from [0, 1]. Some or all of the above-mentioned optical parameters can be used as input parameters of the error proxy model. At the same time, the optical parameter information used in this embodiment may include, but is not limited to, the above parameters.
[0032] Silicon wafer processing requires various techniques. Focusing sensors are primarily used in the exposure or inspection stages of silicon wafers. The silicon wafer under test typically has multiple process layers. The process parameters for these layers mainly include at least one of the following: the layer structure of the silicon wafer, the material type of each layer, the refractive index n and extinction coefficient k of each layer, and the layer thickness. The layer structure parameters describe the number of layers on the surface of the silicon wafer under test and the function of each layer. Typical layers include: photoresist layer, top or bottom anti-reflection layer, hard mask layer, and substrate layer. The material type refers to the material model of each layer; different models of the same type of layer may have significantly different optical properties. Some or all of the above-mentioned process parameters can be used as input parameters for an error surrogate model. Furthermore, the process parameter information used in this embodiment may include, but is not limited to, the above parameters.
[0033] The highly process-dependent error is an evaluation value calculated by a surrogate model, given the optical parameters of the focusing sensor and the silicon wafer film layer process parameters as inputs. The error surrogate model in this embodiment is established using machine learning methods, such as... Figure 4 As shown, the machine learning methods include at least one of the following: neural networks, decision trees, random forests, and support vector regression; when using neural network methods, the neural network model constructed can be any one of the following models: multilayer perceptron (MLP), residual neural network (RNN), convolutional neural network (CNN), and deep neural network (DNN), and this embodiment does not impose specific limitations.
[0034] S20: Build a training dataset to train the error surrogate model until the model converges.
[0035] Once the error surrogate model is constructed, it can be trained using a training dataset to reflect the real physical model. Specifically, the training dataset can be established through optical model simulation calculations, experimental data, or a combination of both. The training dataset constructed in this embodiment should at least include the optical parameters of the focusing sensor used in the simulation or experiment, as well as the process parameters of the silicon wafer film. It should also include simulation or experimental errors corresponding to highly process-dependent errors under different optical and process parameter settings. The error surrogate model is then trained using the training dataset until the model converges. Specific convergence conditions can be set according to the amount of data or accuracy requirements; this embodiment does not impose specific limitations.
[0036] In practice, the training dataset can be divided into training samples and test samples. The training samples are used for model training, and the test samples are used to evaluate and test the trained model, and the model is fine-tuned based on the evaluation results.
[0037] S30: Based on the preset error target and parameter optimization range, the optical parameters of the focusing sensor and the silicon wafer film process parameters are globally optimized using an error surrogate model to obtain the optimal parameter set that meets the preset error target.
[0038] Parameter optimization refers to the global optimization of the optical parameters of the focusing sensor and the silicon wafer film layer process parameters under given optimization objectives and constraints, to obtain the optimal parameter set that minimizes the optimization objective. This embodiment implements the above parameter optimization function by calling a parameter optimization model. The inputs are the parameter optimization ranges of the focusing sensor optical parameters and the silicon wafer film layer process parameters, as well as a preset optimization objective for highly process-dependent errors. During the parameter optimization process, an error proxy model is also called to assist in error calculation and verification, ultimately outputting the optimal parameter set. In some embodiments, the preset optimization objective is a set of functions related to highly process-dependent errors, describing the magnitude or rate of change of these errors. The optimal parameter optimization set can provide at least one solution set that satisfies the above optimization objectives and constraints. Operators can select one solution set from multiple sets based on actual needs as the analysis and optimization result for each parameter.
[0039] In some embodiments, the parameter optimization model executes a parameter optimization algorithm to achieve the above optimization process. The parameter optimization algorithm includes at least one of the following methods: genetic algorithm, evolutionary algorithm, and particle swarm optimization algorithm.
[0040] In some embodiments, after training the error surrogate model, a global sensitivity analysis step for the parameters may be included. Specifically, by performing a global sensitivity analysis on all parameters, the influence of each parameter on highly process-related errors can be determined, and the degree of influence can be described by a sensitivity index. After determining the sensitivity indices of all parameters, at least one parameter with a sensitivity index greater than a preset threshold is selected as a key parameter. In subsequent parameter optimization, only the optimization range of the key parameters is set, and the error surrogate model is used to perform global optimization on the key parameters. This ensures that the error calculated from the optimal set of key parameters meets the preset error target. Since the key parameters have a significant impact on highly process-related errors, only these parameters are optimized, while parameters with less influence are ignored, thereby reducing the complexity of the model and algorithm and improving the computational speed.
[0041] Furthermore, after determining the key parameters, the training dataset can be further filtered based on these parameters. For example, only the sample data corresponding to the key parameters can be retained in the training dataset, while data for other parameters with less impact on error can be directly deleted. The filtered training dataset can then be used to optimize the error surrogate model, improving its fit to the key parameters, reducing the amount of data used in training the model, and increasing optimization speed and model accuracy. In practical applications, the above steps of global sensitivity analysis can be implemented by calling a global sensitivity analysis model, which can be constructed based on variance methods or density function methods.
[0042] The following is combined with Figures 5 to 12 One possible implementation flow of the analysis and optimization method in this embodiment will be described.
[0043] The highly process-dependent error analysis and optimization process comprises three parts: an error surrogate model for highly process-dependent errors, a global sensitivity analysis model, and a parameter optimization model. The error surrogate model calculates the highly process-dependent error assessment value using machine learning methods. Inputs are the optical parameters of the focus sensor and the silicon wafer film layer process parameters; output is the highly process-dependent error. The global sensitivity analysis model quantitatively analyzes the influence weights of the focus sensor optical parameters and silicon wafer film layer process parameters on the highly process-dependent error. Inputs are the focus sensor optical parameters, the silicon wafer film layer process parameters, and the highly process-dependent error surrogate model; output is the sensitivity index of these parameters, i.e., the influence weights on the highly process-dependent error. This model uses an error surrogate model to calculate the highly process-dependent error. The parameter optimization model, given the optimization objective and constraints, performs global optimization on the focus sensor optical parameters and silicon wafer film layer process parameters to obtain the optimal parameter set that minimizes the optimization objective. Inputs are the optimization range of the focus sensor optical parameters and silicon wafer film layer process parameters, the preset error optimization objective, and the error surrogate model; output is the optimal parameter set. This model also uses an error surrogate model to calculate the highly process-dependent error.
[0044] To better illustrate the analysis and optimization method of this embodiment, a typical four-layer silicon wafer film structure is used as an example. From top to bottom, the silicon wafer films consist of a photoresist layer, an anti-reflection layer, a hard mask layer, and a substrate layer. The optical parameters of the focusing sensor include: numerical aperture (NA), incident light angle, incident light center wavelength, incident light wavelength bandwidth, incident light spectral distribution, and polarization coefficient. The silicon wafer film process parameters include: the material and thickness of each film layer, and the output parameter is a highly process-dependent error. The machine learning proxy model is established using a deep neural network method.
[0045] Figure 5 , Figure 6 and Figure 7 These are schematic diagrams showing how the height of the process-dependent error varies with the thickness of the photoresist, the anti-reflective layer, and the hard mask, while keeping other parameters constant. It is clear that the silicon wafer process structure parameters have a significant impact on the height of the process-dependent error. Figure 8 This is a schematic diagram showing how the height process-related error changes with the incident angle while keeping other parameters constant. The optical parameters of the focusing sensor will also cause changes in the height process-related error. Figure 9 This is a schematic diagram showing the variation of highly process-dependent error with photoresist thickness when the optical parameters of the focusing sensor are fixed and different anti-reflective layer thicknesses and hard mask thicknesses are used. The diagram illustrates the complex coupling relationship between the three parameters of photoresist thickness, anti-reflective layer thickness, and hard mask thickness for highly process-dependent error. Figure 10 This diagram illustrates how the highly process-dependent error varies with photoresist thickness when different optical parameters of the focusing sensor are used, with fixed silicon wafer film layer process parameters. The diagram shows that for highly process-dependent errors, there are also complex coupling relationships between the optical parameters of the focusing sensor.
[0046] To quantitatively analyze the weight of the influence of focusing sensor optical parameters and silicon wafer film process parameters on highly process-dependent errors, a global sensitivity analysis model was used. Figure 11 This is the result of a global sensitivity analysis using a variance-based method. The figure quantifies the influence weights of all the parameters mentioned above, allowing us to determine their relative magnitudes. Based on the global sensitivity analysis results, we can optimize the design of the error surrogate model and parameter optimization model by ignoring parameters with small influence weights, thereby reducing the complexity of the model and algorithm and improving computational speed.
[0047] For the parameter optimization process, the design optimization objectives are to minimize the absolute value of the high correlation error and the statistical sum of the high correlation errors, respectively. The design constraints are the design range of the focusing sensor optical parameters and the silicon wafer film layer process parameters. The above statistical sum is designed using a weighted sum of the mean and standard deviation of the high correlation errors.
[0048] Figure 12 This diagram illustrates the Pareto front of a genetic algorithm used to perform multi-objective optimization of the optical parameters of the focusing sensor and the silicon wafer film processing parameters. In the diagram, f1 is the absolute value of the highly process-dependent error, and f2 is the weighted sum of the mean and standard deviation. For ease of display, the value of f1 is magnified by a factor of 1000. This diagram represents the set of all optimal solutions that minimize the above optimization objectives under the constraints described. The optimal parameter set can be selected from this set of solutions.
[0049] This embodiment establishes an error surrogate model based on machine learning for highly process-dependent errors. In principle, it can approximate the physical model of the focusing sensor and the multi-layer structure of the silicon wafer with high precision. The training dataset formed by training with a large amount of simulation or actual test data can be used to train and optimize the error surrogate model, thereby improving the accuracy of the error surrogate model in evaluating the real situation. At the same time, it can achieve optimal parameter design by combining parameter optimization, without relying on the analytical calculation of the physical model, making the parameter analysis and optimization process more feasible and faster in terms of calculation speed.
[0050] Based on the same inventive concept, the second embodiment of this disclosure provides an analysis and optimization device for optical parameters of a focusing sensor and process parameters of silicon wafer film layers, the schematic diagram of which is shown below. Figure 13 As shown, it mainly includes: a surrogate model construction module 10, which is used to construct an error surrogate model with the optical parameters of the focusing sensor and the silicon wafer film layer process parameters as inputs and the highly process-related error as output; a training module 20, which is used to construct a training dataset to train the error surrogate model until the model converges; and a parameter optimization module 30, which is used to perform global optimization of the optical parameters of the focusing sensor and the silicon wafer film layer process parameters using the error surrogate model according to the preset error target and parameter optimization range, so as to obtain the optimal parameter set that meets the preset error target.
[0051] Specifically, the optical parameters of the focusing sensor include at least one of the following parameters: numerical aperture, incident light angle, incident light center wavelength, incident light wavelength bandwidth, incident light spectral distribution, and incident light polarization coefficient; the silicon wafer film layer process parameters include at least one of the following parameters: silicon wafer film layer structure, material type of each film layer, refractive index and extinction coefficient of each film layer, and film layer thickness.
[0052] In some embodiments, the device further includes a sensitivity analysis module 40, specifically configured to perform global sensitivity analysis on all parameters, and determine at least one key parameter among all parameters whose sensitivity index is greater than a preset threshold. Furthermore, when the sensitivity analysis module 40 determines the key parameter, the parameter optimization module 30 can perform global optimization of the key parameter using the error surrogate model according to a preset error target and the key parameter optimization range, to obtain an optimal set of key parameters that meets the preset error target. Additionally, when the sensitivity analysis module 40 determines the key parameter, the training module 20 can also perform sample screening on the training dataset according to the key parameter, and optimize the error surrogate model based on the screened training dataset.
[0053] Specifically, the sensitivity analysis module 40 implements its function by calling a global sensitivity analysis model, which is constructed based on the variance method or the density function method. Specifically, the parameter optimization module 30 implements its function by calling a parameter optimization model, which executes a parameter optimization algorithm, including at least one of the following methods: genetic algorithm, evolutionary algorithm, and particle swarm optimization algorithm. Specifically, the error proxy model is established using machine learning methods, including at least one of the following methods: neural network, decision tree, random forest, and support vector regression.
[0054] The specific functions and principles implemented by each functional module in this embodiment have been described in detail in the first embodiment, and will not be repeated here.
[0055] This embodiment establishes an error surrogate model based on machine learning for highly process-dependent errors. In principle, it can approximate the physical model of the focusing sensor and the multi-layer structure of the silicon wafer with high precision. The training dataset formed by training with a large amount of simulation or actual test data can be used to train and optimize the error surrogate model, thereby improving the accuracy of the error surrogate model in evaluating the real situation. At the same time, it can achieve optimal parameter design by combining parameter optimization, without relying on the analytical calculation of the physical model, making the parameter analysis and optimization process more feasible and faster in terms of calculation speed.
[0056] Based on the same inventive concept, the third embodiment of this disclosure provides a storage medium storing a computer program, which, when executed by a processor, implements the steps of the analysis and optimization method for the optical parameters of the focusing sensor and the silicon wafer film layer process parameters described in the first embodiment of this disclosure.
[0057] Based on the same inventive concept, the fourth embodiment of this disclosure provides an electronic device, including at least a memory and a processor. The memory stores a computer program, and when the processor executes the computer program in the memory, it implements the steps of the method for analyzing and optimizing the optical parameters of the focusing sensor and the silicon wafer film layer process parameters described in the first embodiment of this disclosure.
[0058] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this disclosure, and are not intended to limit them. Although this disclosure has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this disclosure.
Claims
1. A method for analyzing and optimizing the optical parameters of a focusing sensor and the process parameters of a silicon wafer film layer, characterized in that, include: An error surrogate model is constructed, which takes the optical parameters of the focusing sensor and the silicon wafer film process parameters as inputs and the highly process-dependent error as output. Construct a training dataset to train the error proxy model until the model converges; Based on the preset error target and parameter optimization range, the error proxy model is used to perform global optimization on the optical parameters of the focusing sensor and the silicon wafer film layer process parameters to obtain the optimal parameter set that meets the preset error target. After training the error proxy model on the constructed training dataset until the model converges, the process further includes: Perform a global sensitivity analysis on all parameters and identify at least one key parameter whose sensitivity index is greater than a preset threshold. Based on the key parameters, the training dataset is sampled, and the error proxy model is optimized based on the selected training dataset. Based on the preset error target and the optimization range of key parameters, the error proxy model is used to perform global optimization of the key parameters to obtain the optimal set of key parameters that meets the preset error target.
2. The analysis and optimization method according to claim 1, characterized in that, The optical parameters of the focusing sensor include at least one of the following parameters: numerical aperture, incident light angle, incident light center wavelength, incident light wavelength bandwidth, incident light spectral distribution, and incident light polarization coefficient. The silicon wafer film layer process parameters include at least one of the following parameters: the film layer structure of the silicon wafer, the material type of each film layer, the refractive index and extinction coefficient of each film layer, and the film layer thickness.
3. The analysis and optimization method according to claim 1, characterized in that, The step of performing global sensitivity analysis on all parameters is achieved by calling a global sensitivity analysis model, which is constructed based on the variance method or the density function method.
4. The analysis and optimization method according to claim 1, characterized in that, The step of using the error proxy model to globally optimize the optical parameters of the focusing sensor and the silicon wafer film layer process parameters according to the preset error target and parameter optimization range, and obtaining the optimal parameter set that meets the preset error target, is implemented by calling the parameter optimization model. The parameter optimization model executes a parameter optimization algorithm, which includes at least one of the following methods: genetic algorithm, evolutionary algorithm, and particle swarm optimization algorithm.
5. The analysis and optimization method according to any one of claims 1 to 4, characterized in that, The error proxy model is established using machine learning methods, which include at least one of the following: neural networks, decision trees, random forests, and support vector regression.
6. A device for analyzing and optimizing optical parameters of a focusing sensor and process parameters of silicon wafer film layers, characterized in that, include: The proxy model building module is used to build an error proxy model that takes the optical parameters of the focusing sensor and the silicon wafer film layer process parameters as inputs and the highly process-dependent error as the output. The training module is used to construct a training dataset to train the error proxy model until the model converges; The parameter optimization module is used to perform global optimization of the optical parameters of the focusing sensor and the silicon wafer film layer process parameters based on a preset error target and parameter optimization range, using the error surrogate model, to obtain the optimal parameter set that meets the preset error target. The sensitivity analysis module is used to perform global sensitivity analysis on all parameters and identify at least one key parameter whose sensitivity index is greater than a preset threshold. The training module is also used to filter samples in the training dataset according to the key parameters, and optimize the error proxy model based on the filtered training dataset. The parameter optimization module is further configured to perform global optimization of the key parameters using the error proxy model based on the preset error target and the key parameter optimization range, so as to obtain the optimal set of key parameters that meets the preset error target.
7. A storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the method for analyzing and optimizing the optical parameters of the focusing sensor and the silicon wafer film process parameters as described in any one of claims 1 to 5.
8. An electronic device, comprising at least a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program on the memory, it implements the steps of the method for analyzing and optimizing the optical parameters of the focusing sensor and the silicon wafer film layer process parameters as described in any one of claims 1 to 5.