A method and apparatus for evaluating overlay mark process perturbations

Abstract

The disclosure provides an evaluation method and device for overlay mark process disturbance, the method comprising: constructing a training data set; training a process disturbance evaluation model to obtain a trained process disturbance evaluation model; inputting geometric parameter information, process parameter information, overlay measurement equipment parameter information, and evaluation indexes of the overlay mark to be evaluated into the process disturbance evaluation model to obtain process disturbance parameters of the overlay mark to be evaluated. The disclosure is based on a machine learning reverse model establishment method, takes the geometric parameter information, process parameter information, measurement equipment parameter information, and evaluation indexes of the overlay mark as input parameters, takes the process disturbance parameters as output parameters, captures complex process disturbance characteristics by training the process disturbance evaluation model of the overlay mark, improves the accuracy of the evaluation model, accurately estimates the process disturbance parameters, helps to complete reverse positioning and error tracing of problems in the process, and improves the overlay accuracy.

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 evaluating disturbances in overlay marking processes. Background Technology

[0002] In semiconductor manufacturing, metrology equipment measures the overlay error of exposed patterns using overlay marks. The fabrication quality and process stability of the overlay marks directly affect the measurement accuracy of the overlay error. During chip manufacturing, the overlay marks undergo complex processes such as deposition, resist coating, photolithography, etching, and chemical mechanical polishing, which can cause unexpected, minute changes in the shape of the overlay marks—a process disturbance. These disturbances cause changes in the geometric parameters of the overlay marks, affecting evaluation indicators such as the diffraction efficiency of the probe beam in metrology equipment, and consequently, the measurement accuracy of the equipment.

[0003] The main evaluation indicators for measuring the design quality of overlay marks include diffraction efficiency (DE), stack sensitivity (SS), and measurement sensitivity (K). Conventional evaluation methods for DE, SS, and K use process disturbances, overlay mark geometric parameters, process stack structure parameters, and optical parameters of the overlay measurement equipment as inputs. These methods employ rigorous coupled-wave analysis (RCWA), slice absorption method (SAM), finite-difference time domain method (FDTD), and finite-difference frequency domain method (FDFD) to obtain the calculated values ​​of DE, SS, and K. However, these methods are all typical forward models. While these analytically based forward models can obtain accurate calculation results and error estimates for each evaluation indicator, the actual process disturbances in the actual process are difficult to obtain directly. Therefore, these forward models struggle to trace the source of actual process disturbances. Summary of the Invention

[0004] The purpose of this disclosure is to provide a method, apparatus, storage medium, and electronic device for evaluating disturbances in overlay marking processes, in order to solve the problems existing in the prior art.

[0005] The embodiments of this disclosure adopt the following technical solution: a method for evaluating process disturbances of overlay markings, comprising: constructing a training dataset, wherein the training dataset includes at least: geometric parameter information of overlay markings, process parameter information of overlay markings, parameter information of overlay measurement equipment, evaluation index of overlay markings, and process disturbance parameters; training a process disturbance evaluation model with the geometric parameter information of overlay markings, the process parameter information of overlay markings, the parameter information of overlay measurement equipment, and the evaluation index of overlay markings as input, and the process disturbance parameters as output, to obtain a trained process disturbance evaluation model; inputting the geometric parameter information of overlay markings to be evaluated, the process parameter information of overlay markings, the parameter information of overlay measurement equipment, and the evaluation index of overlay markings to the process disturbance evaluation model to obtain the process disturbance parameters of the overlay markings to be evaluated.

[0006] This disclosure also provides an evaluation device for overlay mark process disturbance, comprising: a dataset construction module for constructing a training dataset, wherein the training dataset includes at least geometric parameter information of the overlay mark, process parameter information of the overlay mark, parameter information of the overlay measurement equipment, evaluation index of the overlay mark, and process disturbance parameters; a training module for training a process disturbance evaluation model using the geometric parameter information of the overlay mark, the process parameter information of the overlay mark, the parameter information of the overlay measurement equipment, and the evaluation index of the overlay mark as input, and the process disturbance parameters as output, to obtain a trained process disturbance evaluation model; and an evaluation module for inputting the geometric parameter information of the overlay mark to be evaluated, the process parameter information of the overlay mark, the parameter information of the overlay measurement equipment, and the evaluation index of the overlay mark into the process disturbance evaluation model to obtain the process disturbance parameters of the overlay mark to be evaluated.

[0007] This disclosure also provides a storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described method for evaluating overlay marking process disturbances.

[0008] This disclosure also provides an electronic device, including at least a memory and a processor, wherein the memory stores a computer program, characterized in that the processor, when executing the computer program in the memory, implements the steps of the above-described evaluation method for overlay marking process disturbances.

[0009] The beneficial effects of the embodiments of this disclosure are as follows: The reverse model building method based on machine learning uses the geometric parameter information of the overlay mark, the process parameter information of the overlay mark, the parameter information of the overlay measurement equipment, and the evaluation index of the overlay mark as the input parameters of the evaluation model, and the process disturbance parameter of the overlay mark as the output parameter of the evaluation model. By training the process disturbance evaluation model of the overlay mark, complex process disturbance characteristics are captured, the accuracy of the evaluation model is improved, the process disturbance parameter is accurately estimated, and the reverse location of problems and error tracing in the process are completed, thereby improving the overlay accuracy. 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 evaluating the disturbance of the overlay marking process in the first embodiment of this disclosure; Figure 2 This is a schematic diagram of the design pattern of the overlay mark in the first embodiment of this disclosure; Figure 3 for Figure 2 A cross-sectional diagram showing the location of the dashed line AA'. Figure 4 This is a schematic diagram of the disturbance in the overlay marking process in the first embodiment of this disclosure; Figure 5 This is a schematic diagram of the shape of the aperture of the overlay measuring device in the first embodiment of this disclosure; Figure 6 This illustrates how the sidewall inclination angle of the overlaid mark varies with DE in the first embodiment of this disclosure; Figure 7 This illustrates how the sidewall inclination angle of the overlaid mark varies with SS in the first embodiment of this disclosure; Figure 8 This illustrates the variation of the sidewall inclination angle of the overprinted mark with the combined effect of DE and SS in the first embodiment of this disclosure; Figure 9 This illustrates how the thickness of the overlay markings varies with DE in the first embodiment of this disclosure. Figure 10 This illustrates the variation of the stack thickness of the overlay markings with SS in the first embodiment of this disclosure; Figure 11 This illustrates how the thickness of the overlay markings varies with the combined effects of DE and SS in the first embodiment of this disclosure. Figure 12 This is a schematic diagram of the structure of the evaluation device for overlay marking process disturbance in the second embodiment of this disclosure. Detailed Implementation

[0012] 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.

[0013] To address the problems existing in the prior art, the first embodiment of this disclosure provides a method for evaluating overlay marking process disturbances, which obtains accurate estimates of process disturbances by using a trained inverse evaluation model of overlay marking process disturbances. Figure 1 The flowchart of the evaluation method for overlay marking process disturbance in this embodiment is shown, which mainly includes steps S10 to S30: S10, Construct the training dataset; S20, taking the geometric parameter information of the overlay mark, the process parameter information of the overlay mark, the parameter information of the overlay measurement equipment, and the evaluation index of the overlay mark as input, and the process disturbance parameter as output, the process disturbance evaluation model is trained to obtain the trained process disturbance evaluation model. S30: Input the geometric parameter information of the overlay mark to be evaluated, the process parameter information of the overlay mark, the parameter information of the overlay measurement equipment, and the evaluation index of the overlay mark into the process disturbance evaluation model to obtain the process disturbance parameters of the overlay mark to be evaluated.

[0014] In this embodiment, the process disturbance parameters evaluated are the changes in the geometric parameters of the overlay mark relative to the design values ​​after undergoing front-end processing. These parameters include at least one or more of the following: stack thickness, etching sidewall tilt angle, top tilt angle, and bottom tilt angle. By analyzing the specific values ​​or trends of these process disturbance parameters, the design and optimization of the overlay mark's related parameters and processes are performed. During the design and implementation of an overlay mark, its shape design, process implementation, and measurement equipment all affect the mark's presentation quality. Therefore, this embodiment constructs a training dataset with the aforementioned overlay mark-related parameters as the training basis for the process disturbance evaluation model. Specifically, the training dataset constructed in this embodiment includes at least the geometric parameter information of the overlay markers, the process parameter information of the overlay markers, the parameter information of the overlay measurement equipment, the evaluation index of the overlay markers, and the process disturbance parameters. The geometric parameter information of the overlay markers, the process parameter information of the overlay markers, the parameter information of the overlay measurement equipment, and the evaluation index of the overlay markers are mainly used as input data when training the process disturbance evaluation model, while the process disturbance parameters are mainly used as output data when training the process disturbance evaluation model.

[0015] In the process of constructing the training dataset, the input parameters can actually be any one or more of the following: geometric parameters of the overlay markers, process parameters of the overlay markers, parameters of the overlay measurement equipment, and evaluation indicators of the overlay markers. For each combination of geometric parameters of the overlay markers, process parameters of the overlay markers, and parameters of the overlay measurement equipment, there should be a corresponding evaluation indicator for the overlay markers, so as to facilitate the training of the process disturbance evaluation model.

[0016] Specifically, the geometric parameter information of the overlay mark is mainly used to describe the structure and parameters that the overlay mark is expected to present during design. It includes at least the geometric structure of the overlay mark and the geometric parameters of the geometric structure. For a conventional overlay mark design, there are two sets of marks located at different levels in the vertical direction, namely the reference layer mark located at the lower level and the current layer mark located at the upper level. The two layers of marks have the same geometric pattern, but can have different subdivisions. Usually, the geometric structure of the overlay mark is formed by at least one grating structure. In this embodiment, the shape and number of grating structures and the positional relationship between multiple grating structures are used as the geometric structure data of the overlay mark. In addition, the geometric structure data should also include the positional relationship between the two layers of marks. For example, based on the different correspondences of the grating line positions between the reference layer mark and the current layer mark, there are two usage scenarios, including at least the structure where the grating line of the reference layer mark is aligned with the grating line of the current layer mark (line-on-line) and the structure where the grating line of the reference layer is aligned with the grating groove of the current layer (line-on-trench). Figure 2A schematic diagram of a design pattern for overlay marking is shown, wherein Figure 2 The left side shows the pattern marked on the current layer. Figure 2 The right side shows the pattern of the reference layer markers. Figure 2 Each black line in the diagram represents a grating line in the grating structure, and the blank spaces between the grating lines correspond to grating slots. Figure 3 Then it shows Figure 2 A cross-sectional diagram showing the location of the dashed line AA' is used to illustrate the vertical positional relationship between the reference layer marker and the current layer marker. Figure 3 The reference layer marker and the current layer marker shown have known, fixed offsets (shifts) in the X and Y directions within their respective planes, used to calculate overlay errors. See details... Figure 2 The overlay marks shown, whether reference layer marks or current layer marks, each layer mark includes at least one grating structure independently set in the X and Y directions within its own plane, to jointly form the overlay marks.

[0017] In this embodiment, the geometric parameters of the geometric structure used as the training dataset mainly correspond to the geometric parameters of the grating structure. For each grating structure, the corresponding geometric parameters include at least one of the following: structure type, marker size, period, duty cycle, etc., and may also include other parameters used to describe the grating structure, such as the subdivision of each grating line in the grating structure, which includes at least: horizontal subdivision, vertical subdivision, and two-dimensional subdivision in both horizontal and vertical directions. The above subdivision can be implemented independently on the reference layer marker and the current layer marker, or it can be implemented simultaneously on both layers.

[0018] The process parameters of the overlay markings mainly describe the processing methods used in the actual fabrication of the overlay markings and the parameters designed in the process implementation. In this embodiment, different processes are required during silicon wafer processing. A typical process in a photolithography exposure cycle includes deposition, coating, exposure, development, and etching. The overlay markings are at different process stages and will undergo multiple photolithography exposure cycles. Each photolithography exposure cycle corresponds to a marking layer, and the type, number, and parameters of the process layers contained in the marking layer can be different.

[0019] Based on the above description, the process parameter information describes various parameters for different process stages, mainly including at least one of the following parameters: process type, material type, process layer thickness, material optical refractive index (n-value), and material optical extinction coefficient (k-value). Usually, the n-value and k-value of the material are used simultaneously. Among them, the process type corresponds to different process stages, including at least one of the following: deposition, coating, exposure, development, and etching. The material type, material optical refractive index, and material optical extinction coefficient are determined according to the material preparation requirements of different layers. The process layer thickness is designed according to the process requirements of each layer preparation.

[0020] An overlay measurement device is used to measure overlay marks and calculate key evaluation indicators. It generates an incident light beam, receives the diffracted light beam formed by the overlay mark, and then calculates various key evaluation indicators based on the amplitude, frequency, and phase information of the diffracted beam. In this embodiment, the parameters of the overlay measurement device include at least: incident light wavelength, incident light bandwidth, incident light angle, incident light polarization state, and aperture shape. For any wavelength, a wavelength bandwidth centered on that wavelength can be set, and the overlay measurement device measures the overlay mark using incident light within that wavelength range. The incident light angle is the angle between the incident light and the normal to the surface of the overlay mark. The incident light polarization state can primarily be set to 0 degrees, 90 degrees, or both. The aperture shape can be set according to measurement requirements. Figure 4 The diagram shows the shapes of several apertures in overlay measurement equipment. In actual measurement, any aperture shape can be used alone on the measurement equipment, or several aperture shapes can be superimposed.

[0021] Overlay mark evaluation metrics are a series of indicators used to evaluate the design quality of overlay marks after their design is completed. These include diffraction efficiency (DE), stack sensitivity (SS), and measurement sensitivity (K). Diffraction efficiency (DE) reflects the ratio of the emitted light intensity to the incident light intensity after the probe beam passes through the current and reference layer overlay marks. Its design aim is to obtain a high value within the probe light wavelength range of the measurement equipment, with minimal fluctuation under process disturbances. Stack sensitivity (SS) reflects the asymmetric sensitivity when there is a relative offset (shift) between the current and reference layer overlay marks. Similarly, SS is designed to obtain a high value within the probe light wavelength range of the measurement equipment, with minimal fluctuation under process disturbances. Measurement sensitivity (K) reflects the sensitivity of the diffracted light intensity difference to the relative offset (shift) between the current and reference layer overlay marks. Similarly, K is designed to obtain a high absolute value within the probe light wavelength range of the measurement equipment, with minimal fluctuation under process disturbances.

[0022] Process disturbance parameters are disturbances that occur during actual process preparation, mainly including at least one of the following: stack thickness, sidewall angle, top tilt, and bottom tilt. Figure 5 A schematic diagram of the perturbation in the overlay marking process is shown, wherein, Figure 5 The left side shows the ideal shape for photolithographic marking. Figure 5 The pattern in the middle represents the sidewall corner perturbation d1 that occurs after actual fabrication. Figure 5 The pattern on the right represents the top tilting disturbance d2 that occurred after actual fabrication.

[0023] It should be noted that all data in the above training dataset can be used as training information for the process disturbance evaluation model. In addition, the training data used by the process disturbance evaluation model during actual training may include, but is not limited to, the above data.

[0024] In some embodiments, when constructing the actual training dataset, considering the scarcity of measured data in semiconductor manufacturing scenarios, this embodiment combines readily available simulation data to simultaneously construct the dataset. Specifically, firstly, the true values ​​of process perturbations, true geometric parameters, true process parameters, true parameters of the overlay measurement equipment, and measured values ​​of evaluation indicators for a first preset number of overlay marks on wafers are acquired. That is, measured data are collected by preparing and measuring the overlay marks on real wafers. Although the quantity of these measured data is limited, they contain systematic deviations and noise characteristics under real process conditions and are irreplaceable benchmarks for subsequent model calibration. To prevent overfitting caused by directly training the model with a small number of measured samples, this embodiment further constructs a statistical distribution model of the process disturbance based on the true value of the process disturbance, and samples and generates multiple sets of different process disturbance setpoints within the preset confidence interval of the statistical distribution model. That is, the statistical law of the real process disturbance is used as a priori guide to ensure that the generated simulation samples are concentrated within a reasonable range that may occur in the actual process. This avoids the waste of computing resources caused by invalid sampling and prevents the introduction of abnormal samples that do not conform to physical reality due to the sampling range being too wide. Thus, the physical credibility of the data is maintained while ensuring data diversity.

[0025] Based on this, using the overlay marking simulation model, forward calculations based on algorithms such as RCWA, SAM, FDTD, and FDFD are performed on multiple sets of different process disturbance setpoints, multiple sets of different set geometric parameter information, set process parameter information, and overlay measurement equipment set parameter information to obtain a second preset number of simulated evaluation index values, and the second preset number is greater than the first preset number. In this way, the training sample size can be rapidly expanded at a very low cost, making up for the lack of quantity of measured data, and allowing the model to learn the complex nonlinear mapping structure between evaluation indexes and process disturbances on large-scale data. Finally, the process disturbance setpoints, set geometric parameter information, set process parameter information, overlay measurement equipment set parameter information, and simulated evaluation index values ​​of the second preset number of overlay marking simulation model are used as a pre-training dataset to drive the model to learn general representations; at the same time, the true values ​​of process disturbances, true geometric parameter information, true process parameter information, true overlay measurement equipment parameter information, and measured evaluation index values ​​of the first preset number are used as a fine-tuning dataset, specifically for correcting systematic errors between simulation and measurement.

[0026] Unlike conventional optical diffraction models, this embodiment employs a machine learning-based process disturbance evaluation model to assess process disturbance parameters. Specifically, the process disturbance evaluation model in this embodiment is a regression model, established using machine learning methods, and includes at least one of the following models: neural network model, decision tree model, random forest model, support vector regression model, Gaussian process regression model, and k-nearest neighbor algorithm model. In actual implementation, any one of these models can be used as the process disturbance evaluation model. In some embodiments, the neural network model includes any one of the following: multilayer perceptron model, residual neural network model, convolutional neural network model, and deep neural network model.

[0027] After constructing the training dataset, the aforementioned process disturbance assessment model can be trained. This involves using the geometric parameters, process parameters, measurement equipment parameters, and evaluation metrics of the overlay marks from the training dataset as input, and the process disturbance parameters of the overlay marks as output. The model is trained to optimize its parameters based on the difference between the actual output and the evaluation metrics in the training dataset, until the minimum number of iterations or the minimum loss function value is achieved. Once trained, when the process disturbance parameters of an overlay mark need to be evaluated, the geometric parameters, process parameters, measurement equipment parameters, and evaluation metrics of the mark to be evaluated are input into the process disturbance assessment model. The corresponding process disturbance parameters for that mark can then be used as the basis for determining the current quality of the overlay mark.

[0028] In some embodiments, the accuracy of the process disturbance evaluation model can be gradually improved by expanding the training dataset capacity, or the process disturbance evaluation model can be optimized based on the evaluation results and the actual measurement results for the overlay marks during the actual evaluation process to achieve more accurate result output.

[0029] In some embodiments, the training of the process disturbance evaluation model can adopt a phased strategy to balance learning efficiency and calibration accuracy. First, the process disturbance evaluation model is pre-trained using a pre-training dataset, allowing the model to learn a general mapping topology from evaluation metrics to process disturbances from large-scale simulation data. Due to the large volume and wide coverage of simulation data, the model can establish stable low-level feature representations at this stage. Subsequently, the low-level feature extractor parameters of the pre-trained model are frozen. Since the general representations captured by the low-level feature extractor, such as signal encoding, have high transferability between simulation and actual measurements, allowing them to be updated during the fine-tuning stage would not only be insufficient to effectively optimize these parameters with a very small amount of actual measurement data, but might also destroy the weight coefficients learned in the pre-training stage. Therefore, this embodiment, by freezing the low-level feature extractor parameters, can concentrate limited fine-tuning computing power on adjusting the shallow regression head, allowing it to focus on learning the deviation correction mapping between simulation and actual measurements, rather than relearning basic features. Finally, the frozen model was fine-tuned using the fine-tuning dataset to obtain a trained process disturbance evaluation model. Through the three-stage training paradigm of pre-training-freezing-fine-tuning, higher prediction accuracy and stronger robustness were achieved than direct end-to-end training under small sample test conditions.

[0030] In the fine-tuning phase, in order to further improve the reliability of the model under limited measured data, this embodiment adopts a hybrid loss function based on a combination of data-driven and physical constraints to update the model parameters. The hybrid loss function in this embodiment is designed to include two parts: a data loss term and a physical consistency loss term. The data loss term is the mean square error or mean absolute error between the process disturbance parameters output by the process disturbance evaluation model and the true value of the process disturbance. Its function is to provide a direct supervision signal and drive the model output to approximate the actual measurement results. The physical consistency loss term is the difference between the predicted evaluation index calculated based on the process disturbance parameters output by the process disturbance evaluation model and the measured value of the evaluation index. In this embodiment, the process disturbance parameters output by the model are re-substituted into the forward physical model to calculate the predicted evaluation index, and then compared with the measured value. This transforms the physical consistency verification into an optimization objective with the same dimensions as the data loss term. This makes the physical consistency loss term naturally within the same optimization framework as the data loss term. There is no need to design complex constraint forms or adjust the weight coefficients between heterogeneous losses. In the process of minimizing the total loss, the model will spontaneously tend to a solution space that conforms to both the training data and the physical laws. This effectively prevents overfitting and non-physical solutions caused by small sample fine-tuning and significantly improves the reliability and stability of the model in practical applications.

[0031] To better illustrate the functionality of the process disturbance assessment model in this embodiment, Figure 2 Taking the overlay mark shown as an example, a deep neural network process perturbation evaluation model was established. The geometric parameters of the overlay mark during training include: the main evaluation indicators for the overlay mark include diffraction efficiency (DE) and stacking sensitivity (SS); the application of the overlay mark (corresponding to...). Figure 2 The overlay markings are of line-on-line structure, and the geometric structure and parameters of the overlay markings are as follows: the process parameters of the overlay markings include: the material and thickness of each process layer, as well as the process disturbance of the marking sidewall corner and the process disturbance of the process layer thickness; the parameters of the overlay measurement equipment include: the incident light wavelength, the incident light bandwidth and the incident light polarization state; the evaluation indexes are the sidewall tilt angle (SWA) and the stack thickness (THK).

[0032] Figures 6 to 11 This shows an example of the evaluation results of the process disturbance assessment model, where, in Figure 6 and Figure 7 In this model, under specific overlay mark geometric parameters, process parameters, and overlay equipment measurement parameters, the corresponding mark sidewall tilt angle SWA process disturbance results can be evaluated based on the input DE and SS. When the input DE and SS vary within a specific range, the evaluated values ​​of the above-mentioned sidewall tilt angle SWA process disturbance change significantly. The evaluation model can assess the magnitude of the process disturbance of the sidewall tilt angle using the overlay mark evaluation index and the corresponding input parameters. Figure 8The variation of wall tilt angle SWA process disturbance with DE and SS was characterized by 3D plots.

[0033] Figure 9 and Figure 10 In this model, under specific overlay marking geometric parameters, process parameters, and overlay equipment measurement parameters, the process perturbation results of the THK laminate thickness can be evaluated based on the input DE and SS. The evaluation model not only obtains the influence of DE and SS on the THK laminate thickness process perturbation, but also... Figure 11 The variation of the stack thickness THK with DE and SS was obtained.

[0034] This embodiment uses a machine learning-based inverse model building method. It takes the geometric parameters of the overlay marks, the process parameters of the overlay marks, the parameters of the overlay measurement equipment, and the evaluation indicators as input parameters of the evaluation model, and the process disturbance parameters of the overlay marks as output parameters of the evaluation model. By training the process disturbance evaluation model of the overlay marks, it captures complex process disturbance characteristics, improves the accuracy of the evaluation model, accurately estimates the process disturbance parameters, helps to complete the reverse location of problems and error tracing in the process, and improves the overlay accuracy.

[0035] Based on the same inventive concept, the second embodiment of this disclosure provides an evaluation device for disturbances in the overlay marking process, the schematic diagram of which is shown below. Figure 12 As shown, it mainly includes: a dataset construction module 10, used to construct a training dataset, which includes at least: geometric parameter information of the overlay mark, process parameter information of the overlay mark, parameter information of the overlay measurement equipment, evaluation index of the overlay mark, and process disturbance parameters; a training module 20, used to train the process disturbance evaluation model with the geometric parameter information of the overlay mark, process parameter information of the overlay mark, parameter information of the overlay measurement equipment, and evaluation index of the overlay mark as input, and the process disturbance parameters as output, to obtain a trained process disturbance evaluation model; and an evaluation module 30, used to input the geometric parameter information of the overlay mark to be evaluated, the process parameter information of the overlay mark, the parameter information of the overlay measurement equipment, and the evaluation index of the overlay mark into the process disturbance evaluation model to obtain the process disturbance parameters of the overlay mark to be evaluated.

[0036] It should be noted that the specific functions and implementation principles of each functional module of the evaluation device in this embodiment have been described in detail in the first embodiment, and will not be repeated here.

[0037] This embodiment uses a machine learning-based inverse model building method. It takes the geometric parameters of the overlay marks, the process parameters of the overlay marks, the parameters of the overlay measurement equipment, and the evaluation index as input parameters of the evaluation model, and the process disturbance parameters of the overlay marks as output parameters of the evaluation model. By training the process disturbance evaluation model of the overlay marks, it captures complex process disturbance characteristics, improves the accuracy of the evaluation model, accurately estimates the process disturbance parameters, helps to complete the reverse location of problems and error tracing in the process, and improves the overlay accuracy.

[0038] Based on the same inventive concept, the third embodiment of this disclosure provides a storage medium that can be installed in any semiconductor manufacturing and measurement equipment. Specifically, it is a computer-readable medium that stores a computer program that, when executed by a processor, implements the method provided in any embodiment of this disclosure.

[0039] The fourth embodiment of this disclosure provides an electronic device, which can be any semiconductor manufacturing and measurement device, and includes at least a memory and a processor. The memory stores a computer program, and the processor implements the method provided in any embodiment of this disclosure when executing the computer program in the memory.

[0040] 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 evaluating disturbances in an overlay marking process, characterized in that, include: Construct a training dataset, which includes at least: geometric parameter information of the overlay mark, process parameter information of the overlay mark, parameter information of the overlay measurement equipment, evaluation index of the overlay mark, and process disturbance parameters; Using the geometric parameters of the overlay mark, the process parameters of the overlay mark, the parameters of the overlay measurement equipment, and the evaluation index of the overlay mark as inputs, and the process disturbance parameters as outputs, the process disturbance evaluation model is trained to obtain a trained process disturbance evaluation model. The geometric parameters of the overlay mark to be evaluated, the process parameters of the overlay mark, the parameters of the overlay measurement equipment, and the evaluation index of the overlay mark are input into the process disturbance evaluation model to obtain the process disturbance parameters of the overlay mark to be evaluated.

2. The method for evaluating disturbances in the overlay marking process according to claim 1, characterized in that, The evaluation indicators for the overlay markings include at least one of the following: diffraction efficiency, stacking sensitivity, and overlay error; The process disturbance parameters include at least one of the following: stack thickness, etching sidewall tilt angle, top tilt angle, and bottom tilt angle.

3. The method for evaluating disturbances in the overlay marking process according to claim 1, characterized in that, The geometric parameter information of the overlay mark includes at least: the geometric structure of the overlay mark and the geometric parameters of the geometric structure; wherein, the geometric parameters include at least one of the following: structure type, mark size, period, and duty cycle; The process parameter information includes at least one of the following: process type, material type, process layer thickness, material optical refractive index, and material optical extinction coefficient; wherein, the process type includes at least one of the following: deposition, coating, exposure, development, and etching; The parameters of the overlay measurement device include at least: incident light wavelength, incident light bandwidth, incident light incident angle, incident light polarization state, and aperture shape.

4. The method for evaluating disturbances in the overlay marking process according to claim 1, characterized in that, The construction of the training dataset includes: Obtain the true value of process disturbance, true geometric parameter information, true process parameter information, true parameter information of overlay measurement equipment, and measured value of evaluation index of the first preset number of wafer overlay markings; Based on the true value of the process disturbance, a statistical distribution model of the process disturbance is constructed, and multiple sets of different process disturbance setpoints are generated by sampling within a preset confidence interval of the statistical distribution model. Using the overlay mark simulation model, forward calculation is performed based on the multiple sets of different process disturbance setting values, multiple sets of different setting geometric parameter information, setting process parameter information, and overlay measurement equipment setting parameter information to obtain a second preset number of evaluation index simulation values, where the second preset number is greater than the first preset number. The process disturbance setting value, set geometric parameter information, set process parameter information, set parameter information of the overlay measurement equipment, and simulated evaluation index value of the second preset number of the overlay mark simulation model are used as the pre-training dataset; The first preset number of true values ​​of process disturbances, true geometric parameter information, true process parameter information, true parameter information of overlay measurement equipment, and measured values ​​of evaluation indicators are used as the fine-tuning dataset.

5. The method for evaluating disturbances in the overlay marking process according to claim 4, characterized in that, The training of the process disturbance assessment model includes: The process disturbance evaluation model is pre-trained using the pre-trained dataset. Freeze the parameters of the low-level feature extractor of the pre-trained model; The frozen model is fine-tuned using the fine-tuning dataset to obtain the trained process disturbance evaluation model.

6. The method for evaluating disturbances in the overlay marking process according to claim 5, characterized in that, The fine-tuning of the frozen model using the fine-tuning dataset includes: The model parameters are updated using a hybrid loss function that combines data-driven and physical constraints. The hybrid loss function includes a data loss term and a physical consistency loss term. The data loss term is the mean square error or mean absolute error between the process disturbance parameters output by the process disturbance evaluation model and the true value of the process disturbance. The physical consistency loss term is the difference between the predicted evaluation index calculated based on the process disturbance parameters output by the process disturbance evaluation model and the measured value of the evaluation index.

7. The method for evaluating the disturbance of the overlay marking process according to any one of claims 1 to 6, characterized in that, The process disturbance evaluation model includes any one of the following: neural network model, decision tree model, random forest model, support vector regression model, Gaussian process regression model, and k-nearest neighbor algorithm model; The neural network model includes any of the following: multilayer perceptron model, residual neural network model, convolutional neural network model, and deep neural network model.

8. An evaluation device for disturbances in an overlay marking process, characterized in that, include: The dataset construction module is used to construct a training dataset, which includes at least the geometric parameter information of the overlay mark, the process parameter information of the overlay mark, the parameter information of the overlay measurement equipment, the evaluation index of the overlay mark, and the process disturbance parameter. The training module is used to train the process disturbance evaluation model by taking the geometric parameter information of the overlay mark, the process parameter information of the overlay mark, the parameter information of the overlay measurement equipment, and the evaluation index of the overlay mark as input, and the process disturbance parameter as output, so as to obtain the trained process disturbance evaluation model. The evaluation module is used to input the geometric parameter information of the overlay mark to be evaluated, the process parameter information of the overlay mark, the parameter information of the overlay measurement equipment, and the evaluation index of the overlay mark into the process disturbance evaluation model to obtain the process disturbance parameters of the overlay mark to be evaluated.

9. 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 evaluation method for overlay marking process disturbances as described in any one of claims 1 to 7.

10. 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 evaluation method for overlay marking process perturbation as described in any one of claims 1 to 7.

Images