An evaluation method and apparatus for alignment mark process perturbations

Abstract

The application provides an evaluation method and device for alignment mark process disturbance. The evaluation method for alignment mark process disturbance comprises the following steps: S1, establishing a process disturbance evaluation model; S2, acquiring a data set and training the process disturbance evaluation model by using the data set; and S3, inputting a measurement parameter into the process disturbance evaluation model to obtain a process disturbance amount. The application uses the process disturbance evaluation model trained by the data set to establish a reverse nonlinear mapping between an alignment mark evaluation index, an alignment mark geometric parameter, a process stack parameter, an alignment sensor parameter and the process disturbance, and the alignment mark process disturbance amount can be acquired with high precision.

Description

Technical Field

[0001] This invention relates to the field of semiconductor manufacturing technology, and in particular to a method and apparatus for evaluating alignment mark process disturbances. Background Technology

[0002] The process stability of alignment marks in semiconductor manufacturing is fundamental to ensuring the alignment accuracy of photolithography equipment and directly affects the overlay accuracy of chip manufacturing. During chip manufacturing, alignment marks undergo complex processes such as thin film deposition, photoresist coating, photolithography exposure, etching, and chemical mechanical polishing. Due to machine parameter drift or process fluctuations, the geometry of the alignment marks (such as sidewall tilt angle and film thickness) may undergo unexpected slight changes (i.e., process disturbances). These disturbances cause changes in the grating structure parameters of the alignment marks, thus affecting the diffraction signal received by the alignment sensor in the photolithography equipment, ultimately resulting in errors in the measured positional accuracy.

[0003] The performance metrics for measuring the stability of alignment marking processes mainly include wafer quality (WQ) and alignment position deviation (APD). Conventional methods for evaluating WQ and APD involve using parameters such as process disturbances as input and employing optical methods like strictly coupled-wave (RCWA) to obtain estimates of WQ and APD; or directly calculating WQ and APD using measurements such as wafer reflectivity. However, while this forward model can obtain APD and thus correct the alignment sensor's measurement results, it fails to capture the process disturbances that cause APD, making it impossible to trace the root cause of these disturbances during the process. Summary of the Invention

[0004] The purpose of this invention is to provide a method and apparatus for evaluating alignment marking process disturbances, which can obtain the amount of alignment marking process disturbances with high accuracy.

[0005] This invention provides a method for evaluating alignment marking process disturbances, comprising the following steps: S1: Establish a process disturbance assessment model; S2: Obtain the dataset and use the dataset to train the process disturbance evaluation model; S3: Input measurement parameters into the process disturbance assessment model to obtain the process disturbance amount.

[0006] The dataset used for training the process disturbance evaluation model may include data from the following different sources: the dataset includes at least one of alignment mark evaluation index, alignment mark geometric parameters, process stack parameters, alignment sensor parameters, and process disturbance amount; wherein, the input of the process disturbance evaluation model includes at least one of alignment mark evaluation index, alignment mark geometric parameters, process stack parameters, and alignment sensor parameters, and the output of the process disturbance evaluation model is the process disturbance amount.

[0007] Alignment mark evaluation metrics are parameters used to evaluate the performance of the mark under different process conditions. They mainly include wafer quality (WQ) and alignment position deviation (APD). One or more of the alignment mark evaluation metrics can be selected as the value of the evaluation metric.

[0008] The alignment mark geometry is primarily a grating structure; correspondingly, the alignment mark geometric parameters describing the alignment mark geometry mainly include at least one of the following: structure type, mark size, period, and duty cycle. The grating lines of the grating structure can be further subdivided, including: horizontal subdivision, vertical subdivision, and simultaneous two-dimensional subdivision in both horizontal and vertical directions. The wafer's X and Y directions each correspond to an independent set of grating structures, collectively forming the alignment mark. Some or all of the alignment mark geometric parameters described above can serve as training input information for a process disturbance evaluation model; it is understood that the alignment mark geometric parameters used in the process disturbance evaluation model include, but are not limited to, the parameters described above.

[0009] Process stack parameters describe various parameters at different process stages, including at least one of process type, material type, process layer thickness, material optical n-value, and k-value. Different processing steps are required during wafer fabrication. A typical process in a photolithography exposure cycle includes: deposition, resist coating, exposure, development, etching, and polishing. Alignment marks, occurring at different process stages, undergo multiple photolithography exposure cycles. Each photolithography exposure cycle corresponds to a marker layer, and the type, number, and parameters of the process layers contained in the marker layer can vary. Some or all of the above-mentioned process stack parameters can serve as training input information for a process perturbation evaluation model; it is understood that the process stack parameters used for the process perturbation evaluation model include, but are not limited to, the parameters described above.

[0010] The function of the alignment mark measuring device described above is to measure and calculate the evaluation index results of the mark. The alignment sensor generates an incident light beam, receives the various diffraction orders of the diffracted light beam formed by the alignment mark, and then calculates various evaluation indexes based on the amplitude, frequency, and phase information of the diffracted light beam.

[0011] Alignment sensor parameters include at least one of the following: incident light wavelength, incident light angle, and incident light polarization state. The incident light angle is the angle between the incident light and the normal to the alignment mark surface. The incident light polarization state can be set to 0 degrees or 90 degrees. To improve process adaptability, the alignment sensor uses multiple incident lights with different wavelengths and polarization states. An independent evaluation index can be calculated for each wavelength and polarization state, thus forming a set of evaluation indices. Some or all of the alignment sensor parameters described above can serve as training input information for the process disturbance evaluation model; it is understood that the alignment sensor parameters used for the process disturbance evaluation model include, but are not limited to, the parameters described above.

[0012] Process perturbations are the changes in the shape of an alignment mark after undergoing a series of processes, including at least one of etching sidewall tilt angle, stack thickness, and top or bottom tilt angle. Some or all of the above-described process perturbations can be used as training input information for a process perturbation evaluation model; it is understood that the process perturbations used for the process perturbation evaluation model include, but are not limited to, the aforementioned indicators. One or more process perturbations corresponding to the X and Y directions of the alignment mark can be simultaneously obtained through the process perturbation evaluation model.

[0013] In this invention, a process disturbance evaluation model is established using machine learning methods, including at least one of neural networks, decision trees, random forests, support vector regression, Gaussian process regression, and K-nearest neighbor regression.

[0014] Furthermore, the neural network includes at least one of the following: multilayer perceptron (MLP), residual neural network (RNN), convolutional neural network (CNN), recurrent neural network (RNN), and deep neural network (DNN).

[0015] The input parameters of the process disturbance evaluation model established using machine learning methods include alignment mark evaluation index, alignment mark geometric parameters, process stack parameters, alignment sensor parameters, etc., and some or all of these parameter information can be used.

[0016] The computer program for the process disturbance evaluation model can be stored in a data storage medium. The computer device can read, calculate and store the data, and includes structural components such as processors, memory and input / output hardware modules.

[0017] The present invention also provides an evaluation apparatus for alignment mark process disturbance, used to perform the above evaluation method, the evaluation apparatus comprising: Memory: Used to store datasets; Processor: Used to build a process disturbance evaluation model and train the process disturbance evaluation model using a dataset; Input hardware module: used to input measurement parameters; Output hardware module: used to output process disturbance quantities.

[0018] Given the extreme complexity of the physical processes involved in alignment mark perturbations, it is difficult to establish an inverse model using analytical methods. This invention proposes a deep learning-based inverse model for key performance parameters of alignment marks, used to quantitatively evaluate process perturbations affecting alignment mark evaluation indicators. This inverse model takes alignment mark evaluation indicators, alignment mark geometric parameters, process stacking parameters, and alignment sensor parameters as input parameters, and the amount of process perturbation as the output parameter. The model is trained using a dataset established from the forward model of alignment mark evaluation indicators or experimental data, thereby quantitatively evaluating process perturbations. This approach can estimate the magnitude of various process perturbations, thereby enabling problem localization and error tracing in the process, and ultimately improving the overlay accuracy of lithography equipment.

[0019] The inverse model building method based on machine learning of this invention does not rely on a precise physical model of alignment mark process perturbations, but can be obtained through forward model simulation or training with experimental data. The machine learning-based process perturbation evaluation model, in principle, can establish an inverse nonlinear mapping between alignment mark evaluation indicators, alignment mark geometric parameters, process stack parameters, alignment sensor parameters, and process perturbation quantities through data-driven methods. This enables high-precision calculation of alignment mark process perturbation quantities, resulting in high evaluation accuracy. Furthermore, the machine learning-based process perturbation evaluation model can be trained on a large amount of test data. As the training dataset continuously increases, the process perturbation evaluation model can be continuously trained and upgraded, thereby obtaining even more accurate evaluation results. The above method is applicable to semiconductor manufacturing equipment and software, especially EDA software related to lithography technology. Attached Figure Description

[0020] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0021] Figure 1 This is a schematic diagram of a process disturbance assessment model; Figure 2 A top view of the alignment mark pattern; Figure 3 A side view of the alignment mark pattern; Figure 4 A schematic diagram of the diffraction order of the alignment mark; Figure 5 This is a schematic diagram of the alignment marking process disturbance. Figure 6 Example diagram of alignment marks; Figure 7This is the result of the sidewall tilt angle changing with alignment position deviation in the process disturbance assessment; Figure 8 This is the result of the sidewall tilt angle varying with wafer quality in the process disturbance assessment; Figure 9 This is the result of the change in laminate thickness with alignment position deviation in the process disturbance assessment; Figure 10 This represents the result of the stack thickness varying with wafer quality during process disturbance assessment. Detailed Implementation

[0022] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0023] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations according to this application. As used herein, the singular form includes the plural form unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this description, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0024] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] Example 1 This embodiment provides a method for evaluating alignment marking process disturbances, the steps of which are as follows: A process disturbance assessment model was established for alignment marking process disturbances. The model architecture diagram is shown below. Figure 1 As shown, the process disturbance assessment model is established using machine learning methods.

[0026] The process disturbance assessment model is used to evaluate the magnitude of the process disturbance of the alignment mark based on the alignment mark evaluation index, alignment mark geometric parameters, process stack parameters, and alignment sensor parameters. It further optimizes the design of the alignment mark geometric parameters, process stack parameters, and alignment sensor parameters to improve the process robustness of the alignment mark.

[0027] The inputs to the process disturbance assessment model include: alignment mark evaluation index, alignment mark geometric parameters, process stack parameters, and alignment sensor parameters.

[0028] The output of the process disturbance evaluation model is the process disturbance of the alignment mark, which is the change in the shape of the alignment mark relative to the design value after undergoing a series of process treatments, such as the etching sidewall tilt angle SWA and the stack thickness THK.

[0029] Alignment mark evaluation metrics are a series of indicators used to evaluate the design quality of the alignment mark itself after the design is completed, such as wafer quality (WQ) and alignment position deviation (APD).

[0030] Wafer quality (WQ) is defined as the percentage of the diffraction intensity of the alignment mark to the diffraction intensity of the reference mark. Its design aims to obtain the highest possible WQ value, and to minimize the variation of this value in the presence of process perturbations, i.e., process insensitivity.

[0031] Alignment position deviation (APD) is defined as the alignment position offset relative to the alignment sensor spindle. Its theoretical value is zero, and the value itself and its fluctuations are kept as small as possible in the presence of process disturbances.

[0032] The geometry of alignment marks is mainly a grating structure. The geometric parameters describing alignment marks mainly include structure type, mark size, period, and duty cycle.

[0033] Figure 2 and Figure 3 This is a set of typical top and side view examples of alignment marks. The raster structure's grid lines can be further subdivided, including horizontal subdivision, vertical subdivision, and simultaneous two-dimensional subdivision in both horizontal and vertical directions. The raster structure subdivision parameters are also part of the aforementioned geometric parameters. Figure 2 In this process, each gate line is subdivided vertically, and the subdivision parameters, including period and duty cycle, are part of the geometric parameters.

[0034] Each of the X and Y directions of the wafer corresponds to an independent set of grating structures, which together form alignment marks used to measure the alignment position of the wafer in the X and Y directions. The parameters of the grating structures in the X and Y directions can be the same or different, and they are independently used as input information for the process disturbance evaluation model.

[0035] All or part of the above alignment mark geometric parameters can be used as input information for the process disturbance evaluation model.

[0036] Process stack parameters describe the process treatments and corresponding parameter settings that a marker undergoes during the process.

[0037] Wafer fabrication requires various processing steps. A typical photolithography cycle includes deposition, resist coating, exposure, development, etching, and polishing. These processes are generally performed sequentially. The alignment marks are typically formed on a silicon substrate, and the processes described above are completed vertically from bottom to top.

[0038] Alignment marks undergo multiple photolithography exposure cycles at different process stages, with each photolithography exposure cycle corresponding to a series of process stacks. Process stack parameters are various parameters describing the different process stages mentioned above, including process type, material type, process layer thickness, material optical n-value, and k-value, etc.

[0039] All or part of the above process stack parameters can be used as input information for the process disturbance evaluation model.

[0040] The function of the alignment sensor applicable to the above alignment marks is to measure the alignment marks and calculate the evaluation index results of the marks.

[0041] The alignment sensor is designed based on the principle of diffraction optics. It first generates an incident beam, then receives the diffracted beam formed by the alignment mark, and finally calculates various evaluation indicators based on the amplitude, frequency and phase information of the diffracted beam.

[0042] Alignment sensor parameters include incident light wavelength, incident light angle, incident light polarization state, and diffraction order.

[0043] The incident angle is the angle between the incident light and the normal to the surface of the alignment mark.

[0044] The polarization state of the incident light can be set to 0 degrees or 90 degrees.

[0045] The order of diffraction light is the order of incident light after it has passed through the alignment mark and diffracted. Generally, the alignment sensor collects odd-numbered orders such as +1, +3, +5, and +7.

[0046] All or part of the alignment sensor parameters mentioned above can be used as input information for the process disturbance evaluation model.

[0047] Figure 4 It is a schematic diagram of the diffraction order formed by the incident light of the alignment sensor after passing through the alignment mark. The alignment sensor collects the diffracted light of the odd order and then calculates various evaluation indicators of the alignment sensor.

[0048] Process disturbance is the amount of change in the shape of an alignment mark after a series of processes, such as the etching sidewall tilt angle (SWA), the stack thickness (THK), and the top or bottom tilt angle.

[0049] Process disturbances can be a single disturbance term or multiple disturbance terms existing simultaneously.

[0050] Figure 5 This is an example diagram of process disturbances, illustrating sidewall corner disturbances and top tilt disturbances. In actual processes, alignment marks may generate one or more of these process disturbances, and these disturbances may occur simultaneously in different process stacks.

[0051] Each process disturbance can be numerically quantified.

[0052] The process disturbance assessment model is a regression model built using machine learning methods. Machine learning models include at least one of the following methods: neural networks, decision trees, random forests, support vector regression, Gaussian process regression, and k-nearest neighbor algorithm.

[0053] The neural network-based process disturbance assessment model is an advanced machine learning model that includes at least one of the following methods: multilayer perceptron (MLP), residual neural network (RNN), convolutional neural network (CNN), recurrent neural network (RNN), and deep neural network (DNN).

[0054] The input parameters of the process disturbance evaluation model established using machine learning methods include alignment mark geometric parameters, alignment mark process parameter information, and alignment sensor parameters, and some or all of the data information can be used.

[0055] The output parameters of the process disturbance assessment model are the process disturbance quantities, which can be one or more of them.

[0056] The dataset for the process disturbance evaluation model is composed of the aforementioned alignment mark evaluation indicators, alignment mark geometric parameters, process stack parameters, alignment sensor parameters, and process disturbance quantities. The accuracy of the process disturbance quantity determines whether the process disturbance evaluation model meets the requirements. Furthermore, the accuracy of the process disturbance evaluation model can be gradually improved by expanding the sample size of the dataset.

[0057] To better illustrate the establishment and training method of the process disturbance evaluation model in this embodiment, the following is adopted: Figure 6 The alignment markers shown are evaluated using a deep neural network model. The main evaluation metrics for alignment markers include wafer quality (WQ) and alignment position deviation (APD); the main geometric parameters of alignment markers include structure type, marker size, period, and duty cycle; the main process stack parameters include process type, material type, process layer thickness, material optical n-value, and k-value; the main alignment sensor parameters include incident light wavelength, incident light polarization state, and diffraction order; and the process perturbation (i.e., evaluation metrics) includes etching sidewall tilt angle (SWA) and stack thickness (THK).

[0058] Figures 7 to 10The process disturbance assessment results are based on machine learning. The input to the process disturbance assessment model includes key measurement parameters such as wafer quality (WQ), alignment position deviation (APD), incident light wavelength, polarization state, and diffraction order. When the input optical measurement signals (alignment position deviation (APD) and wafer quality (WQ)) and alignment sensor parameters (such as wavelength and polarization) are different, the obtained process disturbance values ​​will have certain differences. This difference characterizes the comprehensive impact of the above parameter changes on the process disturbance.

[0059] Figure 7 and Figure 8 In this model, under specific incident light wavelengths and diffraction orders, the corresponding sidewall tilt angle process disturbance results can be evaluated based on the input APD and WQ signals. When the input APD and WQ signals vary within a specific range, the evaluated values ​​of the sidewall tilt angle change significantly, thus allowing the process disturbance amount of the sidewall tilt angle to be evaluated using the process disturbance evaluation model and input parameters.

[0060] Figure 9 and Figure 10 This represents the evaluation results of the APD and WQ optical signals for each layer thickness (layer thickness 1, layer thickness 2, and layer thickness 3) under specific incident light wavelengths and diffraction orders. Corresponding to the variation characteristics of the input optical signal, the evaluation fluctuations for different layer thicknesses differ. This differentiated variation trend characterizes the varying sensitivities of the measured optical signals to process disturbances, thus allowing the evaluation of the process disturbance amount of the layer thickness through the evaluation model and input parameters.

[0061] The process disturbance evaluation model in this embodiment can be implemented using a computer program, with a computer device as the physical carrier. The computer device generally includes a processor, memory, and input / output hardware; wherein, the memory is used to store the dataset used for training the process disturbance evaluation model and the database of the process disturbance evaluation model; model training and process disturbance evaluation are performed by the processor; and the input / output hardware is used for user interface, evaluation result display, and other communication functions.

[0062] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention 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; and these 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 the present invention.

Claims

1. A method for evaluating alignment marking process disturbances, characterized in that, Includes the following steps: S1: Establish a process disturbance assessment model; S2: Obtain the dataset and use the dataset to train the process disturbance evaluation model; S3: Input measurement parameters into the process disturbance evaluation model to obtain the process disturbance amount.

2. The evaluation method according to claim 1, characterized in that, The dataset includes at least one of the following: alignment mark evaluation metrics, alignment mark geometric parameters, process stack parameters, alignment sensor parameters, and process disturbances.

3. The evaluation method according to claim 2, characterized in that, The alignment mark evaluation index includes at least one of wafer quality and alignment position deviation.

4. The evaluation method according to claim 2, characterized in that, The alignment mark geometry parameters include at least one of structure type, mark size, period, and duty cycle.

5. The evaluation method according to claim 2, characterized in that, The process stacking parameters include at least one of process type, material type, process layer thickness, material optical n-value, and k-value.

6. The evaluation method according to claim 2, characterized in that, The alignment sensor parameters include at least one of the following: incident light wavelength, incident light incident angle, incident light polarization state, and diffraction order.

7. The evaluation method according to claim 1, characterized in that, The process disturbance includes at least one of the following: etching sidewall tilt angle, stack thickness, top tilt angle, and bottom tilt angle.

8. The evaluation method according to claim 1, characterized in that, Establishing a process disturbance assessment model includes: using machine learning methods to establish the process disturbance assessment model, where the machine learning methods include at least one of neural networks, decision trees, random forests, support vector regression, Gaussian process regression, and K-nearest neighbor regression.

9. The evaluation method according to claim 8, characterized in that, The neural network includes at least one of the following: multilayer perceptron, residual neural network, convolutional neural network, recurrent neural network, and deep neural network.

10. An evaluation apparatus for alignment marking process disturbance, characterized in that, The evaluation apparatus for performing the evaluation method according to any one of claims 1-9 includes: Memory: Used to store datasets; Processor: Used to build a process disturbance evaluation model and train the process disturbance evaluation model using a dataset; Input hardware module: used to input measurement parameters; Output hardware module: used to output process disturbance quantities.

Images